Multivariate Methods in Chromatography: A Practical Guide

Multivariate Methods in Chromatography: A Practical Guide Tibor Cserháti WILEY Multivariate Methods in Chromatograph...

Author: Tibor Cserhati

121 downloads 1686 Views 4MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

Multivariate Methods in Chromatography: A Practical Guide

Tibor Cserháti

WILEY


´ TIBOR CSERHATI Research Institute of Materials and Environmental Chemistry Chemical Research Center, Hungarian Academy of Sciences Budapest, Hungary

A John Wiley and Sons, Ltd, Publication

Multivariate Methods in Chromatography


´ TIBOR CSERHATI Research Institute of Materials and Environmental Chemistry Chemical Research Center, Hungarian Academy of Sciences Budapest, Hungary

A John Wiley and Sons, Ltd, Publication

C 2008 Copyright

John Wiley & Sons Ltd, The Atrium, Southern Gate, Chichester, West Sussex PO19 8SQ, England Telephone

(+44) 1243 779777

Email (for orders and customer service enquiries): [email protected] Visit our Home Page on www.wileyeurope.com or www.wiley.com All Rights Reserved. No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, scanning or otherwise, except under the terms of the Copyright, Designs and Patents Act 1988 or under the terms of a licence issued by the Copyright Licensing Agency Ltd, 90 Tottenham Court Road, London W1T 4LP, UK, without the permission in writing of the Publisher. Requests to the Publisher should be addressed to the Permissions Department, John Wiley & Sons Ltd, The Atrium, Southern Gate, Chichester, West Sussex PO19 8SQ, England, or emailed to [email protected], or faxed to (+44) 1243 770620. Designations used by companies to distinguish their products are often claimed as trademarks. All brand names and product names used in this book are trade names, service marks, trademarks or registered trademarks of their respective owners. The Publisher is not associated with any product or vendor mentioned in this book. This publication is designed to provide accurate and authoritative information in regard to the subject matter covered. It is sold on the understanding that the Publisher is not engaged in rendering professional services. If professional advice or other expert assistance is required, the services of a competent professional should be sought. The Publisher and the Author make no representations or warranties with respect to the accuracy or completeness of the contents of this work and specifically disclaim all warranties, including without limitation any implied warranties of fitness for a particular purpose. The advice and strategies contained herein may not be suitable for every situation. In view of ongoing research, equipment modifications, changes in governmental regulations, and the constant flow of information relating to the use of experimental reagents, equipment, and devices, the reader is urged to review and evaluate the information provided in the package insert or instructions for each chemical, piece of equipment, reagent, or device for, among other things, any changes in the instructions or indication of usage and for added warnings and precautions. The fact that an organization or Website is referred to in this work as a citation and/or a potential source of further information does not mean that the author or the publisher endorses the information the organization or Website may provide or recommendations it may make. Further, readers should be aware that Internet Websites listed in this work may have changed or disappeared between when this work was written and when it is read. No warranty may be created or extended by any promotional statements for this work. Neither the Publisher nor the Author shall be liable for any damages arising herefrom. Other Wiley Editorial Offices John Wiley & Sons Inc., 111 River Street, Hoboken, NJ 07030, USA Jossey-Bass, 989 Market Street, San Francisco, CA 94103-1741, USA Wiley-VCH Verlag GmbH, Boschstr. 12, D-69469 Weinheim, Germany John Wiley & Sons Australia Ltd, 42 McDougall Street, Milton, Queensland 4064, Australia John Wiley & Sons (Asia) Pte Ltd, 2 Clementi Loop #02-01, Jin Xing Distripark, Singapore 129809 John Wiley & Sons Ltd, 6045 Freemont Blvd, Mississauga, Ontario L5R 4J3, Canada Wiley also publishes its books in a variety of electronic formats. Some content that appears in print may not be available in electronic books. Library of Congress Cataloging-in-Publication Data Cserháti, Tibor, Multivariate methods in chromatography : a practical guide / Tibor Cserháti. p. cm. Includes bibliographical references and index. ISBN 978-0-470-05820-6 (cloth : alk. paper) 1. Chromatographic analysis. 2. Multivariate analysis. I. Title. QD79.C4C79 2008 543 .8–dc22 2008011000 British Library Cataloguing in Publication Data A catalogue record for this book is available from the British Library ISBN 978-0-470-05820-6 Typeset in 10/12pt Times by Aptara Inc., New Delhi, India. Printed and bound in Great Britain by Antony Rowe Ltd, Chippenham, Wiltshire.

Contents

Preface Introduction List of Abbreviations, Acronyms and Symbols 1

2

vii ix xi

Fundamentals 1.1 Multilinear and Nonlinear Regression Analyses 1.2 Stepwise Regression Analysis and Partial Least Squares Method 1.3 Two- and Three-dimensional Principal Component Analysis, Various Factor Analytical Techniques 1.4 Canonical Correlation Analysis 1.5 Discriminant Analysis 1.6 Spectral Mapping 1.7 Nonlinear Mapping 1.8 Cluster Analysis 1.9 Other Multivariate Techniques 1.10 Measured and Calculated Physicochemical Parameters of Chromatographic Systems and Analytes References

1 1 2

Gas Chromatography 2.1 Theory and Practice of Gas Chromatography 2.2 Comparison of Gas Chromatography Stationary Phases Using a Homogenous and Nonhomogenous Set of Analytes 2.3 Elucidation of Similarities and Dissimilarities Among Samples 2.3.1 Human Health and Pharmaceuticals 2.3.2 Forensic Analyses 2.3.3 Biology and Agrobiology 2.3.4 Food and Food Products

9 9

2 4 4 5 5 5 5 6 7

12 26 28 32 36 46

vi

Contents

2.3.5 Environmental Analyses 2.3.6 Other Synthetic Compounds 2.3.7 Miscellaneous Applications References 3

4

76 97 104 107

Liquid Chromatography 3.1 Thin-layer Chromatography 3.1.1 Theory and Practice of Thin-layer Chromatography 3.1.2 Multidimensional Classification of Thin-layer Chromatography Stationary and/or Mobile Phases 3.1.3 Relationships Between Molecular Parameters and Thin-layer Chromatography Retention of Analytes 3.1.4 Relationship Between Thin-layer Chromatography Retention Parameters and Biological Activity of Analytes 3.1.5 Miscellaneous Applications 3.2 High-performance Liquid Chromatography 3.2.1 Theory and Practice of High-performance Liquid Chromatography 3.2.2 Multivariate Classification of High-Performance Liquid Chromatography Stationary and/or Mobile Phases 3.2.3 Differentiation Between Homologous and Nonhomologous Sets of Analytes References

113 113 113

Electrically Driven Systems 4.1 Theory and Practice of Electrically Driven Systems 4.2 Gel Electrophoretic Techniques 4.2.1 Theory and Human Health Aspects 4.2.2 Microorganisms 4.2.3 Microbial Communities 4.2.4 Plant Tissues 4.3 Capillary Zone Electrophoresis 4.3.1 Human Health and Pharmacology 4.3.2 Other Applications 4.4 Micellar Electrokinetic Chromatography and Related Technologies References

265 265 267 267 274 286 298 300 300 302 309 315

Index

115 117 136 138 140 141 144 153 256

325

Preface

Each real chromatographer dislikes chemometrics. Nobody knows whether it is a poisonous snake or not. Do not worry, it is not. Chemometrics itself is a dead and impassive tool. You have to ask it intelligently, to evaluate the answers intelligently and to give a chromatographic meaning to the numbers (to fill the numbers with chromatographic meaning). This is what this book is intended for: how to use chemometrics for the solution of problems hardly soluble otherwise without the exact knowledge of the underlying, sometimes complicated mathematical apparatus. This book was written by a practising chromatographer for practising chromatographers – a bridge between everyday chromatographic practice and theoretical mathematics. Chemometrics is not a monster, everybody can dominate it with more or less effort than me, a chemist. I strongly hope that with the help of this book my colleagues all over the world will find such delight in the chromatographic application of various methods of chemometrics as I found during my professional career. ´ Tarlós and Ms Esther Bartha for their valuable technical The author is grateful to Ms Eva assistance. Tibor Cserháti

Introduction

Chromatographic methods have been developed and applied for the separation and quantitative determination of a wide variety of organic and inorganic compounds present in complicated accompanying matrices even at the trace level. The development of new computer software to interpret retention data matrices of considerable dimensions has been one of the major advances in analytical chemistry during the last decades. The most conclusive characteristics of this rapidly developing field are the high-speed automated chromatographic instruments and computers and a considerable number of mathematical–statistical methodologies. The extraction of maximal information of large data sets (i.e. retention times of numerous compounds determined on several chromatographic columns using eluents of different composition and elution strength) is cumbersome and practically impossible by the traditional linear regression model. The up-to-date multivariate mathematical–statistical methods make possible the simultaneous assessment of a considerable number of variables (generally chromatographic parameters) facilitating the solution of both theoretical and practical problems. Multivariate methods have been frequently applied in chromatography to identify basic factors having a marked impact on solute–solvent and solute–stationary phase interactions and to study the clustering of solutes, supports and solvents into groups exposing similar retention characteristics. As each multivariate mathematical–statistical method highlights only one or two aspects of the chromatographic problem to be solved, the simultaneous application of more than one mathematical–statistical method is rather a rule than an exception. The objectives of the book are the collection, concise description and evaluation of the various multivariate mathematical–statistical methods applied for the assessment of retention data sets. It has to be emphasized that the book was written for practical purposes, it avoids the meticulous theoretical treatment of the calculation methods which is not necessary for their successful application in chromatography. The principles of the individual methods are discussed as simply as possible using practical examples taken equally from thin-layer chromatography (TLC), high-performance liquid chromatography (HPLC), supercritical fluid chromatography (SFC), gas chromatography (GC), and electrically driven

x

Introduction

systems such as capillary zone electrophoresis (CZE), capillary gel electrophoresis (CGE), micellar electrokinetic chromatography (MEKC), capillary isotachophoresis, and capillary isoelectric focusing. I strongly hope that the book will be a valuable help for both scientist and serious students employed by research and development who are interested in the application of mathematical–statistical methods for the evaluation of large data matrices in chromatography. The general aspects of chemometrics have been previously discussed in detail [1,2] and the newest results in the application of chemometrics in analytical chemistry have been recently reviewed [3,4]. The employment of mathematical–statistical methods in chromatography has been described in the excellent books of Professor R. Kaliszan [5,6].

References [1] Mardia, K.V., Kent, J.T., and Bibby, J.M. Multivariate Analysis, Academic Press, London, 1979. [2] Vandeginste, B.G.M., Massart, D.L., Buydens, L.M.C., De Jong, S., Lewi, P.J., and SmeyersVerbeke, J. Handbook of Chemometrics and Qualimetric. Data Handling in Science and echnology, Vol.20, Elsevier, Amsterdam, 1998. [3] Lavine, B.K. and Workman, J. Chemometrics. Anal. Chem. 74 (2002) 2763-2769. [4] Lavine, B.K. and Workman, J. Chemometrics. Anal. Chem. 76 (2004) 3365-3372. [5] Kaliszan, R. Quantitative Structure-Chromatographic Retention Relationship, Wiley, New York, 1987. [6] Kaliszan, R., Structure and Retention in Chromatography:A Chemometric Approach, Harwood Academic Publishers, Australia, 1997.

Abbreviations, Acronyms and Symbols

ε m p+o AA ABTS AK AED AFLP AFS ALS ANN AOB APCI apSASA APTES ARD B1 and B4 BCDP BE BMC

dielectric constant zeta potential bulk viscosity Hansch-Fujita’s substituent constants characterizing hydrophobicity Hammett’s constant characterizing the electron-withdrawing power of the substituents at meta position Hammett’s constant characterizing the electron-withdrawing power of the substituents at para+ortho positions amino acid 2,2 -azinobis(3-ethylbenzothiazoline)-6-sulfonic acid adenylate kinase atomic emission detector amplified fragment length polymorphism atomic fluorescence spectrometry alternating least squares artifical neural network ammonia-oxidizing bacteria atmospheric pressure chemical ionization apolar solvent accessible surface area 3-aminopropyl)triethoxysilane apple (Malus domestica) replant disease Sterimol width parameters determined by distance of substituents at their maximum point perpendicular to attachment. ß-cyclodextrin polymer binding energy biopartitioning micellar chromatography

xii


BP-ANN CA CAE CCA CCIE CD CDA CE CFA cGC CGE CIEF CITP CLPP CoMFA DGE DPPH DTD CZE DA DAD DDF DGGE DIP DF DTD E ECD EFA ELISA ELSD EMLC EOF Es ESI EVOO F FA FCA FID FSMW-EFA FTIR GC GCA GDH

back-propagation neutral network cluster analysis cellulose acetate electrophoresis canonical correlation analysis core-core interaction cyclodextrin canonical discriminant analysis capillary electrophoresis correspondence factor analysis capillary gas chromatography capillary gel electrophoresis capillary isoelectric focusing capillary isotachophoresis community level physiological profiles comparative molecular field analysis denaturing gel electrophoresis 1,1-diphenyl-2-pictylhydrazyl direct thermal desorption capillary zone electrophoresis discriminant analysis diode array detector dry deposit flux denaturing gradient gel electrophoresis dipole moment discriminant function direct thermal desorption electric field electron capture detector evolving factor analysis enzyme-linked immunosorbent assay evaporative light scattering detection electrochemically modulated liquid chromatography electroosmotic flow Taft’s constant characterizing the steric effects of substituents electrospray ionization extra virgin olive oil Swain and Luton’s electronic parameter characterizing the inductive effect factor analysis fragmental constant approach flame ionization detector fixed size moving window-evolving factor analysis Fourier-transform infrared spectroscopy gas chromatography group contribution approach glutamate dehydrogenase


GLC GPTMS GRAM GSC H-Ac HbAlc H-Do HAC HELP HOF HOMO HPAE HPLC HPTLC HS-SPME GPA GRNN I IAM I.D. IEC IEF IGC ICP-AES IRMS K mw L LC LDA LDL LSER LSST LUMO MAE MALDI-TOFMS MCR-ALS MDS MEKC MFA MLEE MLR MLST MNLR MOS MOSFET M-RE

gas-liquid chromatography 3-(glycioxypropyl)trimethylsilane generalized rank annihilation method gas-solid chromatography indicator variable for proton acceptor properties hemoglobin Alc indicator variable for proton donor properties hierarchical ascending classification heuristic evolving latent projection heat of formation the highest occupied molecular orbital high-performance anion-exchange chromatography high performance liquid chromatography high performance thin-layer chromatography headspace solid phase micro-extraction general procrustes analysis generalized regression neural network effective length of the capillary used for CE separation immobilized artifical membrane internal diameter ion exchange chromatography isoelectric focusing inverse gas chromatography inductively coupled plasma atomic emission spectrometry isotope ratio mass spectrometry water-SDS micelle partition coefficients total length of the capillary used for CE separation liquid chromatography linear discriminance analysis low-density lipoprotein linear solvation energy relationship linear solvent strength theory energy of the lowest unoccupied molecular orbitals microwave-assisted extraction matrix-assisted laser desorption/ionization time-of-flight MS multivariate curve resolution-alternate least square multidimensional scaling analysis micellar electrokinetic chromatography multiple factorial analysis multilocus enzyme electrophoresis multilinear (multiple) regresion analysis multilocus sequence typing multi (multiple) nonlinear regression analyses metal oxide semiconductor metal oxide semiconductor field effect transistors molar refractivity

xiii

xiv


MRSA MS MSPME NECEEM NLM NMR NOM NPAH OPA OPLC OPR ORC clog P PAD PAGE PAH PARAFAC PCO PCR PE PEP B and PEP D PGC PGM PLE PLFA PLS PLS-DA PLs PDO PFGE PLE PMF pSASA PSA PU PUF Py-GC Q+ QFF QMAX QSAR QSRR QTOT R

methicillin-resistant Staphylococcus aureus mass spectrometry micellar solid-phase microextraction nonequilibrium capillary electrophoresis of equilibrium mixtures nonlinear mapping nuclear magnetic resonance natural organic matter nitropolycyclic aromatic hydrocarbon orthogonal projection approach overpressured-layer chromatography orthogonal projection resolution oxygen radical absorbance capacity logarithm of the peptide’s calculated n-octanol-water partition coefficient pulsed amperometric detection polyacrylamide gel electrophoresis polycyclic aromatic hydrocarbons parallel factor analysis principal co-ordinate analysis principal component regression polyethylene peptidase B and D porous graphitized carbon phosphoglucomutase pressurized liquid extraction phospholipid fatty acid profiling partial least squares partial least squares discriminant analysis phospholipids protected denomination of origin pulsed field gel electrophoresis pressurized liquid extraction positive matrix factorization polar solvent accessible surface area polar surface area polyurethane polyurethane foam pyrolysis gas chromatography the maximum of the net atomic charge on the H atoms quartz fibre filters the maximum of the net atomic charge on the C atom quantitative structure-activity relationship quantitative structure-retention relationship the sum of positive charge on C atoms Swain and Luton’s electronic parameter characterizing the resonance effect


RAPD RBFNN REP Rep-PCR RFLP RMSE RP-HPLC RP-TLC SASA SCD SDA SDS SEC SFA SFC SFE SIMCA SNR SPME SRA log SumAA SVM TCA TCM TG TE TOFMS TQS tR T-RFLP TSP TTGE UPGMA V log VDWVol. VER VOC WA WPC WSF

xv

randomly amplified polymorphic DNA radial basis function neutral network repetitive estragenic palindromes repetitive sequence polymerase chain reaction restriction fragment length polymorphism root-mean-square error reversed-phase high performance liquid chromatography reversed-phase thin layer chromatography solvent accessible surface area method of the sum of coefficient of determination stepwise discriminant analysis sodium dodecylsulfate size exclusion chromatography sub-window factor analysis supercritical fluid chromatography supercritical fluid extraction soft-independent modelling of class analogy signal-to-noise ratios solid phase microextration stepwise regression analysis logarithm of the sum of retention times of the amino acids composing the peptide support vector machine trichloroacetic acid typical-conditions model triacylglycerol total energy time-of-flight mass spectrometry total quality score retention time terminal-restriction fragment length polymorphism total suspended particulate temporal temperature gradient gel electrophoresis unweighted pair group method using arithmetic averages applied voltage logarithm of the van der Waals volume of the peptide varimax extended rotation volatile organic compounds wiener index whey protein concentrate water soluble fraction

1 Fundamentals 1.1

Multilinear and Nonlinear Regression Analyses

Traditional simple linear regression analysis calculates the correlation between one dependent (i.e. log k value of a solute) and one independent variable (i.e. concentration of organic modifier in the eluent C vol%). The general formula can be described by: Y = a + b1 · X1

(1.1)

where Y = log k , a = constant, X1 = C and b1 is the coefficient of regression. However, when the retention time of a set of solutes has been determined on more than one HPLC column, in mobile phases containing various organic modifiers at different concentrations and at various temperatures and we are looking for the influence of each chromatographic parameter on the log k’ value, each of them has to be included in equation (1.1), which modifies to: Y = a + b i · Xi + · · · + b i · Xi + · · · + b k · X k

(1.2)

Multivariate (multiple) linear regression (MLR) analysis deals with the solution of equations containing minimally two independent variables. Moreover the parameters included in equation (1.2) software generally calculate some additional mathematical-statistical values such as standard deviation of regression coefficients (sbi , where ‘i’ indicates any of the independent variables included in MLR), the path coefficients (bi ), calculated F values, and the coefficient of determination (r 2 ). The path coefficients are dimensionless numbers indicating the relative impact of a given independent variable (stationary phase characteristics, mobile phase composition, column temperature, etc.) on the dependent variable (generally retention time of the set of analytes). A higher path coefficient value (close to 1) indicates a higher influence of the given independent variable on the dependent one. The calculated F value shows the fit of the equation to the experimental data. When it is higher than the tabulated value corresponding to the same number of observations and independent variables, the relationship between the variables is significant. Calculated F values can be Multivariate Methods in Chromatography: A Practical Guide Tibor Cserháti C 2008 John Wiley & Sons, Ltd

2


found in every statistical handbook. The coefficient of determination indicates the ratio of the change of the dependent variable explained by the change of the independent variables. In other words, r 2 = 0.9243 indicates that 92.43% of the change of the dependent variable can be explained by the change of the independent variables. The number of papers dealing with the application of multiple nonlinear regression (MNLR) is fairly low. This may be due to the fact that linear models generally adequately describe the relationship between dependent and independent variables making it unnecessary to use the more complicated nonlinear (mainly exponential and logarithm) regression analysis.

1.2

Stepwise Regression Analysis and Partial Least Squares Method

Besides the traditional MLR technique some new computerized regression analysis methods have been developed based on theoretical considerations. It is well known that in the traditional multivariate regression analysis the presence of independent variables (chromatographic parameters or physicochemical characteristics) that exert no significant influence on the dependent variable (retention behaviour) lessens the significance level of the independent variables that significantly influence the dependent variable. To overcome this difficulty, stepwise regression analysis (SRA) automatically eliminates from the selected equation the insignificant independent variables increasing in this manner the information power of the calculation. The form and the statistical parameters computed by SRA are identical to those discussed in Section 1.1 and their statistical meaning is the same. The partial least squares (PLS) technique is actually a method of preference for carrying out regression analysis. The advantage of the method is that it can be successfully employed when the independent variables are highly intercorrelated. PLS relates two different sets of variables using one data set as predictor variables. The dimensionality of the original data set is reduced by estimating one or more underlying background variables. PLS forms the score vectors as linear combinations of the original sets of variables. The theory of PLS has been previously discussed in detail [1]. PLS has been readily accepted and extensively used in up-to-date mathematical-statistical studies. The theory of various regression analysis models is discussed in detail in Mager [2].

1.3

Two- and Three-dimensional Principal Component Analysis, Various Factor Analytical Techniques

In many cases the chromatographer is not interested in the dependence of one retention parameter on the chromatographic or physicochemical characteristics (see above multiple regression methods), but rather wishes to find the relationship between all parameters without one being the dependent variable. Both PCA and factor analysis (FA) comply with this requirement. The main advantages of these computation techniques in chromatography are: r elucidating the similarity and dissimilarity among the variables (clustering chromatographic systems or solutes according to their retention behaviour);

Fundamentals

3

r the possibility of the extraction of one or more background (theoretical) variables having concrete physicochemical meaning for the theory and practice of chromatography; r making it possible to reduce the number of variables (chromatographic systems or solutes) to the minimum necessary for the solution of a problem. Traditional two-dimensional principal component analysis (2D-PCA), a versatile and easy-to-use multivariate mathematical-statistical method has been developed to contribute to the extraction of maximal information from large data matrices containing numerous columns and rows. PCA makes possible the elucidation of the relationship between the columns and rows of any data matrix without one being the dependent variable. 2D-PCA is a so-called projection method representing the original data in smaller dimensions. It calculates the correlations (similarities and dissimilarities) between the columns of the data matrix and classifies the variables according to the coefficients of correlations taking into consideration simultaneously the magnitude and sign of the coefficients of correlation. As the resulting matrices of principal component loadings and variables (scores) are also multidimensional, the visual evaluation of such matrices is cumbersome or even impossible. The plot of principal component loadings and/or scores of the first versus the second principal component has been frequently used for the evaluation of the similarities and differences among the observations. This method takes into consideration only the variance explained in the first two principal components and ignores the impact of variances explained by the other principal components on the distribution of the matrix elements. The use of this approximation is only justified when the first two principal components explain the overwhelming majority of variance, which is not probable in the case of large original data matrices. Theoretically PCA can be used for the analysis of any data matrices; however, its inadequate application may lead to serious misinterpretation of the results. As it was previously mentioned, PCA calculates the similarities and dissimilarities among the variables and observations according to the differences among the coefficients of correlation. This means that the distribution of principal component loadings and scores will be similar in the theoretical cases when each coefficient of correlation is in the range of 0.1, 0.5 or 0.9. While the scattering of points calculated from the coefficient of correlation 0.1 does not contain any useful (significant) information, a similar distribution of points marks significant relationships when it is calculated from the correlation matrices of 0.9. The publication of a table containing each coefficient of correlation overcomes the difficulty emerging from the similar scattering of points. PCA software generally calculates the so-called ‘eigenvalue’, the ratio of variance explained by the individual principal component loadings and scores, and the numerical values of principal component loadings and scores in the principal components. The total variance explained by PCA has to be previously fixed and arriving at this limit value the computation stops. Depending on the character of the original data matrix the total variance explained is generally fixed at 95 or 99%. Eigenvalues indicate the ratio of variance explained by the individual principal components; their value decreases monotonously with increasing number of principal components. It is generally assumed that only eigenvalues more than 1 contain valuable information; eigenvalues less than 1 only refer to the error of measurement. Similarly to eigenvalues, the ratio of variance also decreases with increasing number of principal components. The numerical values of principal component loadings

4


and scores indicate the impact of a given point of observation on the individual principal component. Traditional PCA is a typical multivariate two-dimensional statistical method unsuitable for the evaluation of three(or more)-dimensional data matrices. A three-dimensional PCA (3D-PCA) model has been developed (Tucker model) to overcome this difficulty. However, 3D-PCA, a very elegant and easy method to carry out, has not found frequent application in the evaluation of chromatographic data matrices. For the exact mathematical treatment of these methods see references [1], [2], [5] and [6] in the Introduction and Malinowski and Howery [3]. As a result of its versatility PCA has been extensively used not only in chromatography but also in other fields of research. Interestingly, FA suitable for similar calculations has been less frequently employed in chemometrics studies than PCA.

1.4

Canonical Correlation Analysis

Sometimes analytical chemists are interested in the simultaneous dependence of more than one retention characteristic or more than one chromatographic parameter on a considerable number of chemical (eluent composition, eluent pH, etc.), physical (temperature, flow rate, etc.) and physicochemical (connectivity and sterical indices, etc.) variables. As the number of dependent variables is higher than one, traditional MLR models cannot be applied in these instances. Canonical correlation analysis (CCA) has been developed for the solution of this type of computational problem; it calculates the correlation between two different data sets. CCA computes linear correlations between the variables of the two data sets, both of them including more than one variable. The variables in the data set with lower number of variables (matrix II, i.e. more than one retention parameter) are considered as dependent variables, while the variables in the data set with higher number of variables (matrix I, i.e. physicochemical parameters of solutes) are considered as independent variables. The correlation coefficients of the linear relationships are computed in such a manner that they explain the maximal information in the data set containing the lower number of variables (matrix II) using the information in the other matrix (matrix I). The number of relationships explaining 100% of the variance in matrix II is equal to the number of original variables in matrix II. The theoretical background of CCA is given in detail in Orloci et al. [4]. Despite its obvious advantages, CCA has found only limited application in chromatography [5–8].

1.5

Discriminant Analysis

Discriminant analysis (DA) can be used for the computation of a hyperplane in the input space minimizing the within class variance and maximizing the between class distance. It is one of the supervised pattern recognition methods frequently used in various fields of chemometric calculations. It finds explicit boundaries between given classes, in order to discriminate them. The latent (combined) variable is the linear combination of the original variables.

Fundamentals

1.6

5

Spectral Mapping

The majority of multivariate methods classify the chromatographic retention data taking into consideration simultaneously the strength and selectivity of the retention, so cannot be applied when the separation of the strength and selectivity of the effect is required. The spectral mapping (SPM) technique, another multivariate mathematical-statistical method, overcomes this difficulty [9]. The method divides the information into two matrices using the logarithm of the original data. The first matrix is a vector containing the potency values related to the overall effect. The second matrix (selectivity map) contains information concerning the spectra of activity (the qualitative characteristics of the effect) [10]. SPM first calculates the logarithm of the members of the original data matrix facilitating the evaluation of the final plots in terms of log ratios. Consecutively, SPM subtracts the corresponding column-mean and row-mean from each logarithmic element of the matrix calculating potency values. The source of variation remaining in the centred data set can be evaluated graphically (selectivity map). SPM has been previously employed for the characterization of stationary phases in HPLC [11].

1.7

Nonlinear Mapping

As mentioned in Section 1.3, the principal components and scores computed by PCA are generally multidimensional, which makes the visual evaluation cumbersome, even impossible. Nonlinear mapping (NLM) has been developed for the reduction of the dimensionality of complicated matrices consisting of numerous columns and rows. The method can reduce the dimensionality of the data matrices to two in such a manner that the distances between the points on the projection plane approximate the distances on the multidimensional space. This means that points (i.e. chromatographic systems) near to each other are similar and points situated far away from each other are markedly different.

1.8

Cluster Analysis

Similarly to NLM various cluster analysis (CA) techniques have been developed and successfully employed for the easy visualization of multidimensional data matrices by reducing dimensionality. Variables with similar characteristics are near to each other on the CA dendograms, while variables with different characteristics are far away from each other. On account of the good visualization of the results, CA is generally combined with other multivariate methods (mainly with PCA). The principles of CA are discussed in detail in Willett [12].

1.9

Other Multivariate Techniques

In addition to the mathematical-statistical methods discussed above, many other computational techniques have been developed and applied in the chemometrical evaluation of

6


chromatographic data sets. However, they have not been frequently used so that their contribution to mathematical-statistical evaluation in chromatography is fairly low.

1.10

Measured and Calculated Physicochemical Parameters of Chromatographic Systems and Analytes

A high number of physical, physicochemical and biophysical parameters have been applied in the chemometrical investigation of chromatographic data. The majority of these characteristics are calculated and not measured values. This can be explained by the fact that the advent of rapid computers with high calculation capacity and complicated software make possible the rapid calculation of numerous parameters, while the measurement of any concrete parameter is time-consuming and sometimes expensive compared with computational results. Since different principles are involved in the various chromatographic techniques, the impact of parameters used in the calculations may also be different. Thus, the importance of a given molecular characteristic may be high in GC and negligible in CZE. The parameters used more frequently in GC are: boiling point, molar volume, molecular mass, molar refraction, octanol-water partition coefficient, various indicator variables (i.e. no side chain = 0, one side chain = 1, two side chains = 2), AM1 total energy, Randic molecular profile, Randic shape profile, mean electrotopological state, number of rotatable bonds, Gutman molecular topological index by valence vertex degree, third component accessibility directional WHIM index/unweighted, H autocorrelation of lag 5/weighted by atomic masses, R maximal autocorrelation of lag 2/weighted by atomic Sanderson electronegativities, H autocorrelation of lag 2/weighted by atomic Sanderson electronegativities, R maximal autocorrelation of lag 4/weighted by atomic van der Waals volume, Balaban-type index from van der Waals weighted distance matrix, Balaban-type index from electronegativity weighted distance matrix, average valence connectivity index chi-2 and chi-0, valence connectivity index chi-1, polarity index, magnitude of dispersive interactions between a methylene group and the stationary phase, partial molar excess Gibbs free energy of solution per methylene group, cohesive energy, solubility parameter, enthalpy of vaporization distance edge vector, standard molecule chemical potential, energy of the lowest unoccupied molecular orbital (LUMO), dipole moment (DIP), the maximum of the net atomic charge on the C atom (QMAX), the sum of positive charge on C atoms (QTOT), topological indexes (CHI-2, CHI-0AV, CHI-2AV, CHI-0a, CHI-2A), wiener index (WA), heat of formation (HOF), total energy (TE), the maximum of the net atomic charge on the H atoms (Q+), binding energy (BE), core–core interaction (CCIE), solvent accessible surface area (SASA), polar solvent accessible surface area (pSASA), and apolar solvent accessible surface area (apSASA). Quantitative structure–retention relationship (QSRR) studies carried out in subcritical chromatography used excess molar refraction, dipolarity/polarizability, hydrogen bond acidity and basicity, and McGowan’s characteristic volume as molecular parameters. The principle of separation in TLC and HPLC is similar, so the physicochemical parameters employed are also similar. Thus, the following molecular characteristics have been extensively applied: Hansch-Fujita’s substituent constants characterizing hydrophobicity (); indicator variables for proton acceptor and proton donor properties (H-Ac and H-Do,

Fundamentals

7

respectively); molar refractivity (M-RE); Swain and Luton’s electronic parameters characterizing the inductive and resonance effects (F and R, respectively); Hammett’s constant characterizing the electron-withdrawing power of the substituents at meta and para+ortho positions (m and p+o , respectively); Taft s constant characterizing the steric effects of substituents (Es); Sterimol width parameters determined by distance of substituents at their maximum point perpendicular to attachment (B1 and B4 ). These molecular parameters were calculated according to the additivity rule from the fragmental constants. Fragmental constants are parameters characterizing elementary molecular substructures. However, it has to be borne in mind that the application of the additivity rule does not take into consideration the possible intramolecular interactions among the substructures in the molecule which may result in inadequate results.

References [1] Geladi, P., and Kowalski, B. R. Anal. Chim. Acta 185 (1986) 1–17. [2] Mager, H. Moderne Regressionsanalyse, Salle, Sauerländer, Frankfurt am Main, 1982. [3] Malinowski, E. R., and Howery, D. C. Factor Analysis in Chemistry, John Wiley & Sons, Ltd, New York, 1980. [4] Orloci, L., Rao, C. R., and Stitiler, W. M. Multivariate Methods in Ecological Work, International Cooperative Publishing House, Fairland, MD, 1979. [5] Forgács, E., Cserháti, T., and Bordás, B. Chromatographia 36 (1993) 19–26. [6] Forgács, E., Cserháti, T., and Bordás, B. Anal. Chim. Acta 279 (1993) 115–122. [7] Cserháti, T., and Forgács, E. Chem. Intell. Lab. Syst. 28 (1995) 305–313. [8] Forgács, E., and Cserháti, T. J. Liq. Chrom. Rel. Technol. 19 (1996) 1849–1858. [9] Lewi, P.J. Arzneim. Forsch. 26 (1976) 1295–1300. [10] Lewi, P. Chemom. Intell. Lab. Syst. 5 (1989) 105–116. [11] Hamoir, T., Cuaste Sanchez, F., Bourguignon, B., and Massart, D. L. J. Chromatogr. Sci. 32 (1994) 488–498. [12] Willett, P. Similarity and Clustering in Chemical Information, Research Studies Press, New York, 1987.

2 Gas Chromatography 2.1

Theory and Practice of Gas Chromatography

The principle of gas chromatography (GC) methods is the distribution of volatile analytes between a stationary and a mobile phase. The mobile phase is a gas and the stationary phase is either solid (gas–solid chromatography, GSC) or a nonvolatile liquid (gas–liquid chromatography, GLC). Although various GC techniques have been extensively used in many fields of research and development for the separation and quantitative determination of a wide variety of analytes, their application is restricted to compounds which are volatile and stable at the temperature of the analysis (column temperature being possibly below 350–400◦ C). A traditional GC instrument consists of a carrier gas delivery system, injector apparatus, a column carrying out the separation and a data processing section. Various types of injectors have been developed for delivering the sample to the head of the separation column with the smallest possible bandwidth. Both vaporization and on-column injectors can be employed in GC. The temperature of vaporization injectors is identical or higher than that of the separation column. Syringes of different construction can be used to introduce the sample into the heated injector. The volatile components of the sample are readily vaporized, mixed with the carrier gas and delivered to the GC column. Nonvolatile components of the sample remain in the injector or at the head of the column decreasing the separation capacity of the system. On-column injectors deposit the sample directly into the column. Headspace sampling, an elegant and simple method, places a liquid or solid sample in a vial sealed with a septum cap. The vial is heated at a given temperature and a ratio of volatile components of the sample is equilibrated between the gas and the liquid phase. The quantity of a given analyte in the headspace depends on the volatility of the analyte and on its ratio in the sample. After reaching equilibrium a sample of the headspace is removed with a needle and injected into the GC column. Both packed and capillary columns can be used for the separation of analytes. Packed columns are prepared of metal or glass, the internal diameter (ID) generally varying between Multivariate Methods in Chromatography: A Practical Guide Tibor Cserháti C 2008 John Wiley & Sons, Ltd

10


2 mm and 4 mm. Columns are filled with small particles of uniform diameter. GLC uses particles coated with a thin layer of high molecular weight nonvolatile polymer. Supports most commonly employed in packed columns are diatomaceous earths, fluorocarbons, graphitized carbon black and glass beads. The polymer coating has to have very low vapour pressure, high chemical stability, possibly low viscosity at the temperature of the separation, selectivity for the analytes to be separated, and good wetting capacity for the surface of the inert support or the inert wall of the column. The length of the packed columns is limited to about 3 m because of the high pressures needed to maintain the flow rate of the carrier gas required for the optimal separation capacity. The advantage of the packed columns is the higher sample capacity (10 to 1000 times greater than that of capillary columns), but their theoretical plate number per metre is lower than that of capillary columns. As capillary columns are markedly longer (10–60 m) than packed ones the total plate number of capillary columns is considerably higher. Capillary columns are a glass or fused-silica tube with an ID of 0.20–0.53 mm. The inner walls of the capillary tubes are coated with a thin layer of nonvolatile polymeric stationary phase. The requirements for the coating are the same as those of packed columns. The majority of coatings are cross-linked and covalently bonded to the inner wall of a fused-silica tube. The film thickness of the coating varies between 0.1 and 5 μm. The retention of analytes increases with increasing thickness of polymer coating. Capillary columns have a very large number of theoretical plates resulting in high separation capacity and high sensitivity. The more advantageous characteristics of capillary columns means that they are more frequently used than packed ones. The separation efficacy of a gas chromatographic column depends on the reliability of the thermostation of the column. Simple separations can be carried out at constant column temperature (isocratic separation mode) or – in the case of complicated analyte mixtures – its temperature can be increased according to a predetermined program (temperature gradient analysis). As the retention of analytes depends considerably on the column temperature, the precise regulation of the column temperature is of paramount importance. Permanent gases with low or negligible adsorption capacity (i.e. hydrogen, helium and nitrogen) can be employed as carrier gas. The type of carrier gas modifies the separation characteristics, separation efficacy and sensitivity of the GC system. The high flammability and the rigorous safety regulations limit the application of hydrogen as carrier gas. Helium is an excellent alternative but its use is restricted by its price. The majority of GC analyses employ inexpensive nitrogen as carrier gas. Other gases have found only limited application and have been used for the solution of specific separation problems or for the elucidation of theoretical problems. Similarly to the thermostation, the stability and reproducibility of the flow rate of the carrier gas is a prerequisite for an effective GC investigation. Analytes separated on the GC column are determined at the column exit. Detectors interact with the eluted compounds; the interaction is converted into an electrical signal which is sent to a recorder or data storage apparatus. The data are presented by plotting the intensity of the signal versus the time of analysis. This plot is the so-called chromatogram. A fairly high number of detectors have been developed and employed in GC. Detectors differ in sensitivity and selectivity. Sensitivity is characterized by the lowest detectable quantity of analyte whereas selectivity can be defined as the differences between the detector responses to various analytes at the same concentration. Detectors such as flame ionization, nitrogen-phosphorous, flame photometric, electron capture, thermal conductivity, atomic emission, electrolytic conductivity, chemiluminescence and various

Gas Chromatography

11

types of mass spectrometric systems have been extensively employed in up-to-date GC analytical processes. Data evaluation has been facilitated by the application of complicated data handling devices and special software based on personal computers. The distribution of analytes between the stationary and mobile phase (carrier gas) can be described by the distribution constant (K D ), which is the ratio of the concentrations of the analyte in the stationary and mobile phases: K D = compound concentration in stationary phase/compound concentration in mobile phase (2.1) The dependence of the retention on the column temperature and on the characteristics of an analyte is defined by: ln K D = −G 0 /RT

(2.2)

where K D is the distribution constant determined by equation (2.1), G 0 is the change in Gibbs free energy for the evaporation of an analyte from the stationary phase, T is the column temperature and R is the ideal gas constant. Equation (2.2) illustrates that the differences in the Gibbs free energy for the evaporation of the analytes from the stationary phase result in their different retention. Retention time of an analyte (tR ) is equal to the time differences between the start of the separation procedure and the maximum of its chromatographic peak. The partition ratio or capacity factor (k ) is the time ratio that an analyte spends in the stationary phase relative to the mobile phase: k = (tR − t0 ) /t0

(2.3)

where t0 is the time needed for a nonretained analyte to travel through the column. The term retention index (I ) was introduced by Professor E. Kováts [1]. Its application markedly enhanced the reliability and reproducibility of the measurement of retention time. The method compares the retention time of a given analyte with the retention times of normal hydrocarbons eluting before and after the analyte. The retention index can be computed by: I = 100 · c + 100 · (log VNx − log VNc ) / (log VNc+1 − log VNc )

(2.4)

where x refers to the analyte, c refers to the number of carbon atoms of the n-hydrocarbon eluting before the analyte, and c + 1 refers to the number of carbon atoms in the n-hydrocarbon eluting after the analyte. VN is the net retention volume. Net retention time can be determined by measuring the distance between the peaks of nonadsorbed and adsorbed analytes. Separation characteristics of a chromatographic analysis can be described by the separation factor (): = k2 /k1 k1

(2.5) k2

is the partition ratio of the earlier eluting peak, and is the partition ratio of where the later eluting peak. The efficacy of the separation of two neighbouring peaks can be characterized by the resolution number (R): R = 1.18 (tR2 − tR1 ) / (wh1 + wh2 )

(2.6)

R = 2 (tR2 − tR1 ) / (wb1 + wb2 )

(2.7)

12


where tR1 and tR2 are the retention times of peaks 1 and 2, respectively; wh1 and wh2 are the peak widths at half-height of peaks 1 and 2, respectively; wb1 and wb2 are the peak widths at the base of peaks 1 and 2, respectively. Theoretical plate number (N ) is related to the separation capacity of the chromatographic column. It can be computed by: N = 5.545 (tR /wh )2

(2.8)

The theory and practice of GC has been discussed in detail in numerous books and book chapters dealing with the retention parameters of stationary phases [2], and mixed stationary phases [3], column switching techniques [4], solvating GC employing packed column [5], theoretical aspects of detectors for capillary GC [6], practical [7] and analytical applications [8], laboratory analysis [9], analysis of air pollutants [10] and natural products [11].

2.2

Comparison of Gas Chromatography Stationary Phases Using a Homogenous and Nonhomogenous Set of Analytes

One of the main fields of the application of mathematical-statistical methods in GC is the classification of stationary phases according to their retention characteristics. On account of their relative simplicity, the majority of studies employ either MLR or other PCA methodologies. In order to elucidate the various aspects of the correlation between stationary phase characteristics and retention parameters of analytes these computing procedures are frequently applied simultaneously. MLR techniques were used to study the correlation between the retention behaviour of 75 polychlorinated dibenzodioxin congeners on various stationary phases and a new molecular distance edge vector (μ). The retention times of analytes were measured on DB-5, SP2100, SE-54, and OV-1701 columns. The data clearly show that the three novel molecular descriptors differ considerably among the analytes; therefore, they can be used in QSRR computations. The parameters of linear regressions between the measured chromatographic parameters and the calculated descriptors are listed in Table 2.1. The good fit of the equation to the experimental data suggests that these descriptors can be successfully employed for the prediction of the retention behaviour of this class of analytes and MLR is an adequate method to calculate the predicted retention values [12]. MLR and CA have been used to determine the differences between the system constants of bis(cyanopropylsiloxane)-co-methylsilarylene HP-88 and poly(siloxane) Rtx-440 stationary phases. It has been established that system constants can be readily calculated by MLR from the solvation parameter model. The CA dendogram of a considerable number of GC stationary phases was calculated and is shown in Figure 2.1. The data can facilitate the selection of a stationary phase with highly different selectivity for the solution of difficult separation problems [13]. A similar method applying MLR, PCA and CA was employed for the study of the similarities and dissimilarities among another set of stationary phases and analytes. The stationary phases were poly(methyl-trifluoropropyl) siloxanes with various trifluoropropyl content and some other commercially available ones. Hydrocarbons, aromatics, alcohols, amines, nitriles, and halogenated and heterocyclic compounds served as analytes (altogether

1657.59 ± 1.71 206.28 ± 7.15 −254.83 ± 8.05 186.06 ± 1.78 41 0.9924

1657.07 ± 65.67 202.91 ± 111.21 −258.13 ± 181.69 186.84 ± 39.64 41 0.9949 0.9945 25.84 1200.1

0.2396 ± 0.6299 5.0266 ± 1.2620 1.2863 ± 2.3020 −0.3228 ± 0.4221 39 0.9569 0.9531 0.357 126.5

1.284 ± 0.028 1.0872 ± 0.0476 0.3598 ± 0.0197 −0.1528 ± 0.0109 15 0.9914

0.018

SP-2100

OV-1701 (CV)

31.43

DB-5 (CVa )

DB-5

SE-54

−1.5942 ± 0.1202 0.6032 ± 0.2409 −0.5189 ± 0.4394 0.3943 ± 0.0806 39 0.9985 0.9983 0.068 3835.6

SP-2100b

1.6739 ± 0.5477 1.5763 ± 0.6548 0.6580 ± 0.5813 −0.3022 ± 0.2285 15 0.9935 0.9918 0.017 280.9

Regression coefficients and statistics of various linear models

0.069

−1.5913 ± 0.0027 0.6216 ± 0.0068 −0.5082 ± 0.0083 0.3909 ± 0.0017 39 0.9969

SP-2100b (CV)

0.022

1.6935 ± 0.0351 1.6102 ± 0.05 0.6687 ± 0.0325 −0.31 ± 0.0139 15 0.9889

SE-54 (CV)

b

CV, the predictive retention data in the cross validation procedure. The result after the transformation. Reproduced from the Journal of Chromatographic Science by permission of Preston Publications, a division of Preston Industries Inc. ref [12].

a

b0 b1 b2 b3 n R EV RMS F

b0 b1 b2 b3 n R EV RMS F

Table 2.1

1.2641 ± 0.3746 1.0534 ± 0.4478 0.3451 ± 0.3975 −0.1444 ± 0.1563 15 0.9965 0.9956 0.012 526.3

OV-1701

14

Multivariate Methods in Chromatography: A Practical Guide DB-1301 DB-624 Rtx-440 Rtx-20 DX-1 HP-5 DB-VRX DB-1 SolGel-1 HP-5TA SPB-Octyl Rtx-50 Rtx-65 DB-608 DB-35 DB-1701 DB-210 DB-200 Cyclodex-B Cyclosil-B DX-3 DB-225 SolGel-Wax HP-20M DB-FFAP DX-4 SP-2340 HP-88 DB-23

Figure 2.1 Average-linkage dendogram for hierarchical CA of representative columns for different stationary phase compositions contained in the database of system constants for the temperature range 60–140◦ C. DP, HP, DX, Cyclodex and CycloSil columns are from Agilent Technologies; Rtx from Restek Corporation; sol-gel from SGE Incorporated; and SP and SPB from Supelco Incorporated. Reprinted with permission from ref. [13]. Copyright Elsevier

64 analytes). MLR computations were carried out using the solvation model. The classification of stationary phase according to the results of CA is depicted in Figure 2.2. It was established that these calculation methods are suitable for the classification of the newly synthesized stationary phases among the well established commercial products. [14]. The molecular descriptors of 50 disulfides were employed for the prediction of Kováts indices on four different stationary phases. The stationary phases included in the computation were Apiezon M, OV-17, Triton X-305 and PEG-1000. The molecular descriptors included

Gas Chromatography

15

DB-200 VB-210 TFPS35 TFPS26(2) TFPS26(1) DB-210 OF-1(1) OV-215 OF-1(2) TFPS50 PSF6 DB-VRX HP-5(1) DB-1 TFPS002 TFPS001 HP-5(2) HP-50 DB-1301 DB-624 Rtx-440 Rtx-20 DB-35 TFPS12 TFPS09 DB-5 OV-195 Rtx-50 Rtx-65 DB-608 TFPS15 DB-1701 BP-10 SP-2340 HP-88 DB-23 BPX70 DB-225

Figure 2.2 Dendogram of the column set by the nearest neighbour method. Reprinted with permission from ref. [14]. Copyright Elsevier

in the calculation were heat of formation (HOF), dipole moment (DIP), the maximum of net atomic charge on the C atom (QMAX), total energy (TE), the lowest unoccupied molecular orbital (LUMO), the highest occupied molecular orbital (HOMO), the maximum of the net atomic charge on the H atoms (Q+), binding energy (BE) and core–core interaction (CCIE). Data matrices were evaluated by both MLR and PCA. The data prove that good

16


Table 2.2

CCIE HOF HOMO LUMO QMAX Q+ DIP TE BE

Result of the PCA Factor loading 1

Factor loading 2

Factor loading 3

Factor loading 4

Factor loading 5

−0.4454 0.4119 0.0274 −0.2605 −0.3339 0.2300 −0.0564 0.4469 0.4428

0.0480 −0.1785 −0.6276 −0.1915 −0.2339 0.3054 0.6168 −0.0493 −0.0661

−0.0348 0.3132 −0.1142 0.7930 0.3120 0.2322 0.2542 0.1427 0.1502

0.1901 −0.0924 0.3048 0.1122 −0.2986 0.8020 −0.2396 −0.1646 −0.1817

−0.1695 −0.1630 0.0210 −0.3891 0.7935 0.3905 −0.0126 0.0829 0.0613

HOMO (eV), the highest occupied molecular orbital; Q+ the maximum of the net atomic charge on the H atoms; BE (kcal), binding energy; TE (kcal mol−1 ), total energy. Reprinted with permission from ref. [15]. Copyright Elsevier.

linear correlations can be found among the retention indices and the calculated physicochemical parameters of analytes on each stationary phase. The fact that the number and relative weight of the independent variables varied according to the polarity of the stationary phases indicates the different retention mechanisms of the columns. It has been supposed that the results of regression analyses can be successfully employed for the prediction of the retention indices of other analytes on these stationary phases. The PCA loadings are compiled in Table 2.2. It was established that five background variables explain 99% of the total variance indicating the similarities between some calculated physicochemical parameters [15]. PLS has also been employed for the study of the correlation between the Kováts retention indices of 35 aliphatic ketones and aldehydes and their physicochemical parameters on four different stationary phases (methyl-, methyl-phenyl-, trifluoropropyl-methyl siloxane and polyethylene glycol). The physicochemical characteristics included in the computation were boiling point, molar volume, molecular mass, molar refraction, and octanol-water partition coefficient. It was found that the PLS method is suitable for the prediction of the retention indices of these classes of analytes on the columns investigated [16]. MLR has also been applied for the characterization of stationary phases in subcritical fluid chromatography. The parameters of the stationary phases are compiled in Table 2.3. The retention factor of 109 analytes was measured on the stationary phases listed in Table 2.3 using carbon dioxide–methanol mixtures (9:1, v/v) at a column temperature of 25◦ C. Analytes were characterized by the excess molar refraction, dipolarity/polarizability, hydrogen bond acidity and basicity, and McGowan’s characteristic volume. Data were analysed by SRA. The parameters of linear relationships between stationary phases and physicochemical parameters of analytes using a linear solvation energy relationship (LSER) are listed in Table 2.4. The significant correlation between the dependent and independent variables indicates that these physicochemical parameters exert a marked influence on the retention of this class of analytes and they can be employed for the prediction of retention of analytes of similar chemical structure. It was further concluded from the calculations that MLR

Gas Chromatography

17

Table 2.3 Stationary phases characterized in this study

Abbr. C4 C8 C12 C18 RPH PE FD

Nature of the stationary phase Butylsiloxane-bonded silica Octylsiloxane-bonded silica Dodecylsiloxane– bonded silica Octadecylsiloxane– bonded silica Octadecyl- and phenylsiloxane– bonded silica Amide-embedded octadecylsiloxanebonded silica Fluorodecylsiloxane– bonded silica

Surface (m2 g−1 )

C%

Interchim

320

7

Uptisphere C8

Interchim

320

9

Synergy Max RP

Phenomenex

475

n.a.

Kromasil C18 100

Eka Nobel

340

19

Uptisphere RPH

Interchim

330

15

Supelcosil ABZ+

Supelco

320

12

Chromegabond Fluorodecyl

ES Industry

n.a.

n.a.

Trade name

Manufacturer

Uptisphere C4

n.a., not available. Reprinted with permission from ref. [17]. Copyright Elsevier.

can be successfully applied for the study of the relationship between the characteristics of analytes and mobile phases [17]. Similar investigations were carried out using aromatic stationary phases and LSER calculations combined with SRA. Also, in this case the highly significant correlations indicate the correct selection of independent variables and suggest the possibility of prediction of retention of other analytes on these stationary phases. Moreover, it proves that MLR can be applied for these types of studies even in subcritical fluid chromatography [18]. SRA was employed for the investigation of the retention characteristics of a 2,3,6-tri-Omethyl--cyclodextrin coated capillary column. The retention time of (+)-g-lactone and (−)-g-lactone was determined at various temperatures (105–165◦ C, in steps of 5◦ C) and flow rates (0.5–1.5 ml min−1 , in steps of 0.25 ml min−1 ). The dependence of the retention parameters (differences between the retention times, separation factor, and peak half width) on the chromatographic conditions (linear, logarithmic, quadratic, and reciprocal values of column temperature and flow rate) was calculated by SRA. The results of SRA are compiled in Tables 2.5–2.7. The data clearly show that the relationship between the retention of analytes and the chromatographic conditions deviates markedly from those established for traditional columns. This discrepancy has been tentatively explained by the formation of inclusion complexes between the cyclodextrin (CD) stationary phase and the analytes [19]. PCA and similar calculation procedures have also often been applied in the elucidation of the relationship between the retention characteristics of various stationary phases due to their popularity and simplicity. Thus, the similarities and dissimilarities among the polarity indicators and stationary phases have been evaluated by PCA. Computations were carried

e

0.209 0.021 0.304 0.018 0.408 0.021 0.577 0.032 0.470 0.022 0.648 0.043 0.282 0.024

c

−1.232 0.039 −0.995 0.031 −0.809 0.037 −0.858 0.057 −0.809 0.039 −1.214 0.073 −1.069 0.053

LSER models

−0.167 0.036 −0.229 0.033 −0.269 0.035 −0.388 0.055 −0.261 0.032 −0.292 0.072 − -

s −0.021 0.027 −0.260 0.024 −0.366 0.027 −0.443 0.038 −0.407 0.024 1.287 0.053 0.457 0.039

a 0.160 0.040 −0.200 0.039 −0.321 0.040 −0.470 0.060 −0.411 0.040 −0.310 0.078 0.961 0.056

b 0.216 0.030 0.257 0.024 0.317 0.028 0.465 0.043 0.373 0.039 0.320 0.056 −0.501 0.060

v

98

77

76

85

86

87

89

n

0.930

0.975

0.992

0.982

0.984

0.969

0.909

R

0.859

0.947

0.983

0.962

0.967

0.936

0.817

2 Radj

0.100

0.121

0.054

0.095

0.060

0.060

0.060

sd

148.8

272.7

863.3

430.5

495.6

282.2

89.4

F

2 n, the number of solutes considered in the regression; R, the multiple correlation coefficient; Radj, the adjusted correlation coefficient; sd, the standard error of the estimate; F , Fisher’s statistic. The numbers in italics represent 95% confidence limit. Reprinted with permission from ref. [17]. Copyright Elsevier.

FD

PE

RPH

C18

C12

C8

C4

Stat. phase

Table 2.4

Gas Chromatography

19

Table 2.5 Dependence of the difference retention between g-lactone isomers ( y = t1−2 ) on the column temperature ( x1 ) and on the flow rate of carrier gas ( x2 ). Results of stepwise regression analysis Y = a + b1. 1/x1 + b2. log x1 + b3. x2 N = 55, a = −141.7, F calc. = 144.8, r 2 = 0.8888 No. of independent variables Parameter

1

b sb Path coefficient %

3910 678 53.27

2

3

53.6 11.8 42.17

−0.88 0.11 4.56

Reprinted with permission from ref. [19].

out on a data matrix consisting of 8 polarity indicators (columns of the original matrix used for PCA) and 30 stationary phases (rows in the same PCA matrix). Calculations were motivated by the fact that a considerable number of polarity indicators has been introduced for the characterization of stationary phases and the conclusions drawn from the application of different polarity parameters are frequently different. Polarity indicators included were the McReynolds polarity, the sum of the first five McReynolds constants, the sum of retention index differences between a given liquid phase and squalene for benzene, n-butanol, 2-pentanone, nitropropane and pyridine; Kováts’ coefficients (Kc ) defined by (−100)times the intercept slope ratio of the logarithm of retention volume versus carbon number lines for n-alkanes; retention polarity, Snyders selectivity parameters b , n and d for butanol, nitropropane and dioxane, respectively; Castello’s DC values defined similarly to Kováts’ coefficients but applying n-alkanols as model compounds, and the combined variable yb = b /( b + ). The data clearly show that four polarity parameters have high loadings in the first factor illustrating their similarity. They explain about 54% of the total variance. The other three indices have high loadings in the second factor explaining 35% of Table 2.6 Dependence of the difference retention between g-lactone isomers ( y = W) on the column temperature ( x1 ) and on the flow rate of carrier gas ( x2 ). Results of stepwise regression analysis Y = a + b1. 1/x1 + b2. log x1 + b3. x2 N = 55, a = −871, F calc. = 154.5, r 2 = 0.8951 No. of independent variables Parameter b sb Path coefficient %

1

2

3

22 466 2662 52.70

333.0 46.1 45.09

−2.49 0.43 2.21


20

Multivariate Methods in Chromatography: A Practical Guide Table 2.7 Dependence of the separation factor ( y = Rs ) of g-lactone isomers (y = t1−2 ) on the column temperature (x1 ) and on the flow rate of carrier gas (x2 ). Results of stepwise regression analysis Y = a + b1 ·1/x1 + b2 ·(x1 )2 + b3 ·x1 ·x2 N = 55, a = 2.11, F calc. = 37.6, r 2 = 0.6708 No. of independent variables Parameter b sb Path coefficient %

1

2

3

−152.9 24.2 43.46

−3.47 × 10−5 4.95 × 10−6 48.27

−5.66 × 10−4 8.44 × 10−5 8.27


the variance. The polarity parameter has the highest loading in the third factor illustrating that its information content considerably deviates from those of other polarity indices. The distribution of the points (stationary phases) on the plane of the two-dimensional map of component scores indicates that they do not form separate clusters according to the eight polarity indices. Furthermore, it can be concluded that stationary phases located at the opposite corner of the maps probably show different polarity indicating different selectivity [20]. PCA, correspondence factor analysis (CFA) and hierarchical ascending classification (HAC) were applied for the study of the retention behaviour of hydrocarbons on three individual columns (SE 30, SE 54, and a polyethylene glycol bonded phase) and on their combinations. The retention time and retention factor of 51 sample components were calculated for each column (A, B and C) and for each column combination (ABC, CBA, BAC and ACB). The data clearly show that the retention times of analytes show marked variations. As the length and ID of the columns were not identical the differences cannot be attributed to the discrepancy among the polarities of the stationary phases. The first principal component explained 97.3% of the total variance when the calculation was carried out on the matrix of retention times. In the case of the matrix of retention indices the first principal component explained 93.8%. This fact indicates the intrinsic similarity between the retention characteristics of columns and column combinations. The results of HAC of analytes carried out only on the individual columns are displayed in Figure 2.3. The calculations indicated that n-alkanes, alkanes, cyclanes and aromatics form different clusters suggesting different retention characteristics on stationary phase of different polarities [21]. Both PCA and MLR were employed for the prediction of the Kováts indices of 98 saturated esters on seven stationary phases of different polarities (SE-30, OV-7, DC-710, OV-25, 100% phenyl, DC-230 and DC-530). The calculated molecular descriptors were energy of the lowest unoccupied molecular orbitals (LUMO), dipole moment (DIP), the maximum of the net atomic charge on the C atom (QMAX), the sum of positive charge on C atoms (QTOT), topological indexes (CHI-2, CHI-0AV, CHI-2AV, CHI-0a, CHI-2A)

Gas Chromatography n-Propylbenzene 1,2-Diethylbenzene iso-Butylbenzene 1,4-Diethylbenzene Benzene Toluene Cumene 1,3,5-Trimethylbenzene p-Xylene o-Xylene 1,2,4-Trimethylbenzene Ethylbenzene m-Xylene Butylcyclopentane Ethylcyclohexane Butylcyclohexane 1,1,2-Trimethylcyclohexane tert-Butylcyclohexane Cyclohexane sec-Butylcyclohexane 2,2,3-Trimetylpentane 2,5-Dimethylhexane 2,4-Dimethylhexane 2,3,4-Trimetylpentane 3,3,5-Trimetylheptane 2,3,3-Trimetylpentane 2-Methyl-3-ethylpentane 2,3,-Dimethylhexane 2,2,-Dimethylheptane 2,2-Dimethylhexane 2,6-Dimethyloctane 3,3-Diethylpentane 2,2,4-Trimethylpentane 2,2,3,3-Tetramethylpentane *1,1,3-Trimethylcyclohexane 2,3,3,4-Tetramethylpentane 3-Ethylhexane 3,3,4-Trimethylhexane n-Nonane *2,2,3-Trimetylhexane n-Octane n-Decane n-Hexane n-Heptane 3-Mehtylheptane 3,4-Dimethylhexane 3-Methyl-3-ethylpentane 2,4-Dimethyl-3-ethylpentane 4-Methylheptane 2-Methylheptane *n-Undecane

21

aromatics

cyclanes

alkanes

n-alkanes

Figure 2.3 Hierarchical ascending classification tree of 51 test compounds according to their retention indices on three different chromatographic columns. For the measurement similarity, chi-square metrics was used. The clusters are linked by the minimal cluster variance criterion. Reprinted with permission from ref. [21]. Copyright Elsevier

22


Table 2.8

Result of the principal component analysis

CHI-OA CHI-2A CHI-0AV CHI-2AV WA QMAX QTOT LUMO DIP Explained variance Proportion of the total variance

Factor loading 1

Factor loading 2

Factor loading 3

Factor loading 4

Factor loading 5

−0.741 −0.783 0.744 0.735 0.894 0.431 0.878 0.510 −0.138 4.290

−0.419 0.274 −0.281 0.317 0.346 −0.810 0.241 −0.355 0.798 2.026

−0.208 −0.0532 −0.274 −0.107 −0.0251 0.177 −0.313 0.726 0.462 1.003

0.439 0.248 0.461 0.408 −0.150 −0.0644 −0.0395 0.212 0.252 0.770

−0.125 0.442 −0.238 0.350 0.132 0.285 −0.0930 −0.0245 −0.192 0.535

0.477

0.225

0.114

0.086

0.059

Reprinted with permission from ref. [22]. Copyright Elsevier.

and Wiener index (WA). The results of PCA are compiled in Table 2.8. It was established that six principal components explain 95.8% of the total variance. The distribution of factor loadings proves the similarities and dissimilarities between the calculated molecular parameters. The parameters of the linear regressions computed by MLR prove that MLR found highly significant correlations between the retention behaviour of saturated esters on each column and some of the molecular descriptors. It was stated that these equations can be applied for the prediction of the retention behaviour of other analytes too [22]. PCA and CA have also found application in the evaluation of the results of inverse gas chromatography (IGC). The polarity and interaction parameters of 21 stationary phases using 16 different test compounds were determined and the data matrix was evaluated by PCA. Stationary phases included in the PCA analysis are listed in Table 2.9. PCA established that the majority of variance inherent in the original data matrix can be explained by three factor loadings (36.2%, 18.9% and 14.0%). It can be concluded from the data that the majority of stationary phases form a well defined cluster on the maps of principal component loading indicating the grade of similarity and dissimilarity between the stationary phases [23]. Another IGC method was applied for the study of polymer–solvent and filler–solvent interactions using nine model analytes. Stationary phases used in the study are listed in Table 2.10. It was concluded from the results that the method is suitable for the classification of stationary phases and analytes [24]. A similar IGC method was employed in the study of the interaction between stationary phases and some volatile analytes. Stationary phases included in the investigation were polyethylene (PE), polyurethane (PU), silica, and silica surface modified by N -2-aminoethyl-3-aminopropyltrimethoxysilane, 3-aminopropyltriethoxysilane, 3-mercaptopropyltrimethoxysilane, and octyltriethoxysilane. The retention parameters of pentane, hexane, heptane, octane, nonane, dichloromethane, chloroform, carbon tetrachloride, and 1,2-dichloroethane were determined on each stationary phase and the data were

Gas Chromatography

23

Table 2.9 Abbreviations used for surface-active agents and for column fillings (support) Abbreviation W1k W1o W1k1o P1o W2k W2o W2k2o W3k W3o W3k3o W4k W4o W4k4o P4o P5o P6o P7o P7o1o P7o2o P7o5o P5o6o

Surface-active agents (stationary phase)

Support chromosorb

1-(2’-hydroxy-5’-methylphenyl)-butan-1-one 1-(2’-hydroxy-5’-methylphenyl)-butan-1-one oxime 1:1 mixture of W1k and W1o (w/w) 1-(2’-hydroxy-5’-methylphenyl)-butan-1-one oxime (2’-hydroxy-5’-t-octyl)-benzophenone (2’-hydroxy-5’-t-octyl)-benzophenone oxime 1:1 mixture of W2k and W2o (w/w) (2’-hydroxy-5’-t-octyl)-benzophenone (2’-hydroxy-5’-t-octyl)-benzophenone oxime 1:1 mixture of W3k and W3o (w/w) 1-(2’-hydroxy-5’-methylphenyl)-octan-1-one 1-(2’-hydroxy-5’-methylphenyl)-octan-1-one-oxime 1:1 mixture of W4k and W4o (w/w) 1-(2’-hydroxy-5’-methylphenyl)-octan-1-one (2’-hydroxy-5’-t-methyl)-acetophenone oxime (2’-hydroxy-5’-t-butyl)-benzophenone oxime (2’-hydroxy-5’-t-methyl)-benzophenone oxime 1:1 mixture of P7o and P1o (w/w) 1:1 mixture of P7o and P20 (w/w) 1:1 mixture of P7o and P5o(w/w) 1:1 mixture of P5o and P6o (w/w)

W W W P W W W W W W W W W P P P P P P P P

Reproduced from the Journal of Chromatographic Science by permission of Preston Publications, a division of Preston Industries Inc. Ref [23].

evaluated by PCA. The three-dimensional plot of PE and PU systems are depicted in Figure 2.4. The first principal component explains the overwhelming majority of variance indicating the strong similarity between the GC systems. It was stated that PCA facilitates the evaluation on the character of interaction between polymers and analytes, and it can be used for the reduction of the number of measurements necessary for such investigations [25]. MLR, PCA and CA were applied simultaneously for the study of the characteristics of different stationary phases. Model compounds were benzene, 1-butanol, 2-pentanone, 1-nitropropane pyridine, 2-methyl-2-pentanol, 1-iodobutane, 2-octyne, 1,4-dioxane, and cis-hydrindane. Analytes were characterized by the following calculated parameters: HOMO, LUMO, DIP, isotropic average polarizability, molecular volume, log P, solvent accessible surface area (SASA), polar solvent accessible surface area (pSASA), and apolar solvent accessible surface area (apSASA). Stationary phases tested were bis(2-butoxyethyl)adipate (BBA), bis(2-ethylhexyl)adipate (DAP), bis(2-butoxyethyl)phthalate (BBP), bis(2-ethylhexyl)phthalate (BEP), bis(2ethoxyethyl)phthalate (BIP), bis(2-ethoxyethoxyethyl)phthalate (BEEP), butyloctylphthalate (BOP), dicyclohexylphthalate (CIC), didecylphthalate (DDP), dinonylphthalate (DNP), bis(2-ethylhexyl)tetrachlorophthalate (DIOC2), bis(2-ethoxyethyl)sebacate (BES), bis(2-ethylhexyl)sebacate (DOS), dinonylsebacate (DNS), octyldecyladipate, N,N,N ,N tetrakis-(2-hydroxyethyl)-ethylenediamine (THEED), cresyldiphenyl phosphate (CDP), tributoxyethyl phosphate (TBP), tris(2-ethylhexyl) phosphate (TEHP), tricresyl phosphate

24


Table 2.10 Abbreviations used Abbreviation

Name

A B C D

Oligomer-polyether-urethane, PU Polyethylene, PE Unmodified silica Silica modified with 1 part of N-2-aminoethyl-3aminopropyltrimethoxysilane Silica modified with 2 parts of N-2-aminoethyl-3aminopropyltrimethoxysilane Silica modified with 3 parts of N-2-aminoethyl-3aminopropyltrimethoxysilane Silica modified with 5 parts of N-2-aminoethyl-3aminopropyltrimethoxysilane Silica modified with 10 parts of N-2-aminoethyl-3aminopropyltrimethoxysilane Silica modified with 1 part of 3-aminotriethoxysilane Silica modified with 2 parts of 3-aminotriethoxysilane Silica modified with 3 parts of 3-aminotriethoxysilane Silica modified with 5 parts of 3-aminotriethoxysilane Silica modified with 10 parts of 3-aminotriethoxysilane Silicamodified with 1 part of 3-mercaptopropyltrimethoxysilane Silica modified with 2 parts of 3-mercaptopropyltrimethoxysilane Silica modified with 3 parts of 3-mercaptopropyltrimethoxysilane Silica modified with 5 parts of 3-mercaptopropyltrimethoxysilane Silica modified with 10 parts of 3-mercaptopropyltrimethoxysilane Silica modified with 1 part of triethoxyoctylsilane Silica modified with 2 parts of triethoxyoctylsilane Silica modified with 3 parts of triethoxyoctylsilane Silica modified with 5 parts of triethoxyoctylsilane Silica modified with 10 parts of triethoxyoctylsilane 363 K 363 K 363 K 403 K

E F G H I J K L M N O P Q R S T U V W 1 2 3 4


(TCP), acetyltributyl citrate (AC), sorbitan monostearate (SOR), sorbitane monooleate (SORM), tetracyanoethylpentaerythritol (TCEPE), and diethylene glycol distearate (DGDS). The results of MLR calculations are compiled in Table 2.11. MLR found significant linear correlations among the stationary phases and the physicochemical parameters indicating that these parameters can be successfully used for the characterization of stationary phases. The results of CA are shown in Figure 2.5. The cluster dendograms clearly show that both descriptors and dependent variables form characteristic groups indicating that this method can also be used for the reduction of the number of analytes without considerable loss of information [26]. CA has also found application in the investigation of the similarities and dissimilarities among various stationary phases. The retention of cuminal, (±)linalol, (±)carvone, estragol, (+)pulegone, (+)citronellol, (+)citronellal, (−)menthol, c-citral, and (−)fenchone was

Gas Chromatography

25

5% 10 % 20 %

3 PC 3 (5.23 %)

2 1 0 −1 −2 −3 −4 5 4 3

PC

2

0.2

2 (1

1

%)

0 −1 −2 −3 −4 −2

0

−1

2 1 %) (79.4 1 C P

5

4

3

(a)

8 7 6 5 4 3 2 1 0 −1 −2

4

2

1 PC 2

1 (8

3

1.9

1 0 −1 −2 −3 −4 −3 −4 −5

%)

3 2

0 (8.7 −1 −2 7% )

PC

) PC 3 (6.86 %

5% 10 % 20 %

(b)

Figure 2.4 Scatter plot for (a) polyurethane and (b) polyethylene systems. Reprinted with permission from ref. [25]. Copyright Elsevier

26


Table 2.11 Results of MLR calculations

BBA DAP BBP BEP BIP BEEP BOF DIC DDP DNP DIOC2 BES DOS DNS ODA THEED CDP TBP TEHP TCP AC SOR SORM TCEPE DGDS

Intercept

LUMO

μ

V

SASA

R2

F

176.39 84.72 237.23 138.34 345.42 546.31 146.02 330.71 196.99 114.56 258.33 351.09 83.77 64.36 85.25 1187.80 540.78 374.60 192.79 405.8 179.14 323.83 334.52 1526.07 235.77

28.14 21.29 24.04 16.84 29.99 28.60 17.49 16.23 22.53 18.86 n.a. 35.00 19.05 20.43 20.32 86.51 n.a. 39.97 40.73 22.44 27.91 29.87 28.15 n.a. 21.12

36.19 26.68 39.72 31.09 48.43 51.31 32.18 37.08 39.55 31.67 9.73 46.96 25.54 26.05 26.50 91.93 24.60 53.62 45.99 43.48 38.75 30.41 30.99 n.a. 21.88

−5.12 −3.33 −4.60 −3.03 −5.85 −3.58 −3.17 −2.13 −4.31 −3.31 −1.59 −2.53 −2.97 −3.13 −3.34 −8.11 −7.09a −2.92 −1.85 −2.64 −5.04 −2.52 −2.52 −20.41a −1.85

1.55 1.05 1.21 0.83 1.47 n.a. 0.86 n.a. 1.20 0.98 n.a. n.a. 0.91 1.02 1.06 n.a. n.a. n.a. n.a. n.a. 1.50 n.a. n.a. n.a. n.a.

0.950 0.936 0.965 0.972 0.962 0.930 0.972 0.941 0.970 0.971 0.915 0.765 0.934 0.943 0.941 0.853 0.896 0.835 0.738 0.936 0.956 0.787 0.796 0.841 0.749

23.98 18.32 34.21 42.90 31.53 26.55 43.86 31.65 40.41 41.94 37.85 6.51 17.68 20.59 19.96 11.62 30.14 10.14 5.63 29.19 26.85 7.38 7.80 42.21 5.98

R 2 , square of correlation coefficient; F , Fischer number; n.a., not available. Descriptor: polarizability. Reprinted with permission from ref. [26]. Copyright Elsevier.

a

measured on 12 modified CD stationary phases. The CD phases were Chiraldex G-PH, APH, B-PH, G-DA, Beta-Dex 325, Chiraldex A-DA, Beta-Dex 225, Cyclodex-B, Cydex-B, Chiraldex-G-PB, G-PN, and B-DA. The cluster dendogram of CD stationary phases is shown in Figure 2.6. (CA adequately differentiates between the retention characteristics of stationary phases facilitating the selection of CD derivatives with highly divergent retention parameters.) It has been established that computations make possible the elucidation of the physicochemical character of the interaction between the ring structure of modified CDs and the analytes included in the investigation [27].

2.3

Elucidation of Similarities and Dissimilarities Among Samples

Natural products, such as human and animal tissues, blood and blood plasma, bacteria, fungi, various plant organs, and oils, contain a considerable number of organic and inorganic compounds. Consequently, the chromatograms of natural products are generally very complicated and contain a considerable number of peaks. The comparison of

Gas Chromatography

27

Ward’s method Euclidean distances 6

Linkage Distance

5

4

3

2

AC

BBA

BEP

DNP

DDP

BIP

BBP

BEEP

TCP

CDP

DIC

BOF

DIOC2

DAP

TCEPE

ODA

DOS

TBP

DNS

BES

TEHP

SOR

THEED

SORM

0

DGDS

1

Figure 2.5 Results of CA (Ward’s method) for the dependent variables. Reprinted with permission from ref. [26]. Copyright Elsevier

Similarity

−89.82

−28.64

36.73

100.00 2

3

11

12

1

8 7 4 Observations

5

10

6

9

Figure 2.6 Hierarchical CA (standardized variables, Euclidean distance, and Ward’s clustering method) of the relative retention data. Observations displayed (left to right) from the CD phases: Chiraldex G-PH (2), A-PH (3), B-PH (11), and G-DA (12), Beta-Dex 325 (1), Chiraldex A-DA (8), Beta-Dex 225 (7), Cyclodex-B (4), Cydex-B (5), Chiraldex G-BP (10), G-PN (6), and B-DA (9). Reproduced from the Journal of Chromatographic Science by permission of Preston Publications, a division of Preston Industries Inc. Ref. [27]

28


such chromatograms is practically impossible with traditional visual methods. Multivariate mathematical-statistical methods offer a unique possibility for the elucidation of the similarities and dissimilarities among the chromatograms facilitating sample classification. 2.3.1

Human Health and Pharmaceuticals

MLR, PLS and PCA have been simultaneously employed for the study of the relationship between the retention of 35 trimethylsilyl derivatives of α-, ß1 -, and ß2 -agonists. The analytes are compiled in Table 2.12. Altogether 143 descriptors were calculated and included Table 2.12 Trimethylsilyl (TMS) derivatives of α-, and β1 -, and β2 -agonists identified by GC-MS during the study No.

Analyte

Derivative

Mra

RT

RRTb

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35

Bambuterol Bromobuterol Bromobuterol Cimbuterol Clenbuterol Clenbuterol Dobutamine Dobutamine Fenoterol Fenoterol Isoetharine Isoproterenol Isoproterenol Isoxsuprine Metaproterenol Metaproterenol Methoxamine Midodrine Midodrine Midodrine Oxymetazoline Phenylephrine Pholedrine Procaterol Reproterol Reproterol Reproterol metab. Ritodrine Ritodrine Salbutamol Salmeterol Salmeterol Terbutaline Tulobuterol Penbutolol

mono-TMS mono-TMS bis-TMS bis-TMS mono-TMS bis-TMS tris-TMS tetrakis-TMS tetrakis-TMS pentakis-TMS tris-TMS tris-TMS tetrakis-TMS bis-TMS tris-TMS tetrakis-TMS bis-TMS tris-TMS tris-TMS tetrakis-TMS bis-TMS tris-TMS bis-TMS tris-TMS tris-TMS tetrakis-TMS tris-TMS tris-TMS tetrakis-TMS tris-TMS tris-TMS tetrakis-TMS tris-TMS mono-TMS mono-TMS

439 438 510 377 349 421 517 589 591 663 455 427 499 445 427 499 355 470 470 542 404 382 309 506 605 677 617 503 575 455 631 703 441 299 363

7.71 5.58 6.16 5.39 4.71 5.44 8.78 9.46 8.44 9.02 4.77 4.50 5.51 7.17 4.59 5.48 3.76 6.35 7.41 7.16 7.11 3.78 3.78 6.70 10.99 10.94 11.47 8.01 8.63 5.09 10.94 11.08 4.71 2.65 5.61

1.37 0.99 1.10 0.96 0.84 0.97 1.56 1.68 1.50 1.60 0.85 0.80 0.98 1.28 0.82 0.98 0.67 1.13 1.32 1.27 1.27 0.67 0.63 1.20 1.96 1.95 2.04 1.43 1.54 0.91 1.94 1.97 0.84 0.47 1.00

a

Relative molecular mass. Relative retention time to penbutolol mono-TMS. Reprinted with permission from ref. [28]. Copyright Elsevier. b

Gas Chromatography

29

Table 2.13 Best regression model found by the application of the MLR method to the second descriptors group Descriptor 3χ p υ 7χ p S1 S2 10χ p Constant

Coefficient

Standard

Descriptor explanation

0.255 −0.0874 −0.127 −0.0696 −0.193 −0.310

0.017 0.014 0.024 0.025 0.096 0.100

Connectivity index third order Connectivity index seventh order Electrotopological index of atom in position Electrotopological index of atom in position Connectivity index tenth order (1.4) Y-intercept


in the computations. The best regression models found by the application of topological and electrotopological descriptors (second descriptors group) are shown in Tables 2.13 and 2.14. The results of MLR calculations proved that the descriptors selected exert a significant influence on the retention of this class of agonist and the equations can be used for the prediction of the retention of other similar compounds. The plot of first and second principal components is shown in Figure 2.7. It was found that the first two principal components accounted for 60% of the total variance present in the original data matrix. It was further established that the presence of a benzene ring, the length of side chain and the number of trimethylsilyl groups in the molecule exert the highest impact on the retention [28]. MLR was used for the determination of the factors influencing the concentration of urinary and salivary cotinine measured by GC, HPLC and enzyme-linked immunosorbent assay (ELISA). The results of MLR are compiled in Table 2.15. It has been established that the cotinine concentration significantly depended on smoking [29]. The relationship between the retention of 37 pharmaceuticals influencing the central nervous system (CNS) and their 261 calculated descriptors was also investigated by MLR. The best fitting equations are compiled in Table 2.16. The results indicated that the reversed form of (Rev ), the maximal electrotopological positive variation (MaxDP), the relative number of hydrogen atoms (RelnH), the maximum partial charge for an oxygen atom (MaxQO) and the reversed form of - (RevDif ) exert a significant impact on the retention. Equations were proposed for the prediction of retention behaviour of other CNS agents [30]. MLR calculations were further employed for the determination of the impact of incubation temperature and time on the recovery of barbiturate derivatives and for the optimization of the derivatization process [31]. Table 2.14 PLS statistics of the derived PLS model for the second descriptors group A

Eig.

R2Y (cum)

Q2Y (cum)

N tr

RMSEtr

N pr

RMSEpr

1 2

16.812 2.806

0.736 0.906

0.717 0.868

24

0.137

7

0.0123

A, PLS component; Eig., component eigenvalue; R2Y(cum), cumulative sum of squares of the Ys (RRT) explained by all extracted components; Q2Y(cum), cumulative fraction of the total variation of the Ys that can be predicted by all extracted components; N tr , number of compounds in the training set; RMSE tr , root-mean-square error for the dependent variable of the training set; N pr , number of compounds in the prediction set; RMSE pr , root-mean-square error for the dependent variable of the prediction set. Reprinted with permission from ref. [28]. Copyright Elsevier.

30

Multivariate Methods in Chromatography: A Practical Guide 10 8 6

10 32

PC2 (60.8%)

4 2 0 −2 −4

9 13 8 20 16 30 33 29 31 7 28 11 25 22 19 4 36 23 18 17 14 34 21 2 5 24

−6

1

−8

27

−10

26

−12

25 −20

−10

0 PC1 (40.3%)

10

20

Figure 2.7 First two components (PC1 versus PC2) from the PCA of the 34 investigated analytes, with the percentage of variance expressed in parentheses. Reprinted with permission from ref. [28]. Copyright Elsevier

PCA has also found application in the evaluation of chromatographic data referring to human health. Thus, PCA was employed in the study of fingerprint characteristics of the emanations from human arm skin using the original sampling system by SPME-GC/MS. The plot of PC1 versus PC2 indicated that the arm skin emanations markedly differ in spring and winter (Figure 2.8) [32]. Table 2.15 MLR models to show factors affecting concentrations of urinary and salivary cotinine for the three measurements Variable

HPLC-urine β (SE)

Gender (female=o) 171.1(300.0) Age (years) Smoking −12.3(14.1) (pack-years) 0–1 (non-smoker = 0) 1183.9(423.4)∗ 1–5 (non-smoker = 0) 2603.2(383.9)∗ >5 (non-smoker = 0) 2059.4(527.1)∗ Smoking at home 184.9(255.2) (No = 0) Drug usage −15.3(497.0) (No = 0) Disease history −204.4(348.6) (No = 0) R2 0.49 P -value < 0.01

ELISA-urine β(SE) 185.0(351.2) −21.2(16.5) 1494.8(495.7)∗ 2822.7(449.6)∗ 2058.3(517.1)∗ 294.2(298.8)

ELISA-saliva β(SE) 0.1(2.4) −0.2(0.1) 12.4(4.1)∗ 9.2(3.0)∗ 14.7(3.3)∗ 0.3(2.0)

GC-NDP-urine β(SE) 162.7(325.7) −18.4(15.3) 1119.1(459.6)∗ 3318.5(416.8)∗ 2650.2(572.2)∗ 78.4(277.0)

8.5(581.9)

−1.2(3.9)

126.1(539.6)

−347.6(408.1)

−2.4(2.7)

−179.1(378.5)

0.46 < 0.01

*P < 0.01. Reprinted with permission from ref. [29]. Copyright Elsevier.

0.34 < 0.001

0.55 < 0.01

Gas Chromatography Table 2.16

31

Summary of the best regression models found for the data set

Descriptors RevDif MaxDP RelnH MaxQO Revχ Intercept Statistical parameters R F SE n

DB-5

Mean effect

8.392 −79.799 −1.571 × 103 1.310 × 103 −2.204 × 104 3.637 × 103

0.124 −0.230 −0.303 −0.128 −0.701

DB-17

Mean effect

13.310 −127.074 −3.378 × 103 −1.538 × 103 −2.244 × 104 4.809 × 103

0.199 −0.366 −0.487 −0.153 −0.624

0.988 253.89 18.8 37

Reprinted with permission from ref. [30]. Copyright Wiley-VCH

0.5 A

0.4 0.3

PC3

0.2 0.1 0.0 −0.1 −0.2 −0.3 0.6

B

0.8 0.6

0.4

0.2

0.4 0 PC2

PC1 −0.2

−0.4

0.2 −0.6

−0.8

0

Figure 2.8 Pattern recognition based on PCA. The rectangular area A represents the segregation of the spring sampling data cluster. The rectangular area B represents the segregation of the winter sampling data cluster. Reprinted with permission from ref. [32]. Copyright Elsevier

32

Multivariate Methods in Chromatography: A Practical Guide 1.0x104

o oo o o oo o

6

5.0x103

++ 0.0

2x10

o oo

+ + ++

o ++

oo

o

−5.0x103

Principal Component 2


o

1x106

0

−1.0x104

o −1.5x104 −1.0x104 −5.0x103 0.0

o

5.0x103 1.0x104

+

−1x10

6

+ ++ +

+

+

−1.0x107 −5.0x106

0.0

5.0x106



(a)

(b)

Figure 2.9 Principal component analysis of total ion chromatograms from GC-MS analysis of headspace above plasma samples from two donors. Scores plot showing the separation of the data along the first two principal components, with Donor A (+) and Donor B (o), (a) before and (b) after alignment. Reprinted with permission from ref. [33]. Copyright Elsevier

Volatile organic compounds (VOCs) were determined in blood and urine samples using the SPME prepurification step and the samples were compared by PCA. The plot of PC1 versus PC2 is shown in Figure 2.9. It was established that the alignment of raw data facilitate the evaluation of chromatograms containing complex mixtures [33]. The volatiles of an extract of the medicinal plant Eurycoma longifolia were analysed by cGC-MS and using a quartz crystal microbalance (QCM). Sensors were prepared of traditional GC stationary phases, namely Apiezon L (APZ-L), polypropylene glycol 1200 (PPG1200), polyethylene glycol 1000 (PEG 1000), polyethylene glycol 4000 (PEG 4000), ethyl cellulose (EC), dioctyl phosphate (DOP), trioctyl methylammonium chloride (TOMA), and oleylamine (OAm). The results of the two methods were compared by PCA. The factor loadings of QCM and GC-MS data are compiled in Tables 2.17 and 2.18. It was found that both GC and QCM are suitable for the differentiation of extracts of spray dried and freeze dried plant material as demonstrated in Figure 2.10 [34]. 2.3.2

Forensic Analyses

Various multivariate mathematical-statistical evaluation methods have found application in forensic analysis. Although the quantitative determination of the mean component does not require complicated statistical procedures, the comparison of the accompanying matrix of pollutions may facilitate the determination of the origin of the product (authenticity test). Interestingly, the majority of studies use PCA for data evaluation. The classification of illicit heroin samples has been carried out using PCA and artificial neural networks (ANNs). It was established that PCA is suitable for the grouping of samples on the basis of their chromatographic profile [35]. Similar results have been achieved by the analysis of illicit heroin samples (undiluted and diluted) as shown in Table 2.19. GC-MS


33

Factor loadings of the varimax rotated PCA of QCM data Principal component

Sensor material APZ-L PPG-1200 PEG-1000 PEG-4000 EC DOP TOMA OAm

1

2

3

4

0.767 0.566 0.538 0.701

0.356 0.530 0.555 0.529

0.432

0.453 0.844 0.829

0.851 0.930 0.398

5 0.542

0.564

0.317 0.971


Table 2.18 Factor loadings of the varimax rotated PCA analysis of GC-MS data Principal component Volatiles Acetic acid Curcumene (R)-(−)-massoilactone 3-Phenoxy-1-propanol Octanoic acid 2-Phenoxy-ethanol 4-Ethynyl-4-hydroxy-3,5,5Trimethyl-2-cyclohex-1-enone Benzoic acid

1

2

0.325 0.711

0.923 0.349 0.626 0.575

0.942

3

4

5

6

0.983 0.848 0.712

0.836

0.315

−0.367 0.313

0.304

0.977


2.0 1.5 1.0

PC 2

.5 0.0 −.5 −1.0 −1.5 −2.0 −2.5 −2.0 −1.5 −1.0 −.5 PC 1

0.0

.5

1.0

1.5

Figure 2.10 PC1 versus PC2 plot of QCM analysis of spray dried () and freeze dried () extracts. Reprinted with permission from ref. [34]. Copyright Elsevier

34


Table 2.19 Samples prepared from batches A, B and C Sample label

Batch A

Batch B

Batch C

Cutting agent

Sample

Cutting agent

Sample

B2 B5

Procaine Citric acid

C3 C6

Citric acid Manitol, caffeine

A7

Sucrose Procaine, caffeine, glucose Paracetamol

B8

Procaine

C9

A10 A13 A16 A19 A22

— — Caffeine Procaine Paracetamol

B11 B14 B17 B20 B23

— — Caffeine Procaine Paracetamol

C12 C15 C18 C21 C24

Procaine, caffeine, glucose — — Caffeine Procaine Paracetamol

A1 A4

Cutting agent

Sample


results were classified by PCA, hierarchical CA and the k-nearest neighbour (k-NN) method. A cluster dendogram is depicted in Figure 2.11 and the distribution of samples in threedimensional PCA space is shown in Figure 2.12. The comparative values of the three methods are compiled in Table 2.20. The data unambiguously proved that each method is suitable for the classification of illicit heroin samples [36].

B8 B2 B11 B5 B14 B17 B23 B20 C6 C9 C3 C18 C24 C15 C12 C21 A1 A4 A7 A10 A13 A16 A19 A22

GROUP B

GROUP C

GROUP A

0

5

10

15

20

Distance

Figure 2.11 Dendogram showing the result of hierarchical CA. Reprinted with permission from ref. [36]. Copyright Elsevier

Gas Chromatography

35

GROUP C GROUP B

PC1-48.81

PC3-13.45 sample C3

PC2 24.66

GROUP A

Figure 2.12 Separation of samples according to their batch in three-dimensional (PC1, PC2, and PC3) measurement space. Reprinted with permission from ref. [36]. Copyright Elsevier

A predictive model was developed and successfully applied for the classification of street samples of heroin. It was found that linear discriminant analysis (LDA) has a higher separation power than CA [37]. PCA and related techniques have also been employed for the evaluation of drug preparations containing amphetamine and amphetamine derivatives. The GC-FID method separated 50 impurity peaks including p-bromotoluene, N -benzyl amphetamine, N -ethyl amphetamine, N -ethyl methamphetamine, phenyl-2-propanone, N ,N -dimethyl amphetamine, and N -formyl amphetamine. CA dendograms demonstrate that a minimum of 40 impurity peaks are needed for the safe classification of samples. Furthermore, the results prove that CA is a reliable method for the discrimination between drug preparations [38]. Similar GC-FID techniques have been employed for the separation of the components of methamphetamine tablets. The following volatile constituents were identified: 1,2-dimethyl-3-phenylaziridine, ephedrine, methylephedrine, N -acetylmethamphetamine, N -formylephedrine, N -acetylephedrine, N ,O-diacetylephedrine, methamphetamine Table 2.20 Classification results obtained by hierarchical CA, PCA and the k-NN method Method

Number of misclassified samples

Classification rate (%)

0 1 0 1a 1

100 95,8 100 95,8 95,8

Hierarchical CA PCA (two dimensions) PCA (three dimensions) k-NN (k = 3, averaging) k-NN (k = 3, majority rule) Reprinted with permission from ref. [36]. Copyright Elsevier.

36


dimer, caffeine, ethyl vanillin, acetylcodeine, monoacetylmorphine, diacetylmorphine, and trans-3,4-diethyl-5-phenyl-2-oxazolidine. It has been stated that the combination of CA with GC-FID is a very valuable tool for the comparison of illicit methamphetamine tablets [39]. The 15 N/14 N isotopic ratio in seized Ecstasy tablets was determined by GC-combustionisotope ratio mass spectrometry (GC-C-IRMS). It was established that this method can be applied for the comparison of the origin of Ecstasy tablets [40]. GC-FTIR (Fourier transform infrared) spectroscopy combined with PCA has also been employed in the analysis of drugs of abuse. The application parameters of the method are presented elsewhere [41,42]. 2.3.3

Biology and Agrobiology

A considerable number of investigations have been carried out to determine the possibility of applying multivariate mathematical-statistical methods for the evaluation of the chromatographic behaviour of samples of biological and agrobiological origin. A new semi-empirical optimum topological index (IET(opt) ) was developed and successfully employed for the prediction of the retention of methyl-branched alkanes produced by insects. This topological index takes into consideration the length of alkyl chain, the position of the methyl groups, and the number of methyl groups attached to the alkyl backbone: IET(opt) = IET − 1/3 (log n 1 + 1) + (log n 2 + 8) + (log n 3 + 24) + (log n 4 + 14) (2.9) where IET is the semi-empirical topological index for alkanes, and n i are the positions of connection of methyl groups to the carbon backbone. The relationship between the retention of analytes and the new topological index was calculated by MLR. A significant linear correlation was found between the experimental retention indexes (RIExp ) and the optimum semi-empirical topological indexes: RI Exp = −39,5251 + 123,161610 × IET(opt) n = 178; r = 0.99998; r 2 = 0.99994; SD = 4.31

(2.10)

It was stated that the strong correlation between dependent and independent variables makes the prediction of other methyl-branched alkanes under similar GC conditions possible [43]. Pyrolysis-gas chromatography (Py-GC) has been employed for the classification of Eucalyptus camaldulensis grown from seeds of various origins. The chromatograms were compared by PCA taking into consideration each component detected (altogether 26 molecules). The first and second principal components explained 33.1 and 27.4% of the total variance (see Figure 2.13). The plot in Figure 2.13 clearly indicates that the method discriminates the Eucaliptus trees well according to their origin, so it can therefore be used for their authenticity test [44]. GC and HPLC have been simultaneously employed for the determination of the composition of the vascular tissue of Douglas fir (Pseudotsuga menziesii). GC-FID was employed for the separation and quantitative analysis of 26 different terpenoids, whereas the content of fructose, glucose, sucrose and the phenolic glycoside coniferin was measured by

Gas Chromatography

37

6 M-2 2nd Principal Component

4

K-2

2 W-1

M-1 P-2

0

M-3 −2

P-3

W-3 K-1

P-1

−4 W-2 −6 −6

−4

−2 0 2 4 1st Principal Component

6

Figure 2.13 Score plot of the first and second principal components for Eucaliptus trees. P, Petford; M, Murchison river; W, Wrotham Park; K, Katherine river. Reprinted with permission from ref. [44]. Copyright Elsevier

anion-exchange HPLC. The clusters generated from the concentration data of terpenoids are shown in Table 2.21. It was suggested that the method can facilitate the selection of Douglas fir less preferred by black bears [45]. Fatty acid composition of spirochaetal isolates was determined by cGC after methylation. The similarities and differences among the isolates were elucidated by CA. The cluster dendogram of the results is shown in Figure 2.14. The combination of cGC analysis with CA has been proposed as a useful tool for the identification of spirochetes [46]. Table 2.21 Clusters generated from quantitative terpenoid data Cluster 1

Cluster 2

Cluster 3

Cluster 4

Cluster 5

α-Thujene Sabinene α-Phellandrene 3-Carene α-Terpinene p-Cymene γ -Terpinene Terpinolene Linalool Terpinen-4-ol Citronellal Citronellyl acetate

α-Pinene Camohene β-Pinene Myrcene Limonene Bornyl acetate Geranyl acetate Longifolene

Glucose Fructose Surcose Coniferin

Fenchyl alcohol α-Terpineol Burneol

Caryophyllene Humulene


38

Multivariate Methods in Chromatography: A Practical Guide BR 135 BR 134 BR 116 BR 173 BR 231 BR 208 BR 151 BR 132 BR 92 BR 122 BR 14 VS 461T BR 194 20047T B31T VS 116T BR 91 0

5

10 15 20 25 30 35 40 45

Euclidian Distance

Figure 2.14 Dendogram demonstrating similarity (Euclidean distance) of fatty acid methyl ester analysis of all strains tested. BR 135, BR 134, BR 116, BR 173, BR 231, BR 208, BR 151; non-identified spirochetes; Br 132, BR 92, BR 122, Br 14, 20047T ; B. garinii; VS 461T : B. afzelii; BR 194, B 31T : B. burgdorferi s.s.; BR 91: ‘Spironema culicis’. Reprinted with permission from ref. [46]. Copyright Elsevier

Analysis of fatty acid methyl esters by cGC-MS and phospholipid determination by LC-ESI-MS were employed for the identification of microbes. The cGC and HPLC data were separately calculated by FA. The plots clearly illustrate that the phospholipid profile has a higher discriminative power than fatty acid analysis [47]. GC has also been used for the separation and qualitative determination of n-alkanes, longchain fatty alcohols and long-chain fatty acids in various plant species. The three data sets were treated separately and the plants were classified by PCA. The two-dimensional plot of the principal components is shown in Figure 2.15. The distribution of plant species on the two-dimensional plot shows marked differences indicating that these markers can be used for the evaluation of the botanical composition of the diet of free-ranging herbivores [48]. Various fungi (edible, poisonous and wood decaying) were also analysed by Py-GC, and by thermally assisted hydrolysis and methylation (THM). The fungi investigated were Amanita citrina, A. phalloides, Amanita muscaria, Amanita pantherina, A. campester, Fomes fomentarius, Trametes versicolor, Gloeophyllum odoratum, Gloeophyllum sepiarium, Serpula lacrymans, Cortinarius violaceus, Lactarius necator, Austroboletus gracilis, Boletus retipes and Spongiporus caesius. The compounds detected in fungi by pyrolysis and by THM are compiled in Tables 2.22 and 2.23. The characteristic chromatograms of fungi were compared by PCA. The comparison of the results of PCA carried out on pyrolysis and THM data sets indicated that THM has a higher discriminating capacity than analytical pyrolysis. It was stated that this combined method is suitable for the identification of fungi [49].

Gas Chromatography

39

4 Stylosan +

Chloris + Eragrost D.acgyp 1 2 + +

Digera m + +Epomea c C.cilia 2 + −2 D.aegyp2 Hordeum + +

−4

Heliotro +

0 0

Xanthium Polycarp C.cilia 1 ++Phaseolu + A. seneg +Ocimum b + C. biflo +

Achyrant + Abutilon + −2

Vigna sp + 2 Aristolo +

Balanite + Celosia A. melli + +

Ziziphus +

Figure 2.15 Two-dimensional plot of the first two principal components (PC1 and PC2) from PCA based on normalized alcohol concentrations conducted on the set of plant species studied. C. cilia1, Cenchrus ciliaris at flowering stage; C. cilia2, Cenchrus ciliaris at maturity; D. aegyp1, Dactyloctenium aegyptum at flowering stage; D. aegyp2, Dactyloctenium aegyptum at maturity. Reprinted with permission from ref. [48]. Copyright Elsevier

The total amount of lipids and fatty acid composition of depot fat were measured in 58 immature green turtles (Chelonia mydas) and the data on fatty acids were evaluated by PCA. It was found that the foraging area influences the fatty acid composition of green turtles, and PCA can be employed for the classification of green turtles according to their foraging area [50]. The monoterpene profile (altogether 18 monoterpenes in four species of conifers) was measured by GC-MS and the chromatograms were compared by PCA. Douglasfir [Pseudotsuga menziesii (Mirb.) Franco], lodgepole pine (Pinus contorta var. latifolia Engelm), interior spruce (Picea engelmannii × glauca) and interior fir (Abies lasiocarpa × bifolia) were included in the investigation. PCA indicated that the terpene profile is different according to the geographical origin [51]. The determination of the fatty acid profile in heart and gill tissues and in skull oil was carried out by cGC-FID and used for the discrimination of three redfish species (Sebastes viviparus, S. marinus and S. Mentella). The data matrix was evaluated by PCA. PCA was employed separately for the fatty acid composition measured in heart and gill tissues and

40

Multivariate Methods in Chromatography: A Practical Guide Table 2.22

Main products from the pyrolysis of fungi

Number 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19

Compound Hydroxyacetone Acetic acid Pyruvic acid methyl ester 2-Furfural Pyrrole Furfuryl alcohol Acetamide 2-Hydroxy-3-methyl-2-cyclopentenone 2,5-Dimethyl-4-hydroxy-3(2H)-furanone Anhydromannitol Glycerol 1,4:3,6-Dianhydroglucopyranose 1H-indene 16:0 Fatty acid 16:1 Fatty acid 18:0 Fatty acid 18:1 Fatty acid 18:2 Fatty acid Levoglucosan (1,6-anhydroglucopyranose)


Table 2.23 Products from the THM analysis of fungi, which were used for PCA plots Number 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37

Compound 3-Deoxy-2,4,5-tri-O-methylribonic acid methyl ester 3-Deoxy-2,4,5-tri-O-methylarabinonic acid methyl ester 3-Deoxy-2,4,5,6-tetra-O-methylgluconic acid methyl ester 3-Deoxy-2,4,5,6-tetra-O-methylgalactonic acid methyl ester 1,2,4-Trimethoxybenzene 3-Deoxy-2,4,5,6-tetra-O-methylmannonic acid methyl ester 3-Deoxy-2,4,5,6-tetra-O-methyltalonic acid methyl ester 3-Deoxy-2-methoxymethyl-2,4,5-tri-O-methylribonic acid methyl ester 3-Deoxy-2-methoxymethyl-2,4,5-tri-O-methylarabinonic acid methyl ester 16:0 Fatty acid methyl ester 16:1 Fatty acid methyl ester 18:0 Fatty acid methyl ester 18:1 Fatty acid methyl ester 18:2 Fatty acid methyl ester 3-(3,4-Dimethoxyphenyl)-2-propenoic acid methyl ester Methylated glucoside Methylated carbohydrate Isomer of 36


41

PC2 (21%)

PC2 (14%)

Gas Chromatography

PC1 (60%)

PC1 (35%) (b)

PC2 (28%)

(a)

PC1 (54%) (c)

Figure 2.16 (a) Principal component plot of three redfish species ( S. viviparus, S. marinus and S. mentella) based on the fatty acid composition in the heart tissues. Each symbol represents one individual fish. The percentage of the total variance along each of the principal components is given. (b) Principal component plot of two redfish species ( S. marinus and S. mentella) based on the fatty acid composition in the heart tissues. Each symbol represents one individual fish. The percentage of the total variance along each of the principal components is given. (c) Principal component plot of two redfish species ( S. marinus and S. mentella) based on the fatty acid composition in the heart tissues. Each symbol represents one individual fish. The percentage of the total variance along each of the principal components is given. Reprinted with permission from ref. [52]. Copyright Elsevier

skull oil. The plot of the first versus second principal components is depicted in Figure 2.16. The data points are well separated according to the species and tissues illustrating that the determination of the fatty acid profile can facilitate the discrimination of these three species [52]. A slightly different procedure has been employed for the investigation of the fatty acid profile of the species mentioned in Pureswaran et al. [51]. The measurement of the otolith oil composition was also performed. The results obtained by PCA of the different tissues are shown in Figure 2.17. The conclusions were identical with those discussed in Joensen and Grahl-Nielsen [52] emphasizing again that fatty acid profiles differ according to the species and tissues, and can be applied for the classification of species [53].

PC2 (18.4%)


PC2 (13.0%)

42

PC1 (50.0%)

PC1 (27.1%) (b)

PC2 (28%)

(a)

PC1 (51.8%) (c)

Figure 2.17 (a) Principal component plot of three redfish species ( S. viviparus, S. marinus and S. Mentella) based on the fatty acid composition in the heart tissues. Each symbol represents one individual fish. The percentage of the total variance along each of the principal components is given. (b) Principal component plot of three redfish species ( S. viviparus, S. marinus and S. mentella) based on the composition of the 11 most discriminating (P < 0.01) fatty acids in the heart tissues. Each symbol represents one individual fish. The percentage of the total variance along each of the principal components is given. (c) Principal component plot of two redfish species ( S. marinus and S. mentella) based on the composition of the eight most discriminating (P < 0.01) fatty acids in the heart tissues. Each symbol represents one individual fish. The percentage of the total variance along each of the principal components is given. Reprinted with permission from ref. [53]. Copyright Elsevier

PCA and CA have been simultaneously employed for the computation of the similarities between four pinus species according to their monoterpene composition. Measurements were carried out on a nonpolar and chiral phase GC column coupled to a MS detector. The species investigated were Pinus pinaster, P. silvestris, P. pinea, and P. halepensis. The CA dendogram taking into consideration only the nonchiral volatile concentrations is shown in Figure 2.18. The dendogram illustrates that the species cannot be entirely separated on the basis of the monoterpene content of their phloem. The two-dimensional plot of PC1 versus PC2 of the enantiomeric analyses is depicted in Figure 2.19. The authors stated that the combined method applied is suitable for the differentiation of Pinus species [54].

Gas Chromatography

43

25

20

15

10

5

P. pinea 1 P. pinea 2 P. pinea 3 P. pinea 4 P. pinea 5 P. pinaster 1 P. pinaster 2 P. syfvestris 1 P. syfvestris 2 P. syfvestris 3 P. syfvestris 4 P. syfvestris 5 P. halepensis 1 P. halepensis 2 P. halepensis 3 P. halepensis 4 P. halepensis 5 P. halepensis 6 P. halepensis 7

0

Figure 2.18 Dendogram using Ward’s method for hierarchical CA. Pine species were grouped using nonchiral volatile concentrations only. Reprinted with permission from ref. [54]. Copyright Elsevier

Direct thermal desorption (DTD) combined with GC-MS has been applied for the separation and quantitative determination of the volatiles in the leaves and flowers of Origanum vulgare. The data show that the amount and composition of volatiles display high variation among the individual samples and leaves and flowers. PCA indicated that the first principal component contains the overwhelming majority of variance (93.2%) which is related to linalool (40.5). The plots of other principal components are given in Figure 2.20. It was suggested that the procedure can be applied for chemotaxonomical purposes [55]. GC and other physicochemical methods were applied for the analysis of water- and dewretted flax fibres and the structure of the data set was studied by PCA. The concentration of rhamnose, arabinose, xylose, mannose, galactose and glucose were measured as alditol acetates. The variance explained was 43 and 16%, respectively. It has been suggested that similar investigations may contribute to the elucidation of the basic differences in the two retting methods [56]. Two-dimensional GC with time-of-flight MS (GCxGC-TOFMS) was applied for the analysis of metabolites in plant samples (basil, Ocimum basilicum; peppermint, Mentha piperita; and stevia, Stevia rebaudiana), and the data were evaluated by PCA. plots of PC1 versus PC2 are shown in Figure 2.21. The plots illustrate that the two first principal components explain the overwhelming majority of variance. It was further established that this computation method separates the plant species well, so that it can therefore be used in authenticity tests [57].

44


2.00000

2.00000

1.00000

1.00000

C3

3.00000

C2

3.00000

0.00000

0.00000

−1.00000

−1.00000

−2.00000

−2.00000

−1.00000

0.00000

1.00000 C1

−1.00000

2.00000

0.00000

1.00000 C1

2.00000

(b)

(a) 3.00000

2.00000

C3

1.00000

0.00000

−1.00000

−2.00000 −2.00000 −1.00000 0.00000 1.00000 2.00000 2.00000 C2

(c)

Figure 2.19 Scatter plots of the principal component analysis performed using the enantiomeric fraction of Pinus spp. phloems. Three components were extracted: (a) factor 1 versus factor 2; (b) factor 1 versus factor 3; (c) factor 2 versus factor 3. () P. sylvestris; (◦) P. halepensis; () P. pinaster; (×) P. pinea. Reprinted with permission from ref. [54]. Copyright Elsevier

Gas Chromatography

Third componnt

2

13 11 34 12 30 24 14 8 3326 32 3121 27 6 9 23 22 20 17 35 28 25 29 36 1

1 0 −1

37

−2 −2

3 18 5 192

16

−1

45

4

10 137

1 0 Second component

2

3

(a) 2.0 1.5

Third componnt

1.0 0.5 0.0 −0.5

37

−1.0

36

−1.5 −2.0 −2.5

16 2

33 24 2731 21 25 37 32 38 29 23 20 22 28

184

9

15 17 8 6

137 25 11 1 35 14

5 3

19

10

12 −1.5

−1.0

0.5 0.5 0.0 1.0 Second component

1.5

2.5

(b)

Figure 2.20 Principal component plot of Origanum vulgare samples (original concentration values). (a) Leaf samples; (b) flower samples. Reprinted with permission from ref. [55]. Copyright Elsevier

Standard samples of humic acids (NRFA and NRHA) as well as humic substances isolated from Lake Islienas, and ground water were transesterified and the volatile products were separated by GC-MS and GC-FID. The data set was evaluated by PCA. The plot of PC1 versus PC2 is shown in Figure 2.22. Samples of humic substances form clear-cut distinct clusters on the plot suggesting that the combined GC-PCA procedure is suitable for the classification of humic substances of different origin [58]. Various wood tissues from transgenic poplar clones and control samples were investigated by analytical pyrolysis GC-MS, and the results were evaluated by PCA. The characteristics of the samples under investigation are listed in Table 2.24. The method separated more than one hundred components which were either from carbohydrate- or lignin-derived pyrolysis. Some results of PCA including all pyrolysis products are listed in Figure 2.23. It was stated that the best separation among transgenic poplar clones and control samples can be achieved by including all pyrolysis products in the PCA and using the plot of PC1 versus PC3 [59].

46


0.01 Scores on PC2

Peppermint Stevia

0 Basil

−0.01

−0.01

0

0.01

0.02

Scores on PC 1 (a) 1.5

Eigenvalue (10.4)

PC 1

1.0

0.5

0

PC 2

0

5

10

15

20

Principal Component (b)

Figure 2.21 (a) Scores plot obtained from PCA of 18 basil, 18 peppermint, and 18 stevia GCxGC-TOFMS m/z 73 chromatograms demonstrates differentiation between species based on metabolite profiles. PC1 captured 61.84% of the variance and PC2 captured 16.78% of the variance. (b) Eigenvalues for the first 20 principal components. PC1 and PC2 capture the significant chemical variations. Reprinted with permission from ref. [57]. Copyright Elsevier

2.3.4

Food and Food Products

Food and food products generally contain a considerable number of organic and inorganic components with very similar chemical structures. Moreover, the concentration of components shows a considerable variation (g kg−1 to μg kg−1 ), which makes it difficult or

Gas Chromatography

47

4

PC[2]

2 A3 A2 A1 0

I3 H2 I2 I1 H1

F1 F3

F2

−2 −4 B1 −6

B2

−6 −5 −4 −3 −2 −1 0 1 2 3 4 5 6 7 8 9 10 11 12 PC[1]

Figure 2.22 Plot of the first two principal components from PCA of chromatographic data. F, Nordic reference fulvic acid; H, Nordic reference humic acid; A, commercial humic acid (Aldrich); B, groundwater humic acid from Latvia (Baltezers); I, aquatic humic acid from Latvia (Lake Islienas). Reprinted with permission from ref. [58]. Copyright Elsevier

even impossible to carry out the complete analysis of a product in one run. The complex chromatographic profiles of food and food products cannot be evaluated and compared with each other using the traditional visual method. The complexity of the problem of food analysis requires the use of multivariate mathematical-statistical methods. 2.3.4.1

Wines

The commercial importance of wines means that their composition has been vigorously investigated using various chromatographic techniques. The majority of methods are used for the identification of the origin of the wine sample (authenticity test) and/or the chemical Table 2.24 Summary of analysed samples Description (clone-line#tree number) esch5#15 (control) esch5#30 (control) e2-5#23 e2-5#24 e2-3#9 e2-3#24 e2-1#1 e2-16#10 e2-16#20 e14-4#6 e14-4#9 e14-18#9 e14-18#15 Reprinted with permission from ref. [59]. Copyright Elsevier.

PCA identifier

Gene construct

C1 C2 A 1a A 1b A 2a A 2b A3 A 4a A 4b B 1a B 1b B 2a B 2b

35 S-rol C 35 S-rol C 35 S-rol C 35 S-rol C 35 S-rol C 35 S-rol C 35 S-rol C rbsc-rol C rbsc-rol C rbsc-rol C rbsc-rol C

48

Multivariate Methods in Chromatography: A Practical Guide Scores

1.5 PC2

• A1b 1.0

• B1b B2b

0.5 • C1

•

A2a

0 • B1a −0.5 −1.0

• A4b

• A1a

• A4a

• C2

• A3

• B2a PC1

−1.0 −0.5 −1.5 RESULT, X-expl 56%, 16%

0

0.5

1.0

1.5

2.0

Figure 2.23 Score plot of principal components PC1 and PC2 after modelling with all pyrolysis products. A and B, Genetically modified samples; C, control samples. Reprinted with permission from ref. [59]. Copyright Elsevier

characterization of the quality of the product. The complicated chromatographic profile of wine has necessitated the application of sophisticated mathematical-statistical procedures. The analysis of the volatile analytes in various Italian wines was carried out by SPMEGC-MS and the data set was evaluated by PCA, CA and DA. The method found 35 volatile components in wines. Table 2.25 shows the variance accounted for by the first five principal components. The relatively high number of principal components containing eigenvalues over 1 indicates the complexity of the composition of wine samples. The solutes having high loadings in PC1 are listed in Table 2.26. It was found that PCA classified the wines well according to their origin. The CA results shown in Figure 2.24 support the classification obtained by PCA [60]. PCA has also been employed for the evaluation of the response of sensors towards red wines. Its good classification ability was demonstrated [61]. GC-FID was used for the analysis of wines fermented by adding a fungal glucosidase enzyme. The data were evaluated by PCA. The three-dimensional plot of PC1–PC3 is shown

Table 2.25 Variance accounted forby the first five principal components in Italian wine analysis Principal component

Eigenvalue

Explained variance (%)

Cumulative explained variance (%)

1 2 3 4 5

9.17 6.20 4.14 1.78 1.70

26.2 17.7 11.8 5.1 4.9

26.2 43.9 55.8 60.9 65.8


Gas Chromatography

49

Table 2.26 Analytes with higher loadings on PC1 (either positive or negative) Compounds Increasing with vintage Ethyl acetate Ethyl 2-methylpropanoate Ethyl lactate Ethyl 2-methylbutanoate Ethyl 3-methylbutanoate Ethyl succinate α-Ionone Decreasing with vintage Ethyl hexagonate Ethyl octanoate Ethyl decanoate Hexyl acetate 3-Methylbutyl hexanoate

Loadings on PC1 −0.77 −0.85 −0.74 −0.70 −0.82 −0.83 −0.75 0.74 0.80 0.83 0.78 0.84


in Figure 2.25. The distribution of data points in the three-dimensional space suggests that the effect of variety on the aroma composition of wines is higher than that of enzyme treatment. The volatile compounds having a high loading in the first three principal components are compiled in Table 2.27 [62]. The terpenoid profile of wines has also been employed for their differentiation. Measurements were performed by dynamic headspace solid-phase microextraction (HS-SPME) coupled to GC-MS. The data set was evaluated by PCA, and the parameters of the first two principal components are listed in Table 2.28. The first two principal components account for about 90% of the total variance and discriminate adequately between the wines according to variety as illustrated in Figure 2.26. It was found that the method separates the varieties well according to their origin [63]. The amount of higher alcohols, fatty acids and carbonyl compounds (altogether 42 variables) in 36 Madeira wines were determined by HS-SPME–GC-MS. PCA was separately carried out on each type of analyte. The computations suggested that PCA is suitable for the classification of Madeira wines [64]. The concentration of monoterpenes, higher alcohols, ethyl esters, fatty acids, acetates, volatile phenols and heavy sulfur compounds was determined by GC-FID and the data set was subjected to PCA and LDA. The plot of the first and second factor loadings is depicted in Figure 2.27. Wines form clear-cut clusters on the map suggesting that the procedure can be successfully applied for the differentiation of this class of white wine [65]. The volatile profile of six Mencia wines (Galicia, NW Spain) has also been studied using a similar method to that described in Falqué et al. [65]. PCA was employed for the elucidation of the dissimilarities among the wine samples; 67 variables were used as the analytical data. The variables included terpenes, norisoterpenes, alcohols, acetates, ethyl esters, fatty acids, volatile phenols, lactones, aldehydes and sulfur compounds measured by

50

Multivariate Methods in Chromatography: A Practical Guide Distances 0

10

20

30

40

50

60

RO69 RO66 RO65 RO73 RO64 RO63 RO61 RO74 RO68 LN44 LN38 LN34 LN33 RO67 RO62 LN38 LN37 LN32 LN31 RO71 LN41 NA19 NA23 NA16 BB12 BB11 NA21 NA18 LN45 LN39 LN35 BR57 NA17 BB5 LN42 NA24 RO72 LN40 BB9 NA20 LN43 NA25 NA22 NA26 BB3 BR52 BR64 BR51 BR49 BR59 BR46 BR53 BB14 BR59 BR56 BR55 BR49 BR50 BB13 BB8 BB15 BB7 BB6 BR47 BB4 BB10 BB2 BB1

Figure 2.24 Dendogram of cases obtained with hierarchical CA on 68 wine samples. Reprinted with permission from ref. [60]. Copyright Elsevier

Gas Chromatography

51

2

PC 3

1

0

albe2 albe1 albe2 albe3 albe2 albe1

me3 me2 me3 me1 me1 me3

aire1 aire2 aire2 aire1

−1

2.0

che1 che2 che2 che3 che3 1.0 PC 1

0.0

−1.0

0.0

1.0

2.0

3.0

PC 2

Figure 2.25 Plot of wines with and without enzyme treatment on the space defined by the first three principal components. Ale, albillo with enzyme treatments; albc, albillo control; aire, airen with enzyme treatment; airc, airen control; me, macabeo with enzyme treatment; mc, macabeo control; che, chardonnay with enzyme treatment; chc, chardonnay control. Reprinted with permission from ref. [62]. Copyright Elsevier Table 2.27 Correlation coefficients for wine volatile components against PC1, PC2 and PC3 Compounda

Correlation coefficient

PCl

2-Phenylethanol γ -Butyrolactone Ethyl 4-hydroxy butyrate Hexanoic acid Octanoic acid 1-Hexanol Decanoic acid Isoamyl acetate Ethyl phenyl acetate

0.97 0.96 0.95 −0.94 −0.90 0.84 −0.79 −0.74 −0.72

PC2

Catechins trans-3-Hexen-1-ol cis-3-Hexen-1-ol Geraniol Total polyphenols

0.96 0.83 0.77 −0.72 0.70

PC3

Benzyl alcohol

−0.88

Principal component

a

Only those compounds with absolute correlation coefficients >0.70 have been included. Reprinted with permission from ref. [62]. Copyright Elsevier.

52

Multivariate Methods in Chromatography: A Practical Guide Table 2.28 Eigenvalues, percentage of variance and cumulative percentage accounted for by the two first principal components Rotation sums of squared loadings Principal component

Eigenvalue

Variance (%)

Cumulative (%)

1 2

3.634 3.562

45.431 44.543

45.431 89.974


GC-MS. The distribution of points on the plot PC1 versuss PC2 indicates that PCA of the original data sets is suitable for the classification of wines according to their geographical origin [66]. A rapid method was developed and applied for the distinction of wines based on the global volatile signature. The data obtained by HS-SPME–GC-MS were analysed by PCA. The good separation ability of the method is demonstrated in the plot PC1 versus PC2 shown in Figure 2.28 [67].

1.5

Verdelho

PC2 (44.5 %)

1.0

0.5 Malvazia 0.0

−0.5 Boal −1.0

Sercial

−1.5

−1.0

−0.5

0.0

0.5

1.0

1.5

PC1 (45.4 %)

Figure 2.26 Principal component analysis for all wines differentiating the varieties projected in the plane defined by the first two factors. Reprinted with permission from ref. [63]. Copyright Elsevier

Gas Chromatography

53

Factor Loadings, Factor 1 vs. Factor 2 Rotation: Unrotated Extraction: Principal components 0.7 L95 L92

L94

0.5

L93

Factor 2

0.3

0.1 A95 A94 −0.1 −0.3

DB94

−0.5 0.66

0.70

T95 T94 T93 T92

A93

DB95

DB93

A92

DB92

0.74

0.78

0.82 Factor 1

0.86

0.90

0.94

0.98

Figure 2.27 Plot of the first principal component with respect to the second principal component, using the principal component based on odour unit values for wine samples. DB, Dona Branca wines; A, Albarino wines; L, Loureira wines; T, Treixadura wines. Reprinted with permission from ref. [65]. Copyright Elsevier 0.3

FP

Ar

0.2

PC2 (25%)

0.1

−0.25 −0.2 −0.15 −0.1 −0.05

0 0

0.05

0.1

0.15

0.2

0.25

−0.1 −0.2

5FP

10FP

15FP

−0.3 PC1 (53%) 20FP 5Ar

10Ar

15Ar

20Ar

Figure 2.28 PC1 versus PC2 plot of the global volatile signature of Fernao-Pires and Arinto wines for 5, 10, 15 and 20 min of extraction. Reprinted with permission from ref. [67]. Copyright Elsevier

54


2.00000 Sercial

PC2 (36.6%)

1.00000

0.00000

Boal

Malvazia

Verdelho −1.00000 −1.00000

1.00000 0.00000 PC1 (54.7%)

2.00000

Figure 2.29 Extracted principal components as a function of seven variables for 36 samples of musts. Reprinted with permission from ref. [68]. Copyright Elsevier

HS-SPME–GC-MS has also been employed for the separation and quantitative determination of flavour compounds in 39 different must samples of different geographical origin. PCA and LDA were used for the evaluation of the data matrix. The first two principal components accounted for the overwhelming majority of variance (54.7 and 36.6%, respectively). Samples are well separated on the two-dimensional plot, as illustrated in Figure 2.29, demonstrating the good separation capacity of the procedure [68]. GC combined with PCA was employed for the selection of aroma compounds as markers of the changes in sherry wines subjected to biological ageing [69]. Electronic nose and GC have been combined and employed for the detection of offflavours in wine. The wines were classified by PCA. The plot PC1 versus PC2 is depicted in Figure 2.30. It was concluded from the distribution of wines on the plot that the method is suitable for the differentiation between control wines and off-flavour doped wines [70]. The advantages of the simultaneous application of GC and HPLC have also been exploited in the instrumental analysis of wines. Thus, the volatile composition of wines from New Zealand was determined by GC. Organic acids and inorganic ions were measured by ion chromatography and sugars were analysed by HPLC. PCA has been used for the optimization of the composition of these wines. The classification of these wines according to the first two principal components shows that the optimal composition is 10 vol%, more than 30 g l−1 sugars and 0.5 bar CO2 (Figure 2.31) [71]. GC-FID, GC-MS and HPLC were employed for the investigation of the effect of winemaking technologies on the phenolic and volatile compounds. It was found that both

Gas Chromatography

55

1.0 0.8 Wine + HEX

Axe 2 (62.9% variance)

0.6 0.4

Wine + TCA

Wine + 4EP

0.2 0.0 Wine + OCT −0.2 −0.4 Wine + AE Control Wine

−0.6

−0.8 −1.2 −1.0 −0.8 −0.6 −0.4 −0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 Axe 1 (62.7% variance)

Figure 2.30 Discrimination by PCA of off-flavours in a red wine. HEX, hexan-1-ol; TCA, 2,4,6-trichloroanisole; AE, ethyl acetate; OCT, oct-1-en-3-ol; 4EP, 4-ethyl phenol. Reprinted with permission from ref. [70]. Copyright Elsevier

30/7/2 o

2

15/7/2 o

AXIS 2

1

15/7/1 o

0

−1

2/5/0 o

45/7/2 15/10/2 o o 45/10/1 30/10/1 15/15/1 oo 30/13/2 o o 30/13/1 15/10/11 o 30/10/2 o 45/10/2 45/10/2 o o 45/10/1 15/13/2 o o 45/10/1 o o

−2 −3

−2

−1

0

1

AXIS 1

Figure 2.31 Arrangement of samples according to PC1 and PC2. The coding of the sample follows the levels of the chemical factors in the order: sugar/alcohol/carbon dioxide. Reprinted with permission from ref. [71]. Copyright Elsevier

56


PCA and CA can be applied for the differentiation of wines according to the winemaking technology [72]. 2.3.4.2

Oils

Both wines and also various oils have been extensively investigated using GC technologies. A wide variety of mathematical-statistical methods has been applied for the facilitation of data evaluation. Thus, the prediction of the retention of fatty acid methyl esters of rice bran oil was performed using MLR [73]. As the chromatographic profiles of oils are complicated, the relatively simple MLR techniques have not often been applied. A lot of effort has been devoted to the GC analysis of extra virgin olive oils (EVOOs). This interest can be explained by the considerable commercial importance and nutritive value of EVOO. The volatile compounds in EVOO have been separated, identified and quantified in 12 EVOO samples by electronic nose and HS-SPME–GC-MS procedures completed by sensory analysis. PCA was separately employed for the data sets obtained by GC and electronic nose. The three-dimensional plot of the first three principal components computed from the GC data is depicted in Figure 2.32. The data points representing individual EVOO samples do not form clear-cut groups on the map suggesting the homogeneity of single cultivars [74]. Another study used direct thermal extraction and GC-MS for the measurement of the volatile content of EVOOs. The data set was evaluated by PCA. The plot of PC1 versus PC2 indicates that Italian and Spanish EVOOs can be well differentiated by the method. It was further established that the lipogenase oxidation products have the highest differentiating power [75].

4

1 - 2CRL 2 - Americano 3 - Leccione 4 - Tisignana 5 - Maremmano 6 - Olivastra di Mortalcino 7 - Leccino 8 - Scarlinese 10 - Morcone 11 - Madonna dell’lmpruneta 12 - Lazzero di Prata 13 - Gremigno di Montalcino

3

PC3 (14%)

2 1 0 −1 −2 −3 −4

−2

0 PC1

2 2 (38% )

4 6

−4

−2

0

4

)

8%

2 (1

PC

Figure 2.32 PCA score plot obtained using the data from the HS-SPME–GC-MS analysis of extra virgin olive oils. Reprinted with permission from ref. [74]. Copyright Elsevier

Gas Chromatography

57

OIL6 OIL12 OIL16

OIL26

OIL6 OIL24 OIL25 OIL13 OIL9 OIL17 OIL26

OIL21 OIL39 OIL40


OIL2 OIL3

0


OIL1 OIL23 OIL33 OIL34

OIL4

30 10

20

Height

40

OIL20 OIL35 OIL38 OIL37 OIL36

50

60

AGNES HCA plot

OIL6 OIL24 OIL25 OIL13 OIL17 OIL28 OIL33 OIL14 OIL29 OIL30 OIL22 OIL27 OIL19 OIL32 OIL21 OIL40

OIL9 OIL12 OIL16

OIL4 OIL35 OIL39 OIL38 OIL37

OIL34

OIL26 OIL8 OIL5 OIL11 OIL10 OIL18 OIL31 OIL7 OIL15

0

OIL2 OIL3

OIL1 OIL23

20

Height

40

OIL36

OIL20

60

DIANA HCA plot

Figure 2.33 (a) AGNES and (b) DIANA hierarchical CA plots of chromatographic profiles (FID, scaled) from retention times 0.8 to 30 min. Reprinted with permission from ref. [76]. Copyright Elsevier

The volatile profile of Australian olive oils was investigated by GC-FID and GC-MS and the results were evaluated by two CA techniques (agglomerative nesting, AGNES and divisive analysis, DIANA). CA dendograms are shown in Figure 2.33. It was concluded from the cluster dendograms that the variety exerted the highest impact on the composition of volatiles while the influence of malaxation time and temperature was of secondary importance [76]. HS-SPME–GC-FID, HS-SPME–GC-MS and sensory analysis were applied for the comparison of volatiles of five plant oils (rapeseed, soybean, peanut, sunflower and olive oil).

58

Multivariate Methods in Chromatography: A Practical Guide Table 2.29 The variances in seven principal components obtained after PCA of 132 samples normalized by mean-centring Principal component 1 2 3 4 5 6 7

%Variance

Total

95.09 2.72 1.98 0.12 0.07 0.01 0.01

95.09 97.80 99.78 99.90 99.98 99.99 100.00


PCA and CA were used for the comparison of the volatile composition of fresh and stored oil samples. PC1 and PC2 explained 36.79 and 22.40% of the total variance, respectively. As the distribution of individual oil samples was similar on the PCA plot and cluster dendogram, it was concluded that both PCA and CA are suitable for the comparison of the volatile profile of oils and can be used for the detection of changes caused by storage [77]. The fatty acid composition of another set of 134 plant oil samples (pumpkin, sunflower, peanut, olive, soybean, rapeseed, corn and mixed oils) was measured by GC-FID after hydrolysis and methylation. The differences among the fatty acid profiles were elucidated by using various multivariate statistical methods such as PCA, CA and counterpropagation neural network. Only seven fatty acids (palmitic, stearic, oleic, linoleic, linolenic, eicosanoic and eicosenoic) were included in the computation. The results of the PCA calculation are compiled in Table 2.29. The first principal component accounts for the overwhelming majority of total variance illustrating the basic similarity among the oil samples. It was stated that the procedure makes identification of the origin of the oil possible and facilitates the detection of adulteration [78]. The composition of essential oils of different origin has also been vigorously investigated by various GC technologies combined with PCA and other multivariate mathematicalstatistical methods less frequently used. Thus, the use of GC-FID and GC-MS for the analysis of essential oil composition in mandarin cultivars has been reported. Data were evaluated by DA. It can be concluded from the distribution of points on the three-dimensional map that the method can be employed for the differentiation of mandarin cultivars [79]. A similar method was applied for the differentiation of various species of mandarin according to the chemical variability of peel and leaf essential oils. Measurements of essential oil composition were performed by GC-MS and 13 C NMR, and the quantitative data were analysed by CA and DA. The cluster dendogram of 56 mandarin peel oils is depicted in Figure 2.34. According to the dendogram the main differentiating features were limonene concentration, limonene/ -terpinene ratio and linalyl acetate/limonene ratio [80]. Another study used GC-FID and GC-MS for the separation and quantitative determination of the volatile components of Thymus serpylloides ssp. gadorensis. The chromatographic profiles were compared by PCA, CA and multidimensional scaling (MDS)

Gas Chromatography

97 96 95 94 93 92 91 90 89 88 87 86 85 84 83 82 81 80 79 78 77 76 75 74 73 72 71 70 69 68 67 66 65 64 63 62 61 60 59 58

III

linalyl acetate/limonene

II

limonene/γ-terpinene

59

IA

I

limonene

IB

0

20

40

60

80

100

120

140

160

180

Figure 2.34 Dendogram obtained from the CA of 56 mandarin peel oils. Samples are clustered using Ward’s technique with a Euclidean distance measure. Reprinted with permission from ref. [80]. Copyright Elsevier

analysis. The CA dendogram is depicted in Figure 2.35. The compounds responsible for the differences observed are marked on the dendogram. The results of MDS support the conclusions drawn from the CA dendogram [81]. The same analytical and computational procedures were applied for the study of the variability of essential oil content and composition of Thymus baeticus. However, the CA dendogram (Figure 2.36) and MDS analysis show a markedly different distribution of volatile compounds [82]. The variability of the essential oil of Tymbra capitata (L.) Cav. was studied using a similar method to that described in Sáez [81, 82]. The results of CA are depicted in Figure 2.37 and they agree with the findings of Sáez [81–83]. 2.3.4.3

Foods of Plant Origin

In addition to wines and oils, a wide variety of other foods and food products of plant origin have been investigated by GC technologies and evaluated by chemometrical methods. Thus, the performance of GC-MS was compared with those of electronic nose and MS-based electronic nose to assess apple quality during shelf life. The first two principal components account for 85 and 10% of the total variance, respectively. The distribution of data points on

60

Multivariate Methods in Chromatography: A Practical Guide A----------Buitre 2 geranicl

Buitre 1 La Ragua 3

linalool + linalyl ac

D------Cambrón 4 Bacares 1

linalool + trans-sabinene hydrate

Bacares 2 Morrón 5 Morrón 2 Morrón 3 Morrón 4 Cambrón 1

linalool

Cambrón 2 C------Cambrón 3 Morrón 1 Tetica 4 Aulago 2

carvacrol

B----------Tetica 1 María 1

thymol

Aitana 1 María 2 Tetica 2 La Ragua 1 La Ragua 2 La Ragua 4 Tetica 3 Aulago 1 Tetica 5 E---Calar M. bajo 1

thymol geranyl ac myrcene caryophyllene ox

Calar Mundo 3 Calar Mundo 1 Calar Mundo 2 Aitana 2

carvacrol + bomeol + 1,8-cineole terpinen-4-ol

Bacares 3 F---Calar M. bajo 2

myrcene + o-terpineol + terpinen-4-ol + 1,8-cineole

Figure 2.35 Cluster analysis of the 34 studied samples with the essential oil component(s) that characterize the major subgroups shown. Reprinted with permission from ref. [81]. Copyright Elsevier

the score plot is depicted in Figure 2.38. Samples are well separated according to the storage conditions confirming the applicability of PCA to assess apple quality during shelf life [84]. The volatile components of fresh potherb mustard and their pickles with different pickling time were measured by HS-SPME–GC-MS. The similarity of the 87 compounds determined by instrumental analysis was evaluated by CA. The CA dendogram is shown in Figure 2.39. It was concluded from the results of computation that the glycosinolates in mustard contain

Gas Chromatography

0.0

DISTANCES

10.0 α-lerpineol + borneol 1,8-cineole linalool geranial

Purchena 1 Purchena 2 Huercal-Ov 1 Velefique 1

linalool + terpinen-4-ol + linalyl ac.

Velefique 2

linalool + terpinen-4-ol + α-terpineol

Cantoria 2 Ocaña 2

61

geranyl ac terpinen-4-ol

Ocaña 1 Cpco. López 1

p-cymene + α-terpinene

Ohanes 1 Cantoria 1

α-terpineol

Enmedio 1 Lubrin 1

bomeol

Ohanes 2 tr-sabinene hydrante Enmedio 2 Lubrin 2

bomeol

Figure 2.36 Cluster analysis of the16 studied samples with the essential oil component(s) that characterize the major subgroups shown. Reprinted with permission from ref. [82]. Copyright Elsevier

allyl, butenyl, isobutanyl and phenylethyl groups; the derivatives with allyl and phenylethyl groups in the side chain were predominant [85]. The products of Maillard reactions between sugars (fructose, glucose, lactose, maltose, rhamnose and xylose) and amino acids (alanine, asparagine, glutamine, glycine, threonine, arginine, cysteine, lysine and glutamate) were separated and quantified by GC. The 199 GC peaks were analysed by parallel factor analysis (PARAFAC) using a genetic algorithm and procrustes analysis. It was stated that the performance of this new computation method is higher than the traditional loading procedure [86]. The composition of the essential oils of hop varieties has also been investigated. Sample components were collected by HS-SPME and analysed by GC-MS. Samples were classified by PCA using 11 analytes as observations. The distribution of 31 hop varieties in the plot of PC1 versus PC2 is depicted in Figure 2.40. The data illustrate that the method can be successfully employed for the clustering of varieties [87]. The essential oils of hop varieties were also analysed by steam distillation followed by GC-FID and GC-MS. Sixteen well defined peaks were selected and subjected to CA and PCA computation. The CA dendogram and the three-dimensional space formed by the first three principal components are depicted in Figures 2.41 and 2.42. It was established that both computation techniques separate the hop varieties well. It was further established that the method can find only limited application in the analysis of hop mixtures [88]. The composition of both green and roasted coffees has been measured using different GC methodologies. Thus, the fatty acid profile of coffee lipid extracts was determined by GC-FID after derivatization and the concentrations of 10 free fatty acids were analysed

62


(a) Thymol

(c) Thymol/Carvacrol

(b) Carvacrol

0

20

40 60 Linkage Distance

80

100

Figure 2.37 Cluster analysis of the 115 studied samples with the essential oil component(s) that characterize the major subgroups shown. Reprinted with permission from ref. [83]. Copyright Elsevier

by PCA and DA. The first two principal components accounted for 62.1% of the total variance (40.7% and 21.4%, respectively). The loadings of variables in PC1 are compiled in Table 2.30. The data in Table 2.30 indicate that the ratio of oleic and linolenic acids separates the Arabica and Robusta coffee varieties. The same conclusions can be drawn from the plot of PC1 versus PC2. The equation for the discriminant function (DF) is: DF = −8.87 × C16 : 0 − 5.40 × C18 : 0 − 3.59 × C18 : 1 −9.26 × C18 : 2 − 1.29 × C18 : 3 − 1.46 × C22 : 0

(2.11)

It was stated that the function has a recognition ability of 100%. It was further established that both varieties and the green and roasted states of coffee can be separated by using two discriminant functions (Figure 2.43) [89].

Gas Chromatography

63

0.15

CA 12+15 0.1

CA 8

PC 2 10%

COOL 8 −0.3

ULO 8

0.05

COOL 12+15

−0.05

ULO 12+15

−0.1

0 0.1 −0.05

0.3

0.5 ULO 5

CA 5

COOL 5 −0.1 CA 1 COOL 1

ULO 1

−0.15

−0.2 PC 1 85%

Figure 2.38 Score plot of PC analysis of the 22 selected volatile compounds of Jonagold apples measured by SPME–GC-MS. The different samples are grouped as a function of the storage atmosphere [() COOL (1◦ C, 20.8 kPa O2 , 0.03 kPaCO2 ); () CA (1◦ C, 3 kPa O2 , 2.5 kPa CO2 ); () ULO (1◦ C, 1 kPa O2 , 2.5 kPa CO2 )] and days of shelf life exposure. The arrows indicate increasing shelf life exposure. Reprinted with permission from ref. [84]. Copyright Elsevier

The performance of HS-SPME coupled with GC-FID and GC-MS was compared with that of gas sensor array analysis. Roasted Arabica and Robusta coffees served as model compounds. The plot of PC1 versus PC2 is shown in Figure 2.44. It was concluded from the data that PCA identified the varieties well but failed in the separation of samples according to their geographical origin. On account of its speed and high identification power, use of the aroma sensing device was proposed [90]. A similar method (HS-SPME coupled with GC-MS) has found application in the analysis and classification of different roasted coffee varieties. The 32 peaks were analysed by PCA. It was established that PC1 and PC2 account for 43.3 and 27.7%, respectively, of the total variance and are suitable for the classification of coffee varieties [91]. The correlation between the consumer preference of tomatoes and the results of instrumental analyses (SPME–GC-MS, headspace fingerprint mass spectrometry, and quartz microbalance-based electronic nose) was evaluated by CA and PCA. Significant relationships were found between the consumer preference and the results of instrumental analyses [92].

64

Multivariate Methods in Chromatography: A Practical Guide Pascaled Distance Cluster Combine CASE

Cluster 5

Cluster 4

Cluster 3

Cluster 2 Cluster 1

Label VAR0058 VAR0086 VAR0019 VAR0001 VAR0069

No

VAR0017 VAR0087 VAR0016 VAR0042 VAR0073 VAR0024 VAR0015 VAR0025 VAR0070 VAR0031 VAR0052 VAR0036 VAR0023 VAR0011 VAR0049 VAR0073 VAR0074 VAR0003 VAR0080 VAR0035 VAR0065 VAR0022 VAR0045 VAR0081 VAR0020 VAR0038 VAR0072

17 87 16 42 73 24 15 25 70 31 52 36 23 11 49 73 74 3 80 35 65 22 45 81 20 38 72

0

5

10

15

20

25

58 86 19 01 69

Branch B

Branch A

Figure 2.39 Dendogram obtained from hierarchical CA of individual volatile compound analysis of potherb mustard and pickles after 10 days and 6 months of storage. The values presented are the means of six replicate trials. Reprinted with permission from ref. [85]. Copyright Elsevier

The VOCs were separated and quantitatively determined in Capsicum chinense sp. peppers using HS-SPME–GC-MS. The similarity of the 34 volatile compounds found in pepper samples was elucidated by PCA. The principal component plot is shown in Figure 2.45. The data indicate that the method can facilitate the identification of pepper varieties [93]. An HS-SPME-GC-MS technique was developed and applied for the detection of statutory potato pathogens such as brown rot (Ralstonia solanacearum, R.S.) and ring rot (Clavibacter michiganensis subsp. sepedonicus, Cms). The volatiles of infected and uninfected tubers were measured and the chromatographic profiles were compared by PCA, electronic nose and ANNs. The discrimination of infected and uninfected tubers by PCA is shown in Figures 2.46 and 2.47. It can be observed that tubers infected with pathogens differ markedly from the uninfected ones. An instrument was developed composed of an automated SPME sampler and 8-metal oxide sensor array for the safe differentiation between infected and uninfected tubers [94].

Gas Chromatography

65

3

2

C2

1

0

−1

−2

−3 −3

−2

−1

0

1

2

3

4

5

C1

Figure 2.40 Principal component analysis of essential oil components (C1, component 1; C2, component 2) performed on 31 hop samples of hop varieties grown in Slovenia: () Aurora; () Celeia; (•) Magnum; and (◦) Savinski golding. Reprinted with permission from ref. [87]. Copyright Elsevier

30

25

SA GO VIN LD JS IN KI G M AG NU M

LE IA CE

BO BE K

AU

10

A

15

RO R

Distance

20

5

0

Figure 2.41 Dendogram obtained by CA (nearest neighbour, squared Euclidean) of 78 hop samples. Reprinted with permission from ref. [88]. Copyright Elsevier


−−−−− − −−−−

Component 3

66

t1

en

on

mp

Co Component 2

Figure 2.42 Principal component analysis of 78 hop samples presented in three- dimensional space: (◦) Aurora; (+) Bobek; (×) Celeia; (−) Magnum; and (∗) Savinski golding. Reprinted with permission from ref. [88]. Copyright Elsevier

The volatile flavour compounds released from red kidney beans under in vitro (model mouth system) and in vivo (in-nose) conditions were measured by GC-MS. The relative dynamic release data were evaluated by PCA. Some results are compiled in Figure 2.48. It was established on the basis of the distribution of points in Figure 2.48 that the in vitro technique adequately simulates the in vivo release [95]. Both GC-FID and high-performance anion-exchange (HPAE) chromatography were applied for the analysis of sugars and for the detection of adulteration of acacia, chestnut, and lavender honey. The differences among the concentrations of sugars were elucidated by PCA. The PCA results are shown in Figure 2.49. It was found that the method was suitable for the detection of the addition of 5–10% exogenous sugars [96].

Table 2.30 Loadings of each variable for PC1 Variable

Loading

C14:0 C16:0 C16:1 C18:0 C18:1 C18:2 C18:3 C20:0 C20:1 C22:0

0.849 −0.122 −0.064 0.369 0.903 −0.849 −0.891 0.480 0.681 −0.409


Gas Chromatography

67

5 4 3

DF 2

2 1 0 −1 −2 −3 −4 −15

−10

−5

0

5

10

DF 1 Ga

Gr

Ra

Rr

Figure 2.43 Plot of the two discriminant functions. Ga, green Arabica; Gr, green Robusta; Ra, roasted Arabica; Rr, roasted Robusta. Reprinted with permission from ref. [89]. Copyright Elsevier

The quantification of ochratoxin A and deoxynivalenol in barley grains was performed by GC-MS and electronic nose and the data set was evaluated by PCA. It was found that PCA is suitable for the separation of samples with normal and off-odour [97]. The influence of the type of organic solvent on the extraction efficacy of alkylresorcinol from wheat grain was studied in detail. Both GC-MS and TLC were employed for the 6 Scores Guatemala

Brazil

5 4 3 2 1 0 −1 Ivorycoast −2

Zaire Angola

Kenya A Kenya B

Uganda

−3 −4

Colombia

−5 −20 −15 −10 −5 arom 0, PC(expl):

0

5

10

15

Figure 2.44 Plot of the PCA scores obtained for the first two principal components that characterize the Arabica and Robusta varieties analysed by SPME–GC-MS. Reprinted with permission from ref. [90]. Copyright Elsevier

68


3

Score

PC1

• Y6

• R5 • R4

2

• R3 R1 ••R2

1

• Y5

Red

0 P5 •• P4

−1

• Y1

• Y4

• Y2

• P2 • P3

−2 −3

• Y3

• P1

Purple

Yellow

−4 −4 −3 RESULT, M-AMPT

0

2

4

PC2 8

6

(a) 0.3

Loading Graph

PC3

• 14

0.2 Red

0.1

• 13

0 −0.1

Purple

−0.2 −0.3 −0.4

• 10

• −20 •

−9

Yellow

• −12

−0.10 −0.11 RESULT, M-AMPT

• 12

−0.06

0

0.06

•

21

• 12• 11 • 21 • 24 • 1114 • 10• 13 • 20•

• •1430 • 11 • 20 • 30 • 3• 1•4 • 16 •2 • 36 •5 • 30• 4 0.10

0.16

0.18

PC4 0.20

(b)

Figure 2.45 Principal component analysis (a) scores and (b) loading graph (see text for details). Reprinted with permission from ref. [93]. Copyright Elsevier

analysis of the extracts and the data were compared by CA as shown in Figure 2.50. It was concluded from the data that cyclohexane is the most appropriate solvent for the Soxhlet extraction of this class of analytes [98]. 2.3.4.4

Miscellaneous Food Products

GC-MS followed by PCA was employed for the identification of Africanized honeybees. GC analysis found 65 separated peaks in the hydrocarbon extracts of 238 honeybees. The

Gas Chromatography PCA

2.0 1.5 1.0 PCA2 (5.72%)

69

Brown Rot

0.5 0.0 −0.5 −1.0 −1.5 −2.0 −2.5 −2.0 −1.5 −1.0 −0.5

0.0

2.0 1.5 1.0 0.5 0.0 0.5 1.0

0.5

PCA1 (92.30%)

1.0

Not Brown Rot

1.5 1.5 2.0 2.0

Figure 2.46 Principal component analysis data of four Ralstonia solanacearum infected cultivars and their control illustrate clear discrimination of the brown rot ) from the not brown rot samples in a display of the first three principal components. Reprinted with permission from ref. [94]. Copyright Elsevier

PCA 2.0 1.5 Ring Rot 1.0

PCA2 (13.49%)

0.5

−2.0 −1.5 −1.0 −0.5

0.5

1.0

1.5

2.0

2.5

−0.5 Not Ring Rot −1.0 −1.5 PCA1 (76.24%)

Figure 2.47 Principal component analysis data for three Clavibacter michiganensis subsp. sepedonicus infected cultivars and their control illustrate a moderate discrimination of the ring rot from the not ring rot samples in a display of the first two principal components. Reprinted with permission from ref. [94]. Copyright Elsevier

70

Multivariate Methods in Chromatography: A Practical Guide in vitro 0 rpm 2 1 0 in vivo free

in vitro 26 rpm

−1 −2

in vivo 52 rpm

in vitro 52 rpm

Figure 2.48 Scores of in vitro and in vivo samples varying in mastication rate on the first principal component. Principal component analysis was conducted on the dynamic release data of mass m/z 33 relative to the maximum concentration measured for a given sample. Reprinted with permission from ref. [95]. Copyright Elsevier

PC2 (30.9%)

points representing European and Africanized honeybees are not well separated on the plot of PC1 versus PC2 indicating that PCA is not suitable for their differentiation. It was further found that a genetic algorithm for pattern recognition separates the two types of honeybees well [99]. An Italian chestnut honey and a commercial cumin seed preparation were selected as model compounds for the investigation of the precision of quantitative data from multicomponent GC-FID or GC-MS analysis. Both HS-SPME and direct thermal desorption were employed for sample pretreatment. Eigenvalues of PCA carried out on honey and cumin samples are shown in Figures 2.51 and 2.52. It was determined that PCA may

C18 0

10

20

25 C23C10 14 PC1 (59.8%)

C35

Figure 2.49 Principal component analysis plot of pure chestnut honey (), honey adulterated with sugar syrup A (×), honey adulterated with sugar syrup B (∗), honey adulterated with sugar syrup C (+) and commercial honey (). Reprinted with permission from ref. [96]. Copyright Elsevier

Gas Chromatography

71

ACETONE HEXANE CYCLOHEXANE CHLOROFORM ETHYL ACETATE DIETHYL ETHER

10

20

30

40

50 60 70 80 % of Euclidean distance

90

100

110

Figure 2.50 Grouping of obtained wheat grain extracts using CA (Euclidean distances, Ward’s amalgamation algorithm) based on relative spot abundance of contaminating ballast substances. Reprinted with permission from ref. [98]. Copyright Elsevier

help in the estimation of factors causing quantitative dispersion and,consequently,loss of precision [100]. The separation and quantitative determination of free amino acids in honeys of various origin (eucaliptus, rosemary, orange and heather) were performed by GC-FID and GC-MS after derivatization. The differences between the amino acid profiles were elucidated by PCA and canonical discriminant analysis. The first four principal components accounted for about 78% of the total variance. According to the data, the loadings of amino acids show high differences between the principal components indicating that they can be used 25

Eigenvalue

20 15 10 5 0 1

2

3

4

5

6

7

PC

Figure 2.51 Eigenvalues obtained for the first seven principal components (experimental data, white bars; simulated data, grey bars) in the PCA analysis of honey volatiles by SPME and GC. Reprinted with permission from ref. [100]. Copyright Elsevier

72

Multivariate Methods in Chromatography: A Practical Guide 14

Eigenvalue

12 10 8 6 4 2 0 1

2

3

4

5

6

7

8

9

PC

Figure 2.52 Eigenvalues obtained for the first nine principal components (experimental data, white bars; simulated data, grey bars) in the PCA analysis of cumin volatiles by DTD–GC-MS. Reprinted with permission from ref. [100]. Copyright Elsevier

for the identification of honeys. The comparison of plots of principal components and DA suggests that the separation capacity of DA is higher than that of PCA, so that it can be used for the assessment of the botanical origin of honey samples [101]. Milk products, especially cheeses, have been frequently investigated by various GC technologies coupled with multivariate mathematical-statistical procedures. Thus, PCA has been applied for the investigation of the influence of ripening temperature on the GC profile and flavour of Cheddar cheese manufactured from raw and pasteurized milk [102] and for the elucidation of the correlation between flavour and microbiological profiles in Serra da Estrela cheese through ripening [103]. The VOCs in Piacentinu Ennese cheese were preconcentrated by SPMA and analysed by GC-olfactometry. The sensory attributes and VOC results were subjected to PCA. PC1 and PC2 explained 53.0 and 18.7% of the total variance. The plot of PC1 versus PC2 is shown in Figure 2.53. It was found that PCA separates the artisanal and industrial cheeses well and the cheeses with different ripening times [104]. The effect of different fibres on the performance of SPME preconcentration has also been investigated using ‘Terrincho’ ewe cheese as model material. GC-MS was applied for the separation of acid and neutral compounds, ketones, aldehydes, esters and other volatile molecules and the correlation between the elements of the data set was evaluated by PCA. The first two principal components accounted for 83.33% of the total variance. The distribution of points on the plot of PC1 versus PC2 indicates that the method allows the classification of cheeses according to the site of production [105]. Another method used dynamic headspace analysis and GC-MS to follow the development of VOCs in processed cheese during various storage conditions. PCA was performed on the data matrix consisting of 29 VOCs and of various storage conditions. It was found from the PCA computations that the first three principal components account for 93% of the total variance (81, 8 and 4%, respectively). It was assumed that PC1, PC2, and PC3 can be identified as storage time, conditions of light or darkness, and storage temperature [106].

Gas Chromatography

73

1 roughness hardness spicy aroma intensity salty butyric disporsion

0.8 0.6

PC2 (18.7%)

0.4 mushroom/earthy terpenses

0.2

unknowns

0 −0.2

total VCC FA esters

−0.4 FFA −0.6 −0.8 −1 −1

−0.8

−0.6

−0.4

−0.2 0 0.2 PC1 (53.0%)

0.4

0.6

0.8

1

Figure 2.53 Principal component analysis plot showing the relationships between variable vectors having the highest correlations with PC1 or PC2 (> |0.6|) and individual sample scores on the same dimensions: , ripened for 2 months; •, ripened for 4 months; , ripened for 6 months. Closed symbols represent the R cheeses; open symbols represent the P+S cheeses. Reprinted with permission from ref. [104]. Copyright Elsevier

The measurement of the triacylglycerol and free fatty acid content by GC-MS has been applied for the detection of milk fat authenticity. MLR was employed for the calculation of the M value (M value is a constant 100 for pure milk fat). The equations for Cabral, blue cheese and Majorero goats milk cheese are: M = −2.7575 × C26 + 6.4077 × C28 + 5.5437 × C30 − 15.3247 × C32 + 6.2600 × C34 + 8.0108 × C40 − 5.0336 × C42 + 0.6356 × C44 + 6.0171 × C46 (2.12) M = 7.733 × C40 + 2.428 × C50 + 21.392 × C52

(2.13)

M = 7.836 × C26 + 6.390 × C28 + 12.828 × C30 − 21.324 × C32 + 5.989 × C34 + 7.264 × C40 − 1.581 × C42 + 0.186 × C44 + 3.077 × C46

(2.14)

It was concluded from the calculations that this computation mode facilitates the detection of foreign fats in these types of cheeses [107]. The VOC content of heated fish powders (herring and blue whiting) was investigated by SPME–GC-MS. The similarities and dissimilarities among the chromatographic patterns were elucidated by PCA. The plots of PC1 versus PC2 for herring and blue whiting are shown in Figure 2.54. It was established that both the length and intensity of heating influence the development of VOCs. It was further established that the best markers of

74

Multivariate Methods in Chromatography: A Practical Guide 15 Herring 10

PC2 (13%)

100 4

5

2

60

0.5

4 1 2

0.5

4/1 2

4

0.5

80

4

0

2

1

1

120 2

0.5

4 1

40

140

0.5

−5

2

−10 −10

−5

5

0

4

10

15

PC1 (70%) (a) 15 Blue whiting 10

PC2 (16%)

100

5

60

0.5 4

0

40

2 1 0.5

1

2

4

4 1 2 0.5

0.5 80

1

0.5 2

120

140

4 1

−5 2

−10 −10

−5

5

0

10

4

15

PC1 (57%) (b)

Figure 2.54 Principal component scores for (a) herring and (b) blue whiting samples. All variables were log-transformed and standardized. Temperature and exposure time are shown in the plots. The background samples (no temperature exposure) are indicated by open circles. Explained variance is given on the axes. Reprinted with permission from ref. [108]. Copyright Elsevier

Gas Chromatography

75

Similanty

−106.28

−30.52

−31.24

−100.00 Variables

Figure 2.55 Dendogram based on cumulative hierarchical clustering by complete linkage (Ward distances) of variables. Reprinted with permission from ref. [109]. Copyright Elsevier

the heat exposure were pyridines, pyrazines, aromatic hydrocarbons, amides, and volatile sulfur compounds [108]. The fatty acid composition in the tissues of rainbow trout (Oncorhynchus mykiss) was determined by GC-FID. Fatty acids were separated and quantitated as their methyl esters. Both FA and CA were applied for the determination of the impact of the fatty acid composition on the dietary feed and that of hatchery. CA dendograms of variables and observations are depicted in Figures 2.55 and 2.56. It was concluded from the data that both FA and CA differentiate between the composition of dietary feed and type of tissue [109]. Another study investigated the relationship between the fatty acid composition and diet in young goat meat. The analytical method (transesterification followed by GC-FID) was similar to those described in Barrado et al. [109]. CA and PCA were employed for the comparison of the chromatographic profiles. The loadings of fatty acids in the first three principal components are compiled in Table 2.31. The first two principal components account for the majority of the variance present in the original data matrix. The loadings of fatty acids and

Similarity

−306.60

−171.07

−36.63

100.00 Observations

Figure 2.56 Dendogram based on agglomerative hierarchical clustering by complete linkage (Ward distances) for the 60 samples analysed. Reprinted with permission from ref. [109]. Copyright Elsevier

76


Table 2.31 Loadings for the first three principal components and their respective variances Components

PC1

PC2

PC3

n-3 n-6 n-6 : n-3 SFA MUFA PUFA PUFA : SFA

−0.42510 0.41611 0.43416 0.29152 −0.35810 0.39722 0.29424

−0.16460 −0.18750 0.10224 −0.57420 0.43388 0.30407 0.56286

0.36437 −0.46830 −0.33290 0.28316 −0.37900 0.42731 0.36128

72.1472 72.1472

23.7765 95.9238

3.8139 99.7377

Variance % Cumulative %

n-3, omega-3 fatty acid; n-6, omega-6 fatty acid; SFA, saturated fatty acid; MUFA, monounsaturated fatty acid; PUFA, polyunsaturated fatty acid. Reprinted with permission from ref. [110]. Copyright Elsevier.

fatty acid ratios are different in different principal components indicating that these parameters can be used for the classification of samples. The CA dendograms of meat samples and variables are shown in Figures 2.57 and 2.58. Computations proved that the fatty acid composition can be used for differentiation between dietary treatments [110]. Static HS-GC-MS, an electronic nose and traditional microbiological analysis were used to investigate dry salami species. The results obtained by the electronic nose were evaluated by PCA. A three-dimensional plot of the first three principal components is depicted in Figure 2.59. It was stated that the method is suitable for differentiation between different salami, between samples with different ripening and between salami prepared from male and female swine [111]. 2.3.5

Environmental Analyses

The application of up-to-date industrial and agrochemical practice not only increases the yield of foods and other products but also enhances environmental pollution causing considerable public concern. The continuous refinement of chromatographic methods coupled with various multivariate techniques allows the detection of environmental pollutants at very low level even in complicated accompanying matrices. 2.3.5.1

Volatile Organic Compounds, Polycyclic Aromatic Hydrocarbons, and Polychlorinated Biphenyls

Metal oxide semiconductor (MOS), metal oxide semiconductor field effect transistors (MOSFETs), GC-MS and sensory analysis were simultaneously applied to study VOC emissions from car seat foams. PCA, multiple factorial analysis (MFA), and general procrustes analysis (GPA) were used to determine the relationships between the samples. The plot of PC1 versus PC2 based on GC profiles is depicted in Figure 2.60. It was found that the information content of the methods shows high deviations, so the problem needs further investigation [112]. The components of VOC emission from landfills were separated and quantitatively determined by SPME–GC-MS. The data were evaluated by PCA and CA, and the biplot

Gas Chromatography 1.0

0.8

0.6

0.4

0.2

77

0.0

32 26 25 31 27 30 29 28 24 19 23 22 18 20 21 17 16 14 15 11 12 9 10 13 8 1 3 6 5 2 7 4

Figure 2.57 Hierarchical CA dendogram for caprine meat submitted to four different dietary treatments. Samples numbered according to dietary treatment: samples 1–8, 9–16, 17–24, and 25–32 represent different treatments. Reprinted with permission from ref. [110]. Copyright Elsevier 1.0

0.8

0.6

0.4

0.2

0.0

PUFA:SFA PUFA

n-6 n-3 n-6

SFA MUFA

n-3

Figure 2.58 Hierarchical CA dendogram for variables. Reprinted with permission from ref. [110]. Copyright Elsevier


1.5

CEO [4] NMO 23 NMI 80 NMI 51

PC 3 (5%)

1.0 FEMALE

0.5 0.0

MALE

−0.5 −1.0

1 2 (88% )

)

%

0

(6

PC 1

2

−1.5−4 −3 −2 −1

1.5 1.0 0.5 0.0 −0.5 −1.0

3

4

5 −1.5

PC

78

Figure 2.59 Principal component analysis three-dimensional plot obtained from the data derived from the repeated exposure of the different salami samples with 15 days of ripening in the sensor array. Reprinted with permission from ref. [111]. Copyright Elsevier

5.0

Factor 2 - 24.18 %

2.5

g

b em

g gm

g

0

e

e

bm b

e

b

c c d cm c dm d d

−2.5 a a am a

−5.0

−8

f f fm f −4 0 Factor 1 - 35.37 %

4

8

Figure 2.60 Principal component analysis scores on PC1 and PC2 for seven foams (a–g) analysed in three replicates by GC-MS. Reprinted with permission from ref. [112]. Copyright Elsevier

Gas Chromatography

79

Biplot 1.3

lmm. north of the landfill toluene

Axis 2

lmm.6 Km ethyl benzene o-xylene lmm.1.5 Km

0.6

p-xylene acetic acid ethyl ester

acetic acid butylester 2 hexanone limonene

Fresh wastes lmm. landfill entrance lmm. 3 Km butanoic acid Old wastes

Scrubber exit

−0.6

−1.3 1.2 dicloro etilene

0.6

1.3 hexanoic acid Scrubber entrance tetrachloroethylene

alpha terpinene hexanal

−0.6

Biogas

eucalyptol camphor

p-cymene Leachate −1.3 Axis 1

Figure 2.61 Biplot from the PCA evaluation of the data relative to emission and ambient air samples. Reprinted with permission from ref. [113]. Copyright Elsevier

of principal components and the CA dendogram are depicted in Figures 2.61 and 2.62. The results of principal component computations indicate that fresh and older wastes can be characterized by limonene and p-cymene, respectively. The CA dendogram illustrates the relationship between emissions from biogas and leachate and between the sampling sites [113]. UPGMA Scrubber exit Scrubber entrance lmm.6 Km lmm.3 Km lmm.1.5 Km lmm. northern of the landfill Leachate Biogas lmm. landfill entrance Old wastes Fresh wastes 6

5

4

3

2 Eclidean

1

0

Figure 2.62 Cluster analysis of emission and ambient air samples. Reprinted with permission from ref. [113]. Copyright Elsevier

80

Multivariate Methods in Chromatography: A Practical Guide Ester Arorratic

Factor 2(20 % of variance)

TGS2610

Terperic

Ether CAP25 CAP01 CAP02

TGS642

Dioxin

Sulfur Alcohol Aldehyde TGS822

Aliphatic TGS2650

Furans

Chiorinated Acid

Ketones NH3

Nitrogen

Factor 1(39 % of variance)

Figure 2.63 Plot of loadings in the plane of the first two factors of a PCA carried out using both sensor signals and relative chemical concentrations. Reprinted with permission from ref. [114]. Copyright Elsevier

The composition of the exhaust air from a compost pile was simultaneously monitored by metal oxide sensors, electronic nose and DTD-GC-MS. The concentrations of ketone, furan, nitrogen compounds, carboxylic acids, chlorinated compounds, ammonia, alcohol, aldehyde, terpene, sulfur compounds, ester, aromatic hydrocarbon, and aliphatic hydrocarbon were measured. The data set was evaluated by PCA; the plot of PC1 versus PC2 is shown in Figure 2.63. It was established that chloride, acids, nitrogen compounds and ammonia form a clear-cut cluster as well as dioxin, sulfur, alcohol and aldehyde. The use of electronic nose for the analysis of exhaust gases is advocated [114]. The presence, concentration and distribution of environmental pollutant polycyclic aromatic hydrocarbons (PAHs) in different matrices have also been frequently investigated. PAHs were measured in surface sediment of a coastal lagoon using Soxhlet and ultrasound extraction followed by GC-MS. The inherent similarities among the elements of the data matrix were elucidated by CA. The CA dendogram of both variables and observations are depicted in Figures 2.64 and 2.65. The results illustrated that the concentration of PAHs is related to the organic matter content. It was further established that the amount of PAHs depends considerably on the sampling site [115]. PAHs have also been measured in reclaimed and surface water using SPME–GC-MS. The distribution of the elements of the data set according to the first two principal components is shown in Figure 2.66. As the plot PC1 versus PC2 illustrated the considerable differences among the surface water stations, it was proposed for the investigation of surface waters [116]. PAHs have also been measured in ambient air by GC-MS and the chromatographic profiles of the samples were compared by FA. The FA results are compiled in Table 2.32. It was concluded from the data that multivariate mathematical statistics may facilitate the characterization and identification of the PAH emission source [117]. PAHs in urban air have also been investigated using GC-MS. The influence of environmental parameters such as temperature, relative humidity, seasonal variation and fossil fuel

Gas Chromatography

81

0.1 0.2

SIMILARITY

0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

1 2 2 7 1 1 1 1 1 2 1 1 2 8 1 3 6 9 1 2 4 5 1 3 1 2 2 1 4 8 7 0 5 6 0 3 9

Figure 2.64 Dendogram for 22 PAHs and organic matter content: (1) naphthalene; (2) 2methylnaphthalene; (3) 1-methylnaphthalene; (4) acenaphthylene; (5) acenaphthene; (6) fluorene; (7) phenanthrene; (8) anthracene; (9) 2-methylanthracene; (10) 9-methylanthracene; (11) fluoranthene; (12) pyrene; (13) 1-methylpyrene; (14) benzo[a]anthracene; (15) chrysene; (16) benzo[b]fluoranthene; (17) benzo[k]fluoranthene; (18) benzo[a]pyrene; (19) perylene; (20) indeno[1,2,3- cd]pyrene; (21) dibenzo[a, h]anthracene; (22) benzo[g, h, i]perylene; (23) organic matter content. Reprinted with permission from ref. [115]. Copyright Elsevier

0.1 0.2

SIMILARITY

0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0

1

2

3

4

5

8

6

7

Figure 2.65 Dendogram for the sampling sites. Reprinted with permission from ref. [115]. Copyright Elsevier

82

Multivariate Methods in Chromatography: A Practical Guide 1.0

S5 0.8

GROUP C

0.2 Factor 2

S4 GROUP B

0.0

S1 S2

S6 S7

−0.2

S3 GROUP A

−1.00

−1.05

−0.5 Factor 1 (a)

−0.4

3.5 3.0

5

Factor 2 (12.52%)

2.5 2.0 1.5 4 2

1.0

8

0.5 0.0 −0.5

7 13 1415 10 3(6) 1216 9(11)

1

−1.0 −4.0

−3.5

−1

0

1

Factor 1 (87.37%) (b)

Figure 2.66 Principal component analysis showing the pattern of PAHs in the surface water: (a) factor loading plot; (b) factor score plot. (1) Naphthalene, (2) acenaphthylene, (3) acenaphthene, (4) fluorene, (5) phenanthrene, (6) anthracene, (7) fluoranthene, (8) pyrene, (9) benzo[ a]anthracene, (10) chrysene, (11) benzo[ b]fluoranthene, (12) benzo[ k]fluoranthene, (13) benzo[ a]pyrene, (14) dibenzo[ a,h]anthracene, (15) indeno[1,2,3-c,d]pyrene and (16) benzo[g,h,i]perylene. Reprinted with permission from ref. [116]. Copyright Elsevier

Gas Chromatography

83

Table 2.32 Factor analysis of fine (PM2,5 ) and coarse (PM2,5−10 ) particulates for the PAHs concentration data set at the Taichung Harbor sampling site Fine (PM2,5 ) particulates PAHs Naphthalene Acenaphthylene Acp Fluorene Phenanthrene Anthracene Fluoranthene Pyrene Cyclopenta[c.d]pyrene Benzo[a]anthracene Chrysene Benzo[b]fluoranthene Benzo[k]fluoranthene Perylene Benzo[e]pyrene Benzo[a]pyrene Indenol[1.2.3−c.d] pyrene DBA BbC Benzo[g.h.i]pyrene Coronene Eigenvalue Variance % Cumulative %

Origin source

Coarse (PM2,5−10 ) particulates

Factor 1

Factor 2

Factor 3

Factor 1

Factor 2

Factor 3

0.11 0.86 —

0.77 0.08 — 0.70 0.95 — 0.28 — 0.73 — —

0.51 — 0.19 0.43 — 0.27 0.58 — 0.76 0.43 0.31 — — 0.32 0.32 — 0.81

0.05 — 0.10 0.76 0.87 0.61 — 0.79 — — — — — 0.09 — 0.15 0.67

— 0.35 — 0.76 0.79 — 0.58 0.87 — — — 0.11 0.37 0.89 — — 0.75

0.48 — — — — — 0.71 0.59 0.75 — 0.78 0.27 — 0.28 — 0.76 0.42

0.02 — 0.72 0.43 — 0.80 — — 0.59 0.67 0.12 — — — — 0.44 —

0.21 0.57 — — 6.49 26.81 26.81

— — 0.75 0.84 3.01 21.87 48.68

— — — — 2.49 11.33 60.01

— — — 0.32 — — 0.87 — — 0.15 0.55 — 7.03 4.07 3.56 23.03 18.91 15.97 23.03 41.94 57.91 Incomplete Diesel Gasoline combustion — vehicle vehicle Coal and pyrolysis — emission emission combustion of fuel Oil burning —

Acp, acenaphthene; DBA,dibenzo[a,h]anthracene; BbC, benzo[b]chrysene. Only factor loadings >0.1 are presented. Factor loadings >0.7 are in bold. Reprinted with permission from ref. [117]. Copyright Elsevier.

usage was elucidated by principal component factor analysis. The varimax rotated factor loadings for individual PAHs and the correlation matrix between specific PAHs and other parameters are compiled in Tables 2.33 and 2.34, respectively. It was established that lowvolatile PAHs were more influenced by wind and solar radiation, while semi-volatile PAHs were more influenced by relative humidity and temperature [118]. Polychlorinated biphenyls (PCBs) have also been measured in ambient air using a well established GC-MS technology. The chromatographic profiles of samples and the environmental conditions (maximum and minimum temperature, wind velocity, particulates) were compared by FA and CA. The FA results are compiled in Tables 2.35 and 2.36. Five factors explained only 77.8% of the total variance indicating the considerable diversity of

84


Table 2.33 Varimax rotated factor loading for particulate PAHs PAHs

Factor 1

Factor 2

Factor 3

Communality

Naphthalene Acenaphthylene Acenaphthene Fluorene Phenanthrene Anthracene Fluoranthene Pyrene Benzol[a]anthracene Chrysene Benzo[b + k]fluoranthene Benzo[a]pyrene Indenolpyrene Dibenzo[a.h]anthracene Benzo[g.h.i]pyrene Variance explained by each factor

0.360 0.684 −0.247 0.211 0.504 0.527 0.674 0.708 0.773 0.820 0.836 0.853 0.914 0.873 0.888

0.087 0.525 −0.003 0.249 0.787 0.804 0.706 0.682 0.573 0.495 0.485 0.484 0.247 0.390 0.440

0.789 0.148 0.914 0.908 0.214 0.115 0.156 0.122 0.125 0.223 0.126 0.099 0.083 0.057 0.035

0.759 0.766 0.897 0.931 0.919 0.937 0.977 0.982 0.942 0.967 0.949 0.971 0.903 0.917 0.984

7.27 48.5

4.02 26.8

2.51 8.0

% Variance

13.8

Factor loadings >0.7 are in bold. Reprinted with permission from ref. [118]. Copyright Elsevier.

Table 2.34 Correlation matrix between specific PAH representatives of the individual factors, carbonaceous aerosols, meteorological variables and vapour-phase PAHs Factor 1 Items Organic carbons Elemental carbons Temperature Relative humidity Wind speed Solar radiation Naphthalenea Acenaphthylenea ACEa Fluorenea Phenanthrenea Anthracenea Fluoranthenea Pyrenea Benzo[a]anthracenea CHRa

Factor 3

B[b + k]F

B[g, h, i]P

PHEN

FLT

ACE

FLU

0.820 0.581 −0.548 −0.054 −0.304 −0.336 0.705 0.850 0.727 0.837 0.895 0.854 0.766 0.441 0.336 0.457

0.817 0.553 −0.463 −0.059 −0.445 −0.327 0.688 0.855 0.684 0.781 0.848 0.810 0.657 0.255 0.170 0.301

0.622 0.381 −0.729 −0.310 −0.006 −0.169 0.811 0.804 0.764 0.794 0.769 0.680 0.607 0.357 0.373 0.371

0.730 0.467 −0.667 −0.243 −0.188 −0.245 0.814 0.888 0.781 0.840 0.848 0.797 0.678 0.376 0.346 0.378

−0.149 −0.093 −0.276 −0.114 0.407 −0.057 0.095 0.066 0.374 0.300 0.120 0.186 0.312 0.627 0.636 0.250

0.192 0.057 −0.569 −0.176 0.183 −0.239 0.446 0.434 0.666 0.633 0.456 0.489 0.522 0.599 0.615 0.336

ACE, acenaphthene; CHR, chrysene. Vapour-phase PAHs. Reprinted with permission from ref. [118]. Copyright Elsevier.

a

Factor 2


85

Factor analysis for ambient air data

Factor

Eigenvalue

Variance (%)

Cumulative variance (%)

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21

7.09 3.35 3.03 1.50 1.12 0.96 0.76 0.60 0.49 0.45 0.33 0.29 0.25 0.21 0.14 0.12 0.096 0.076 0.064 0.039 0.024

33.78 15.96 14.43 7.15 5.32 4.58 3.63 2.86 2.34 2.13 1.58 1.40 1.19 1.01 0.65 0.59 0.46 0.36 0.30 0.18 0.12

33.8 49.7 64.2 71.3 77.8 81.2 84.8 87.7 90.0 92.2 93.7 95.1 96.3 97.3 98.0 98.6 99.0 99.4 99.7 99.9 100.0


Table 2.36 Rotated factor values Variable

Factor 1

Factor 2

Factor 3

Factor 4

Factor 5

PCB 8 PCB18 PCB28+31 PCB 44 PCB 52 PCB 66 PCB 70 PCB 77 PCB 97 PCB 101 PCB 105 PCB 118 PCB 138 PCB 151 PCB 153 PCB 170 PCB 180 Max. temperature Min. temperature Wind velocity Particulate

−0.042 0.054 0.215 0.480 0.101 0.849 0.527 0.707 0.393 0.882 0.303 0.593 0.434 0.804 0.652 −0.027 0.009 0.235 0.202 0.042 0.243

0.902 0.927 0.900 0.642 0.849 0.395 0.651 0.302 0.049 0.123 0.147 −0.146 0.098 0.099 −0.102 0.061 0.133 −0.019 0.065 0.093 −0.102

0.285 0.207 0.027 0.066 −0.102 0.056 −0.146 0.197 0.232 0.112 0.743 0.597 0.773 0.255 0.579 0.556 0.743 −0.020 −0.114 −0.147 −0.229

−0.067 −0.112 0.084 0.329 0.154 0.193 0.036 0.202 0.017 0.083 −0.300 0.026 0.105 0.134 0.047 −0.543 −0.056 0.921 0.922 −0.044 −0.256

0.047 0.145 0.225 −0.012 −0.11 0.037 0.107 −0.062 0.557 0.136 0.046 0.086 −0.007 −0.092 0.049 −0.286 0.113 −0.059 0.121 0.771 −0.608


86


Table 2.37 Principal component analysis results for the first campaign (factor loadings after varimax rotation) Factor 1

Factor 2

Factor 3

0.5 are underlined. Reprinted with permission from ref. [110]. Copyright Elsevier.

proves that PCA differentiates successfully between the PL profiles of treated and control rats [112]. The retention time of 33 purine nucleobases was measured under RP-HPLC conditions and the relationship between the retention parameters and physicochemical characteristics of analytes was elucidated using MLR, PLC and comparative molecular field analysis (CoMFA). The best correlation obtained by MLR is: log k = 0.0006 × I + 0.017 × MV − 0.003 × DIP − 0.010 × PSA − 0.871 n = 33, R 2 = 0.790, s = 0.294, F = 26.33, RMSECV = 0.871

(3.59)

where I is the moment of inertia, MV is the molecular volume, DIP is the dipole moment, and PSA is the polar surface area. It was stated that the results of CoMFA are superior to other computation procedures included in the study [113]. 3.2.3.3

Agrobiochemistry

HPLC coupled with PCA has been employed for the differentiation between wild-type and transgenic alfalfa (Medicago sativa L.) plants using the chromatographic profiles of phenolic metabolites. It was stated that PCA does not differentiate leaf metabolite profiles between modified and wild-type plants of the same genetic background, but stem phenolic

F2 150 x log k0b x z3

x ba1

x ba2

F1 60

log k0a x 230 z2 x

x Z1 x bb2

bb1 x 20

Figure 3.24 Relationship between the physicochemical parameters of peptides and their retention behaviour on polyethylene-coated alumina stationary phase. Two-dimensional NLMAP of principal component loading. Number of iterations: 103; maximum error: 3.89 × 10−2 . Subscripts a and b refer to acidic and basic mobile phases, respectively. Reprinted with permission from ref. [110]. Copyright Elsevier 6 4

PC2 (31%)

2 0 −2 −4 −6 −8

−6

−4

−2 0 PC1 (50%)

2

4

6

Figure 3.25 Principal component analysis score plot separating the four variants of lactate dehydrogenase studied: beef heart (); hog muscle (◦); pig heart (•); rabbit muscle (). Reprinted with permission from ref. [111]. Copyright Elsevier

164

Multivariate Methods in Chromatography: A Practical Guide 6 4

PC1 (31%)

2 0 −2 −4 −6 −8

−6

−4

−2 0 PC1 (50%)

2

4

6

Figure 3.26 (A) Multivariate analysis of the PL HLPC profile of rat sera from animals treated with amiodarone, cf. controls. (B) Chromatograms showing the peaks that were identified by PCA to vary in peak ratio. Reprinted with permission from ref. [112]. Copyright Elsevier

profiles were markedly different. It was further established that the method allows the monitoring of the biochemical phenotypes of transgenic plants [114]. The flavonic pattern of 404 plants of Astragalus caprinus (Fabacae) was determined by RP-HPLC. The similarities and dissimilarities between the chromatographic profiles were assessed by correspondence analysis, CA and DA. The CA dendograms are depicted in Figure 3.27. CA separated four chemotypes well proving that the technique can be used for the chemotaxonomic study of this class of plants [115]. The RP-HPLC method was applied for the measurement of the amount of flavonoids, phenolic acids and sesquiterpenoids in Leontodon autumnalis (Asteraceae, Lactuceae). The chromatographic profiles of 183 individual samples were compared by PCA. The biplot of PC1 versus PC2 computed from the relative quantification data matrix is depicted in Figure 3.28. The scattering of data points shows that the RP-HPLC technique followed by PCA differentiates between the two chemotypes [116]. A similar study was performed employing the concentration of flavonoids and phenolic acids in the flowerhead as markers. The amount of markers was determined by RP-HPLC and the similarities and dissimilarities between the samples were assessed by PCA. The first three principal components accounted for 56.6, 23.8 and 9.2% of the total variance, respectively, indicating the basic similitude between the specimens. The three-dimensional plot of the three principal components is shown in Figure 3.29. It was found that PCA forms five clear-cut separated clusters, which are taxonomically interpretable and agree well with the morphological based system of the genus Leontodon [117]. The concentration of floral anthocyanidines (cyanidine, peonidin, delphinidin, petunidin, and malvidin) was measured in 195 commercial petunias. The data matrix consisting of the anthocyanidines and the plant samples was evaluated by PCA. The results of PCA are

Liquid Chromatography

Chemotypes

4

3

2

165

1

(a)

Patterns within chemotypes

4a 4b 4c

3

2b

2a 1b

1a

(b)

Figure 3.27 Clustering of 404 plants of Astragalus caprinus into four genotypes using Euclidean distance and Ward aggregation on the first three principal components of correspondence analysis. (a) Complete dendogram representing the four principal clusters corresponding to the four chemotypes. (b) Expanded dendogram highlighting the patterns within the chemotypes. Reprinted with permission from ref. [115]. Copyright Elsevier

166


PCAG2r

−10

−20

−30

−40 0

40

80

PCAG1r

Figure 3.28 Scatter plot of the scores for PC1 versus PC2 obtained from the relative quantification data matrix of flavonoids, phenolic acids, and sesquiterpenoids. Reprinted with permission from ref. [116]. Copyright Elsevier

15

L . croceus

10 L. rilaensis L. autumnalis

5 L. pyrenaicus

PCS2 (23.8%)

0 −5

NW-European samples alpine samples

L. montaniformis −10 L. montanus −15

subsp. subsp. metonotirichus montanus

−20 L. duboisil

−25 −30 −35 35

20

PCS1

L. helveticus 5

−10 −25 (56.5 %)

0

5

20 25 10 15 ) (9.2% PCS3

30

Figure 3.29 Scores for the first three principal components from the PCA based on relative quantification data. Reprinted with permission from ref. [117]. Copyright Elsevier


167

Table 3.29 Results of PCA on anthocyanidine contents of commercial petunias Principal component 1 2 3

Eigenvalue

Variance (%)


2.07542 1.83133 0.71175

41.5 36.6 14.2

41.5 78.1 92.4


compiled in Tables 3.29 and 3.30. The data show that the first three principal components explain the overwhelming majority of variance suggesting a strong intercorrelation between the elements of the original data set. Anthocyanidines have high loadings in different principal components indicating that they can be employed for the differentiation of petunias. It was established that PCA of RP-HPLC differentiates well between the phenotypes; therefore, its application for the separation of phenotypes is advocated [118]. RP-HPLC was applied for the determination of the concentration of toxins (T-2 toxin, HT-2 toxin, diacetoxyscirpenol and neosolaniol) produced by Fusarium sporotrichioides. The fungi were grown on maize, wheat and rice as substrates under various conditions. The chromatographic profiles were classified by PCA. The plot of PC1 versus PC2 is depicted in Figure 3.30. The scattering of points on the map indicates that PCA cannot separate the samples well according to the type of substrate [119]. Humic materials of different origin were characterized with various analytical procedures, such as elemental analysis, size exclusion chromatography (SEC), UV absorption ratios, and liquid state 13 C-NMR spectroscopy. The elements of the data set were compared by CA. The CA dendograms obtained with the different analytical procedures are depicted in Figure 3.31. The dendograms show high diversity between each other indicating that the information content of the analytical methods is markedly different [120]. Gel permeation (Sepharose 4B column) and affinity chromatography (hyroxyapatite columns) together with other methods were employed to investigate the efficacy of extraction and purification techniques for the analysis of the microbial community profile in compost. The data were evaluated by PCA and it was found that PCA successfully differentiates between two diverse compost samples [121]. Table 3.30 Factor loadings of anthocyanidine contents of commercial petunias for each principal component Principal component Anthocyanidine Cyanidin 2 Peonidin 3 Delphinidin 4 Petunidin 5 Malvidin 6

1

2

3

0.90852 −0.00282 0.51923 −0.50474 −0.85184

0.26240 −0.91135 0.65866 0.63493 0.30814

−0.25653 0.39845 0.34606 0.49127 −0.35508


168


PC2

4

3

2

1

−3

−2

−1

1

2

3

4

5 PC1

−1

−2

−3

Figure 3.30 Scatter diagram of the experimental data points obtained by PCA: (x) maize; (o) rice; () wheat. Reprinted with permission from ref. [119]. Copyright Elsevier

3.2.3.4

Microbiology

RP-HPLC and an electronic tongue were employed for the differentiation of moulds in culture media. The chromatographic profiles and the data obtained by the electronic tongue were separately evaluated by PCA. The distribution of points on the map indicates that moulds are separated according to the species and culture medium [122]. RP-HPLC was applied for the separation and quantitative determination of PLs in methanotropic bacteria. Analytes were detected by ESI-MS. Bacteria included in the investigations were Methylomonas methanica, Methalomonas rubra, Methylomicrobium album BG8 (type I), Methylosinus trichosporium OB3b and CSC1 (type II), Methylococcus capsulatus Bath and Methylocella palustris (type X). The intact PL profiles were used for PCA as observations with the various bacteria being the variables. The plot of PC1 versus PC2 is depicted in Figure 3.32. Bacteria form well separated clusters on the two-dimensional map proving the good classification power of the RP-HPLC-PCA procedure [123]. Another HPLC technique has been applied for the separation and quantitative determination of algal pigments in sea water (SW), interstitial water/sediment system (IW) and near bottom water (NW). Pigments included in the analysis were chlorophyll a, chlorophyll b, chlorophyll c, fucoxanthin, zeaxanthin, peridinin, 19 -hexanoylofucoxanthin, 19 butanoyloxyfucoxanthin, pheophytin a and pheophytin b. The relationship between the amount of chlorophyll a and the other pigments was assessed by MLR. The parameters of the significant equations involved are compiled in Table 3.31. The equations revealed that the total chlorophyll a biomass depends significantly on the concentration of


169

DE72

Fasurface NMR data

Kranichsee purAIHA

SEC data

Kranichsee

DE72

GohyHA GohyHS purAIHA AldHAH

GohyHA

Pond

GohyHS

Fasurface 0.00.10.20.30.40.50.60.70.80.91.0

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 (a)

(b)

GohyHA

Kranichsee

AldHAH

GohyHA

GohyHA Kranichsee

purAIHA AldHAH

DE72

GohyHS

Fasurface

Pond

Pond

Fasurface 0.00.10.20.30.40.50.60.70.80.91.0 (c)

All data

DE72

UV data

purAIHA

0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 (d)

Figure 3.31 Results ofhierarchical CA showing the complete linkage dendogram of the rangenormalized data based on (a) liquid-state 13 C NMR, (b) size exclusion chromatography, (c) UV-visible spectroscopy, and (d) all data. The abscissa represents the Euclidean distances. Reprinted with permission from ref. [120]. Copyright Elsevier

other pigments but the character of correlation depended considerably on the site of sampling [124]. The RP-HPLC technique was employed for the analysis of diarrhetic shellfish poisoning toxins in 19 Prorocentrum lima isolates. The isolates were compared by CA using toxins as observations. The CA dendogram is depicted in Figure 3.33. The distribution of the isolates on the dendogram demonstrates that the geographical origin exerts a marked influence on the similarity between the dinoflagellates included in the investigation [125]. Their marked economical importance means that the various aspects of growth of phytoplanktons (spatial and temporal variabilities, dynamics of developments, vertical distribution) were vigorously investigated using HPLC technologies followed by multivariate mathematical-statistical analysis of the chromatographic results. The chlorophyll and carotenoid pigment composition of phytoplankton communities was measured by RP-HPLC for the determination of the spatial and temporal variabilities. The

170


1.0 6 Factor 3

0.6 0.2

2 7

–0.2

1

0.

5

4

6

3

.2 –0 r2 2 0. acto F

–0

.6

–1.0

–0.6

–0.2

0.2 1 ctor

0.6

Fa

Figure 3.32 Factor loading plot showing variation in intact PL profiles among type I (1, M. rubra; 2, M. methanica; and 3, M. album BC8), type II (4, M. trichosporium OB3b; and 5, M. trichosporium CSCI) and type X (6, M. capsulatus Bath) methanotrophs and a recently isolated acidophilic methanotroph (7, M. palustris). Reprinted with permission from ref. [123]. Copyright Blackwell Publishing

relationship between chlorophyll a characterizing biomass (dependent variables) and other pigments (independent variables) was assessed by MLR. The significant equations are: Alaskan : [chl.a] = 1.12 × [fuco] + 1.09 × [hexa] + 2.31 × [chl.b] + 0.059 n = 74,

r 2 = 0.783,

P < 0.01

(3.60)

Bering : [chl.a] = 1.52 × [fuco] + 0.66 × [hexa] + 1.80 × [chl.b] + 0.150 m = 60,

r 2 = 0.938,

P < 0.01

(3.61)

Kuril : [chl. a] = 1.68 × [fuco] + 0.91 × [hexa] + 2.90 × [chl.b] + 0.055 n = 90,

r 2 = 0.996,

p < 0.01

(3.62)

where chl. a is chlorophyll a, fuco is fucoxanthin, hexa is 19 -hexanoyloxyfucoxanthin, and chl.b is chlorophyll b. It was stated that the results of computation facilitate the study of the structure of phytoplankton communities [126]. A similar HPLC method was applied for the measurement of pigment composition in the investigation of plankton dynamics. The relationship between the various analytical data was computed by the Pearson product-moment correlation analysis. Calculations were performed on the results of the total population and of the 0.001 and the corresponding cumulative % of explained variance Eigenvalue V1

V2

V3

V4

V5

V6

V7

V8

V9

V10

V11

Explained 8.236 3.119 1.558 1.087 0.732 0.600 0.249 0.134 0.100 0.075 0.043 variance (eigenvalue size) Cumulative 51.4 % of explained variance

71.0

80.7

87.5

92.1

95.8

97.4

98.2

98.8

99.3 99.6


the dependence of the selectivity index (SI) on the structural descriptors are: SI = 0.243 × qO − 0.190 × q + 0.946 × MW − 0.285 × hSvdw + 0.255 × MOv + 0.343 × log P − 0.452 × pK m = 50, n = 7,

P = 6,

Radj = 0.756,

s2 = 0.255

(3.99)

where MW is molecular weight and pK is the phenolic dissociation constant, It was stated that PCA considerably facilitated the evaluation of the complicated data matrix [225]. Phenyl acetate, methyl p-toluene and methyl benzoate were applied as model analytes for the investigation of the resolution of on-flow LC 1 H NMR. PCA was applied for the detection of the number of analytes in the sample and CCA was used for the determination of the number and relationship of analytes between spectroscopic clusters [226]. A new computational method has been developed and successfully applied for the mathematical resolution of severely overlapping peaks. The first step of the varimax extended rotation (VER) method is the pre-processing of data (normalization), the second step involves PCA, the third step defines the so-called pure regions in the peak, and the last step applies ALS to improve the estimates of the concentration of analytes. Chromatograms of naphthalene, ethylbenzene, and xylene, as well as simulated chromatograms, were evaluated by the method. It was stated that this novel technique is simple and efficient [227]. Another study compared the efficacy of the linear solvation energy relationship (LSER), linear solvent strength theory (LSST), and typical-conditions model (TCM) for the prediction of the retention of a nonhomologous series of analytes in RP-HPLC. Measurements included 22 analytes, five stationary phases, and three types of mobile phases with four different compositions. The data show that the fit of the TCM to the experimental data is better than those of LSER and LSST. In order to improve the fit of the equations describing the relationship between the concentration of the organic component in the mobile phase and the log k value, the linear equations were recalculated including the square of the organic component as an independent variable in the computation. The results illustrate

236


6

#52

#48 #49

#48 2

0

PCP

PCP

PC3

PC2

3 #49

#52

0 −2

−4

−8

−6

−4

−3

−2 0 PC1 (a)

2

4

6

−4

−2

0 PC2 (b)

2

6

6 #48 #49 3 PCP PC3

#52

0

−3

−8

−6

−4

−2 0 PC1 (c)

2

4

6

Figure 3.89 (a–c) Scores plots of the PCA of the molecular descriptor data set. (◦) Chlorophenols; () alkyl/arylphenols; () other phenols. Note, 48, 49 and 52 are outliers. Reprinted with permission from ref. [225]. Copyright Elsevier

that the inclusion of the quadratic function in the relationships considerably enhanced their predictive capacity [228]. The solvent strength parameters (ε) of 39 analytes on alumina were fit to the solvation parameters model employing MLR. It has been stated that this type of calculation is suitable for the characterization of the surface adsorption properties of both alumina and silica stationary phases. Moreover, the solvation parameter model makes possible the prediction of the solvent strength parameters from the structure of solvents [229]. A QSRR study was carried out to assess the relationship between the log kw value of 21 analytes measured on four different RP-HPLC columns and their physicochemical parameters. The physicochemical parameters employed were the same as those used in LSER

Liquid Chromatography 0.6 0.4

Δq μ2

hSvdw, hSAS

0.2

μ

2

LUMO

0.3

Svdw, SAS

MOv

RG

Δq

0.0

Δorb

MW MOMO

LUMO

−0.2

PC3

PC2

logP

0.0

qH

−0.3

qO

−0.4

237

MOv

qO

hSvdw HOMO hSAS logP pK MW Svdw, SAS RG Δorb qH

pK

−0.6 0.0

0.1

0.2 PC1

0.3

0.4

0.0

0.1

(a)

0.2 PC1

0.3

0.4

(b)

0.6 Δq

0.3

qO

μ2

LUMO

MOv

PC3

logP

0.0

pK HOMO

MW

Δorb

−0.3

hSAS hSvdw

RG

Svdw, SAS

qH

−0.6

−0.4

−0.2

0.0 PC2

0.2

0.4

(c)

Figure 3.90 (a–c) Loading plots of the PCA of the molecular descriptor data set. Reprinted with permission from ref. [225]. Copyright Elsevier

studies. The equations describing the relationship between the dependent and independent variables on the four columns (I–IV) are: Column I P = ≤ 0.0925

P ≤ 0.0053

P ≤ 0.0400

log kw = −0.7933(±0.4369) + 1.1462(±0.3426) × R2 − 0.8270(±0.3625) × πH 2 H − 2.0340(±0.3678) × β2 + 2.6287(±0.3880) × Vx P ≤ 0.0001

P ≤ 0.0001

n = 18, R = 0.9546,

s = 0.3720, F = 33.37,

P ≤ 0.0001

(3.100)

238


ColumnII P = ≤ 0.5840

P ≤ 0.0159

P ≤ 0.0244

log kw = −0.1572(±0.2799) + 0.6320(±0.2281) × R2 − 0.7196(±0.2828) × 2H − 0.5392(±0.2199) × 2H + 1.8479(±0.2637) ×

2H + 1.4756(±0.2530) × Vx P ≤ 0.0291 n = 19,

P ≤ 0.0001

P ≤ 0.0001

R = 0.9696, s = 0.2396,

F = 40.85,

p ≤ 0.0001

(3.101)

Column III P ≤ 0.7236

P ≤ 0.0001

P ≤ 0.0001

log kw = −0.1481(±0.4123) − 3.4085(±0.3576) × R2 − 0.8270(±0.3625) βH × πH 2 2 + 2.2547(±0.1298) × Vx P ≤ 0.0001 n = 21, R = 0.9534, s = 0.5114,

F = 89.8,

P < 0.00001

(3.102)

Column IV P = ≤ 0.0025

P ≤ 0.012

P ≤ 0.0001

log kw = −0.9499(±0.2489) + 0.7262(±0.1728) × 2H − 2.0879(±0.1745) ×

2H + 3.0967(±0.3296) × Vx P ≤ 0.0001 n = 16,

R = 0.9737,

s = 0.1818,

F = 73.08,

P ≤ 0.0001

(3.103)

where R2 is the excess molar refraction, 2H is the dipolarity/polarizability parameter, 2H is the effective hydrogen bond basicity, Vx is the characteristic volume of McGowanand 2H is the effective hydrogen bond acidity. It was concluded from the results of computations that the structure and type of the stationary phase and the length and density of bonded ligands exert the highest impact on the retention [230]. Airborne carbonyl materials were collected and separated by RP-HPLC–APCI-MS as 2,4-dinitrophenylhydrazones. Airborne carbons were also analysed by the radiocarbon (14 C) method. PCA was carried out with the variables being the percentage of fossil carbon calculated from the total amount of carbon minus the amount of carbon deriving from distinct biogenic carbonyls. PCA found only one significant component. It was concluded from the data that the background levels of airborne carbonyl compounds are mainly of biogenic origin [231]. A novel parallel column RP-HPLC method was developed using a single multiwavelength absorbance detector and a nonhomologous series of analytes (18 compounds). The separation power of the parallel system was found to be lower than that of the traditional


0.4 MW MV

LC50 fish

PC2

0.2

LC50 daph EC50 GA ChV GA LC50 SW EC50 daph

239

Pr MR Polarizability

ChV fish

logP logBCF

Melting point

0 logk α

−0.2 Water solubility

logPapp

−0.4 −0.3

−0.2

−0.1

0

0.1

0.2

0.3

0.4

PC1

Figure 3.91 Loading plot of PC1 versus PC2 from PCA. Reprinted with permission from ref. [233]. Copyright Elsevier

single column arrangement but the chemical selectivity is higher. The GRAM technique has been successfully applied for the deconvolution of overlapping peaks [232]. Biopartitioning micellar chromatography (BMC) has been employed for the prediction of ecotoxicity. BMC involves a RP-HPLC system using a mobile phase consisting of phosphate buffer and the nonionic surfactant Brij 35 [polyoxyethylene(23) lauryl ether]. The retention data were correlated with the toxicity values taken from the literature, and with the computed structural descriptors. Except for one case the relationship between BMC and toxicity values was significant; the significance level was over 99.9%. The PCA of the toxicity parameters indicated that 78% of the total variance can be accounted for by the first two principal components. The distribution of variables on the plot of PC1 versus PC2 is shown in Figure 3.91. The scattering of points on the two-dimensional map indicates the high similarity between the toxicity values. The results of QSAR computations are compiled in Table 3.53. Calculations prove the highly significant relationship between measured and calculated physicochemical parameters and toxicity data. The QSAR model was proposed for the screening of new bioactive compounds and for the selection of the less toxic derivatives [233]. A separate QSRR study was devoted to the assessment of the retention behaviour of ionic and nonionic analytes in BMC. The retention time of 151 analytes of widely different molecular structure and polarity was measured under RP-HPLC conditions in mobile phases containing various concentrations of Brij 35. The following hydrophobic, electronic and sterical parameters were included in the computations as independent variables: log P, the anionic molar charge (A ), the positive molar charge (B ), polarizability, molar refractivity (MR), molar volume (MV) and parachor (Pr). The relationship between the measured and calculated hydrophobicity values of analytes was elucidated by PCA. The biplot of PC1

b ± ts

−1800 ± 200

−1900 ± 300

−370 ± 40

1180 ± 130 −90 ± 10

−230 ± 20

−112 ± 13

−1.0 ± 0.9

a ± tsb

1770 ± 10

1370 ± 140

300 ± 20

860 ± 70

72 ± 5

169 ± 12

88 ± 7

0.5 ± 0.5

1.3 ± 04

36 ± 6

77 ± 12

28 ± 5

400 ± 60

120 ± 20

625 ± 12

629 ± 8

c ± ts

0.78(0.77)

0.91(0.90)

0.95(0.94)

0.96(0.96)

0.90(0.90)

0.96(0.96)

0.85(0.85)

0.90(0.90)

R 2 (adjusted r 2 )

0.499

7.39

12.1

5.01

74.1

21.0

146

104

SE

112

247

331

322

242

320

159

282

F

0.488

7.17

11.2

4.76

72.0

19.9

142

101

RMSEC

0.512

7.53

14.9

5.36

86.2

22.7

165

126

RMSECV

0.509

7.58

11.7

5.19

71.9

21.6

143

97.9

RMSECVi

Statistical analysis and predictive features of the toxicity parameters–retention model ecotoxicity parameter = a + b(log k) +

b

LC50 in fish after 96 h. Confidence interval at 95%. SW, salt water; ChV, chronic value; BCF, bioconcentration factor; SE, standard error of the estimate; F , F -ratio; RMSEC, root mean square error of calibration; RMSECV, root mean square error of cross-validation (leave-one-out); RMSECVi, root mean square error of cross-validation (leave-one-out) for interpolated data. Reprinted with permission from ref. [233]. Copyright Elsevier.

a

LC50 (fish) (n = 66) LC50 (daph) (n = 58) LC50 (fish) SW (n = 31) EC50 (GA) (n = 54) EC50 (daph) (n = 31) ChV (fish) (n = 40) ChV (GA) (n = 53) Log BCF (n = 66)

Ecotoxicity parametera

Table 3.53 c(log k)2


241

versus PC2 indicated that hydrophobicity values computed by KOWWIN software were nearest to the experimental values; therefore, it can be used for the calculation of log P values. PLS has been employed for the selection of variables having the highest impact on the retention of analytes in BMC. Calculations indicated that log P, the absolute values of A and B , and the concentration of nonionic surfactant in the mobile phase exert the most significant influence on the retention behaviour. The relationships between the independent variables pre-selected by PLS and the log k values of analytes were computed with MLR and nonlinear models. The statistical parameters prove that the equation fits well to the measured values in each case, with the significance level being always over 95%. The results clearly show that nonlinear models have a slightly higher predictive power than the linear ones but their application is more complicated than that of MLR models [234]. BMC investigations were performed not only on the ODS stationary phase but also on the octylsilica column. The results prove the applicability of this stationary phase for BMC investigations [235]. As a result of the higher separation capacity and the possibility of the measurement of the retention time of analytes with markedly different retention characteristics in the same run the application of gradient elution RP-HPLC systems for QSRR and LSER studies has been vigorously investigated. Besides the application of LSER in isocratic separation mode [236], these computational procedures have been applied for the prediction of retention in gradient LC. The first step included the determination of the retention time of 15 preselected analytes in two different gradients. Components of the unrelated model analytes were benzamide, 4-cyanophenol, indazole, benzonitrile, indole, 2-naphthole, anisole, benzene, 1-naphthylacetonitrile, benzyl chloride, naphthalene, biphenyl, phenanthrene, pyrene, and 2,2 -dinaphthyl ether. Then the RP-HPLC gradient retention time was correlated with the three structural descriptors of the analytes (total dipole moment, electron excess charge of the most negatively charged atom, and water-accessible molecular surface area). The high significance level of the relationship between retention parameters and molecular descriptors indicates that the results of computations can be applied for the approximate prediction of analytes in gradient chromatography performed in a RP-HPLC system previously calibrated [237]. The same theoretical background was employed for the prediction of the gradient retention time using again the linear solvent strength (LSS) model and QSRR. The computations were carried out also by ANN. It was established that the predictive power of ANN is similar to that of MLR; therefore, both methods are suitable for the approximate prediction of the gradient retention time of analytes in a precalibrated RP-HPLC system [238]. Not only gradient elution based on the continuous increase of the concentration of organic modifier in the mobile phase but also gradient elution based on the continuous change of the pH of the mobile phase have been applied for the rapid determination of pK a values of bioactive compounds containing dissociable polar substituents [239]. It was stated that the novel RP-HPLC method is suitable for the determination of the dissociation constants of various molecules [240]. The theory of double gradient RP-HPLC has also been elaborated for the simultaneous determination of hydrophobicity and dissociation constants of various analytes. It was established that this novel technology may increase the versatility of the traditional RPHPLC procedure and promotes the better understanding of the underlying physical and physicochemical procedures [241].

242


Table 3.54

w w pK a ,

α, k1 , k2 , s1 and s2 parameters obtained by nonlinear fitting

Analyte

Log k1

s1

Log k2

s2

pw wKa

α

pw w K a lit.

Aniline 2-Amino-5-nitropyridine n-Methylaniline n-Ethylaniline 2,4,6-Collidine Brucine p-Nitrophenol Diethylbarbituric acid 2-Chloro-4-nitrophenol 2,6-Dimethyl-4-nitrophenol 1-Naphthylacetic acid N -benzyldimethylaniline

0.1475 0.7021 0.6517 1.0878 0.7706 2.4450 1.6293 2.1288 2.2621 2.6512 3.0254 0.8094

4.6086 7.6669 7.5187 7.3918 5.5581 7.4991 2.8834 4.1571 3.2836 3.9115 4.6549 5.5187

1.0505 1.4462 1.6717 2.1172 2.4737 3.2924 0.5039 1.0526 0.9746 1.1569 1.9609 2.5130

2.6217 3.4756 2.7553 3.1734 3.8209 5.4369 3.3592 4.2978 2.2537 3.2326 3.9972 3.5549

4.57 7.40 4.86 5.32 7.54 7.68 7.15 7.61 5.76 7.13 4.11 8.76

0.0504 −2.5788 −0.4371 −0.7674 −1.3646 0.6424 0.6623 1.2044 −0.4950 1.1498 2.5463 −0.0568

4.63 7.22 4.85 5.12 7.43 8.26 7.15 7.43 5.45 7.07 4.26 8.91

w pw w K a , literature pw K a values. Reprinted with permission from ref. [242]. Copyright American Chemical Society.

The theory of the simultaneous application of pH and organic phase gradient has been discussed in detail, and the practical use of the double gradient method in RP-HPLC has been demonstrated. It was stated that the technique is suitable for the determination of the lipophilicity of the ionized and nonionized forms of the analyte. Moreover, it makes possible the measurement of the pK a values of the compounds under investigation. It was stated that the method allows the simultaneous determination of hydrophobicity and pK a values of the analytes and may facilitate the separation of analytes containing dissociable polar substituents (Table 3.54) [242]. The retention time of 14 structurally unrelated organic compounds was measured on nine RP-HPLC columns using gradient elution. The columns were characterized by three theoretical models, such as the calculated logarithm of the octanol–water partition coefficient (clog P)-based model (model I), the application of structural descriptors (model II), and the hydrophobic-substraction model (model III). Model I calculates the univariate relationship between the retention time and clog P value. Model II applies MLR with the retention time being the dependent variable and the total dipole moment, electron access charge of the most negatively charged atom, and the water-accessible molecular surface area being the independent variables. Model III uses column selectivity as the dependent variable calculated by: k log α ≡ = Hη − S∗ + Aβ + Bα + Cκ (3.104) kEB where k is the retention factor of a given analyte, kEB the value of k for a nonpolar reference analyte (for example, ethylbenzene), determined on the same column under the same conditions, and the remaining selectivity-related symbols represented either eluentor temperature-independent properties of the column (H, S*, A, B, and C). The column parameters denote the following column properties: H, hydrophobicity; S, steric resistance to insertion of bulky analyte molecules into the stationary phase; A, column hydrogenbond acidity, mainly attributable to nonionized silanols; B, column hydrogen-bond basicity,


243

hypothesized to result from sorbed water in the stationary phase; C, column cation-exchange activity due to ionized silanols. The parameters , , , , and denote complementary properties of the analyte: , hydrophobicity; , molecular ‘bulkiness’or resistance of the analyte to its insertion into the stationary phase; , hydrogen-bond basicity; , hydrogen-bond acidity; and eˆ , approximate charge (either positive or negative) on the analyte molecule. It must be emphasized here that the values of each analyte parameter are relative to the values for ethylbenzene, the reference analyte, for which all the analyte parameters are zero. The theory and application of the method have been previously described [240,241]. The similarities and differences between the numerical results of models I–III were assessed by PCA. The application of model I resulted in highly significant linear correlations between the retention time and clog P values, with the coefficient of correlation being over 0.949. This result indicates that this simple model can be successfully employed for the prediction of the retention behaviour of this class of analytes on each RP-HPLC column. The coefficients of correlation are similar to those obtained by model I proving the similar prediction power of the methods. The data indicate that the first three principal components account for the overwhelming majority of variance; therefore, the retention characteristics of columns can be described by three background variables. The scattering of columns on the biplot of PC1 versus PC2 calculated from the results of the models is depicted in Figure 3.92. The distribution of the points illustrates that each model can be applied for the characterization of the retention behaviour of columns and the differences obtained by the various models are of secondary importance [243].

3 Symmetry C18

PC 2: 25.19%

2

1

Inertsil ODS-3 XTerra MS C18 Discovery HS F5

0

Nova-Pak C18 LiChrospher 60RP-select 8 Aqua C18 Supelcosil LC-18 Chromolith

−1

−2 −4

−3

−2

−1

0 1 PC 1: 68.72%

2

3

4

Figure 3.92 Projection of nine RP-HPLC columns onto the plane of the first two principal components (PC1 and PC2) from PCA of regression coefficients of the QSRR models of retention based on clog P and on the structural descriptors from molecular modelling. Reprinted with permission from ref. [243]. Copyright Elsevier

244


3.2.3.13

Other Environmental Pollutants

The effect of various environmental conditions and those of the chemical composition of oils on the aldehyde emission from linseed oil paints has been studied in detail. The influence of ambient temperature, air humidity, air exchange rate, light intensity and fatty acid composition on the emission was investigated. The data were evaluated by PCA and PLS. The biplot of PC1 versus PC2 is depicted in Figure 3.93. The distribution of points on the plot illustrates that the aldehyde emission considerably depends on the fatty acid composition of linseed oil. The PLS loading plot is depicted in Figure 3.94. It was concluded from the data that temperature and humidity accelerate the aldehyde formation while light intensity and air exchange rate have a negligible effect on the emission [244]. The concentration and composition of glycerophosphatidic acid, glycerophosphatidylglycerol, glycerophosphatidylethanolamine, glycerophosphatidylcholine, glycerophosphatidylserine, and glycerophosphatidylinositol in particulate matter in outdoor air were measured by normal-phase HPLC and the data were classified by CA. The CA dendogram of fungi common in outdoor air is shown in Figure 3.95. The use of the method was proposed for epidemiological studies and for the investigation of the nature of organic carbon in fine particulate samples [245]. The aromatic compounds in the heavy petroleum fraction were determined by normalphase HPLC using DAD. On account of the strong overlapping of peaks the chromatograms were evaluated by evolving FA and k-means clustering. It was stated that the method can be applied for the determination of aromatic compounds in heavy petroleum [246]. The contamination of mollusc samples by methylmercury (MeHg) and total mercury (HgT) was investigated by atomic fluorescence spectrometry (AFS) and HPLC-AFS. The molluscs included in the experiment were Rapana venosa (Rap), Neverit didyma

3 2

t[2]

1 0 −1 −2 −3 −4

−3

−2

−1

0 t[1]

1

2

3

4

Figure 3.93 Principal component analysis score plot. There are two groups in the data with respect to the fatty acid constituents. Observations with paint 1 () groups are on the left and with paint 2 () groups on the right. Reprinted with permission from ref. [244]. Copyright Elsevier


245

Time 0.80

0.60 c6 c5 w*c[2]

0.40

c2

c1

c3 0.20 Air humidity 0.00 Light Intensity Air exchange

Linoloic acid

−0.20 Temperature −0.40 −0.80 −0.60 −0.40 −0.20 0.00 0.20 0.40 w*c[1]

0.60

0.80

1.00

Figure 3.94 Partial least squares loading plot. The first component describes the difference in fatty acid constituents. c1–c6, n-aldehydes methanal to hexanal. As the Y matrix was transformed with the negative logarithm, the relationship between X and Y is inverted. Reprinted with permission from ref. [244]. Copyright Elsevier G. zene

Cladesporium

A. alternata

E. nigram

G. applanata

A. aneraistifolia 0

1

2

3

Distances

Figure 3.95 Cluster analysis of phospholipid content of fungi common in outdoor air (Euclidean distance, Ward minimum variance method). Reprinted with permission from ref. [245]. Copyright Elsevier

246

Multivariate Methods in Chromatography: A Practical Guide Rescaled Distance Cluster Combine CASE 0 Label Num CRA MER ANU RID KAC REV SCA MYI RAP

5

10

15

20

25

6 7 5 8 9 2 3 4 1

Dendragram using Average Linkage (Between Groups) (a) Rescaled Distance Cluster Combine CASE 0 Label Num RID KAC ANU RAP NEV SCA MYI MER CRA

5

10

15

20

25

8 9 5 1 2 3 4 7 6 (b)

Figure 3.96 Dendogram derived from hierarchical CA. (a) Plot using the data matrix of total mercury (HgT) in molluscs of the first sampling; (b) plot using the data matrix of methylmercury (MeHg) in molluscs of the first sampling. Reprinted with permission from ref. [247]. Copyright Elsevier

(Nev), Scapharca subscretana (Sca), Mytilus edulis (Myt), Amusium (Amu), Crassostrea talienwhanensis (Cra), Meretix meretrix (Mer), Ruditapes philippinarum (Rud), and Mactra veneriformis (Mac). The results were evaluated by CA and PCA. The CA dendograms are depicted in Figure 3.96. The distribution of mollusc types on the dendogram indicates that considerable differences can be observed between the concentrations of HgT and MeHg according to the species investigated. The results of PCA are compiled in Table 3.55.The data entirely support the conclusions drawn from the CA dendogram demonstrating the diverse capacity of molluscs to store HgT and MeHg. The sampling sites are randomly distributed on the biplot of PC1 versus PC2; no correlation was observed between the geographical positions of sampling sites [247]. In addition to organic environmental pollutants the presence, concentration and distribution of elements, inorganic ions and ionic species have also been vigorously investigated. Thus, the determination of 31 elements in PM2.5 particles was performed using inductively coupled plasma mass spectrometry (ICP-MS) and ion chromatography. FA and MLR were

0.800 0.482 0.846 0.977 0.198 0.930 0.988 0.756 0.703 R = −0.050b

Ea

0.171 −0.230 0.197 0.864 0.827

0.883 0.687 0.920

1

HgT for Sample 1

0.724 0.702 0.715 0.910 0.344 0.787 0.964 0.842 0.659 R = 0b

−0.145 0.102

−0.137

0.988 0.411 0.936 0.974

Ea

2 0.386 0.823 0.690 0.609 0.436 0.499 0.879 0.845 0.665

1 −0.758 −0.153 −0.489 −0.734 0.392 0.733 −0.437 −0.358 −0.456

a

2

MeHg for Sample 1

0.720 0.780 0.685 0.481 0.393 0.890 0.772 0.555 0.207 R = 0b

Ea

0.904 −0.611 0.476 −0.309

0.558 0.881 −0.736 0.423

1

2

−0.379 −0.549 0.623 0.270 0.631 −0.573 −0.334

0.640

HgT for Sample 2

Components

Communalities and component score coefficient matrix of PCA

Principal component loadings >0.5 are in bold. Extraction: initial: 1.000. Extraction method: PCA. b Correlation coefficient between PC1 and PC2 (component score covariance matrix). Reprinted with permission from ref. [247].

Rap Nev Sca Myt Amu Cra Mer Rud Mac

Species

Table 3.55

0.459 0.847 0.330 0.268 0.449 0.744 0.497 0.657 0.649 R = −0.050b

Ea

−0.202 0.677 0.229 0.129 0.808 0.612

0.671 0.791

1

2

0.565

−0.831 0.693

−0.470 0.574 −0.477

MeHg for Sample 2

248


Table 3.56 Factor analysis results Variable Al Ca Cd Co Cr Cu Fe K Li Mg Mn Mo Na Ni Pb Ti V Zn Cl F SO4 SO2 Eigenvalue Variance % Source

Factor 1

Factor 2

Factor 3

Factor 4

Factor 5

0.48 0.63 0.85 0.46 −0.02 0.16 0.56 0.93 0.50 0.36 0.46 −0.04 0.69 −0.04 0.80 −0.07 0.06 0.10 0.92 0.17 0.20 0.19 9.3 42.2 Steel works, Sintering plant

0.07 0.01 −0.05 0.46 0.94 0.71 0.14 −0.00 0.04 0.21 0.11 0.94 0.17 0.96 −0.03 0.15 −0.04 0.05 0.01 −0.09 −0.01 −0.010 3.7 16.6 Mechanical engineering factory

0.87 0.56 −0.01 0.03 0.04 0.24 0.04 0.09 0.01 0.70 0.09 0.07 0.10 0.07 0.04 0.86 0.47 0.01 0.08 0.07 −0.08 0.09 2.3 10.5 Soil and street

0.51 0.18 0.24 0.68 −0.03 0.16 0.73 0.27 0.78 0.31 0.81 −0.04 0.57 0.03 0.44 −0.08 0.12 0.88 0.27 0.73 0.14 0.67 1.5 6.7 Steel works, Steel smelting plant

0.19 0.15 0.07 0.10 −0.11 0.18 0.14 0.11 0.10 −0.11 0.15 −0.06 −0.03 −0.01 0.31 −0.02 0.72 0.08 −0.00 −0.10 0.88 0.20 1.3 5.8 Long-range transportation


used for the elucidation of the emission sources. The results of FA are compiled in Table 3.56. The first five factors accounted for 81.1% of the total variance of the original data matrix. Moreover, it was established that the emissions sources responsible for the contamination were long-range transportation (44%), the sintering plant of the steel works (11%), the steel smelting plant of the steel works (3%), and soil and street dust (7%). The parameters of the MLR model predicting PM2.5 are compiled in Table 3.57. The model fitted well to the experimental values the significance level being over 99.9%. The good relationship between measured and calculated values means that the method can be used for prediction purposes [248]. Another study measured the composition of water-soluble atmospheric particulate matter in airborne urban particles using ion chromatography and other physicochemical analytical methods. The data matrix consisting of the sampling site and data as well as the concentration of various inorganic ions was evaluated by PCA, CA and positive matrix factorization. The plot of the first three principal components is depicted in Figure3.97. The scattering of points suggested that marine, industrial, urban and building-related sources may contribute to the pollution. The loadings of the first four principal components are compiled in


249

Table 3.57 Results of multiple regression Variable N = 96, R 2 = 0.88 Intercept Al Cu K Zn SO4

β

Standard error of β

P -value

3.85 22.2 −208 4.42 10.7 231

0.401 5.98 93.6 0.505 3.12 13.4

0.000 0.000 0.028 0.000 0.001 0.000

0.373 5.74 0.507 3.18 13.5

0.000 0.003 0.000 0.001 0.000

N = 96, R 2 = 0.87 Intercept Al K Zn SO4 Reprinted with permission from ref. [248]. Copyright Elsevier.

Table 3.58. The first four principal components explained 36.7, 21.8, 16.3 and 9.3% of the total variance, respectively, indicating the diversity of the data. The CA dendogram is shown in Figure 3.98. The differentiation of the samples does not follow any recognizable pattern indicating that more than one factor influences the similarities and dissimilarities between the sampling sites and data [249]. The ion composition of fresh snow samples was also measured by ion chromatography. Sampling was performed at pre-monsoon and late monsoon periods at geographically diverse sites. The data matrix consisting of the sampling site and times as well as the

3

24.00 1.00

2

PC3

27.00

1

31.00 36.00 32.00 70.00 6.00 6.00

11.00 28.00 10.00 2.00 20.00 47.00 15.00 15.00 7.00 5.0025.00 21.00 33.0040.00 32.00 35.00 13.00 17.00 4.00 3.00 16.00 30.00

0 −1

4

3

2

1 0 PC1

−1

−1 0

1 2 PC2

3

Figure 3.97 Plot of the first three principal components (score supspace, autoscaled data, varimax rotated). Reprinted with permission from ref. [249]. Copyright Elsevier

250


Table 3.58 Loadings encountered for the first four principal components (autoscaled data, varimax rotated) Variable Ca Cl K Mg Na NH4 NO3 SO4 Explained variance (%) Total (%)

PC1

PC2

PC3

PC4

−0.265 0.904 −0.084 0.942 0.741 −0.401 −0.520 −0.594 36.7 36.7

−0.063 0.260 −0.149 0.303 0.523 0.822 0.283 0.725 21.8 58.5

0.705 0.181 0.770 0.168 0.110 −0.104 0.362 −0.019 16.3 74.9

−0.651 0.064 0.484 0.055 −0.136 −0.069 0.206 0.093 9.3 84.1


concentration data of cations and anions was evaluated by PCA. The first three principal components accounted for 47.2, 35.4 and 7.6% of the total variance indicating the basic similarities between the data. The factor loadings are compiled in Table 3.59. It was concluded from the loading values that ion deposits depend mainly on the local transport and monsoon circulation. The data in the map of factor loadings clearly show that ions form

CASE 14 29 33 16 30 3 9 10 15 18 13 25 23 26 24 1 27 4 21 17 28 11 20 19 2 5 8 7 31 6 32

0

5

10

15

20

25

C3B +

C2 +

+ C2

C1 +

Figure 3.98 Dendogram obtained after CA (autoscaled data, squared Euclidean distance, complete linkage). Reprinted with permission from ref. [249]. Copyright Elsevier


251

Table 3.59 Results of PCA carried out on all data: factor loadings, eigenvalues and percentage explained variance

Cl− NO− 3 SO2− 4 Na+ NH+ 4 K+ 2+ Mg Ca2+ Eigenvalue Explained variance (%)

F1

F2

F3

−0.145 0.903 0.926 −0.168 0.853 −0.010 0.666 0.939 3.77 47.2

0.948 −0.143 −0.095 0.899 0.197 0.877 0.535 −0.010 2.83 35.4

−0.258 −0.076 −0.213 −0.390 0.160 0.408 0.324 −0.196 0.61 7.6


two separate clusters. The distribution of sampling sites does not follow the geographical origin but the time of sampling according to the monsoon circulation [250]. Ion chromatography has also been employed for the analysis of ionic species in the watersoluble fraction of total suspended particulate (TSP) and in dry deposit flux (DDF). The data matrix consisting of 40 samples taken at various sites and time and the concentration of the eight ions was evaluated by PCA. The loadings are listed in Tables 3.60 and 3.61. It was concluded from the data that the concentration of pollutants considerably depends on the season, and the quality of soil, secondary aerosol and marine salts also influence the ion concentration [251]. Table 3.60 Factor analysis of DDF of soluble ions at TH and Wuchi sampling sites ( N = 40) TH Species Cl− NO− 3 SO2− 4 NH+ 4 Mg2+ Ca+ Na+ Eigenvalue Proportion of variance (%) Cumulative (%) Source indicated

Wuchi

Factor 1

Factor 2

Factor 1

Factor 2

Factor 3

0.71 0.42 0.73 0.54 0.71 0.70 0.59 2.67 38.13 38.13 Soil, secondary aerosol, marine salt

— 0.80 — 0.630 0.194 — — 1.46 20.86 59.00 Primary aerosol

0.30 0.73 0.71 — 0.16 0.42 0.61 2.03 29.00 29.00 Secondary aerosol

0.32 − — 0.40 0.72 0.70 0.22 1.35 19.32 48.32 Soil

0.76 0.10 — 0.34 — — 0.43 1.16 16.51 64.83 Marine salt

Only factor loading values with moduli >0.1 are presented. Factor loading values with moduli >0.7 are in bold. Reprinted with permission from ref. [251]. Copyright Elsevier.

252


Table 3.61 Factor analysis of TSP of soluble ions at TH and Wuchi sampling sites ( N = 40) TH Species Cl− NO− 3 SO2− 4 NH+ 4 Mg2+ Ca+ Na+ Eigenvalue Proportion of variance (%) Cumulative (%) Source indicated

Wuchi

Factor 1

Factor 2

Factor 1

Factor 2

0.58 0.86 0.85 0.87 — — 0.81 3.95 56.44 56.44 Secondary aerosol, marine salt

0.50 0.18 0.21 — 0.78 0.74 0.08 1.05 15.00 71.44 Soil

0.44 0.88 0.86 0.92 0.11 0.20 0.23 3.56 50.81 50.81 Secondary aerosol

0.61 0.14 — 0.18 — 0.77 0.78 1.32 18.92 69.74 Marine salt, soil

Only factor loading values with moduli >0.1 are presented. Factor loading values with moduli >0.7 are in bold. Reprinted with permission from ref. [251]. Copyright Elsevier.

3.2.3.14

Miscellaneous HPLC Applications

A considerable number of other chromatographic measurements has also been evaluated by multivariate mathematical-statistical methods. As various chromatographic techniques are applied for a wide variety of purposes, the application of chemometrics is also widespread in chromatographic practice. Thus, polysaccharides in ancient wall paintings have been characterized by anionexchange HPLC after acidic hydrolysis. Galactose, fucose, arabinose, rhamnose, mannose, xylose, glucuronic acid, and glucose were separated and detected with pulsed amperometry. PCA has been employed for the classification of raw materials according to the concentration of sugars in the samples. The first three principal components accounted for 93.5% of the total variance. It was stated that the combination of IEC with PCA makes the identification of the polysaccharide binders in the samples possible [252]. An LSER study was carried out for the modelling of the retention of 28 unrelated neutral and ionizable analytes on a polymeric column using methanol–water mixtures as mobile phases. MLR calculations were applied for the elucidation of the relationship between the solute descriptors and chromatographic parameters. The equations computed for neutral and ionic species are: Neutral analytes log kw = −0.76 − 0.22 × E + 0.18 × S − 0.82 × A − 4.75 × B + 6.69 × V SD = 0.23,

R = 0.983,

F = 63

m = −0.23 + 0.17 × E − 0.36 × S − 0.42 × A − 4.31 × B + 6.34 × V SD = 0.11,

R = 0.995,

F = 206

(3.105)


253

Ionizable analytes log kw = −2.29 − 1.30 × E + 1.74 × S − 0.78 × A − 3.07 × B + 1.80 × V SD = 0.51,

R = 0.882,

F=6

m = 3.35 + 1.13 × E − 5.08 × S − 1.05 × A + 2.74 × B − 2.71 × V SD = 0.86,

R = 0.838,

F=4

(3.106)

The application of the global solvation model resulted in: log k = (1.35 − 1.85) + (1.89 − 1.95)E + (−1.33 + 1.19)S + (−1.44 + 0.8)A +(−4.83 + 4.25)B + (4.03 − 2.98)V + log[1 − D(1 − f)] n = 789,

SD = 0.17,

R = 0.967,

F = 1027

(3.107)

where E is the excess molar refraction, A is the effective hydrogen-bond acidity, V is the McGowan molar volume, S is the dipolarity/polarizability, B is the hydrogen-bond basicity, D is the the degree of ionization of the solute; f is a retention derived parameter defined as the ratio of the retention factors of the ionized species and the neutral species of the solute, and is the volume fraction of the organic modifier in the mobile phase. It was concluded from the computations that the global LSER can be successfully applied for the study of the retention behaviour of both neutral and ionizable analytes in RP-HPLC [253]. Blue ballpoint inks were analysed by RP-HPLC using simultaneous detection at four wavelengths. Both PCA and LDA were employed for the differentiation between the inks. The first two principal components accounted for 55.4 and 20.1% of the total variance. It was found that PCA differentiated correctly 96.4% of the pen pairs. The classification obtained by LDA was slightly better that of PCA [254]. A SEC technique coupled with viscometry has been employed for the molecular structure characterization of linear and branched polystyrene blends. The coexistence of linear and branched polystyrenes meant that the chromatograms often contained overlapping peaks. A mathematical procedure was developed for the deconvolution of these peaks. It was stated that the computation technique makes possible the calculation of the purity of the sample, the intrinsic viscosity of the polystyrene sample and the molecular mass distribution [255]. Information theory and FA have been employed for the optimization of a two-dimensional HPLC system for the separation of oligostyrene isomers. It was established that the best results can be obtained on an ODS column with methanol as mobile phase (first dimension) and a carbon clad zirconia column with acetonitrile as mobile phase (second dimension). It was further found that the use of a heat cutting technique enhances considerably the performance of the system. The method allows the separation of 27 of the 32 oligostyrene isomers [256–258]. A novel computation method has been developed for the easy determination of the kinetic parameters of enzyme catalysed reactions. A new steady-state constraint for multivariate curve resolution-alternating least squares (MCR-ALS) was introduced. It was stated that the computation method allows the determination of the Michaelis constant, the maximum velocity and the inhibition constant [259].

254


Table 3.62 Results of eigenvalues, logarithm of eigenvalues and eigenvalue ratio methods

True rank Function EV Log of EV Ratio of EV

Data set 1

Data set 2

Data set 3

3

3

3

2 3 3

2 3 3

3 3 3

Data set 4

Data set 5

3 3 Components identified 3 2 3 — 3 3

Data set 6

Data set 7

Data set 8

4

7

8

3 — 5

7 7 7

5 — 9

EV, eigenvalue. Reprinted with permission from ref. [260]. Copyright Elsevier.

A wide variety of chemometrical techniques has been employed for the determination of the significant components in HPLC-NMR spectroscopy. Non-homologous series of 12 unrelated analytes were applied as model compounds in RP-HPLC. The performance of the following computational methods have been tested: eigenvalues, log eigenvalues, eigenvalue ratios computed by PCA; error indicator functions (residual sum of squares, residual standard deviation, ratio of successive residual standard deviation, root mean square error, imbedded error, factor indicator functions, scree test and Exner function), together with their ratio of derivatives, F tests (Malinowski and Faber–Kowalski) and modified Faber–Kowalski; cross validation, morphological score, purity-based approaches including orthogonal projection approach (OPA) and SIMPLISMA, correlation and derivative plots, evolving PCA (EPCA) and evolving principal component innovation analysis (EPCIA), and subspace comparison. The results of eigenvalues, logarithm of eigenvalues and eigenvalue ratio methods are compiled in Table 3.62. The data in Table 3.62 indicate that the application of the logarithm of eigenvalues or their ratio produce better results than the application of the original eigenvalues. It was concluded from the data that the determination of the number of significant components depends considerably on the method applied; therefore, the results of the different calculation procedures have to be treated cautiously [260]. The acid–base catalysed hydrolysis of (3-glycidoxypropyl)trimethylsilane (GPTMS) and (3-aminopropyl)triethoxysilane (APTES) was followed by RP-HPLC with inductively coupled plasma atomic emission spectrometry (ICP-AES) detection. The scheme of the hydrolysis process of trialkoxysilanes is given by: k3 k2 k1 R − Si(OR )3 −−−→ Si OR 2 (OH) −−−→ R − Si(OR ) (OH)2 −−−→ R − Si(OH)3 A nonlinear multivariate equation was fitted to the experimental data to determine the rate constant of the hydrolysis reactions and for the assessment of the catalytic constants for acidand base-catalysed hydrolysis. The rate constants for consecutive hydrolysis of GPTMS are compiled in Table 3.63. The low standard deviation of the measured rate constants indicates the reliability of the RP-HPLC analysis. The catalytic constants determined under acidic and alkaline conditions are compiled in Table 3.64. It was concluded from the results that the method is suitable for the study of the hydrolysis rate of trialkoxysilanes and can be employed for the assessment of the effect of pH on the decomposition rate [261].


255

Table 3.63 Rate constants for consecutive hydrolysis reactions of GPTMS obtained by nonlinear regression analysis (24.5◦ C unless indicated otherwise) pH

Buffer (mM)a

k1 (×10−4 s−1 )

k2 (×10−4 s−1 )

k3 (×10−4 s−1 )

19.6 196 10.2 64.5 3.11 49.2 2.49 25.0 2.18 2.45 2.02

0.399(0.007)b 1.37(0.02) 0.852(0.011) 1.74(0.04) 2.71(0.07) 4.55(0.10) 7.51(0.14) 9.40(0.14) 14.3(0.4) 4.33(0.17) 35.7(0.8)

1.51(0.15) 2.91(0.14) 4.66(0.50) 5.81(0.81) 14.5(3.1) 19.2(2.7) 50.7(8.3) 51.4(5.7) 93.9(19.2) 25.4(6.7) 239(50)

2.75(0.42) 6.17(0.51) 8.38(1.52) 11.0(2.4) 30.2(12.2) 42.5(11.2) 928(25.9) 108(22) 158(52) 41.0(18.6) 407(141)

6.97 7.00 7.59 7.57 8.19 8.18 8.71 8.71 8.94 8.97c 9.00d

Tris/Tris-HCl pK a = 8.21 (25◦ C, 0.2 M). Uncertainties reflect 95% confidence interval. c 10.1 ◦ C. d 36.9 ◦ C. Reprinted with permission from ref. [261]. Copyright Elsevier.

a

b

The classification and regression trees (CART) method was developed and employed for the prediction of enantioselectivity. Chirality codes were applied as explanatory variables. Chirality codes were defined as a set of molecular descriptors that combine different parameters and are able to distinguish between enantiomers. Enantiomers were separated on two chiral stationary phases (teicoplanin and bianthracene-based). It was established that the chiral descriptors have a high predictive power and they can be used for the prediction of retention of enantiomers on chiral stationary phases. It was further stated that the reliability of the CART method is similar to that of neural network [262]. A QSRR study was performed for the assessment of the best equation describing the retention of ring-substituted phenol derivatives in RP-HPLC. The calculated log P value, and the dissociation constants derived from atomic partial charge computed by three different Table 3.64 Catalytic constants for acid- and base-catalysed reactions of GPTMS and APTES GPTMS Catalyst (units)

k1

k2

APTES k3

k1

k2

k3

H3 O+ (M−1 s−1 ) 131(5)a 310(21) 354(54) 23.1(6.0) 71.1(22.3) 132(82) CH3 CO2 H (×10−3 9.87(2.78) 62.9(10.9) 89.6(28.5) 2.43(1.43) 15.9(5.3) 51.9(19.6) M−1 s−1 ) 144(11) 957(66) 1850(160) 125(1) 1130(110) OH− (M−1 s−1 ) Tris (×10−3 9.59(2.53) 15.0(14.9) 65.9(35.6) 0.32(0.14) 4.75(4.99) M−1 s−1 ) a

Uncertainties reflect 95% confidence interval from regression analysis. Reprinted with permission from ref. [261]. Copyright Elsevier.

256


methods were included in the analysis. In the majority of cases a significant univariable relationship was found between the pKa values and the atomic partial charges. The correlation between the sum of Hammett’s constant was described by bivariate equations: B = (−7.260 × pK a − 129.947) × (partial charge of hydrogen) (3.108) B = (6.137 × pK a − 215.626) × (partial charge of oxygen) (3.109) where B is a constant for individual groups of compounds, and is Hammett’s constant measured by LC or titration. It was concluded from the calculation that the procedure is suitable for the prediction of the retention of this class of analytes in a RP-HPLC system at a given pH value [263]. A similar study was performed and it was established that the retention of phenol derivatives in RP-HPLC at various pH can be predicted by the dissociation constant and the atomic partial charge [264]. The retention behaviour of benzoic acid derivatives was predicted by the dissociation constants and the conclusions were the same as in two references [263,264] by Hanai [265].

References [1] Sherma, J., and Fried, B. Eds. Handbook of Thin-Layer Chromatography, 3rd edition, Marcel Dekker, Inc., NewYork, 2003. [2] Hahn-Deinstrop, E. Applied Thin-Layer Chromatography – Best Practice and Avoidance of Mistakes, Wiley-VCH,Weinheim, 2000. [3] Sherma, J. Anal. Chem. 76 (2004) 3251–3261. [4] Siouffi, A. M. In: A Century of Separation Science (Ed. Issaq,H.J.), Marcel Dekker, Inc., New York, 2002, pp. 69–85. [5] Pyka, A., and Bober, K J. Plan. Chromatogr. 15 (2002) 332–340. [6] Cserháti, T., Forgács, E., Morais, H., and Mota, T. J.Biochem. Biophys. Methods 45 (2000) 221–229. [7] Forgács, E., Cserháti, T., Balogh, S., Kaliszan, R., Haber, P., and Nasal, A. Biomed. Chromatogr. 15 (2001) 348–355. [8] Iqbal, S. H., Rohrschneider, L., Rathore, H. S., and Mital, S. Ind. J. Chem. Technol. 7 (2000) 1–6. [9] Pyka, A. J. Plan. Chromatogr. 14 (2001) 152–159. [10] Pyka, A. J. Plan. Chromatogr. 17 (2004) 275–279. [11] Pyka, A., and Klimczok, W. J. Plan. Chromatogr. 18 (2005) 300–304. [12] Pyka, A., and Sliwiok, J. J. Liq. Chromatogr. Rel. Technol. 27 (2004) 785–798. [13] Vrakas, D., Panderi, I., Hadjipavlou-Litina, D., and Tsantili-Kakoulidou, A. QSAR Comb. Sci. 24 (2005) 254–260. [14] Pyka, A., and Niestrój, A. J. Liq. Chromatogr. Rel. Technol. 24 (2001) 2399–2413. [15] Pyka, A. and Sliwiok, J. J. Chromatogr. A 935 (2001) 71–76. [16] Pyka, A., Bober, K., Gurak, D., and Niestroj, A. Acta Pol. Pharm. 59 (2002) 87–91. [17] Pyka, A., and Bober, K. J. Liq. Chromatogr. Rel. Technol. 25 (2002) 1301–1315. [18] Sarbu, C., Djakovic-Sekulic, T., and Perisic-Janjic. J. Pharm. Biomed. Anal. 30 (2002) 739– 745. [19] Djakovic-Sekulic, T. L., Sarbu, C., and Perisic-Janjic, N. U. J. Plan. Chromatogr. 18 (2005) 432–436.

Liquid Chromatography [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40] [41] [42] [43] [44] [45] [46] [47] [48] [49] [50] [51] [52] [53] [54] [55] [56] [57] [58] [59]

257

Sarbu, C., and Todor, S. J. Chromatogr. A 822 (1998) 263–269. Cserháti, T., and Forgács, E. Eur. J. Pharm. Biopharm. 51 (2001) 39–44. Forgács, E., and Cserháti, T. Anal. Lett. 37 (2004) 1897–1908. Gere-Pászti, E., Cserháti, T., Forgács, E., Deyl, Z., Miksik, I., Eckhardt, A., and Illés, Z. J. Liq. Chromatogr. Rel. Technol. 28 (2005) 2619–2632. Forgács, E., and Cserháti, T. J. Sep. Sci. 25 (2002) 101–104. Cserháti, T., and Forgács, E. Amino Acids 28 (2005) 99–103. Pyka, A., and Bober, K. Ind. J. Chem. 42A (2003) 1360–1367. Niestroj, A., Pyka, A., Klupsch, J., and Sliwiok, J. J. Liq. Chromatogr. Rel. Technol. 27 (2004) 2449–2461. Pyka, A., and Bober, K. Riv. Ital. Sost. Grasse 78 (2001) 477–482. Pyka A. J. Plan. Chromatogr. 14 (2001) 439–444. Niestroj, A., Pyka, A., and Sliwiok, J. J. Plan. Chromatogr. 15 (2002) 177–182. Pyka, A., and Bober, K. J. Plan. Chromatogr. 15 (2002) 59–66. Pyka, A., and Dolowy, M. Acta Pol. Pharm. 61 (2004) 407–413. Sarbu, C., Kuhajda, K., and Kevresan, S. J. Chromatogr. A 917 (2001) 361–366. Pyka, A., and Dolowy, M. I. J. Liq. Chromatogr. Rel. Technol. 26 (2003) 2741–2750. Pyka, A., and Dolowy, M. J. Liq. Chromatogr. Rel. Technol. 28 (2005) 297–311. Pyka, A., and Dolowy, M. J. Liq. Chromatogr. Rel. Technol. 28 (2005) 1765–1775. Pyka, A., Dolowy, M., and Gurak, D J. Liq. Chromatogr. Rel. Technol. 28 (2005) 2705–2717. Pyka, A., Dolowy, M., and Gurak, D. J. Plan. Chromatogr. 18 (2005) 465–470. Pyka, A. J. Plan. Chromatogr. 16 (2003) 131–136. Pyka, J. Plan. Chromatogr. 17 (2004) 58–60. Perisic-Janjic, N., Djakovic-Sekulic, T. Lj., and Popov-Pergal, K. J. Plan. Chromatogr. 17 (2004) 192–195. Janicka, M., Oscik-Mendyk, B., and Tarasiuk, B. J. Plan. Chromatogr. 17 (2004) 186–192. Janicka, M., Perisic-Janjic, N. U., and Rozylo, J. K. J. Plan. Chromatogr. 17 (2004) 468–475. Gozalbes, R., Julián-Oriz, J. V. de, Antón-Fos, G. M., Gálvez-Alvarez, J., and GarciaDomenech, R. Chromatographia 51 (2000) 331–337. Pyka, A., and Miszczyk, M. Chromatographia 61 (2005) 37–42. Zhang, L., Zhang, M., Wang, L. X., and Wang, Q. S. Chromatographia 52 (2000) 306–308. Forgács, E., Cserháti, T., and Barta, I. Amino Acids 18 (2000) 69–79. Cserháti, T., Forgács, E., and Oros, Gy. J. Liq. Chromatogr. Rel. Technol. 23 (2000) 411–421. Cserháti, T., Forgács, E., Darwich, Y., Oros, Gy., and Illés, Z. J. Inclusion Phenom. Macrocycl. Chem. 42 (2002) 235–240. Cserháti, T., and Forgács, E. Pest. Manag. Sci. 56 (2000) 809–812. Cserháti, T., Forgács, E., Deyl, Z., Miksik, I., and Eckhardt, A. J. Chromatogr. A 910 (2001) 137–145. Forgács, E., Cserháti, T., Deyl, Z., Miksik, I. and Eckhardt, A. J. Chromatogr. A 917 (2001) 287–295. Cserháti, T., Forgács, E., and Illés, Z. J. Liq. Chromatogr. Rel. Technol. 26 (2003) 2751–2761. Cserháti, T., and Forgács, E. J. Liq. Chromatogr. Rel. Technol. 26 (2003) 2303–2313. Rozylo, J. K., Matysiak, J.,Niewiadomy, A., and Zabinska, A. J. Plan. Chromatogr. 13 (2000) 176–181. Zabinska, A. Rozylo, J. K., Matysiak, J., and Niewiadomy A. J. Plan. Chromatogr. 13 (2000) 420–425. Janicka, M., Kwietniewski, L., and Matysiak, J. J. Plan. Chromatogr. 13 (2000) 285–289. Janicka, M. Chromatographia 57 (2003) 395–403. Nieto, M. J., Alovero, F. L., Manzo, R. H., and Mazzieri, M. R. Eur. J. Med. Chem.40 (2005) 361–369.

258


[60] Sarbu, C., Casoni, D., Darabantu, M., and Maiereanu, C. J. Pharm. Biomed. Anal. 35 (2004) 213–219. [61] Pyka, A., Kepczynska, E., and Bojarski, J. Ind. J. Chem. 42A (2003) 1405–1413. [62] Bjelland, T., and Thorseth, I. H. Chem. Geol. 192 (2002) 81–98. [63] Csomós, E., Héberger, K., and Simon-Sarkadi, L. J. Agric. Food Chem. 50 (2002) 3768– 3774. [64] Santos, S. C., Costa, W. F., Ribeiro, J. P.,Guimaraes, D. O., Ferri, P. H., Ferreira, H. D., and Seraphin, J. C. Fitoterapia 73 (2002) 292–299. [65] Izzotti, A., Bagnasco, M., Cartiglia, C., Longobardi, M., Camoirano, A., Tampa, E., Lubet, R. A., and De Flora, S. Mutat. Res. 591 (2005) 212–223. [66] Laurinavicius, S., Kakela, R., Bamford, D. H., and Somerharju, P. Virology 326 (2004) 182–190 [67] Rossi, D.T., and Sinz, M.W. Eds. Mass Spectrometry in Drug Discovery, Marcel Dekker, Inc., New York, 2002. [68] Cazes, J., and Scott, R.P.W. Chromatography Theory, Marcel Dekker, Inc., New York, 2002. [69] Cazes, J. Ed. Encyclopedia of Chromatography, Marcel Dekker, Inc., New York, 2001. [70] Poole, C.F. The Essence of Chromatography, Elsevier, Amsterdam, 2003. [71] Fritz, J.S., and Hahn-Deinstrop, D.T.G. Eds. Ion Chromatography, Wiley-VCH, Weinheim, 2000. [72] Kromidas, S. Practical Problem Solving in HPLC, Wiley-VCH, Weinheim, 2000. [73] Nollet, L.M.L. Ed. Food Analysis by HPLC, 2nd edition, Marcel Dekker, Inc., New York, 2000. [74] Gooding, K.M., and Regnier, F.E. Eds. HPLC of Biological Macromolecules, 2nd edition, Marcel Dekker, Inc., New York, 2002. [75] Aguilar, M.-I. Ed. HPLC of Peptides and Proteins: Methods and Protocols. Methods in Molecular Biology, Vol. 251 (Series Ed. Walker, J.M.), Humana Press, Totowa, State NJ, 2004. [76] Kim, I. W., Lee, H. S., Lee, Y. K., Jang, M. D., and Park, J. H. J. Chromatogr. A 915 (2001) 35–42. [77] Diana, J., Visky, D., Roets, E., and Hoogmartens, J. J. Chromatogr. A 996 (2003) 115–131. [78] Visky, D., Heyden, Y. V., Iványi, T., Baten, P., De Beer, J., Kovács, Zs., Noszál, B., Dehouck, P., Roets, E., Massart, D. L., and Hoogmartens, J. J. Chromatogr. A 1012 (2003) 11–29. [79] Iványi, T., Heyden, Y. V., Visky, D., Baten, P., De Beer, J., Lázár, I., Massart, D. L., Roets, E., and Hoogmartens, J. J. Chromatogr. A 954 (2002) 99–114. [80] Euerby, M. R., and Petersson, P. J. Chromatogr. A 1088 (2005) 1–15. [81] Van Gyseghem, E., Jimidar, M., Sneyers, R., Redlich, D., Verhoeven, E., Massart, D. L., and Heyden, Y. V. J. Chromatogr. A 1042 (2004) 69–80. [82] Forgács, E., and Cserháti, T. Chem. Intell. Lab. Syst. 50 (2000) 115–119. [83] Forgács, E., and Cserháti, T. Anal. Sci. 17 (2001) 307–312. [84] Whelan, T. J., Gray, M. J., Slonecker, P. J., Shalliker, R. A. and Wilson, M. A. J. Chromatogr. A 1097 (2005) 148–159. [85] Pirogov, A. V., Platonov, M. M., and Shpigun, O. A. J. Chromatogr. A 850 (1999) 53–63. [86] Turowski, M, Morimoto, T., Kimata, K., Monde, H., Ikegami, T., Hosoya, K., and Tanaka, N. J. Chromatogr. 911 (2001) 177–190. [87] Vonk, E. C., Lewandowska, K., Claessens, H. A., Kaliszan, R., and Cramers, C. A. J. Sep. Sci. 26 (2003) 777–792. [88] Put, R., Van Gyseghem, Coomans, D., and Heyden, Y. V. J. Chromatogr. A 1096 (2005) 187– 198. [89] Forlay-Frick, P., Fekete, J., and Héberger, K. Anal. Chim. Acta 536 (2005) 71–81.


259

[90] Detroyer, A., Schoonjans, V., Questier, F., Heydan, Y. V., Borosy, A. P., Guo, Q., and Massart, D. L. J. Chromatogr. A 897 (2000) 23–36. [91] Le Mapihan, K., Vial, J., and Jardy, A. J. Chromatogr. A 1030 (2004) 135–147. [92] Van Gyseghem, E., Van Hemelryck, S., Daszykowski, M., Questier, F., Massart, D. L., and Van der Heyden, Y. J. Chromatogr. A 988 (2003) 77–93. [93] Van Gyseghem, E, Dejaegher, B., Put, R., Forlay-Frick, P., Elkihel, A., Daszykowski, M., Héberger, K., Massart, D. L., and Van der Heyden. J. Pharm. Biomed. Anal. 41 (2006) 141– 151. [94] Van Gyseghem, E., Crosiers, I., Gourvénec, S., Massart, D. L., and Van der Heyden, Y. J. Chromatogr. A 1026 (2004) 117–128. [95] Van Gyseghem, E., Jimidar, M., Sneyer, R., Redlich, D., Verhoeven, E., Massart, D. L., and Van der Heyden, Y J. Chromatogr. A 1074 (2005) 117–131. [96] Forlay-Frick, P., Van Gyseghem, E., Héberger, K., and Van der Heyden, Y. Anal. Chim. Acta 539 (2005) 1–10. [97] Baczek, T., Markuszewski, M., Kaliszan, R., Van Straten, M. A. and Claessens, H. A. J. High Resolut. Chromatogr. 23 (2000) 667–676. [98] Lopes, P. A., Santos, M. C., Vicente, L., and Viegas-Crespo, A. M. Clin. Chim. Acta 348 (2004) 49–55. [99] Govorukhina, N. I., Reijmers, T. H., Nyangoma, S. O., van der Zee, A. G. J., Jansen, R. C., and Bischoff, R. J. Chromatogr. A 1120 (2006) 142–150. [100] Yang, J., Xu, G., Kong, H., Zheng, Y., Pang, T., and Yang, Q. J. Chromatogr. B 780 (2002) 27–33. [101] Bohnstedt, K. C., Karlberg, B., Wahlund, L.-O., Jönhagen, M. E., Basun, H., and Schmidt, S. J. Chromatogr. B 796 (2003) 11–19. [102] Yang, Y., Xu, G., Zheng, Y., Kong, H., Pang, T., Lv, S., and Yang, Q. J. Chromatogr. B 813 (2004) 59–65. [103] Williams, R. E., Lenz, E. M., Evans, J. A., Wilson, I. D., Granger, J. H., Plumb, R. S., and Stumpf, C. L. J. Pharm. Biomed. Anal. 38 (2005) 465–471. [104] Heidbreder, C. A., Weiss, I. C., Domeney, A. M., Pryce, C., Homberg, J., Hedou, G., Feldon, J., Moran, M. C., and Nelson P. Neuroscience 100 (2000) 749–768. [105] Silva, M. F., Chipre, L. F., Raba, J., and Luco, J. M. Chromatographia 53 (2001) 392–400. [106] Bungers, J., Cummings, J. J., Ebert, M. B., Federici, M. M., Gledhill, L., Gulati, D., Hilliard, G. M., Jones, B. H., Lee, K. R., Mozdzanowski, J., Naimoli, M., and Burman, S. J. Pharm. Biomed. Anal. 21 (2000) 1099–1128. [107] Baczek, T., Wiczling, P., Marszall, M., Vander Heyden, Y.,and Kaliszan,.J. Proteome Res. 4 (2005) 555–563. [108] Baczek, T. J. Sep. Sci. 29 (2006) 547–554. [109] Kaliszan, R., Baczek, T., Cimochowska, A., Juszczil, P., Wisniewska, K., and Grzonka, Z. Proteomics 5 (2005) 409–415. [110] Forgács, E., and Cserháti, T. J. Chromatogr. A 948 (2002) 69–75. [111] Bylund, D., Samskog, J., Markides, K. E., and Jacobsson, S. P. J. Am. Soc. Mass Spectrom. 14 (2003) 236–242. [112] Mortuza, G. B., Neville, W. A., Delaney, J., Waterfield, C. J., and Camilleri, P. Biochim. Biophys. Acta 1631 (2003) 136–146. [113] Luo, H., and Cheng, Y.-K. QSAR Comb. Sci. 24 (2005) 968–975. [114] Chen, F., Durn, A. L., Blount, J. W., Sumner, L. W., and Dixon, R. A. Phytochemistry 64 (2003) 1013–1021. [115] Semmar, N., Jay, M., Farman, M., and Chemli, R. Biochem. Syst. Ecol. 33 (2005) 187–200. [116] Grass, S., Zidorn, C., Blattner, F. R., and Stuppner, H. Phytochemistry 67 (2006) 122–131. [117] Zidorn, C., and Stuppner, H. Biochem. Syst. Ecol. 29 (2001) 827–837.

260


[118] Ando, T., Takahashi, M., Nakajima, T., Toya, Y., Watanabe, H., Kokubun, H., and Tatsuzawa, F. Phytochemistry 65 (2004) 2219–2227. [119] Mateo, J. J., Mateo, R., and Jiménez, M. Int. J. Food Microbiol. 72 (2002) 115–123. [120] Thomsen, M., Lassen, P., Dobel, S., Hansen, P. E., Carlsen, L., and Mogensen, B. B. Chemosphere 49 (2002) 1327–1337. [121] PersonNameProductIDLa MontagneLa Montagne, M. G., Michel, F. C. Jr, Holden, P. A., and Reddy, C. A. J. Microbiol. Meth. 49 (2002) 255–264. [122] Söderström, C., Borén, H., and Krantz-Rulcker, C. Int. J. Food Microbiol. 97 (2005) 247–257. [123] Fang, J., Barcelona, M. J., and Semrau, J. D. FEMS Microbiol. Lett. 189 (2000) 67–72. [124] Metaxatos, A., and Ignatiades, PlaceNameplaceL. NameEst. PlaceTypeCoast. Shelf Sci. 55 (2002) 415–426. [125] Bravo, I., Fernández, M. L., Ramilo, I., and Martinez, A. Toxicon 39 (2001) 1537–1545. [126] Obayashi, Y., Tanoue, E., Suzuki, K., Handa, N., Nojiri,Y., and Wong, C. S. Deep-Sea Res. I 48 (2001) 439–469. [127] Redalje, D. G., Lohrenz, S. E., Verity, P. G., and Flagg, C. N. Deep-Sea Res. II 49 (2002) 4511–4531. [128] Wulff, A., and Wängberg, S.-A. Deep-Sea Res. II 51 (2004) 2701–2713. [129] Yan, S., Luo, G., Wang, Y., and Cheng, Y. J. Pharm. Biomed. Anal. 40 (2006) 889–895. [130] Yan, S., Xin, W.-f., Luo, G., Wang, Y., and Cheng, Y.-y. J. Chromatogr. A 1090 (2005) 90–97. [131] Wan, J. B., Yang, F. Q., Li, S. P., Wang, Y. T., and Cui, X. M. J. Pharm. Biomed. Anal. 41 (2006) 1596–1601. [132] Woo, S.-S., Song, J.-S., and Lee, J.-Y. Phytochemistry 65 (2004) 2751–2761. [133] Yan, F., Liang, Z., Jianna, C., Zhengtao, W., Losahan, X., and Zhengxing, Z. Talanta 53 (2001) 1155–1162. [134] Gan, F., and Ye, R. J. Chromatogr. A 1104 (2006) 100–105. [135] Xiaohui, F., Yi, W.,and Yiyu, C. J. Pharm. Biomed. Anal. 40 (2006) 591–597. [136] Hendriks, M. M. W. B., Cruz-Juarez, L., De Bont, D., and Hall, R. D. Anal. Chim. Acta 545 (2005) 53–64. [137] Iizuka, A., Iijima, O. T., Kondo, K., Itakura, H., Yoshie, F., Miyamoto, H., Kubo, M., Higuchi, M., Takeda, H., and Matsumuya, T. J. Ethnopharmacol. 91 (2004) 89–94. [138] Gong, F., Liang, Y.-Z., Fung, Y.-S., and Chau, F.-T. J. Chromatogr. A 1029 (2004) 173–183. [139] Forgács, E., Cserháti, T. Miksik, I., Eckhardt, A., and Deyl, Z. J. Chromatogr. B 8000 (2004) 259–262. [140] Forgács, G., Forgács, E., and Cserháti, T. Chem. Int. Lab. Syst. 72 (2004) 269–275. [141] Matysia, J., and Niewladomy, A. JAOAC Int. 87 (2004) 579–586 [142] Nasal, A., Wojdelko, A., Baczek, T., Kaliszan, R., Cybulski, M., and Chilmonczyk, Z. J. Sep. Sci. 25 (2002) 273–279. [143] Detroyer, A., Van der Heyden, Y., Cambré, I., and Massart, D. L. J. Chromatogr. A 986 (2003) 227–238. [144] Wiberg, K., Andersson, M., Hagman, A., and Jacobsson, J. Chromatogr. A 1029 (2004) 13–20. [145] StateMontana, M. P., Pappano, N. B., Debattista, N. B., Raba, J., and Luco, J. M. Chromatographia 51 (2000) 727–735. [146] Vrakas, D., Giaginis, C., and Tsantili-Kakoulidou, A. J. Chromatogr. A 1116 (2006) 158–164. [147] Verhoeckx, K. C. M., Bijlsma, S., Jespersen, S., Ramaker, R., Verheij, E. R., Witkamp, R. F., van der Greef, J., and Rodenburg, R. J. T. Int. Immunopharmcol. 4 (2004) 1499–1514. [148] Verdu-Andrés, J., Herráez-Hernández, R., and Campins-Falcó, P. J. Chromatogr. A 930 (2001) 95–107. [149] Chan, E. C. Y., Tan, W. L., Ho, P. C., and Fang, L. J. J. Chromatogr. A 1072 (2005) 159–168.


261

[150] Stensrud, G., Sande, S. A., Kristensen, S., and Smistad, G. Int. J. Pharm. 198 (2000) 213– 228. [151] Quinones-Torrelo, C., Sagrado. S., Villanueva-Camanas, R. M., and Medina-Hernández, M. J. Biomed. Chromatogr. 18 (2004) 427–435. [152] Kallithraka, S., Arvanitoyannis, I. S., Kefalas, P., and El-Zajouly, A. Food Chem. 73 (2001) 501–514. [153] Rodriguez-Delgado, M.-A., Gonzólez-Hernández, G., Conde-González, J.-E., and PérezTrujillo, J.-P. Food Chem. 78 (2002) 523–532. [154] Boselli, E., Minardi, M., Giomo, A., and Frega, N. G. Anal. Chem. Acta 563 (2006) 93–100. [155] Nikfjardam, M. S. P., Márk, L., Avar, P., Figler, M., and Ohmacht, R. Food Chem. 98 (2006) 453–462. [156] Atanasova, V., Fulcrand, H., Cheynier, V., and Moutounet, M. Anal. Chim. Acta 458 (2002) 15–27. [157] Hernández, T., Estrella, I., Carlavilla, D., Martin-Alvarez, P. J., and Moreno-Arribas, M. V. Anal. Chim. Acta 563 (2006) 116–125. [158] Fernández-Pachón, M. S., Villano, D., Troncoso, A. M., and Garcia-Parilla, M. C. Anal. Chim. Acta 563 (2006) 101–108. [159] Alonso-Salces, R. M., Herrero, C., Barranco, A., López-M a´ rquez, D. M., Berrueta, L. A., Gallo, B., and Vicente, F. Food Chem. 97 (2006) 438–446. [160] Heberger, K., Csomós, E., Sarkadi, S.: J. Agric. Food Chem. 51 2003) 8055–8060. [161] Dugo, G., La Pera, L., Pellicanó, T. M., Di Bella, G., and D’Imperio, M. Food Chem. 91 (2005) 355–363. [162] Califano, A. N., and Bevilacqua, A. E. J. Food Compos. Anal. 13 (2000) 949–960. [163] Benfeldt, C., and Sorensen, J. Int. Dairy J. 11 (2001) 567–574. [164] Pripp, A. H., Stepaniak, L., and Sorhaug, T. Int. Dairy J. 10 (2000) 249–253. [165] Hannon, J. A., Sousa, M. J., Lillevang, S., Sepulchre, A., Bockelmann, W., and McSweeney, P. L. H. Int. Dairy J. 14 (2004) 871–880. [166] Sihufe, G. A., Zorrilla, S. E., and Rubiolo, A. C. Food Chem. 93 (2005) 305–310. [167] Sihufe, G. A., Zorrilla, S. E., and Rubiolo, A. C. Food Chem. 96 (2006) 297–303. [168] Poveda, J. M., Sousa, M. J., Cabezas. L., and McSweeney, P. L. H. Int. Dairy J. 13 (2003) 169–178. [169] Di Cagno, R., Banks, J., Sheehan, L., Fox, P. F., Brechany, E. Y., Corsetti, A., and Gobbeti, M. Int. Dairy J. 13 (2003) 961–972. [170] Pripp, A. H., Rehman, S.-U., McSweeney, P. L. H., Sorhaug, T., and Fox, P. F. Int. Dairy J. 10 (2000) 25–31. [171] Mota, M. V. T., Ferreira, I. M. P. L. V. O., Oliveira, M. B. P., Rocha, C., Teixeira, J. A., Torres, D., and Goncalves, M. P. Food Chem. 94 (2006) 278–286. [172] Bustamante, M. A., Virto, M., Aramburu, M., Barron, L. J. R., Pérez-Elortondo, F. J., Albisu, M., and de Renobales, M. Int. Dairy J. 13 (2003) 547–557. [173] González, A. G., Pablos, F., and Martin, M. J. Food Chem. 73 (2001) 93–101. [174] Liang, Y., Lu, J., Zhang, L., Wu, S., and Wu, Y. Food Chem. 80 (2003) 283–290. [175] Van Nederkassel, A. M., Daszykowski, M., Massart, D. L., and Van der Heyden, Y. J. Chromatogr A 106 (2005) 177–186. [176] Niemenak, N., Rohsius, C., Elwers, S., Ndoumou, D. O., and Lieberei, R. J. Food Comp. Anal. 19 (2006) 612–619. [177] Lee, D.-S., Lee, E.-S., Kim, H.-J., Kim, S.-O., and Kim, K. Anal. Chim. Acta 429 (2001) 321–330. [178] Jakab, A., Nagy, K., Héberger, K., Vékey, K., and Forgács, E. Rapid Commun. Mass Spectrom. 16 (2002) 2291–2297.

262


[179] Aranda, F., Gómez-Alonso, S., Statedel Alamo, R. M. R., Salvador, M. D., and Fregapane, G. Food Chem. 86 (2004) 485–492. [180] Hakala, M., Lapveteläinen, A., and Huopalahti, R. J. Food Comp. Anal. 16 (2003) 67–80. [181] Vieira, R. F., Grayer, R. J., Paton, A. J., and Simon, J. Biochem. Syst. Ecol. 29 (2001) 287–304. [182] Vieira, R. F., Grayer, R. J., and Paton, A. J. Phytochemistry 63 (2003) 555–567. [183] Grayer, R. J., Vieira, R. F., Price, A. M., Kite, G. C., Simon, J. E., and Paton, A. J. Biochem. Syst. Ecol. 32 (2004) 901–913. [184] Ferreira, I. M. P. L. V. O., Pestana, N., Alves, M. R., Mota, F. J. M., Reu, C., Cunha, S., and Oliveira, M. B. P. P. Food Control 15 (2004) 291–295. [185] Masino, F., Chinnici, F., Franchini, G. C., Ulrici, A., and Antonelly, A. Food Chem. 92 (2005) 673–679. [186] Slejkovec, Z., Bajc, Z., and Doganoc, D. Z. Talanta 62 (2004) 931–936. [187] Nemati, M., Oveisi, M. R., Abdollahi, H., and Sabzevari, O. J. Pharm. Biomed. Anal. 34 (2004) 485–492. [188] Morais, H., Ramos, A. C., Cserháti, T., and Forgács, E. J. Chromatogr. A 936 (2001) 139–144. [189] Cserháti, T., Forgács, E., Darwish, Y., Morais, H., Mota, T., and Ramos, A. C. J. Chromatogr. A 949 (2002) 269–273. [190] Morais, H., Rodrigues, P., Ramos, C., Almeida, V., Forgács, E., Cserháti, T., and Oliveira, J. S. Food Sci. Technol. Int. 8 (2002) 55–59. [191] Morais, H., Rodrigues, P., Ramos, C., Forgács, E., Cserháti, T., and Oliveira, J. Nahrung/Food 46 (2002) 308–310. [192] Csiktusnádi Kiss, G., Forgács, E., Cserháti, T., Mota, T., Morais, H., and Ramos, A. J. Chromatogr. A 889 (2000) 41–49. [193] Forgács, E., and Cserháti, T. Anal. Lett. 39 (2006) 2775–2785. [194] Careri. M., Furlattini, A. M., Musci, M., Anklam, E., Theobald, A., and von Holst, C. J. Chromatogr. A 912 (2001) 61–71. [195] Morais, H., Ramos, C., Forgács, E., Cserháti, T., and Oliviera, J. J. Chromatogr. B 770 (2002) 297–301. [196] Sanderson, J. S., Daniels, R. D., Donald, A. M., Blennow, A., and Engelsen, S. B. Cabohydr. Polym. 64 (2006) 433–443. [197] Atanassova, I., and Brümmer, G. W. Geodema 120 (2004) 27–34. [198] Gimeno, R. A., Comas, E., Marcé, R. M., Ferré, J., Rius, F. X., and Borrull, F. Anal. Chim. Acta 498 (2003) 47–53. [199] Pino, V., Ayala, J. H., Afonso, A. M., and González, V. Talanta 54 (2001) 15–23. [200] Masih, A., and Taneja, A. Chemosphere 65 (2006) 449–456. [201] De Ribeiro F. A. L., and Ferreira, M. M. C. J. Mol. Struct. (Theochem) 663 (2003) 109–126. [202] Mooibroek, D., Hoogerbrugge, R., Stoffelsen, B. H. G., Dijkman, E., and Berkhoff, C.J. J. Chromatogr. A 975 (2002) 165–173. [203] Ravindra, K., Bencs, L., Wauters, E., de Hoog, J., Deutsch, F., Roekens, E., Bleux, N., Berghmans, P., and Van Grieken, R. Atmos. Environ. 40 (2006) 771–785. [204] Tang, N., Hattori, T., Taga, R., Igarashi, K., Yang, X., Tamura, K., Kakimoto, H., Mishukov, V., Toriba, A., Kizu, R., and Hayakawa, K. Atmos. Environ. 39 (2005) 5817–5826. [205] Garrido Frenich, A. G., CityMartinez Galera, M., Martinez Vidal, Massart, D. L., TorresLapasió, De Brackeleer, K., Wang, J.-H., and Hopke, P. K. Anal. Chim. Acta 411 (2000) 145–155. [206] Hogendoorn, E. A., Huls, R., Dijkman, E., and Hoogerbrugge, R. J. Chromatogr. A 938 (2001) 23–33. [207] Ruggieri, F., D’Archivio, A. A., Carlucci, G., and Mazzeo, P J. Chromatogr. A 1076 (2005) 163–169. [208] Peré-Trepat, E., Lacorte, S., and Tauler, R. J. Chromatogr. A 1096 (2005) 111–122.


263

[209] Delaunay-Bertoncini, Pichon, V., and Hennion, M.-C. J. Chromatogr. A 999 (2003) 3–15. [210] Comas, E., Gimeno, R. A., Ferré, J., Marcé, R. M., Borrull, F., and Rius, F. X. J. Chromatogr. A 988 (2003) 277–284. [211] Djakovic-Sekulic, T., Perisic-Janjic, N., and Pyka, A. Chromatographia 58 (2003) 47–51. [212] Sychov, C. S., Ilyin, M. M., Davankov, V. A., and Sochilina, K. O. J. Chromatogr. A 1030 (2004) 17–24. [213] Kánya, Z., Cserháti, T., and Forgács, E. Chromatographia 57 (2003) 451–456. [214] Kánya, Z., Forgács, E., Cserháti, T., and Illés, Z. Chromatographia 63 (2006) 129–134. [215] Forgács, E., Cserháti, T., Deyl, Z.,and Miksik, I. J. Chromatogr. B 753 (2001) 79–86. [216] Cserháti, T., Forgács, E., Deyl, Z., Miksik, I., and Eckhardt, A. J. Chromatogr. A 987 (2003) 403–408. [217] Airiau, C., Shen, H., and Bereton, R. G. Anal. Chim. Acta 447 (2001) 199–210. [218] Wang, Y., Zhang, X., Yao, X., Gao, Y., Liu, M., Hu, Z., and Fan, B. Anal. Chim. Acta 463 (2002) 89–97. [219] Djakovic-Sekulic, T. Lj., Petrovic, S. M., Perisic-Janjic, N. U., and Petrovic, StateS.D. Chromatographia 54 (2001) 60–64. [220] Cserháti, T., and Forgács, E. J. Chromatogr. A 869 (2000) 41–48. [221] Zagyi, M., Forgács, E., Prodan, M., Cserháti, T., and Berek, D. Anal. Sci. 19 (2003) 245–247. [222] Zagyi, M., and Cserháti, T. J. Liq. Chromatogr. Relat. Technol. 30 (2007) 1–12. [223] Davankov, V. A., Sychov, C. S., Ilyin, M. M., and Sochilina, K. O. J. Chromatogr. A 987 (2003) 67–75. [224] Euerby, M. R., and Petersson, P. J. Chromatogr. A 994 (2003) 13–36. [225] Baggiani, C., Anfossi, L., Giovannoli, C., and Tozzi, C. J. Chromatogr. B 804 (2004) 31–41. [226] Shen, H., Airiau, Y., and Brereton, R. G. Chemom. Intell. Lab. Syst. 62 (2002) 61–78. [227] Lavine, B. K., Ritter, J. P., and Voigtman, E. Microchem. J. 72 (2002) 163–178. [228] Wang, A., and Carr, P. W. J. Chromatogr. A 965 (2002) 3–23. [229] Poole, S. K., and Poole, C. F.. Chromatographia Suppl. 53 (2001) S-162- S-166. [230] Jezierska, M., Cendrowska, I., Markuszewski, M., Kaliszan, R., and Buszewski, B. Chromatographia 51 (2000) 111–118. [231] Larsen, B. R., Tudos, A., Slanina, J., Van der Borg, K., and Kotzias, D. Atmos. Environ. 35 (2001) 5695–5707. [232] Gross, G. M., Prazen, B. J., and Synovec, R. E. Anal. Chim. Acta 490 (2003) 197–210. [233] Escuder-Gilabert, L., Martin-Biosca, Y., Sagrado, S., Villanueva-Camanas, R. M., and MedinaHernández, M. J. Anal. Chim. Acta 448 (2001) 173–185. [234] Escuder-Gilabert, L., Sagrado, S., Villanueva-Camanas, R. M., and Medina-Hernández, M. J. Biomed. Chromatogr. 19 (2005) 155–168. [235] Garcia, M. A., Vitha, M. F., Sandquist, J. Mulville, K., and Marina, M. L. J. Chromatogr. A 918 (2001) 1–11. [236] Espinosa, S., Bosch, E., and Rosés, M. J. Chromatogr. A 945 (2002) 83–96. [237] Baczek, T., and Kaliszan, R. J. Chromatogr. A 962 (2002) 41–55. [238] Kaliszan, R., Baczek, T., Bucinski, A., Buszewski, B., and Sztupecka, M. J. Sep. Sci. 26 (2003) 271–282. [239] Kaliszan, R., Haber, P., Baczek, T., Siluk, D., and Valkó, K. J. Chromatogr. A 965 (2002) 117–127. [240] Markuszewski, M. J., Wiczling, P., and Kaliszan, R. Chem. High Troughp. Screen. 7 (2004) 281–289. [241] Kaliszan, R., and Wiczling, P. Anal. Bioanal. Chem. 382 (2005) 718–727. [242] Wiczling, P., Markuszewski, M. J., Kaliszan, M., and Kaliszan, R. Anal. Chem. 77 (2005) 449–458. [243] Baczek, T., Kaliszan, R., Novotná, K., and Jandera, P. J. Chromatogr. A 1075 (2005) 109–115.

264


[244] Fjällström, P., Andersson, B., Nilsson, C., and Andersson, K. Ind. Crops Prod. 16 (2002) 173–184. [245] Womiloju, T. O., Miller, J. D., Mayer, P. M., and Brook, J. R Atmos. Environ. 37 (2003) 4335–4344. [246] Pasadakis, N., Gaganis, V., and Varotsis, N. Fuel 80 (2001) 147–153. [247] Yawei, W., Lina, L., Jianbo, S., and Guibin, J. Environ. Pollut. 135 (2005) 457–467. [248] Oravisjärvi, K., Timonene, K. L., Wiikinkoski, T., Ruuskanen, A. R., Heinänen, K., and Ruuskanen, J. Atmos. Environ. 37 (2003) 1013–1022. [249] Prendes, P., Andrade, J. M., López-Mahia, P., and Prada, D. Talanta 49 (1999) 165–178. [250] Marinoni, A., Polesello, S., Smiraglia, C., and Valsecchi, S. Atmos. Environ. 35 (2001) 3183– 3190. [251] Fang, G.-C., Wu, Y.-S., Chang, S.-Y., Rau, J.-Y., and Huang, S.-H. Chemosphere 64 (2006) 1253–1263. [252] Colombini, M. P., Ceccarini, A., and Carmignani, A. J. Chromatogr. A 968 (2002) 79–88. [253] Canals, I., Portal, J. A., Rosés, M., and Bosch, E. Chromatographia 56 (2002) 431–437. [254] Kher, A., Mulholland, M., Green, E., and Reedy, B. Vibr. Spectrosc. 40 (2006) 270–277. [255] Kaivez, A., Gallez, X. A., Daoust, D., Devaux, J., and Godard, P. Polymer 43 (2002) 3181–3190. [256] Gray, M. J., Dennis, G. R., Slonecker, P. J., and Shalliker, R. A. J. Chromatogr. A 1015 (2003) 89–98. [257] Gray, M. J., Dennis, G. R., Slonecker, P. J., and Shalliker, R. A. J. Chromatogr. A 1028 (2004) 247–257. [258] Gray, M. J., Dennis, G. R., Slonecker, P. J., and Shalliker, R. A. J. Chromatogr. A 1041 (2004) 101–110. [259] Sánchez-Ponce, R., and Rutan, S. Chemom. Intell. Lab. Syst. 77 (2005) 50–58. [260] Wasim, M., and Brereton, R. G. Chemom. Intell. Lab. Syst. 72 (2004) 133–151. [261] Kozerski, G. E., Gallavan, R. H., and Ziemelis, M. J. Anal. Chim. Acta 489 (2003) 103–114. [262] Caetano, S., Aires-de-Sousa, Daszykowski, M., and Van der Heyden, Y. Anal. Chim. Acta 544 (2005) 315–326. [263] Hanai, T. J. Chromatogr. A 985 (2003) 343–349. [264] Hanai, T. J. Chromatogr. A 1027 (2004) 279–287. [265] Hanai, T. J. Chromatogr. A 1087 (2005) 45–51.

4 Electrically Driven Systems Although capillary electrophoresis and related techniques are relatively new separation methods they have rapidly found application in many fields of up-to-date research and development, such as biology, biochemistry, protein analysis, synthetic chemistry, and environmental protection studies.

4.1

Theory and Practice of Electrically Driven Systems

Capillary electrophoresis (CE) separates analytes in a capillary tube under the effect of an electric field applied at the two ends of the capillary. The different magnitudes and directions of migration velocities of analytes result in their separation. The migration characteristics can be easily changed by modifying the electrophoretic media, the concentration, type and pH of the buffer. The charge of the polar substructures of analytes, the degree of their ionization, the association between the analytes and the various additives in the electrophoretic buffer, and the properties of the capillary surface may also have an impact on the efficacy of separation. Common CE techniques use a narrow bore capillary (10–100 m ID) and a high driving potential (up to 30 kV). The use of capillaries with small ID allows efficient heat dissipation, which permits the employment of high voltages without warming up the CE system. The theoretical plate number of CE techniques is higher than any other traditional chromatographic procedure. In the last decade numerous books and reviews have been published on the various aspects of CE analysis. Books have been published on the clinical and forensic applications [1], the use of CE in pharmaceutical analysis [2], and the theoretical and practical bases of CE separations [3]. The analyses of small molecules [4], natural organic matter [5], metabolites [6], and biomolecules [7] have also been reviewed. The application of micellar electrokinetic chromatography (MEKC) has also been evaluated [8]. The instrumentation of CE is relatively simple. The capillary tube responsible for the separation is filled with a buffer, and both its ends are dipped into two buffer reservoirs Multivariate Methods in Chromatography: A Practical Guide Tibor Cserháti C 2008 John Wiley & Sons, Ltd

266


that are kept at the same level. A high voltage is employed across the capillary. Generally the analytes are introduced at the anodic end and the detector is placed at the opposite (cathodic) end. Various CE techniques have been developed and used for the separation of both charged and neutral analytes. In capillary zone electrophoresis (CZE) the sample is placed in the capillary between two similar buffer solutions, and a potential is applied across the capillary. The migration of charged analyte molecules in the sample depends on their charge-to-mass ratios. Analytes with higher charge have increased migration velocity; they are less retained in the capillary. Capillary gel electrophoresis (CGE) employs a gel-filled capillary. The pores of the gel act as sieves, and the migration of the analytes depends on both their charge and size. Instead of gels, polymer solutions can also be used in CGE. This technique has been successfully applied for the separation of oligonucleotides, peptides and proteins. MEKC has been developed for the separation of neutral analytes without any dissociable polar substructures. An anionic surfactant over its critical micellar concentration is mixed into the background electrolyte. Analytes are partitioned between the apolar core of the surfactant micelles and the polar aqueous buffer according to their hydrophobicity. Capillary isotachophoresis (CITP) employs two different electrolytes, the so-called leading and terminating elecrolytes. After sample injection a potential is applied across the two different buffers. As a result of the two different buffers, the electric field along the length of the column is not constant. The bands of the individual sample components are sharpened under the effect of the changing electric field, and they will migrate at the same velocity. The method can be also be used for the preconcentration of dilute samples. Capillary isoelectric focusing (CIEF) is suitable for the separation of amphoteric analytes in a pH gradient. A continuous pH gradient is built up in the column by using ampholytes under a potential field. Amphoteric analytes migrate to the point where their net charge is equal to zero; they form stationary and sharply focused zones. Electroosmotic flow (EOF) is an important parameter in CE techniques. It is defined as the bulk flow of solvent in the capillary under the effect of potential field. The value characterizing the magnitude of EOF can be defined as: EOF = (ε/)

(4.1)

where EOF is the EOF mobility, ε is the dielectric constant, is the zeta potential, and is the bulk viscosity. Equation (4.1) indicates that EOF does not depend on the electric field applied. The separation characteristics (migration time, separation capacity, etc.) in electrically driven systems can be described in a similar way to that in GC and HPLC. In contrast to other CE systems, the capacity factor has to be calculated differently in MEKC: k = (tR − t0 ) / [t0 (1 − t1 /tM )] = K (VS /VM )

(4.2)

where tR is the retention time of the analyte, t0 is the retention time of unretained analyte moving at EOF , tM is the retention time of the micelle, K is the partition coefficient, VS is the volume of the micellar phase, and VM is the volume of the mobile phase. The apparent mobility of an analyte (a ) can be calculated by: μa = I /t E = I L/t V

(4.3)

Electrically Driven Systems

267

where I is the effective length of the capillary, L is its total length, E is the electric field and V is the voltage employed. The apparent mobility can also be computed by: a = e + EOF

(4.4)

where μe is the effective mobility. The unreacted hydroxyl groups on the surface of the fused silica capillaries dissociate when the pH of the buffer is higher than 2.5. The negatively charged surface attracts the cations of the buffer, the layer of positive charges form a double layer which creates a potential difference near to the capillary wall (zeta potential). The solvated cations migrate towards the cathode transporting the solvated solvent molecules with them. Electrophoretic flow (ep ) is the flow of ions. Under the potential applied cations and anions move to the cathode and anode, respectively. EOF is larger than the electrophoretic flow. On account of the dominating role of EOF each analyte (cationic, anionic and neutral) has a tendency to move towards the cathode. The cation migration velocity is composed of the electrophoretic migration and the EOF. Neutral analytes are not separated under common CE conditions; they move together with the EOF. The migration velocity of anions is the difference between EOF and electrophoretic migration. Consequently, their migration velocity is lower than that of both neutral and cationic compounds. One of the most decisive factors modifying CE separation is the composition of the buffer. The change of the pH of the buffer can increase or decrease the migration velocity, influencing in this manner the separation of charged analytes with various charge-to-size ratios in CZE. The separation of neutral compounds by MEKC can be achieved by the addition of anionic, cationic or neutral surfactants, i.e. sodium dodecyl sulfate (SDS), ethoxylated long chain alcohols, or cetyltetrabutylammonium bromide. The efficacy of CE separation techniques can be further increased by adding modifiers to the buffer, such as methanol, ACN, bile salts, cyclodextrins and cyclodextrin derivatives, hydroxypropylcellulose, hydroxypropylmethylcellulose, and tetraalkylammonium salts. Traditional CE techniques apply aqueous buffers for separation which are unable to dissolve strongly hydrophobic analyte components. These molecules can be precipitated in the aqueous media during the separation process. The addition of organic modifiers can overcome this problem; however, they lengthen the analysis time and reduce the velocity of the EOF.

4.2 4.2.1

Gel Electrophoretic Techniques Theory and Human Health Aspects

The separation of a considerable number of proteins plays a decisive role in proteomic techniques. On account of the high separation power, two-dimensional (2D) gel electrophoresis is a method of preference for the analysis of complicated protein mixtures. The main advantages and disadvantages of 2D gel electrophoresis have been discussed previously [9]. One of the main problems of the evaluation of 2D gels is the matching of gel images. Various automated image registration techniques were developed and applied for comparison [10,11]. An automatic method of fuzzy alignment for the matching of 2D gel electrophoretic images was also developed. The effect of the number of iterations on the degree of fuzziness is shown in Figure 4.1. It was stated that the fuzzy alignment of features makes the efficient automated matching of 2D gel images possible [12].

268

Multivariate Methods in Chromatography: A Practical Guide 04 04 03

03

02

02

01 x2

x2

01 0

0

–01

–01

–02

–02

–03

–03 04

x1

–01 x1

(a)

(b)

–02

0

02

–05

04

03

03

02

02

01

01 x2

x2

06

0

–01

–02

–02 –02

–01

0

01

02

–04

03

(d)

02

02

01

01

0

0

–01

–01

–02

–02 –02

–01

(c) 03

–03

–02

x1

03

–04

–03

x1

x2

x2

–03

–01

0

01

02

03

0

01

0

01

02

03

0

01

02

03

02

03

04

0

–01

–04

–04 –03 –02

–04

–03

–02

–01

x1

x1

(e)

(f)

Figure 4.1 Degrees of fuzziness, determined by the σ of the Gaussian function and represented by the different radii of the circles for (a) the initial sets, and those after (b) 1 iteration, (c) 10 iterations, (d) 20 iterations, (e) 30 iterations, and (f) 40 iterations. Reprinted from ref. [12] with permission from American Chemical Society

Kaczmarek et al. investigated the image transformation methods used in matching 2D gel electrophoresis images and two types of grid mapping were compared for their efficacy [13]. As noise reduction is a prerequisite for the successful matching of 2D images, a considerable number of methods has been developed and applied for this purpose. A separate study has been devoted to the comparison of noise reduction methods, such as linear filtering, e.g. the mean and Gaussian filtering, the nonlinear method, i.e. median filtering, spatially adaptive linear filtering and filtering in the wavelet domain. The main signal-to-noise ratio


269

Table 4.1 Mean values of SNR obtained for 50 synthetic images with different levels of noise using four different methods of filtering and three different sizes of neighbourhood

Type of filter Mean Gaussian Median Wiener

Standard deviation of noise

Size of neighbourhood

10

20

30

3×3 5×5 7×7 3×3 5×5 7×7 3×3 5×5 7×7 3×3 5×5 7×7

35.44a 33.57 29.84 35.06 35.19a 32.10 33.50a 32.63 29.12 35.18 35.66b 33.56

30.31 31.45a 29.14 29.67 32.03b 30.86 28.29 30.33a 28.67 29.26 31.50a 30.75

26.94 28.90a 27.78 26.30 28.97b 28.87 25.02 28.04a 27.79 25.52 28.16 28.16a

a

Best results obtained for specific method. Best results obtained for all methods. Reprinted with permission from ref. [14]. Copyright Wiley-VCH b

(SNR) values obtained are compiled in Table 4.1. It was concluded from the results that the performance of the Wiener filter is the best for images with low noise level while for images with high noise level the efficacy of the Gaussian filter was the best. The data obtained with different wavelet, different methods of threshold estimation, and different number of decomposition levels differed considerably. It was concluded that the filtering of gel images in the wavelet domain using the Bayes Thresh method of threshold value determination was the best preprocessing procedure [14]. The applicability of the baseline reduction method on regression was also studied. The influence of baseline reduction on the quality of images was illustrated. The method was very robust and suitable for baseline reduction enhancing the reliability of the evaluation of 2D images [15]. Temporal temperature gradient gel electrophoresis (TTGE) of bacterial 16S rRNA was applied for the characterization of the faecal microbial community and for the elucidation of its correlation with the cholesterol–coprostanol conversion in human gut. The investigation was motivated by the fact that many intestinal bacteria are not cultivable by classical procedures [16,17] and rDNA analysis may help in the determination of community diversity [18,19]. TTGE separation methods [20], statistical analysis of gel electrophoretic data [21] and PCA for the description of the microbial community [22] have been previously employed for similar purposes. The gels were evaluated by densitometer and the peaks were applied as variables for CA and PCA. CA separated the samples in two clear-cut groups. It was established that the mean faecal coprostanol-to-cholesterol ratio was significantly different indicating the relationship between these two parameters. The mean parameters of PCA are compiled in Table 4.2. The ratio of variance explained by the individual principal components is surprisingly low illustrating the considerable diversity between

270


Table 4.2 Contribution of principal components to the overall variation and correlations with coprostanol-to-cholesterol ratio in fresh stools Principal component

Individual contribution (%)

Cumulative contribution (%)

Correlation with coprostanol-to-cholesterol ratio

18.3 13.4 11.3 8.3 7.7 6.6 6.4 5.8 5.0 4.6 3.9 3.7 2.9 2.1

18.3 31.7 43.0 51.4 59.0 65.7 72.0 77.8 82.8 87.4 91.3 95.0 97.9 100.0

−0.79 0.22 −0.23 0.11 0.10 0.22 0.10 0.00 0.19 0.24 0.08 −0.29 0.04 0.09

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


the samples. It was further established that only the first principal component correlated with the coprostanol-to-cholesterol ratio. The biplot of PC1 versus PC2 is shown in Figure 4.2. The scattering of points on the map suggests again the marked correlation between the faecal microbial community structure and cholesterol–coprostanol conversion in the human gut [23]. A combined transcriptomics and proteomics technique was developed and applied for the study of bromobenzene hepatoxicity in rats. Two-dimensional electrophoresis was C02

C24 C21

0

C16

C06 C14

C11

C25

C27 C15

C22

PC2

C12 –2.5

C04

C18

–5.0

C19

–7.5 –4

–2

0

2

4

PC1

Figure 4.2 Projection of the original TTGE data set onto the first two principal components. Correlation of PC1 with coprostanol-to-cholesterol ratio was −0.79. The diameters of the filled black circles are proportional to the coprostanol-to-cholesterol ratio. Reprinted with permission from ref. [23]. Copyright Elsevier


271

4

3

2

Scores on PC# 2 (12.14%)

1

MUT, Rat4 CO, Rat22 UT, Rat2

BB, Rat28

UT, Rat2,cy5 UT, Rat8,cy5 CO, Rat20 CO, Rat20,cy5

BB, Rat26

UT, Rat6,cy5

0

–1

BB, Rat28,cy5

CO, Rat22,cy5

BB, Rat26,cy5

CO, Rat22,cy5 vs. BB, Rat26

BB, Rat30,cy5

UT, Rat4,cy5

–2

CO, Rat22 vs. BB, Rat26,cy5

BB, Rat32

BB, Rat32,cy5

CO, Rat18,cy5 UT, Rat6

–3

UT, Rat8 CO, Rat22 vs. BB, Rat26,cy5

–4

–5

–6 –3

CO, Rat18

–2

–1

0

1

2

3

4

5

Scores on PC# 1 (17.00%)

Figure 4.3 Principal component analysis of transcriptome data. Objects in the plot represent expression profiles from individual rat livers. The two principal components explaining the majority of the variation in the data set are plotted. The expression profiles of rat liver after treatment are distinct from the controls. The injection of corn oil only as vehicle control did not result in aberrant gene expression profiles compared with the untreated sample. UT, untreated (squares); CO, corn oil vehicle control (circles); BB, bromobenzene treated (triangles); and CO and BB, directly compared CO with BB samples (inverted triangles). Individual rat numbers are indicated, followed by ’Cy5’ when the liver RNA sample was labelled with Cy5 fluorofore (and the reference with Cy3). In the other samples, the fluorofore incorporation was swapped. Reprinted with permission from ref. [24]. Copyright Elsevier

employed for the elucidation of the differences between the protein profile of treated and untreated rats. The data obtained by both transcriptomics and proteomics technologies were evaluated by PCA. The biplot of PC1 versus PC2 carried out on transcriptome data is depicted in Figure 4.3. It was stated that the points representing treated and untreated rats are well separated on the map. The plot of PC1 versus PC2 of PCA performed on proteomes is shown in Figure 4.4. Similar to the distribution of points in Figure 4.3, the liver protein profiles of bromobenzene-treated rats and controls show marked differences indicating the good separation power of PCA [24]. Two-dimensional gel electrophoresis and PCA of the digitized results were employed for the study of proteomic changes associated with healthy and tumoral murine samples in neuroblastoma. The data set consisted of 20 samples belonging to control and diseased nude mouse adrenal cells. The results of PCA carried out on the overall data set are compiled in Table 4.3. The first four principal components accounted for the overwhelming

272

Multivariate Methods in Chromatography: A Practical Guide 0.3 BB,Rat28

CO, Rat22

BB,Rat32

UT,Rat6

0.2

UT,Rat8 UT,Rat4

0.1 CO, Rat20 Scores on PC# 3

0

–0.1 BB,Rat30 UT,Rat2

–0.2

CO,Rat18

–0.3

–0.4 –0.3

–0.2

–0.1 0 Scores on PC# 1

0.1

0.2

Figure 4.4 Principal component analysis of proteomes. PCA was applied using spot volume data obtained from untreated, corn oil control and bromobenzene-treated animals. The two principal components explaining the majority of variation in the data set are plotted. Individual gels of bromobenzene-treated samples are indicated by triangles, corn oil controls are indicated by circles and untreated samples by squares. The three gels from the same animal are connected by lines. Protein patterns from bromobenzene-treated animals are clearly distinct from corn oil controls and untreated samples. A slight difference in the pattern of corn oil controls compared with untreated rats can be concluded from this figure. Reprinted with permission from ref. [24]. Copyright Elsevier Table 4.3 Results of PCA (for the first four principal components) performed on the overall data set

PC1 PC2 PC3 PC4



47.3 14.5 10.6 8.3

47.3 61.8 72.4 80.7



30

273

Scores

PC2

HEA4

20

LL1 LL7 LL4 LL5

10

HEA7 HEA10 HEA9

LL9 LL5 LL10 LL2

0

HEA1 HEA6 HEA3 HEA5 HEA8

LL8

–10

LL3

HEA2

–20

PC1 –20

Figure 4.5 Elsevier

–15

–10

–5

0

5

10

15

20

Score plot of PC1 versus PC2. Reprinted with permission from ref. [25]. Copyright

majority of variance inherent in the original data matrix. The plot of PC1 versus PC2 is depicted in Figure 4.5. The healthy and control samples are well separated proving the good classification capacity of PCA. The results of PCA performed separately on the control and diseased group are similar to those obtained in the case of the overall data as demonstrated in Table 4.4. It was concluded from the results that 2D gel electrophoresis followed by PCA is a promising tool in clinical chemical research facilitating the identification of proteins responsible for the disease [25]. Another 2D gel electrophoretic technique was applied for the differentiation between healthy human lymph nodes and non-Hodgkin lymphomas. Separations were carried out by isoelectric focusing (IEF, first dimension) and by sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE, second dimension). First the images were digitalized, fuzzificated, and decoded by PCA. The results of PCA were compared with those of LDA. Table 4.4 Results of PCA (for the first four principal components) performed on the control and diseased classes separately Control class

PC1 PC2

Diseased class





43.3 29.5

43.3 72.8

46.0 31.7

46.0 77.7


274

Multivariate Methods in Chromatography: A Practical Guide Table 4.5 Linear discriminant analysis results for each level of the fuzzification parameter σ

σ

NER (%)

Wrong classification

— 0.25 0.75 1.25 1.75 2.25 3.50a 4.75a

87.5 87.5 87.5 87.5 100 100 100 87.5

HEA4 HEA4 HEA4 HEA4 — — — HEA

NER, non-error rate. Three principal components in the LDA model. Reprinted with permission from ref. [26]. Copyright Elsevier.

a

The data illustrate that the fuzzification parameter exerts a considerable influence on the results of PCA. The impact of the fuzzification parameter on the efficacy of LDA is demonstrated in Table 4.5. The data clearly show that the optimal value of varies between 1.75 and 2.25 (100% non-error rate) [26]. Two dichotomous (gender, hypolipidemic treatment) and 19 continuous parameters of diabetic patients were measured by using various analytical techniques among them nondenaturing polyacrylamide gradient electrophoresis of plasma. FA was employed for the selection of risk factors of the 21 variables included in the calculation. FA found 10 risk factors namely age, body mass index, systolic blood pressure, smoking, HbAlc (haemoglobin Alc ), high-density C-reactive protein, lipoprotein (a), LDL (low-density lipoprotein), cholesterol and LDL particle size. Computation further proved that LDL particle size is the best risk factor for the presence of coronary heart disease [27]. The efficacy of chemopreventive agents was investigated by measuring 4858 genes in lung and liver of Sprague–Dawley rats unexposed or exposed to environmental cigarette smoke. RNA structure integrity was assessed by gel electrophoresis, DNA adducts were measured by multidirectional TLC and electronic autoradiography and the amount of 518 proteins was measured by a glass microarray. The similarity between the samples was calculated by CA. The following chemopreventive agents were included in the experiments: N -acetyl-l-cysteine, oltipraz, 5,6-benzoflavone, phenethyl isothiocyanate, and indole 3carbinol. The results indicated that efficacy and safety of the agents differ considerably (Figure 4.6) [28]. 4.2.2

Microorganisms

Multilocus enzyme electrophoresis (MLEE) and other electrophoretic technologies have been frequently applied to study the grouping and genetic variability of the Candida albicans population [29–33]. The application of multivariate mathematical-statistical methods for the comparison of DNA fingerprinting of infectious fungi has also been previously reported [34]. A separate study was devoted to the comparison of the efficacy of the various CA

Electrically Driven Systems 20 CYCLIN-RELATED GENES

25 STRESS RESPONSE GENES

ECS

ECS SHAM ECS ECS ECS ECS ECS ECS +NAC +OPZ +5.6-BF +OPZ +13C +PEITC +PEITC +NAC +13C

ECS

ECS ECS ECS ECS SHAM ECS ECS ECS +5.6-BF +OPZ +NAC +OPZ +13C +PEITC +PEITC +NAC +13C

ECS ECS ECS SHAM ECS ECS ECS ECS +5.6-BF +OPZ +13C +PEITC +PEITC +NAC +OPZ +NAC +13C

11 CYTOKERATINRELATED GENES

19 HEAT SHOCK PROTEINRELATED GENES

ECS

275

ECS

ECS ECS ECS SHAM ECS ECS ECS ECS +OPZ +5.6-BF +13C +PEITC +PEITC +NAC +OPZ +13C +NAC

Figure 4.6 Hierarchical cluster analysis of the expression of genes belonging to four functional categories in the lung of sham-exposed or ECS-exposed rats, either untreated or treated with chemopreventive agents. Treatments characterized by similar multigene expression profiles are linked together. Reprinted with permission from ref. [28]. Copyright Elsevier

techniques for the evaluation of MLEE results. The three CA methods were: A, genetic interpretation, genetic distance matrix and unweighted pair group method using arithmetic averages (UPGMA); B, genetic interpretation, similarity matrix and UPGMA; C, numerical interpretation, similarity matrix and UPGMA. The group formation of Candida isolates on the dendograms is slightly different according to the method of computation. This discrepancy was supported by the data. On account of its high discriminative power the method was proposed for the classification of C. albicans isolates [35].

276


CA has also been employed in the comparative sequence analysis of the internal transcribed spacer 1 of Ochrobactrum species [36]. The bacterial pathogen Brucella ovis infects the reproductive tract of sheep and farmed red deer stags [37,38]. The differences between the isolates have been demonstrated but epidemiology did not support this statement [39]. The method of variable number tandem repeats has been previously applied for the differentiation of B. abortus isolates [40]. The polymorphism among B. ovis isolates was proved by employing pulsed field gel electrophoresis (PFGE). The performance of the rare-cutting restriction enzymes Xbal and Swal was also compared. The PFGE patterns of Xbal and Swal digested genomic DNA from B. ovis isolated from ovine and cervine origin showed marked differences indicating the diversity of the effect of rare-cutting restriction enzymes. The analysis of the electrophoretic data by CA indicated the presence of two diverse strain types. It was further established that the discriminatory power of the restriction endonuclease Swal was higher than that of Xbal [41]. The differences between the isolates of Acidithiobacillus ferrooxidans was investigated by gel electrophoresis of the randomly amplified polymorphic DNA (RAPD) and the samples were classified according to the composition of bands by CA. The CA dendogram is depicted in Figure 4.7. The CA dendogram shows the high diversity of Acidithiobacillus isolates with the similarity varying between 5.49% and 85.14%. The data further illustrate that CA analysis of RAPD profiles can be used for the determination of the variability in this bacterium [42]. PFGE has also found application in the investigation of the diversity of Burkholderia pseudomallei strains. This environmental Gram-negative bacillus can cause encephalomyelitis syndrome [43], parotid abscesses [44], shows differential virulence and tropism [45,46], and induces central nervous system infection [47]. It was further established that the severity and pattern of disease may depend on the ribotype [48], the intensity of rainfall [49], and a single clone can cause a variety of disease manifestations [50]. PFGE was also employed for the comparison of Escherichia coli strains [51]. In addition to PFGE, 0

5

10

15

20

25

30

35

40

45

50

55

60

65

70

75

80

85

90

95

100

Cc Wb 31A 33 15A 16A 33A Pa A1 W1 15B ATCC

Figure 4.7 Dendogram representing the relationship between 12 A. ferrooxidans strains based on UPGMA CA of the RAPD profiles derived from 13 primers using Jaccard’s similarity coefficient (%) ( r = 0.9616). Reprinted with permission from ref. [42]. Copyright Elsevier


277

multilocus sequence has also been applied for the differentiation between Staphylococcus aureus [52], and Bartonella henselae strains [53]. The composition of isolates from 114 patients was measured by PFGE, the individual bands were measured and the numerical data were evaluated by CA [54]. According to the results of CA isolates were divided into 71 strain types indicating the marked diversity of strains. It was further established that the strain type does not determine the disease presentation [55]. On account of its importance in human health care the behaviour of Neisseria meningitis has been vigorously investigated [56–58]. The population structure of N. meningitis serogroup A was investigated by MLEE, RAPD and MLST using 84 isolates. The data were evaluated in each instance by CA. The structure of the dendograms differs according to the computation method applied emphasizing again the considerable role of the software used for the evaluation of the original data matrix [59]. A considerable number of studies was applied for the typing and subtyping of E. coli isolates [60,61]. The diversity of E. coli isolates has been assessed many times [62,63]. Such investigations employed various techniques, including amplified fragment length polymorphism (AFLP) analysis [64], repetitive-element PCR fingerprinting [65], and lactase dehydrogenase release assay [66]. The objectives of the developments in the methods included the elucidation of the cause of the gene instability [67] and finding the target gene for virulence [68]. A combined method was employed for the subtyping of 54 foodborne and environmental isolates of E. coli. The techniques applied were multiplex-PCR for toxin genes, PFGE, AFLP, ribotyping and DNA fingerprinting methods, such as repetitive oestrogenic palindromes and Box primers (repetitive sequence polymerase chain reaction, Rep-PCR). PCR amplification products were analysed by agarose gel electrophoresis (1.5 and 2 % w/v agarose). Gel electrophoresis was also applied for ribotyping. The classification of samples according to the electrophoretic profiles was performed by CA using the UPGMA method. It was concluded from the distribution of samples on the dendograms that both techniques differentiate well between the O157 and other isolates but the classification of other isolates is not clear cut. Similar results were obtained using PFGE fingerprints. It was further observed that the results of the three CA analyses were similar but not identical; therefore, the parallel application of the methods may enhance the reliability of the investigation [68]. Various Salmonella strains can infect both animals [69] and the human population [70] causing outbreaks of human salmonellosis. The genetic variation means that their genotyping is of considerable importance. A wide variety of methods has been applied for this purpose; the use of controlled PCR [71], RAPD analysis [72,73], DNA microarrays [74] and automated ribotyping has been previously reported [75]. The identification of the source of salmonellosis and the pathway of transmission has also been investigated in detail [76]. The transmission pattern on swine farms was assessed [77,78], and it was established that the application of the comparison of serotypes has a poor discriminative power [79]. The detection of gene differences in Salmonella is frequently carried out by PFGE [80] or by Rep-PCR [81]. PFGE, RAPD analysis and automated ribotyping were simultaneously employed for genotyping 32 isolates of Salmonella livingstone. The components of the samples were separated by various electrophoretic techniques and the relationship between the fingerprints was elucidated by CA. The dendogram computed from the fingerprints obtained by RiboPrinter is depicted in Figure 4.8. Although each method was suitable for

100

95

90

85

80


278

4 5 26 32 27 25 22 17 7 16 15 14 13 21 11 10 31 30 19 3 2 12 18 6 24 23 20 28 9 29 8 1

Figure 4.8 Dendogram showing RiboPrinter fingerprints from all 32 strains. The dotted line indicates where the division into types was made at 96% similarity with the Dice coefficient. Reprinted with permission from ref. [82]. Copyright Elsevier

the typing of samples the results slightly differed. This discrepancy may be due to the inherent diversity of the techniques applied [82]. The discriminative power of Rep-PCR and PFGE was compared in the investigation of the genetic diversity and transmission of Salmonella. The similarity between the samples was computed by CA separately for the fingerprints obtained by Rep-PCR and PFGE. The corresponding dendograms are shown in Figures 4.9 and 4.10. It was found that the


279

P3[M-RN-22]FLR P3[M-RN-4]FLY P3[M-RN-14]SW6 P4[GR-4-4]FLR P4[GR-4-3]SG15 P4[M-RS-1]SWB P4[M-RS]FLY P3[M-HN]FLY P4/BT1 P3[M-RS-14]FLY P3[M-HN-21]SN3 P3[M-RN-22]SW17 S4/BT6 S2[RN-4-K]SN85 W4[FNE-3-2]FLR F2[RN-F1]MUS1 F2[RN-F1]FLY F3[RN-N5-8]SN13 F2[RN-N4]MU35 F3[GE-B49]SS9 F1[RN-F4]MUS1 F3[RN-N5-R]FLR F3[RN-N5]FLY F1[RN-F4-9]SW77 F3/BRD1

AGONA AGONA AGONA AGONA AGONA AGONA AGONA AGONA AGONA AGONA AGONA AGONA DERBY DERBY DERBY UGANDA UGANDA UGANDA UGANDA UGANDA UGANDA UGANDA UGANDA UGANDA SENFTENBERG

F3[GE-B49]FLR F3[GE]CAT1 R1[M-BG]FLY R1[M-BG-515]FLR R3[M-BG]FLY R3[M-GR3-7]SG4 R3[M-GR3-6]FLR R3[M-N5-10]FLR R3[M-GR3]MUS185

SENFTENBERG SENFTENBERG WORTHINGTON WORTHINGTON WORTHINGTON WORTHINGTON WORTHINGTON LONDON LONDON

R3[M]MUS195 R3[M-N5-9]SN1 G4[FR-E]FLY

LONDON LONDON INFANTS

E2[FN-E-13]SF9 R3[M-N]CAT3 R1[M-BG-719]SS11 R3/BRDI H1[M-GR]SPD W4[RR]MUS5 W4[RR]CAT1 W4[RR]MUS4 W4[RR]MUS3 W4[RR]SPD L1[M-GE-95]FLR L4/H2O L3[FR-D-10]SW27 L3[M-GE-C1]FLR

-----------AGONA MONTEVIDEO MONTEVIDEO -----------WORTHINGTON WORTHINGTON WORTHINGTON WORTHINGTON WORTHINGTON AGONA ANATUM DERBY MBANDAKA

L3[GE-N]FLY L1[M-GE-C39]SS12 H4[M-GE]FLY H4[M-FR-TR]SW19

DERBY DERBY DERBY INFANTIS

H4/BT8 L1[M-GE-N]FLR H4[M-GE-14]FLR H4[M-GE-14]SS1

DERBY DERBY DERBY DERBY

G4/BT1 G2[GS-E-1]SG2 G4/BT6

INFANTIS ANATUM DERBY

G4[GS-E-5]SN15 G4[SO-6]SS10 E1[OF-S-5]SFI E4/BT6 E1[OF-S-5]SF3 E3/BRDD

INFANTIS INFANTIS DERBY DERBY DERBY DERBY

0.1

Figure 4.9 Cluster analysis of Salmonella isolates based on genetic distances calculated from three-primer composite Rep-PCR fragment size matching. Sample identification indicates farm (letter in column 1), farm visit (number in column 2), location on farm (building-room-pen) and ecological compartment sampled (SS, sow; SW, suckling pig; SN, nursery pig; SG, grower pig; SF, finisher pig; FLR, pen floor; FLY, fly; BT, bout; MUS, mouse; BRD, environmental deposit of bird faeces; CAT, cat; SPD, spider; H2 O, water). Serotypes are listed after sample identification. Isolates to the right of the vertical dashed line are in ’tight’ clusters, as defined in the text. The distance scale in terms of proportion of fragments matching is indicated on the lower left of the figure. Reprinted with permission from ref. [83]. Copyright Elsevier

280

Multivariate Methods in Chromatography: A Practical Guide P3[M-RN]FLY P3[M-RN-14]SW6 P3[M-RN-22]FLR P4[M-RS-I]SWB P4[GR-4-3]SG15 P4[GR-4-4]FLR

AGONA AGONA AGONA AGONA AGONA AGONA

P3[M-HN]FLY P4/BTI P3[M-RS-14]FLR P4[M-RS]FLY P3[M-HN-21]SN3 P3[M-RN-22]SW17 R3[M-N]CAT3 F1[RN-F4-9]SW17 F2[RN-F2]FLY

AGONA AGONA AGONA AGONA AGONA AGONA AGONA UGANDA UGANDA

F2[RN-F1]MUS1 F3[RN-N5-8]SN13 F1[RN-F4]MUS1 F2[RN-N4]MUS5 F3[GE-B09]959

UGANDA UGANDA UGANDA UGANDA UGANDA

F3[RN-N5-B]FLR F3[RN-N5]FLY F3[M-N5-10]FLR F3[M-GR3]MUS165 R1[M]MU195

UGANDA UGANDA LONDON LONDON LONDON

R3[M-N5-9]SN1 P3[GE]CAT1 P3[GE-B44]FLR P3/BRD1 G4/BTI G2/SG2 G4[GS-E-5]SN15 G4/BT6 L1[M-GE-F4]FLR

LONDON SENFTENBERG SENFTENBERG SENFTENBERG INFANTIS ANATUM INFANTIS DERBY DERBY

H4[M-GE-14]S31 H4[M-GE-14]FLY H4[M-GE-8]FLR H4[M-FR-18]SW19

DERBY DERBY DERBY INFANTIS

H4/BT8 E1[OF-S-5]SF1 E4/BT6 E1[OF-9-5]SF3 E3/BRD1

DERBY DERBY DERBY DERBY DERBY

S4/BT6 S2[RN-4-8]SNR5 W4[FNE-3-2]FLR R1[M-BG-515]FLR R3[M-GR3-7]SG4 R3[M-BG]FLY R1[M-BG]FLY R3[M-GR3-6]FLY

DERBY DERBY DERBY WORTHINGTON WORTHINGTON WORTHINGTON WORTHINGTON WORTHINGTON

W4[RR]MUS3 W4[RR]MUS4 W4[RR]MUS5 W4[RR]CAT1

WORTHINGTON WORTHINGTON WORTHINGTON WORTHINGTON

W4[RR]STD WORTHINGTON G4[SO-6]SS10 INFANTIS L1[M-GE-C39]SS12 DERBY R1[M-8G-719]SS11 MONTEVIDEO R3/BRD1 H1[M-GR]SMD G4[FR-E]FLY

MONTEVIDEO -------------------INFANTIS

E2[FN-E-11]SF9 L1[M-GE-F5]FLR L4/H2O

-------------------AGONA ANATUM

L3[FR-D-10]SW27 L3[M-GE-C1]FLR

DERBY MBANDAKA

E3[GE-N]FLY

DERBY

0.1

Figure 4.10 Cluster analysis of Salmonella isolates based on genetic distances calculated from three-enzyme composite PFGE fragment size matching. For details see Figure 4.9. Reprinted with permission from ref. [83]. Copyright Elsevier

reliability of both genotyping methods is equally high; however, because of its greater discriminating capacity, the application of Rep-PCR was proposed [83]. Cellulose acetate electrophoresis (CAE) was also employed for identification of microorganisms [84]. It was used for the classification of Fusarium species according to their isozyme profile [85]. The isozyme profile of Fusarium cerealis, F. culmorum, F. graminearum and F. pseudograminearum isolates has been studied in detail. The following


F. graminearum

100

90

9% Simillarity 80 70

281

60

CBS 166.57 CBS 389.62 70106 7137 72235 72323 022016 BUF 55 SUF 1021 NRRL 26155 NRRL.26999 Fg 7.20 Fg 7.9 Fg 11 Item 644 FG01 Iapava 85.1 Fg 7.17

F. cerealis

Fg 7.3 NRRL 254.91 Item 667 Item 619 Item 662 NRRL 13721 Fokw 1 Fokw 9 Item 664 Item 1097 NRRL 28442 Fokw 2 SUF 570

F. culmorum

NRRL 25805 NRRL 25475 NRRL 29158 NRRL 29140 NRRL 29141 CBB 251.52 72186 72187 72202 72303 022205 SUF 995 Item 345 Item 354 Item 741 Item 627

F. pseudograminearum

NRRL 29139 1250 NRRL 23069 NRRL 28069 NRRL 25334 NRRL 13821 NRRL 28388 NRRL 28060 NRRL 28061 Fg 7.12 Fg 7.15 Iapava 218 NRRL 28333 NRRL 28438

Figure 4.11 UPGMA CA dendogram based on AK, FUM, G6PDH, GDH (NADP), GPI, IDH, MDH, PEP A, PEP D, PGM and 6PGD isozyme data for 60 isolates of F. cerealis, F. culmorum, F. graminearum and F. pseudograminearum. Reprinted with permission from ref. [86]. Copyright Elsevier

enzymes were involved in the investigation: adenylate kinase (AK), NADP dependent glutamate dehydrogenase (NADP GDH), peptidase B (PEP B), peptidase D (PEP D), phosphoglucomutase (PGM), fumarate hydratase (FUM), glucose-6-phosphate isomerase (GPI), isocitrate dehydrogenase (IDH) and 6-phosphogluconate dehydrogenase (6PGDH). The typing of isolates was performed by CA. The CA dendogram of isolates based on the isozyme profile is depicted in Figure 4.11. It was found that the geographic origin

282


of the isolates exerts a negligible effect on the isozyme profile. It was further concluded that CAE followed by CA provides an excellent tool for the identification of Fusarium species [86]. Various gel electrophoretic techniques have been extensively applied for the study of the human pathogen Staphylococcus aureus. The objectives of the investigations included the study of the clonal association among various isolates, the determination of genetic variability [87–90], the assessment of the inter-species interference [91], the elucidation of the correlation between genetic background and virulence factors [92–94], and the evaluation of the differences between methicillin-resistant and sensitive strains [95]. The majority of investigations applied PFGE [96,97], multilocus sequence typing [98] or both methods together [99]. PFGE, coagulase typing, ß-lactamase production and antimicrobial susceptibility of methillicin-resistant S. aureus (MRSA) strains were measured and the data were evaluated by CA. The objective of the investigation was the study of the epidemiology of the strains isolated in hospitals. The CA dendogram separated five clusters which correlated with the type of coagulase and the production of ß-lactamase. It was stated that this combined analytical procedure is suitable for the epidemiological study of MRSA [100]. The multiple drug susceptibility towards 11 drugs and the PFGE profile of 71 strains of MRSA was measured and the similarity between the strains was compared by the application of CA to the data matrix. The CA dendogram is depicted in Figure 4.12. The results indicated that CA is a rapid and cost-effective method for the evaluation of epidemiological data [101]. A combined method was proposed for the rapid strain designation of S. aureus. The method consisted of the measurement of the allelic variations of the regulatory operon agr (groups I–IV) and the cap polymorphism (capsular types 5 and 8). Agr subgroups were resolved by restriction fragment length polymorphism (RFLP). PFGE was employed for the analysis of 219 isolates and the electrophoretic profiles were compared with CA. It was stated that the application of CA may facilitate the designation of S. aureus by agr and cap polymorphism typing and may help the delineation of agr diversification [102]. Another study has also used PFGE for the determination of the electrophoretic profile of 130 MRSA isolates. The similarities and dissimilarities between the PFGE profile of isolates was elucidated by simultaneously using CA and PCA. The comparison of PFGE data resulted in 31 different PFGE profiles, which can be clustered in eight groups by CA. PC1 and PC2 explained 64% and 23% of the total variance, respectively. The biplot is depicted in Figure 4.13. The higher classification power of PCA means it was proposed for the analysis of PFGE profiles to promote the epidemiological study of infectious diseases [103]. Natural immobilized-cells frequently occur in biological systems [104] and they can be entrapped in various polymer matrices to promote the biological degradation of environmental pollutants [105,106]. The physiological features of suspended and immobilized cells differ considerably [107]; immobilized cell systems are more resistant to antibiotics [108], and biocides [109]. In addition to their advantageous characteristics immobilized cells can cause infections [110]. It has been established that the physiology of immobilized cells differs from that of suspended ones [111]; however, the molecular basis of the difference is not entirely understood [112,113]. It was further found that the gene expression level in

•••••• HIERARCHICAL CLUSTER ANALYSIS •••••• Dendrogram using Average (Between Groups)

CASE Label Item C C C C C C C C C C C C C C C C C E E E E E E

A-1 A-1 A-1 A-1 A-1 A-1 A-1 A-1 A-1 A-1 A-1 A-1 A-1 A-1 A-1 A-1 A-1 A-2 A-2 A-3 A-3 A-3 A-3 A-3 A-3 A-3 A-3

E B B B A A A A A B E B B B B B B D D D D D D D D D D D D F

A-4 A-4 A-4 A-5 A-5 A-5 A-5 A-5 A-6 A-6 A-6 A-6 A-6 A-6 A-6 A-6 A-7 A-7 A-7 A-7 A-7 A-7 A-7 A-7 A-7 A-7 A-7

F

A-8 A-8 A-8

G G

A-9 A-9

0

6

Rescaled Distance Cluster Combine 10 15 20

25

21 23 12 15 60 19 61 16 20 33 17 62 18 13 22 3 24 14 30 52 56 29 53 27 50 68 10 57 69 54 32 56 37 64 44 1 2 4 7 5 9 51 11 26 28 5 31 25 43 48 39 42 47 65 45 46 38 48 40 63 41 34 67 36 35 64 68 69 70 71 72

Figure 4.12 Computer print out of the dendogram for 71 MRSA strains created by the diameter of growth inhibition circles in 11 drugs. Label, typing by PFGE; A-1–A-9, superimposed data describing CA grouping; Num, number of strains. Reprinted with permission from ref. [101]. Copyright Elsevier

284


0.5

3 Factor 2

4 0.0 2 −0.5

−1.0

−0.9 −0.8 −0.7 −0.6 −0.5 −0.4 −0.3 Factor 1

Figure 4.13 Plot of PC1 (factor 1) versus PC2 (factor 2) of the binary profiles generated from 130 MRSA isolates after PFGE. Numbered clusters indicate comparable clustering observed with UPGMA CA. Group 1 comprises profile types A, B, C, D and U3-U17, group 2 comprises F, G, H, U20 and U21, group 3 comprises E, U18 and U19 and group 4 comprises U1, U2, U22 and U23. Reprinted with permission from ref. [103]. Copyright Elsevier

the cultures is slightly modified [114–117]. Proteomic analyses have been frequently used for the elucidation of the differences between suspended and biofilm organisms and the marked deviations between the isolates have been demonstrated [118–127]. CA and PCA have been simultaneously applied for the evaluation of the 2-D gel electrophoretic profiles of planktonic and immobilized Pseudomonas aeruginosa cells. The CA dendogram showing the distribution of incubation conditions is depicted in Figure 4.14.

Squared Euclidean distance

(X 1000) 1 0.8 0.6 0.4 0.2

A48

A18

CB48

CB18

GW48

GW18

FA48

FCB48

FGW48

FCB18

FA18

FGW18

0

Variables (incubation conditions)

Figure 4.14 Dendogram resulting from CA of the 12 variables (incubation conditions). The further neighbour method, squared Euclidean distances, and standardized variables were used. GW, glass wool; CB, clay beads; A, agar-entrapped cultures; F, free-cell (planktonic) cultures; 18 and 48, incubation times (h). Reprinted with permission from ref. [128]. Copyright Elsevier


285

Incubation conditions form two well separated clusters proving the good differentiation capacity of CA. The first four principal components accounted for 78.75% of the total variance demonstrating the inherent diversity of the data in the original matrix. The scattering of elements on the biplots illustrates that PCA separates effectively the data of planktonic and immobilized growth and between the incubation conditions. The study demonstrated that significant difference can be found between the protein content of attached and entrapped organisms and CA and PCA can be successfully employed for the classification of both incubation conditions and type of growth [128]. Lactic acid bacteria play a considerable role in the fermentation activity of sourdough [129]; among other advantageous characteristics they decrease the development of mould [130] and improve the rheological characteristics of pizza dough [131]. Lactobacillus plantarum is an important species isolated from fermented foods and from starter cultures [132] and is able to control rope-producing strains of Bacillus sp. [133]. Up-to-date analytical procedures, such as restriction endonuclease analysis and PFGE, have been previously used in the study of lactic acid bacteria [134]. The technological characteristics and molecular diversity of 30 Lactobacillus plantarum isolates were investigated by restriction endonuclease analysis and PFGE. The amylase, protease, phytase and antirope activity of the isolates were also measured and the differentiation between the isolates was performed by PCA. The first two principal components explained 45% and 25% of the total variance. The plot of PC1 versus PC2 is depicted in Figure 4.15. It was stated that PCA makes possible the discrimination of various typologies of dough, may help the better understanding of their influence on the nutritional value and the quality of dough [135]. 3

17

3 7

2 Factor 2 (25%)

RATIO

8

2 1

4

LOGBAT 12

1 2

1

pH

0 −1 −1

25

22

9

−1.5

20 31 19 15 23 13 18

10 14 30

TTA 11

29

21

−2

24 3

5

−2 −2 −2.5

6

−1

−0.5 0 0.5 Factor 1 (45%)

26 27 TIME

1

1.5

28

2

2.5

Figure 4.15 Score plot [( ) starters with Saccharomyces cerevisiae T22; ( ) starters with Saccharomyces cerevisiae F1] and loading plot (arrows) of first and second principal components after PCA of microbial contents and fermentation properties of doughs leavened by different L. plantarum strains (doughs 1–28). Reprinted with permission from ref. [135]. Copyright Elsevier

286

4.2.3


Microbial Communities

The investigation of the microbial community in complicated accompanying matrices using nucleic acid technologies necessitates efficient and reliable DNA extraction techniques. It has been indicated on numerous occasions that the elements of the matrix, such as blood [136], faeces [137], soil [138,139], and sediment [140], may influence the efficacy of the extraction procedure. Various nucleic acid techniques have been applied for the characterization of microbial communities. The use of denaturing gradient gel electrophoresis banding pattern [141], evaluation of terminal-restriction fragment length polymorphism (T-RFLP) [142], and the analysis of terminal restriction fragment profiles of 16S rRNA genes [143] have been previously reported. Various extraction and purification methods were compared to obtain the PCRamplifiable DNA from compost for the investigation of the microbial community. T-RFLP, gel permeation and affinity chromatography, and agarose gel electrophoresis were applied to monitor the efficacy of purification steps. The similarities between the T-RFLP patterns were elucidated by PCA carried out on the peak heights and on the presence or absence of peaks. Plots of PC1 versus PC2 are depicted in Figure 4.16. It was concluded from the results of PCA that the T-RFLP profiles of the various extraction techniques did not differ considerably but the presence of humic acids may result in modified T-RFLP profiles [144]. The effect of pesticide pollution of soil on the diversity and composition of the soil microbial community has also been investigated in detail. It was established that pesticide exposure affects bacterial diversity in soil [145,146], and the same effect was also demonstrated by mercury [147,148]. Similar results were obtained by studying the impact of low herbicide concentration on bacterial composition [149]. It was further found that the degradation of pesticides depends considerably on the bacterial composition in soil and shallow aerobic aquifer [150,151]. Denaturing gradient gel electrophoresis (DGGE), colony morphology typing and sole carbon source utilization were employed for the investigation of the bacterial diversity and community structure of a sub-surface aquifer exposed to realistic low herbicide concentrations. CA was applied for the determination of the similarity between the samples. CA dendograms were computed separately from the colony morphology typing, sole carbon source utilization and DGGE analysis. The different dendograms are shown in Figure 4.17. The distribution of samples on the dendograms shows considerable differences indicating that the information content of the analytical methods markedly differs. It was further illustrated that the exposure to herbicides may modify the composition of the overall bacterial community [152]. The bacterial communities of petroleum hydrocarbon-contaminated and noncontaminated soils were also compared using DGGE and community level physiological profiles (CLPP). The similarities between the samples were evaluated by CA performed separately on the data of DGGE and CLPP. The two CA dendograms are depicted in Figures 4.18 and 4.19, respectively. Interestingly, the CA dendograms demonstrated that the results obtained by culture-dependent and -independent measuring methods are similar; that is, the geographical origin of the samples is the decisive factor and the influence of the hydrocarbon pollution is negligible [153]. It has been established many times that the crop production and the pesticide pollution equally influence the fertility of soil [154,155] and the diversity and viability of the bacterial community [156–158]. A wide variety of analytical technologies were employed for


287

0.4 0.3

PC2-18%

0.2 0.1 0.0 −0.1 −0.2 −0.3 −0.4 0.270 0.275 0.280 0.285 0.290 0.295 0.300 0.305 0.310 PC1-57% (a) Presence/absence 0.6

PC2-34%

0.4 0.2 0.0 −0.2 −0.4 0.270 0.275 0.280 0.285 0.290 0.295 0.300 0.310 PC1-67% (b) Peak heights

Figure 4.16 Principal component analysis of T-RFLPs. (The prefix refers to the extraction method and the suffix refers to compost A or B). Variability explained by each principal component is indicated. (a) Variation in number of peaks shared. PCA of variability reduced from similarity indices calculated by the presence/absence of T-RFPL peaks. (b) Variation in peak heights. PCA of variability reduced from correlation coefficients calculated from the heights of reproducible peaks. Reprinted with permission from ref. [144]. Copyright Elsevier

the evaluation of the composition of bacterial communities [159,160]. The application of PCR-DGGE [161,162] and a novel membrane-bound particulate methane monooxygenasetargeted real-time PCR assay [163] have been previously reported. Group-specific PCR-DGGE was employed for the study of the effect of long-term herbicide application on the bacterial community structure and function in an agricultural soil. The similarities between the DGGE profiles were elucidated by CA. The distribution of

Colony morphology of R2A isolates Nx2 Nx2 Nx2 C2 Nx4 Nx4 Nx4 C2 C2 B2 A2 A2 A1 A2 A1 A1 B2 B2 C1 C1 C1 B1 B1 B1 Nx1 Nx1 Nx1 Nx3 Nx3 Nx3

0

0.1

0.2

0.3 0.4 0.5 0.6 0.7 Distance to K-Measure Nearest Group

0.8

0.9

0.8

0.9

(a) Ecoplates (72 h) B2 B2 C2 C2 C2 A1 A1 A1 C1 Nx3 Nx3 Nx3 A2 A2 A2 Nx1 Nx1 Nx1 C1 B2 C1 Nx4 Nx4 Nx4 B1 B1 B1 Nx2 Nx2 Nx2

0

0.1

0.2

0.3 0.4 0.5 0.6 0.7 Distance to K-Measure Nearest Group

1

(b) DGGE (338-510) B2 A1 A2 C2 Nx2 Nx3 C1 Nx1 B1 Nx4

0

0.05

0.1 0.15 0.2 0.25 Distance to K-Measure Nearest Group

0.3

0.35

(c)

Figure 4.17 Relatedness of the microbial communities in Vejen aquifer shown by CA derived from data of (a) colony morphology typing, (b) sole carbon source utilization in EcoPlates after 72 h of incubation and (c) DGGE analysis using the primer pair 338f and 518r. The CA was performed on PCA analysed data. Reprinted with permission from ref. [152]. Copyright Blackwell Publishing


289

0.00

SAB

0.25 0.50 0.75 1.00

A1

Ac

A3

A2

K2

K1

K3

K4

Samples

Figure 4.18 Cluster analysis of DGGE banding patterns based on position of bands using unweighted pair groupings of similarity coefficient (S AB ) matrix. Samples are clustered geographically, with contamination levels having no apparent influence. Reprinted with permission from ref. [153]. Copyright Blackwell Publishing

the samples on the dendograms proves that the technology is suitable for the differentiation between herbicide-treated and non-treated soils, the highest differences being observed in the methanotropic community [164]. The relationship between the diversity and abundance of bacteria and their functional role has been frequently demonstrated [165]. This correlation has been investigated in various ecosystems [166], such as in methanogenic bioreactor communities [167] and proteobacterial ammonia oxidizer population [168]. It was further found that the conditions of plant growth [169,170] and that of rhizosphere also influence the structure of microbial

0.00

SAB

0.25

0.50

0.75

1.00

A1

Ac

A2

A3

K2

K4

K1

K3

Samples

Figure 4.19 Cluster analysis of Biolog GN substrate oxidation rates analysed using Manhattan and squared Euclidean algorithm. Samples are clustered geographically, with contamination levels having no apparent influence. S AB , similarity coefficient. Reprinted with permission from ref. [153]. Copyright Blackwell Publishing

290


communities [171–173]. It has also been illustrated that root exudates and plant composition have also a marked impact on bacterial community [174,175]. The influence of the soil improvement treatments on the bacterial community structure has been investigated in detail. Temporal temperature gradient electrophoresis (TTGE) of 16S rRNA gem fragments amplified with primers selective for eubacteria and ßproteobacterial ammonia-oxidizing bacteria (AOB) was employed for the assessment of the effect of the addition of sewage sludge and/or lime to the soil. The TTGE profiles of the various samples were compared with CA. The CA dendograms depicting the similarity of TTGE profiles of the predominant bacterial population are shown in Figure 4.20. The distribution of samples on the dendograms illustrates that the influence of the addition of sludge and lime to the soil on the eubacterial community is negligible. A similar conclusion was drawn from the CA dendogram of AOB; that is, the treatments do not greatly influence the AOB community (Figure 4.21) [176]. CA together with other multivariate techniques [177] has been frequently used for the comparison of the results of various gel electrophoretic techniques applied for the study of diverse bacterial communities. In some cases the results of CA depend considerably on the cluster algorithm. To overcome this difficulty a significance test has been developed for comparing complex microbial community fingerprints [178].

Figure 4.20 Cluster analysis of UPGMA of Sorensen similarity coefficients calculated from TTGE profiles of 165 rDNA fragments from the predominant bacterial populations in Sourhope soil. Lime (solid black lines), sludge + lime (dashed black lines), control (solid grey lines), and sludge (dashed grey lines). Numbers adjacent to dendogram denote average within treatment similarity ±2 SE. Reprinted with permission from ref. [176]. Copyright Blackwell Publishing

Electrically Driven Systems 0

25

50

75

100

t1

70.9 ± 6.2

t3

45.1 ± 8.4

t16

291

47.7 ± 6.6

t30

26.8 ± 6.0

t58

27.9 ± 14.2

Figure 4.21 Cluster analysis of UPGMA of Sorensen similarity coefficients calculated from TTGE profiles of 165 rDNA fragments from autotrophic ammonia oxidizer populations present in Sourhope soil. Lime (solid black lines), sludge + lime (dashed black lines), control (solid grey lines), and sludge (dashed grey lines). Numbers adjacent to dendogram denote average within treatment similarity±2 SE. Reprinted with permission from ref. [176]. Copyright Blackwell Publishing

Compost is considered as a safe and relatively cheap amendment improving the quality of soil organic matter [179] and supplying a wide range of microorganisms [180]. Many analytical techniques have been developed and applied for the determination of the fingerprints of microbial communities. RFLP, amplified rDNA restriction endonuclease analysis (ARDRA), single strand conformation polymorphism [181,182], DGGE, and real-time PCR [183,184] have been equally employed for this purpose. AOB play a considerable role in soil fertility [185–187]; therefore, their community structure has been frequently investigated [188,189]. Analytical methods such as cation-exchange membranes [190] and real-time PCR have been employed for the study of AOB communities [191]. The AOB population has been investigated in composts, compost-treated soils and mineral-fertilized soils employing PCR-DGGE and cloning of 16S rDNA fragments to quantitative real-time PCR. DGGE data were evaluated by CA. The CA dendogram illustrated that the method differentiates between the microbial communities of compost,

292


compost-treated and mineral-fertilized soils. It was concluded from the results that compost addition did not directly influence the microbial profile of soil but an indirect effect was demonstrated [192]. The composition and diversity of microorganisms of the living surface of marine invertebrates have also been investigated [193,194]. It was found that the bacterial epi- and endobiosis may be advantageous [195] or disadvantageous for the host organism [196,197]. The production of bioactive compounds in various epi- and endobiosis has also been observed [198,199], and their site-specific character was demonstrated [200,201]. The bacterial communities associated with four encrusting bryozoans were investigated by DGGE; Aspidelectra melolontha, Conopeum reticulum, Electra monostachys and Electra pilosa were studied. The similarities between the DGGE fingerprints were assessed by CA. The CA dendograms showing the clustering of bryozoans and the distribution of the samples of C. reticulum are depicted in Figure 4.22. It was concluded from the distribution 20

Similarity

40

60

AMI1

RI1

RII1

AMI11

AMI12

AMI13

AMI15

EMI1

AMI14

EMI12

EMI11

CRI13

EPI12

CRI12

EPI13

CRI2

EPI12

100

CRI1

80

(a) 20

Similarity

40

60

CRI2

RII1

CRI13

RI1

CRI2

100

CRI1

80

(b)

Figure 4.22 Dendogram showing the relatedness of bacterial communities on (a) bryozoan and (b) reference surfaces. The dendograms were constructed using the similarity matrix determined by Bray–Curtis coefficients and UPGMA. Reprinted with permission from ref. [202]. Copyright Elsevier

Electrically Driven Systems 1.0 0.9

293

ROCS CSUS

0.8

ROUS

Factor 2

0.7 0.6 0.5 COCS

0.4

CSCS SEUS

0.3

COUS SECS

0.2 0.1 0.0

0.1

0.2

0.3

0.4 0.5 Factor 1

0.6

0.7

0.8

0.9

Figure 4.23 Principal component analysis of the different soil samples. COCS, hydrocarboncontaminated Coalbrook soil; COUS, Coalbrook uncontaminated soil; CSCS, hydrocarboncontaminated CSIR soil; CSUS, CSIR uncontaminated soil; ROCS, hydrocarbon-contaminated Rosslyn soil; ROUS, Rosslyn uncontaminated soil; SECS, hydrocarbon-contaminated Secunda soil; SEUS, Secunda uncontaminated soil. Reprinted with permission from ref. [206]. Copyright Elsevier

of samples on the CA dendograms that the bacterial communities differ according to the type of host organisms and they are also site-specific [202]. It has been proved many times that the contamination of soil with petroleum hydrocarbons results in human health risk [203] and in the modification of the diversity and composition of the microbial community [204,205]. The influence of the geographic origin and the hydrocarbon pollution of the soil on the diversity and function of the microbial community were elucidated by PCR-DGGE and CLPP. The data matrix of PCR-DGGE was evaluated by both PCA and CA. Computations were separately performed on the uncontaminated soils too. The biplot of PC1 versus PC2 of the uncontaminated soil samples, and the CA dendogram are depicted in Figures 4.23 and 4.24, respectively. The scattering of samples on the biplot and on the CA dendogram

SECS ROUS ROCS CSUS SEUS CSCS COCS COUS 4.0

4.5

5.0

5.5

6.0

6.5

7.0

7.5

Linkage Distance

Figure 4.24 Cluster analysis of the different soil samples from different geographical locations. For definitions see Figure 4.23. Reprinted with permission from ref. [206]. Copyright Elsevier

294

Multivariate Methods in Chromatography: A Practical Guide 1.0 0.9

COUN CSUN

0.8

COCOIL COMO CSMO

Factor 2

0.7 0.6 0.5

CSCOIL

0.4

CSD COD

0.3 0.2 0.1 0.0

0.2

0.4

0.6

0.8

1.0

Factor 1

Figure 4.25 PCA of the different soils contaminated by different hydrocarbons. COD, dieselcontaminated Coalbrook soil; COUN, uncontaminated Coalbrook soil; COCOIL, crude oil contaminated Coalbrook soil; COMO, mineral oil contaminated Coalbrook soil; CSMO, mineral oil contaminated CSIR soil; CSD, diesel-contaminated CSIR soil; CSUN, uncontaminated CSIR soil; CSCOIL, crude oil contaminated CSIR soil. Reprinted with permission from ref. [206]. Copyright Elsevier

is very similar indicting that both methods can be successfully employed for the differentiation between soil types. The results of the calculations including the data obtained from contaminated soils are shown in Figures 4.25 and 4.26. PCA and CA separated the samples well according to both the geographical origin and contamination. However, it was stated that the geographical origin is the decisive factor for the determination of the diversity and function of microbial communities [206]. The influence of grassland management regime on the community structure of selected bacterial groups in soil was investigated in detail using phospholipid fatty acid (PLFA)

COCOIL CSCOIL CSD COD CSMO COMO COUN CSUN 3.0

3.5

4.0

4.5 5.0 Linkage Distance

5.5

6.0

6.5

Figure 4.26 Cluster analysis of the soils contaminated by different hydrocarbons. For definitions see Figure 4.25. Reprinted with permission from ref. [206]. Copyright Elsevier


295

0.8

PC 2

0.6

−3.0

0.4 0.2 −2.0

−1.0

0 0.0 −0.2

1.0

2.0

3.0

−0.4 −0.6 −0.8 −1

PC 1

Figure 4.27 Plot of ordination of means (±standard error) of PC1 and PC2 produced from the multivariate analysis of extracted microbial community PLFAs from undrained-unfertilized (O), drained-unfertilized ( •), drained-N fertilized, () and undrained-N fertilized ( ) grassland soils. Reprinted with permission from ref. [210]. Copyright Blackwell Publishing

profiling [207,208] and various DNA techniques [209]. PCR-DGGE was employed for the study of the changes in the communities of eubacteria, actinomycetes, ammonia oxidizers and pseudomonads. The DGGE fingerprints of the various bacterial groups were evaluated using principal coordinate analysis (PCO). PLFA data were analysed by PCA. The distribution of soil samples on the biplots clearly demonstrates that in the majority of cases the grassland management regimes exert a significant influence on the bacterial community. Similar conclusions can be drawn from the plot of PC1 versuss PC2 computed from the PLFA data (Figure 4.27). It has been further established that the soil organic matter, pH, total C and total N also influence the microbial community structure [210]. It has been frequently established that the petroleum contamination [211,212] and the metal content of the soil exerts a considerable impact on the diversity and function of the microbial community [213]. The influence of heavy metals has been demonstrated in, for example, meiofaunal community [214], AOB [215], and Actinobacteria [216]. The fate of Zn in the soil has also been vigorously investigated [217]; its hyperaccumulation [218,219] has also been reported. PCA has been employed for the evaluation of the influence of heavy metal contamination and phytoremediation on a microbial community. The 16S rRNA genes of Bacteria, ßProteobacteria and the amoA gene were PCR amplified and analysed by DGGE. The biplots of PC1 versus PC2 of the various sample sets are depicted in Figures 4.28–4.30. The ratios of variance explained were similar in each case being 65.7%, 71.7% and 69.2%, respectively. It was established from the scattering of points on the plots that heavy metal contamination exerted the highest impact on the microbial community; the role of the other parameters was negligible [220]. Although the overwhelming majority of investigations used CA or PCA for the elucidation of the similarity or difference between the elements of the original data matrix, PLS and back-propagation artificial neural network (BP-ANN) have also found application in data structure analysis. They were employed for the assessment of the relationship between

296

Multivariate Methods in Chromatography: A Practical Guide 3 2

PC2 (18.1%)

1 0 −1 −2 −3 −4 −2

−1

0

1

2

PC1 (47.6%)

Figure 4.28 Principal component analysis on the DGGE data of the Bacteria 165 rDNA PCR products. The increasing size of the symbols represents the different sampling points (3, 6, 9 and 12 months). ( ) Unplanted contaminated; ( •) rhizospheric contaminated. Reprinted with permission from ref. [220]. Copyright Elsevier

3 2

PC2 (17.3%)

1 0 −1 −2 −3 −4 −2

−1

0

1

2

PC1 (54.4%) Figure 4.29 Principal component analysis on the DGGE data of the ß-Proteobacteria 165 rDNA PCR products. The increasing size of the symbols represents the different sampling points (3, 6, 9 and 12 months). For symbols see Figure 4.28. Reprinted with permission from ref. [220]. Copyright Elsevier


297

2

PC2 (28%)

1

0

−1

−2 −2

−1

0

1

2

3

PC1 (41.2%)

Figure 4.30 Principal component analysis on the DGGE data of the amoA gene PCR products. The increasing size of the symbols represents the different sampling points (3, 6, 9 and 12 months). For symbols see Figure 4.28. Reprinted with permission from ref. [220]. Copyright Elsevier

the characteristics of soil (% sand, % clay, % nitrogen, % organic carbon and g DNA g−1 soil) and the gel electrophoretic fingerprint. The performance of PLS, BP-ANN and PCA followed by BP-ANN was compared. The root-mean-square error (RMSE) values are compiled in Table 4.6. It was stated that the method is suitable for the prediction of the properties of soil. The results in Table 4.6 indicate that the best results can be achieved by the application of the combined PCA–BP-ANN procedure [221]. Methanogenesis is the final procedure in the degradation process of organic matter in an oxic environment, such as paddy fields. The character of methanogenic communities has been frequently investigated [222,223]; the structure [224], activity and dynamics of the methanogenic archeal community [225] were assessed. It has been previously established that the geographical origin of paddy soil [226], oxic conditions [227,228], and period of cultivation [229] influence methanogenesis. The major members of the methanogenic Table 4.6 Summary of the RMSE values for the three calibration models (test set) Soil properties Sand (%) Silt (%) Clay (%) Nitrogen (%) Organic carbon (%) DNA (μg g−1 soil)

RMSE PLS

RMSE BP-ANN

RMSE PCA-BP-ANN

21.0 7.88 12.7 0.201 2.59 2.43

16.5 5.54 7.69 0.203 2.40 2.50

13.8 6.37 8.12 0.130 1.66 1.40

Reprinted with permission from ref. [221]. Copyright Elsevier

298


archeal communities were determined [230,231], and the syntropic association of butyrate and propionate oxidizers with methanogenic archea was observed [232,233]. A new primer set was developed and applied for the PCR-DGGE investigation of the community structure of methanogenic archaea in paddy field soil under double cropping rice (Oryza sativa L.) and wheat (Triticum aestivum L.). The similarity between DGGE fingerprints was assessed by PCA and CA. The distribution of samples on the dendograms indicates that the community structure is stable during the cultivation period but showed marked deviations according to the geographical origin of the samples. The biplots of principal components entirely supported the conclusions drawn from the CA dendograms. It was established that CA and PCA can be successfully employed for the differentiation between paddy fields according to the soil type. The impact of cultivation period and fertilizer treatments exerted a negligible effect on the structure of the community of methanogenic archaea [234]. Apple (Malus domestica) replant disease (ARD) occurs frequently in new orchards on old sites [235]. It was established that biotic [236] and abiotic factors are equally involved in ARD. Methyl bromide and other fumigants have been employed to control ARD [237,238]. It was demonstrated that compost can suppress soil pathogens [239], and some rootstocks can be tolerant to ARD [240]; the mode of cultivation also influences ARD [241]. Moreover, it was found that apple rootstocks modify the microbial communities of their rhizosphere [242], and fumigation decreases the diversity of the microbial community [243], causing a shift in the composition of the bacterial community [244]. It was further demonstrated that the addition of compost to the soil did not influence ARD [245,246]. The effect of pre-plant soil treatments, rootstock genotype and changes in the orchard planting position on the control of ARD was investigated; methyl bromide fumigation and untreated soil served as controls. The nematode count, soil nutrient availability, soil respiration, counts of culturable soil bacteria and Pseudomonas spp. were determined. The microbial communities were compared by CA of the PCR-DGGE fingerprints. The dendograms illustrate that bacterial communities were different in fumigated and nonfumigated soils confirming the marked impact of fumigation on the bacterial community. The CA dendograms demonstrated that the rootstock genotype considerably influenced the composition of the microbial community. It was further found that rootstock genotype was the dominant factor for tree growth and yield [247]. 4.2.4

Plant Tissues

The determination of the isozyme profile of various plant species has been frequently applied for the investigation of generic diversity and divergence [248]. Isozyme electrophoresis was employed for the measurement of the isozyme composition of the bromegrass species of the Bromus madritensis complex, including B. madritensis s.I. and B. rubens. Separation of analytes was performed in vertical polyacrylamide gel slabs and the similarity of the isoenzyme fingerprints was assessed by the CA UPGMA method. The CA dendogram of 40 cases is shown in Figure 4.31. The CA dendogram separated two clearcut clusters corresponding to the two species. It was concluded from the distribution of the points on the dendogram that CA may promote the identification of the B. madritensis species [249].


299

M1 M5 M6 M7 M8 M9 M16 M17 M18 M23 M27 M2 M12 M4 M3 M13 M24 M10 M11 M14 M15 M19 M26 M20 M25 M22 M21 MR1 R4 R9 R3 R5 R2 R6 MR2 R1 MR3 MR4 R8 R7

0

2

4

6 8 Linkage Distance

10

12

14

Figure 4.31 UPGMA dendogram of Manhattan distances for the diploid annual bromes, based on the presence/absence data matrix of 25 shared and 14 unique allozymes of 10 heterozymes. Scale relative; zero is minimal. Reprinted with permission from ref. [249]. Copyright Elsevier

The determination of the isozyme profile of Boesembergia (Zingieraceae), Kaempferia and Scaphochlamys was also carried out by PAGE. The experiments were motivated by the fact that the extract of rhizomes showed marked antitumour and antimutagenic activity [250]. Moreover, the differentiation of Boesembergia spp. according to the morphological characteristics is uncertain [251]; therefore biochemical markers are needed for their exact classification [252,253]. The following enzymes were separated and used for the classification of the samples: peroxidase, superoxide dismutase [254], glutamate dehydrogenase [255], malate dehydrogenase [256], shikimate dehydrogenase, ß-esterase, -esterase, acid phosphatase and alkaline phosphatase. It was found that the first four enzymes are suitable as biomarkers for the differentiation of species under investigation. The data were evaluated by both CA and PCA. The CA dendogram is depicted in Figure 4.32. The distribution of sample points demonstrates that Boesenbergia and Scaphochlamys are closer to each other than Boesenbergia and Kaempferia. Moreover, less differentiated subgroups can be also observed on the dendogram. The scattering of points on the map entirely supports the conclusions drawn from the dendogram [257].

300

Multivariate Methods in Chromatography: A Practical Guide Similarity (%)

95

80

60

40

20

0

S1 S2 B5 B6 B10 B11 B2 B3 B4 B5 B6 B7 B9 K1 K3 K5 K2 K4 K6

Figure 4.32 Dendogram based on UPGMA analysis of genetic similarity obtained from isozyme data showing the relationships between individual plants. Reprinted with permission from ref. [257]. Copyright Elsevier

4.3 4.3.1

Capillary Zone Electrophoresis Human Health and Pharmacology

A validated CZE method was developed for the simultaneous determination of ephedrine, pseudoephedrine, norephedrine (phenylpropanolamine) and norpseudoephedrine (cathine) in urine samples without extraction. The optimization of the separation was carried out by means of experimental design buffer concentration, buffer pH and the concentration of dimethyl-ß-cyclodextrin. A second degree MLR was applied for the calculation. It was found that the method is suitable for the quantization of these compounds in urine [258]. The concentration of 15 nucleosides in the urine samples of healthy and thyroid cancer patients was measured by CZE. The similarity and dissimilarity among the elements of the data matrix were evaluated by various multivariate mathematical-statistical methods, such as PCA, CA, stepwise discriminant analysis (SDA) and canonical discriminant analysis (CDA). The canonical plot of the patients is depicted in Figure 4.33. The scattering of sample points on the biplot of the first and second canonical functions illustrates that the method separates the healthy and thyroid cancer patients well according to the urinary nucleoside profile. The first two principal components accounted for 63% of the total variance. As for the canonical correlation, the healthy and cancer patients are correctly differentiated on the biplot of PC1 versus PC2 as illustrated in Figure 4.34. The CA results support the conclusions drawn from the distribution of elements presented in Figure 4.35. This computation method separated the healthy and cancer patients. It was stated that multivariate pattern recognition methods can increase the diagnostic value of the CZE analysis of urinary nucleosides [259].


301

2

Can2

1 0

N10

−1

C6

C3 C5

C2

C4 C10

N8

C8 C7C11

C9 N11

−2 −3

C12

N3 N6 N12 N5 N2 N4 N9

C1 −6

−5

−4

−3

−2

−1

0

1

2

3

4

5

6

7

Can1

Figure 4.33 Plot of the first and second canonical functions of 14 nucleoside variables for the 12 thyroid cancer patients (C1–C12) and 12 normal female subjects (N1–N12). Variables: pseudouridine, dihydrouridine, uridine, cytosine, 5-methyluridine, 3-methyluridine, inosine, N4 acetylcytidine, adenosine, xanthosine, N2 -methylguanosine, N2 ,N2 -dimethylguanosine, and N6 -methyladenosine. Reprinted with permission from ref. [279]. Copyright Elsevier

The application of various data evaluation methods in the new genomic technologies has been reviewed. Techniques, such as CA, PCA, Shannon entropy [260–262], and k-means [263], are discussed shortly [264]. Experimental design and ANN were employed for the optimization of the CZE separation of the bioactive imperatorin and isoimperatorin in traditional Chinese medicinal 4 C11 2

C8 C4 C6 C9 C3 C2 C10 C12

N5 N10 N4 N9 N3 N6 N8 N11 N2 N1

PC2

0

−2

C1 C7

N12

−4

−6

C5

−6

−4

−2

0

2

4

6

8

PC1

Figure 4.34 Plot of the first and second principal component variables of 14 nucleoside data for the 12 thyroid cancer patients (C1–C12) and 12 normal female subjects (N1–N12). For variables see Figure 4.33. Reprinted with permission from ref. [279]. Copyright Elsevier

302

Multivariate Methods in Chromatography: A Practical Guide C1 C10 C2 C9 C12 C3 C4 C6 C8 C11 C5 C7 N1 N2 N4 N5 N7 N8 N9 N10 N6 N11 N12 N3 0.00

Thyroid cancer

Normal

0.25

0.50

0.75 1.00 1.25 1.50 1.75 Maximum Distance Between Clusters

2.00

2.25

2.50

Figure 4.35 Dendogram based on the first eight principal component variables obtained by PCA of 12 thyroid cancer patients (C1–C12) and 12 normal female subjects (N1–N12) employing the complete linkage method with Euclidean distance calculation. Reprinted with permission from ref. [279]. Copyright Elsevier

preparation. Optimization factors were the concentration of borate and SDS, the ratio of organic modifier and the buffer pH. It was established that ANN is suitable for the prediction of the migration time of these bioactive analytes and can replace the traditional analytical procedures [265]. A CZE method was optimized by a central composite design for the separation of the enantiomers of atropine. A model was developed using MLR of a second-degree mathematical expression. It was found that the optimized method separates littorine and the enantiomers of atropine in plant extracts and pharmaceutical preparations. It was further established that the regression model is suitable for the prediction of the retention of the analytes [266]. The bioactive compounds in extracts of Glycyrrhiza glabra L., G. uralensis Fisch. and G. inflata Bat. (Leguminosae) and in commercial liquorices were separated and quantitated by CZE and the results were compared with those obtained by RP-HPLC. The similarities and differences between the samples were elucidated by PCA using the concentration of the bioactive components; glycyrrhizin, glabridin, glycyrrhetic acid, liquiritin and licochalcone A were used as variables. The biplot of PC1 versus PC2 is depicted in Figure 4.36. The distribution of sample points on the plot illustrates that CZE followed by PCA can be successfully used for the separation of samples according to their geographical origin [267]. 4.3.2

Other Applications

The peptide fingerprints resulting from the deep enzymatic fragmentation of tendons of 6-month-old female Wistar rats was determined by CZE. The objective of the investigation


303

2.5

2nd principal component

1.5 0.5

51-Iran

−0.5 −1.5

66-G. inflata

58-Daltou

50-Italy

−2.5

65-Xinjiang

−3.5 −4.5 −2.5

54-Turkey −1.5

−0.5

0.5

1.5

2.5

3.5

4.5

1st principal component

Figure 4.36 Scatter plot of the first and second principal components of European and Chinese samples obtained by PCA on the basis of the CZE peak area data of glycyrrhizin, glabridin, glycyrrhetic acid, liquiritin and licochalcone A. First principal component: eigenvalue 1.7755, contribution 35.51%; second principal component: eigenvalue 1.2396, contribution 24.79%. ( ) G. glabra, Europe; ( ) commercial liquorice, Europe; ( •) G. uralensis, China; ( ◦) commercial liquorice, China; ( ) Xinjiang-Gancao, China; ( ) G. inflata, China; No. 26 (G. echinata) was excluded from PCA. Reprinted with permission from ref. [267]. Copyright Elsevier

was the elucidation of the differences between the control group (group C), hereditary hyperglyceridemic (HTG) rats (group H), HTG rats kept on high fructose diet (group F) and HTG rats cured with gemfibrozil (group G). The peptide profile was divided into seven fractions to facilitate evaluation. The presence of overlapping peaks meant that the efficacy of the valley-to-valley (A) and baseline integrations (B) was compared. PCA was performed four times: (I) average peak areas, average peak areas -2SD, average peak areas +2SD (integration method A) treatments being the variables; (II) the reciprocal arrangement of observations and variables as in (I); (III) identical to (I) but using the results of integration method B; (IV) identical to (II) but using the results of integration method B. The two-dimensional NLMAP was computed in each instance. The first principal component of PCAI accounted for 98.72% of the total variance indicating the strong similarities between the peptide fingerprints. The two-dimensional NLMAP of treatments illustrated that treatment F (rats kept on high fructose diet) is slightly different from the others. The main parameters of PCAII are compiled in Table 4.7. Three principal components explained the total variance present in the original data matrix suggesting that the seven CZE fractions can be reduced to three background variables. The NLMAP of loadings of PCAII is depicted in Figure 4.37. The scattering of points on the map indicates the basic similarities between the information content of the peptide fractions 3, 4, 5, and 7. The first principal component explained 99.51% of the total variance using PCAIII; therefore, it cannot be employed for the comparison of peptide fingerprints. The results of PCAIV are compiled in Table 4.8 and Figure 4.38. The ratio of variance was similar to that calculated by PCAII

304

Multivariate Methods in Chromatography: A Practical Guide Table 4.7

Similarities and dissimilarities between the peptide sections

No. of principal component 1 2 3

Eigenvalue

Variance explained (%)

Total variance explained (%)

10.61 6.07 4.32

50.51 28.90 20.58

50.51 79.42 100.00

Principal component loading No. of principal components Peptide section 1 2 3 4 5 6 7

1

2

3

−0.12 0.58 −0.88 0.91 0.90 −0.15 −0.87

−0.99 −0.30 0.43 −0.35 0.35 0.78 −0.37

0.10 0.75 −0.14 −0.20 −0.27 0.61 0.32

Results of PCA (valley-to-valley integration). Reprinted with permission from ref. [268]. Copyright Elsevier.

F2 180

2

7 5 4

+

70

6

F1 240

3

1 20

Figure 4.37 Similarities and dissimilarities between peptide sections taking into consideration simultaneously each treatment. Two-dimensional NLMAP of absolute values of principal component loadings. Results of PCAII. No. of iterations: 80; maximum error: 2.98 × 10−4 (valley-to-valley integration). Numbers refer to peptide sections. Reprinted with permission from ref. [268]. Copyright Elsevier

Electrically Driven Systems Table 4.8

305

Similarities and dissimilarities between the peptide sections

No. of principal component 1 2 3

Eigenvalue

Variance explained (%)

Total variance explained (%)

12.08 5.34 3.58

57.51 25.44 17.05

57.51 82.95 100.00

Principal component loading No. of principal components Peptide section 1 2 3 4 5 6 7

1

2

3

−0.97 −0.91 0.99 0.77 0.62 0.41 −0.79

0.08 0.24 0.06 0.51 0.63 0.14 −0.60

0.21 −0.34 −0.11 0.40 0.46 −0.90 0.11

Results of PCA (baseline integration). Reprinted with permission from ref. [268]. Copyright Elsevier

F2 1

140

3

6 2

F1 60

220 4

5

7 60

Figure 4.38 Similarities and dissimilarities between peptide sections taking into consideration simultaneously each treatment. Two-dimensional NLMAP of absolute values of principal component loadings. Results of PCAIV. No. of iterations: 95; maximum error: 1.99 ×10 −4 (baseline integration). Numbers refer to peptide sections. Reprinted with permission from ref. [268]. Copyright Elsevier

306


PHOSPHATE 4 TRE S1

2

S2 S5 S4 S3 AUR MAR

0

HEO

BIR

−2 −4 −6

HUM 20

GJL GJU HEM

26 28 30 22 24 First principal component (97.5%)

Second principal component (1.4%)


but the distribution of peptide fractions on the map is markedly different. This discrepancy was tentatively explained by the difference between the methods of integration used for the evaluation of the electrophoregrams [268]. The dependence of the CZE fingerprint of natural organic matter (NOM) on the buffer additives, such as phosphate, dimethyl sulfoxide, acetonitrile and urea, was investigated. The objectives of the study were the selection of the optimal CZE conditions to produce fingerprints suitable for the differentiation of samples according to source and geographical origin. The effect of buffer additives on the distribution of samples was elucidated by PCA performed separately for each additive. The biplots of PC1 versus PC2 computed for the additives are depicted in Figure 4.39. The first principal component accounts for the overwhelming majority of variance in each instance (phosphate, 97.5%; dimethyl sulfoxide, 96.4%; acetonitrile, 97.4%; and urea, 97.7%) but the distribution of sample elements depended considerably on the character of the additive. It was concluded that PCA facilitates the differentiation between allochthonous and autochthonous organic matter [269]. The casein fractions of ternary mixtures of cow’s, ewe’s and goat’s milk were analysed by CZE and the data set was evaluated by PLS, PCR and MLR. It was concluded from the

DIMETHYL SULFOXIDE 6

HEM

4

TRE

2

AUR HEO

MAR

0

BIR

−2

GJL GJU

−4 −6

HUM 20

22 24 26 28 30 First principal component (97.5%) (b)

ACETONITRILE 5 TRE HEO AUR

3 1

MAR

−1

GJU

−3 −5

BIR GJL HEM HUM

20

22 24 26 28 30 First principal component (97.4%) (c)



(a)

UREA 7

MAR

5 TRE

3

AUR

1 GJL

−1 −3 −5

BIR GJU HEO HUM HEM

20

22 24 26 28 30 First principal component (97.7%) (d)

Figure 4.39 Biplots of PC1 versus PC2 obtained by PCA. Reprinted with permission from ref. [269]. Copyright Elsevier


as1_9

3

as1i_9 as1i_8 b2 b1

Component 2

.5 pk 0.0

as2

as1_8

−.5

−1.0 −1.0

2 Component 2

1.0

f52 f24

1

f06

0 f04

−1

−.5

307

0.0 .5 Component 1 (a)

1.0

−2

−2

−1

0 1 Component 1

2

3

(b)

Figure 4.40 Loadings plot (a) and scores plot (b) from PCA with data from determination of casein by CE of 33 cheeses, 20 from dairy A ( ) and 13 from dairy B ( ). Samples marked with an ‘r’ were the reference cheeses from dairy A, the number referring to the production week. The variance explained by the first and second principal components was 45% and 28%, respectively. Labels in (a): p-k casein (pk), α S1 casein 8P (αs1 8), α S1 casein 9P (αs1 9), αS2 casein (αs2), ß-casein A1 (b1), ß-casein A2 (b2), αS1 - casein 8P (αs1i 8) and αS1 -I casein 9P (αs1i 9). Reprinted with permission from ref. [271]. Copyright Elsevier

data that each method can be applied for the evaluation of the milk composition but the use of MLR allowed the more precise prediction [270]. A combined method including traditional wet analytical techniques, RP-HPLC of peptides, GC-MS of aroma compounds, CZE of casein fractions and sensory analysis was applied for the comparison of the ripening characteristics of Danbo cheeses. PCA was separately performed on the results of wet analysis, sensory analysis, CZE, RP-HPLC and GC. The biplots of PC1 versus PC2 of scores and loadings of CZE data are shown in Figure 4.40. The results of computations demonstrated that the composition of caseins and peptides have the best differentiation power; the difference in the sensory attributes was negligible [271]. The electrophoretic mobility of 26 substituted aromatic acids was measured in running buffers containing various concentrations of methanol and ethanol. A QSRR study was carried out using 131 calculated descriptors. The dielectric constants of mixed solvents and the energy of the highest occupied molecular orbital of methanol and ethanol were also included in the computations. Calculations were carried out with the support vector machine (SVM), radial basis function neural networks (RBFNNs) and SRA. Some results are shown in Table 4.9. It was established that the nonlinear SVM has the highest predictive power and can be used for the elucidation of the structural features influencing mobility [272]. An optimized CZE technique was developed for the separation and quantization of 10 elements in various marbles. In order to find the similarities between the samples the data matrix consisting of 25 samples and 10 elements (Ag, Fe, Cr, Mn, Cd, Co, Pb, Ni, Zn, and Cu) was evaluated by both PCA and CA. The eigenvalues and variances explained by PCA are compiled in Table 4.10. The first five principal components accounted for 90.515%

308


Table 4.9

Results of SRA for mobility in four different aqueous ethanolic solvents

μ

% (v/v) EtOH

μ1 μ2

0 20

μ3

40

μ4

60

Proposed model μ1 = −0.113 × WPSA1 + 39.031 μ2 = −4.123 × KH10 − 5.682 × MIA − 0.08 × WPSA1 + 0.0669 × PPSA2 − 0.159 × HDSA1 + 37.185 μ3 = −1.700 × WPSA3 + 753.186 × Qmax − 2.473 × KS13 − 0.520 × HOMO − 7.1484 μ4 = 3.138 × TEIP + 40.681 × Qmin − 0.042 × VOL − 47.011 × HDSA1/TMSA − 11.070 × MIA + 38.179

R

F

SE

0.680 0.975

20.699 78.167

1.4622 0.3158

0.949

47.881

0.4332

0.971

65.744

0.3385

KH10, Kier and Hall index (order 0); WPSA, weighted PPSA; MIA, moment of inertia A; PPSA2, total charge weighted PPSA; HDSA1, HA dependent HDSA-1; KS13, Kier shape index (order 3); TEIP, topographic electronic index; VOL, molecular volume; TMSA, total molecular surface area. Reprinted with permission from ref. [272]. Copyright Elsevier.

of the total variance emphasizing the relatively high variability between the samples. The scattering of sample points indicates that PCA differentiates between the marble samples according to their geographical origin. The results of CA entirely support the conclusions drawn from the biplots of PCA. The CA dendogram indicates that the marble samples can be differentiated according to their composition (Figure 4.41). It was established that CA and PCA are useful tools for the differentiation of marbles according to their composition and allow the selection of elemental characteristics according to the geographical origin of the samples [273]. CZE has also found application in the analysis of lignin samples subjected to alkaline cupric oxide oxidation. The data were evaluated by PCA, PCR and PLS. It was stated that Table 4.10 Eigenvalues and variances of principal components No. of principal component 1 2 3 4 5 6 7 8 9 10

Eigenvalue

Variance (%)


119.543 39.847 25.440 17.673 14.732 7.902 4.924 3.917 3.211 2.810

49.810 16.603 10.600 7.364 6.138 3.293 2.052 1.632 1.338 0.171

49.810 66.413 77.013 84.376 90.515 93.807 95.859 97.491 98.829 100



309

100

Similarity

80 60 40

0

22 20 21 23 13 12 9 8 18 11 7 25 6 10 24 15 19 17 16 5 14 3 4 2 1

20

Observations

Figure 4.41 Dendogram resulting from complete linkage CA based on standardized data. Reprinted with permission from ref. [273]. Copyright Elsevier

CZE followed by PCA can be employed for the classification of lignin samples taken from the Kraft process [274]. CZE in wide bore capillary tubes with fibre-coupled diode array detection was employed for the separation of model compounds, such as p-sulfanilic, sorbic and naphthalene-2sulfonic acid, tryptophan and asulam. In the case of overlapping peaks the data were evaluated by target transformation FA, fixed size moving window-target transformation FA, FSMW-EFA and OPA. It was stated that the method can be employed for the detection of analytes at low concentration in complicated biological matrices [275].

4.4

Micellar Electrokinetic Chromatography and Related Technologies

As a result of its specific separation characteristics, MEKC has been frequently used in up-to-date analytical practice. The various theoretical aspects of MEKC have also been vigorously investigated. The LSER technique has been applied for the study of solute–solvent interaction and for the characterization of the selectivity of MEKC systems. PCA has been employed for the classification of surfactants included in the experiments (sodium dodecyl sulfate, lithium dodecyl sulfate, lithium perfluorooctanesulfonate, sodium cholate, sodium deoxycholate, tetradecyltrimethylammonium bromide and hexadecyltrimethylammonium bromide). The principal component loadings and the variance explained by the individual principal components are compiled in Table 4.11. Four principal components account for the total variance present in the original data matrix; the first principal component accounts for 67% of the variance. The inclusion of compounds with highly different chemical structure and migration capacity into the experiments was motivated by the assumption that the diversity of analytes allows wide generalization of the results. The biplots of PC1 versus PC2 are depicted in Figure 4.42. The distribution of surfactants on the biplots indicates that PCA classifies the surfactants according to their chemical structure [276]. Another study employed the fragmental constant approach (FCA) and group contribution approach (GCA) for the calculation of water-SDS micelle partition coefficients (K mw )

310


Table 4.11 Contribution of the solute descriptors to the principal components (loading matrix) and percentage of variance explained for each component

PC1 PC2 PC3 PC4

E

S

A

B

V

Variance (%)

0.02 −4.22 2.78 −2.16

2.07 −3.14 −3.00 2.58

−6.13 1.82 1.85 2.25

−2.40 2.03 −3.17 −2.91

6.45 3.51 1.54 0.23

67 19 9 5


and for the prediction of the retention of uncharged analytes in MEKC. The observed and calculated values of K mw for some aromatic compounds and drugs are compiled in Table 4.12. It was stated that FCA can be used for the prediction of retention characteristics of analytes in MEKC from their chemical structure [277]. MLR and BP-ANN were employed for the prediction of the migration behaviour of cocaine and related compounds, fluvoxamine and impurities, and alkylphenones. It was established that the performance of both methods was very similar; the relative prediction error of the effective mobility was ca. 2.5% for cocaine, 1–7% for alkylphenones and 1.5–2.2% for fluvoxamine [278]. The quantification of overlapped peaks in MEKC has been performed by using PCA followed by RBFNN, GRNN and linear ANN. Computations indicated that the ANN models based on PCA are a promising tool for the quantification of overlapping CZE peaks by shortening analysis time [279]. Similarly to CZE, MEKC techniques have also been employed in studies dealing with human health care and pharmaceutical investigations. Thus, 13 urinary nucleosides were 1.5

1.0

LPFOS

0.5 LDS SDS SDC SC TTAB HTAB

0.0 −0.5 −1.0 0.70

0.75

0.80 0.85 PC1 (93%) (a)

0.90

PC2 (5%)

PC1 (7%)

1.0

1.5 LPFOS

0.5

LDS

SDC

0.0

SC

−0.5 −1.0 0.5

SDS

TTAB

0.7

0.9 1.1 PC1 (95%)

HTAB 1.3

1.5

(b)

Figure 4.42 Scores plots of the two principal components of the coefficients of the solvation parameter model applied to MEKC systems: (a) after row normalization; (b) without any pretreatment. LPFOS, lithium perfluorooctanesulfonate; LDS, lithium dodecyl sulfate; SDS, sodium dodecyl sulfate; SDC, sodium deoxycholate; SC, sodium cholate; TTAB, tetradecyltrimethylammonium bromide; HTAB, hexadecyltrimethylammonium bromide. Reprinted with permission from ref. [276]. Copyright Elsevier

2.93(0.02) 2.60(0.03) 2.02(0.02) 1.96(0.02) 2.32(0.02) 2.36(0.21) 3.97(0.05) 2.36(0.02) 3.85(0.05) 2.41(0.05) 3.15(0.06) 3.64(0.05) 3.90(0.06) 3.95(0.05) 2.93(0.06)

log K mw (FCA) predicted 3.02 2.73 2.22 2.09 2.51 2.37 4.19 2.36 3.24 1.88 3.00 2.40 3.31 3.53 2.46

log K mw (GCA) predicted 9.46(0.06) 6.61(0.19) 5.63(0.02) 4.59(0.02) 6.17(0.03) 6.59(0.05) 11.32(0.06) 5.82(0.00) 17.50(0.27) 11.28(0.16) 15.94(0.24) 15.14(0.23) 17.25(0.18) 18.76(0.17) 17.33

t R observed

9.81(0.03) 6.81(0.17) 5.18(0.05) 4.65(0.01) 5.86(0.02) 6.00(0.03) 11.22(0.59) 5.97(0.01) 17.69(0.25) 11.68(0.09) 15.26(0.18) 16.51(0.10) 18.36(0.15) 18.45(0.15) 16.48

t R (FCA) predicted

Numbers in parentheses are the standard deviations. MEKC conditions for the aromatic solutes: 40 mM SDS in 20 mM NaH2 PO4 (pH 7.0 or pH 12.0); for β blockers: 40 mM SDS in 20 mM NaH2 PO4 (pH 12.0). Reprinted with permission from ref. [277]. Copyright American Chemical Society.

a

20.87(0.01) 2.57(0.01) 2.15(0.01) 1.94(0.00) 2.40(0.02) 2.50(0.00) 4.14(0.02) 2.32(0.00) 3.90(0.02) 2.59(0.01) 3.31(0.01) 3.15(0.01) 3.46(0.01) 4.28(0.02) 3.04(0.01)

log K mw observed

Comparison of the FCA, GCA predicted and observed log Kmw and MEKC retention times for representative aromatic solutes

Butyrophenone 4-Propoxyphenol 3-Acetylbenzonitrile 3-Fluorobenzaldehyde Methyl 4-formylbenzoate α, α, α−Trifluoro- p-cresol 1,2,3,4-Tetrachlorobenzene 3-Bromobenzyl alcohol Alprenolol Atenolol Metoprolol Nadolol Acebutolol Propanolol Pindolol

Name

Table 4.12 and drugsa

312


separated and quantitatively evaluated by MEKC. Nucleosides were measured in the urine of 41 healthy controls, 20 patients with benign breast tumours, and 26 breast cancer patients. The differentiation of the samples was carried out by PCA. Computations proved that PCA differentiates 73% of breast cancer patients as illustrated on the biplot of PC1 versus PC2 (Figure 4.43). PCA did not differentiate well between the patients with benign breast tumours and controls. It was assumed that the measurement of urinary nucleosides followed by PCA promotes the better understanding of breast cancer and facilitates the better understanding of the efficacy of surgical treatment [280]. MEKC has also been applied for the separation and quantitative determination of the bioactive components in radix Salviae miltiorrizae and in various pharmaceutical preparations. Cryptotanshinone was well separated from thanshinone IIA and tanshinone I but the last two compounds eluted in overlapping peaks. Second-order electrophoregrams were applied for the deconvolution of overlapping peaks. Regression analyses found good linear correlations between the concentration of the bioactive components in the samples and the peak heights in second-order derivative electrophoregrams. It was stated that the method can be used for the quantitative determination of these compounds by MEKC [281]. The retention of 21 basic pharmaceuticals was measured on an IAM column, with micellar liquid chromatography (MLC) and MEKC. The similarities between the retention data were elucidated by PCA. It was found that hydrophobicity is the decisive molecular property governing retention in IAM, MLC and MEKC, and that each method can be applied for the elucidation of the relationship between retention characteristics and biological activity [282].

4.0 3.0 2.0 1.0

PC2

0.0 −1.0 −2.0 −3.0 −4.0 −5.0 −6.0 −1.0

−0.3

−0.4

1.1

1.8

2.5 PC1

3.2

3.9

4.6

5.3

6.0

Figure 4.43 PCA based on 13 nucleoside concentrations from healthy controls (+) and breast cancer patients ( ◦). The positions of patients with benign breast tumours (*) were marked based on the classification from healthy controls and breast cancer patients. Reprinted with permission from ref. [280]. Copyright Elsevier


313

A QSRR study was performed to compare the retention characteristics of RP-HPLC stationary phases under pressure and in electrically driven (CE) conditions. The retention times of 25 analytes were measured, extrapolated to zero concentration of organic modifier and used for QSRR computations. The log P value, various structural descriptors and the traditional LSER model were employed for the QSRR study, and the differentiation between the stationary phases was performed by PCA. Good significant linear correlations were found between the log kw and log P values on each RP stationary phase demonstrating that hydrophobicity is the decisive factor in the RP-HPLC retention of this class of analytes. The data clearly show that both models describe accurately the relationship between the retention characteristics and structural parameters. The plots of PC1 versus PC2 are depicted in Figure 4.44. The distribution of the stationary phases on the plot illustrates that the Plot of Component Weights

Component 2

0.86 0.66

SpheriODSHPLC

0.46

SpheriODSCEC

0.26 0.06 −0.14 0.19

0.21

0.23 Component 1

0.25

0.27

(a) Plot of Component Weights 0.65

SpheriC8CEC

Component 2

0.45

UniPhenHPLC SpheriC8HPLC UniPhenHPLC

0.25

UniPhenCEC HypC18HPLC

0.05

HypC18GEG UniC8HPLC HypPhenCEC HypC8HPLC UniC18HPLC UniC8CEC

−0.15

HypC8CEC

−0.35 266

266.3

266.6

266.9 267.2 Component 1

UniC18CEC

267.5

267.8 (x 0.001)

(b)

Figure 4.44 Plots of the first two components resulting from PCA of log kw data determined in (a) all the separation systems studied and (b) with Spherisorb ODS stationary phases excluded. SpherisorbODS, Spherisorb ODS; Spheric C8 , Spherisorb C8 ; HypC18 , Hypersil C18 ; HypC8 , Hypersil C8 ; HypPhen, Hypersil Phenyl; UniC18 , Unimicro C18 ; UniC8 , Unimicro C8 ; UniPhen, Unimicro Phenyl. CEC or HPLC after the name of the column indicates performed under CEC conditions or HPLC conditions, respectively. Reprinted with permission from ref. [283]. Copyright Elsevier

82.00 68.67 57.67 75.00

Background+ edge 7.33 4.33 7.00 5.67

Background 7.33 7.67 8.00 4.33

Background+ edge

Optimized no. of principal components

0.95 0.87 0.92 0.91

Background

0.96 0.93 0.91 0.91

Background+ edge


9.17 14.37 10.75 11.53

Background

8.23 10.80 11.90 11.10

Background+ edge

RMSEP (%)

a

RMSEP, root-mean-square error of prediction. Modelling was done by PCR and validated by leverage correction. Figures are averages from data obtained from models based on individual sample sets (n = 25). b Percentage of variance in the composition of the samples accounted for in the model. c ‘Background’ refers to models based on data from images that had been background filtrated using a median filter as described in the text; ‘background+edge’ refers to models based on data from images that additionally had been optimized by horizontal edge filtration. Reprinted with permission from ref. [284]. Copyright Elsevier.

76.00 63.33 65.00 66.67

Background

Explained γ -variance (%)b

Effect on performance of multivariate models from different image filtration proceduresa

Production meat MRM1 MRM2 Head meat

Filtration techniquec

Table 4.13

Electrically Driven Systems Table 4.14

315

Performance of multivariate model based on three different sample setsa

Production meat MRM1 MRM2 MRMc Head meat

Explained γ -variance (%)b

Optimized no. of principal components


RMSEP (%)

84 62 70 82 77

7 5 9 11 8

0.91 0.77 0.82 0.89 0.87

12.1 18.8 16.6 15.9 14.6

RMSEP, root-mean-square error of prediction. Modelling was done by PLS regression and validated by full cross validation (n = 75). b Percentage of variance in the composition of the samples accounted for in the model. c Model was based on data treating MRM1 and MRM2 as one quality. Reprinted with permission from ref. [284]. Copyright Elsevier.

a

retention characteristics of the Spherisorb C18 stationary phase differ considerably from those of the other stationary phases [283]. IEF was employed for the determination of the water-soluble protein profiles of mechanically recovered meat and head meat from cattle in ground beef mixtures. Multivariate calibration was performed by PCR, and model optimization was carried out by PLS. The results of calculations are compiled in Tables 4.13 and 4.14. It was concluded from the data that these multivariate methods can be applied for the detection of mechanically recovered meat and head meat from cattle in ground beef mixtures with reasonable accuracy [284]. Electrochemically modulated liquid chromatography (EMLC) employs a conductive material as stationary phase (PGC) and as a working electrode. Analyte retention can be modified by changing the electrical potential applied to the stationary phase. The theoretical principles of the method and the applications have been discussed in detail elsewhere [285].

References [1] Petersen, J. R., and Mohammad, A. A. Eds. Clinical and Forensic Applications of Capillary Electrophoresis, Humana Press, Inc., New Jersey, 2001. [2] Lunn, G. Ed. Capillary Electrophoresis Methods for Pharmaceutical Analysis, WileyInterscience, New York, 2000. [3] Krull, I. S., Mistry, K., Stevenson, R. L., and Schwarz, M. E. Capillary Electrochromatography and Pressurized Flow Capillary Electrochromatography: an Introduction, HMB, New York, 2000. [4] Altria, K. D., and Elder, D. J. Chromatogr. A 1023 (2004) 1–14. [5] Schmitt-Koplin, P., and Junkers, J. J. Chromatogr. A 998 (2003) 1–20. [6] Britz-McKibbin, P., and Terabe, S. J. Chromatogr. A 1000 (2003) 917–934. [7] Garcia-Campana, A.M., Gamiz-Gracia, L., Baeyens, W.R.G., and Barrero, F.A. J. Chromatogr. B 793 (2003) 49–74. [8] Welsch, T., and Michalke, D. J. Chromatogr. A 1000 (2003) 935–952. [9] Harry, J. L., Wilkins, M. R., Herbert, B. R., Packer, N. H., Gooley, A. A. and William, K. L. Proteomics: Electrophoresis 21 (2000) 1071–1081.

316


[10] Smilansky, Z. Electrophoresis 22 (2001) 1616–1626. [11] Veeser, S., Dunn, M. J.,and Yang, G. Z. Proteomics 1 (2001) 856–870. [12] Kaczmarek, K., Walczak, B., de Jong, S., and Vandeginste, B. G. M. J. Chem. Inf. Comput. Sci. 42 (2002) 1431–1442. [13] Kaczmarek, K., Walczak, B., de Jong, S., and Vandeginste, B. G. M. Acta Chromatogr. 13 (2002) 7–21. [14] Kaczmarek, K., Walczak, B., de Jong, S., and Vandeginste, B. G. M. Proteomics 4 (2004) 2377–2389. [15] Kaczmarek, K., Walczak, B., de Jong, S., and Vandeginste, B. G. M. Acta Chromatogr. 15 (2005) 82–96. [16] Tannock, G. W. Am. J. Clin. Nutr. 73 (2001) Suppl. 2, 410S–414S. [17] Hayashi, H., Sakamoto, M., and Benno, Y. Microbiol. Immunol. 46 (2002) 535–548. [18] Tannock, G. W., Munro, K., Harmsen, H. J. M., Welling, G. W., Smart, J., and Gopal, P. K. Appl. Environ. Microbiol. 66 (2000) 2578–2588. [19] Vanhoute, T., Huys, G., de Brandt, E., and Swings, J. FEMS Microbiol. Ecol. 48 (2004) 437–446. [20] Seksik, P., Rigottier-Gois, L., Granet, G., Sutren, M., Pochart, P., Marteau, P., Jian, R.,and Doré, J. Gut 52 (2003) 237–242. [21] Fromin, N., Hamelin, J., Tarnawski, S., Roesti, D., Jourdain-Miserez, K., Forestier, N., TeyssierCuvelle, S., Gillet, F., Aragno, M., and Rossi, P. Environ. Microbiol. 4 (2002) 634–643. [22] Dollhopf, S. L., Hashsham, S. A., and Tiedje, J. M. Microbiol Ecol. 42 (2001) 495–505. [23] Veiga, P., Juste, C., Lepercq, P, Saunier, K., Bguet, F., and Gérard, P. FEMS Microbiol. Lett. 242 (2005) 81–86. [24] Heijne, W. H. M., Stierum, R. H., Slijper, M., van Bladeren, P. J., and van Ommen, B. Biochem. Pharmacol. 65 (2003) 857–875. [25] Marengo, E., Robotti, E., Righetti, P. G., Campostrini, N., Pascali, J., Ponzoni, M., Hamdan, M., and Astner, H. Clin. Chim. Acta 345 (2004) 55–67. [26] Marengo, E., Robotti, E., Righetti, P. G., and Antonucci, F. J. Chromatogr. A 1004 (2003) 13–28. [27] Berneis, K., Jeanneret, C., Muser, J., Felix, B., and Miserez, A. Metab. Clin. Exp. 54 (2005) 227–234. [28] Izzotti, A., Bagnasco, M., Cartiglia, C., Longobardi, M., Camoirano, A., Tampa, E., Lubet, R. A., and De Flora, S. Mutat. Res. 591 (2005) 212–223. [29] Boriollo, M. F. G., Rosa, E. A. R., Bernardo, W. L. C., Spolidorio, D. M. P., Goncalves, R. B., and Höfling, J. F. Rev. Bras. Epidemol. 8 (2005) 1–16. [30] Rosa, E. A. R., Rosa, R. T., Pereira, C. V., and Höfling, J. F. Int. J. Syst. Evol. Microbiol. 50 (2000) 1343–1349. [31] Rosa, E. A. R., Rosa, R. T., Pereira, C. V., Boriollo, M. F. G., and Höfling, J. F Mem. Inst. Oswaldo Cruz 95 (2000) 801–806. [32] Rosa, E. A. R., Rosa, R. T., Pereira, C. V., and Höfling, J. F. Rev. Iberoam. Micol. 18 (2001) 60–64. [33] Rosa, E. A. R., Rosa, R. T., Boriollo, M. F. G., Bernardo, W. L. C., and Höfling, J. F. Rev. Argent. Microbiol. 35 (2003) 24–28. [34] Soll, D. R. Clin. Microbiol. Rev. 13 (2000) 322–370. [35] Boriollo, M. F. G., Rosa, E. A. R., Goncalves, R. B., and Höfling, J. F. J. Microbiol. Meth. 64 (2006) 346–365. [36] Lebuhn, M., Bathe, S., Achouak, W., Hartmann, A., Heulin, T., and Schloter, M. Syst. Appl. Microbiol. 29 (2006) 265–275. [37] Ridles, A. L., West, D. M., Stafford, K. J., Wilson, P. R., and Fenwick, S. G. N. Z. Vet. J. 48 (2000) 57–59.


317

[38] Ridler, A. L. Surveillance 28 (2001) 6–8. [39] Tcherneva, E., Rijpens, N., Jersek, B., and Herman, L. M. F. J. Appl. Microbiol. 88 (2000) 69–80. [40] Bricker, B. J., Ewalt, D. R., and Haling, S. M. BMC Microbiol. 3 (2003) 15–27. [41] Ridles, A. L., Leyland, M. J., Fenwick, S. G., and West, D. M. Vet. Microbiol. 108 (2005) 69–74. [42] Waltenbury, D. R., Leduc, L. G., and Ferroni, G. D. J. Microbiol. Meth. 62 (2005) 103–112. [43] Currie, B. J., Fisher, D. A., Howard, D. M., Burrow, J. N., Lo, D., and Selva-Nayagam, S. Clin. Infect. Dis. 31 (2000) 981–986. [44] Faa, A. G. and Holt, P. J. Commun. Dis. Intell. 26 (2002) 279–283. [45] Uleu, G. C., Currie, B. J., Clair, T. W., Mayo, M., Ketheesan, N., Labrooy, J., Gal., R., Norton, R., Smith, C. A., Barnes, J., Warner, J.,and Hirst, R. G. Microbes Infect. 3 (2001) 621–631. [46] Godoy, G., Randle, G., Simpson, A. J., Aanensen, D. E., Pitt, T. L., Kinoshita, R., and Spratt, B. G. J. Clin. Microbiol. 41 (20068) 2068–2079. [47] Currie, B. J., Fisher, D. A., Howard, D. M.,and Burrow, J. N. Acta Trop. 74 (2000) 145–151. [48] Pitt, T. L., Trakulsomboon, S., and Dance, D. A. Acta Trop. 74 (2000) 181–185. [49] Currie, B. J., and Jacups, S. P. Emerg. Infect. Dis. 9 (2003) 1538–1542. [50] Currie, B. J., Mayo, M., Anstey, N. M., Doinohoe, P., Haase, A., and Kemp, D. J. Am. J. Trop. Med. Hyg. 65 (2001) 177–179. [51] Davis, M. A., Hancock, D. D., Besser, T. A., and Call, D. R. J. Clin. Microbiol. 41 (2003) 1842–1849. [52] Grundmann, H., Hori, S., Enright, N. C., Webster, C., Tami, A., Feil, E. J., and Pitt, T. J. Clin. Microbiol. 40 (2002) 4544–4546. [53] Iredell, J., Blanckenberg, D., Arvand, M., Grauling, S., Feil, E. J., and Birtles, R. J. J. Clin. Microbiol. 41 (2003) 5071–5079. [54] Grundmann, H., Hori, S., and Tanner, G. J. Clin. Microbiol. 39 (2001) 4190–4192. [55] Cheng, A. C., Day, N. P. J., Mayo, M. J., Gal, D. and Currie, B. J. Microb. Infect. 7 (2005) 104–109. [56] Feil, E. J., Holmes, E. C., Bessen, D. E., Chan, M.-S., Day, N. P. J., Enright, M. C., Goldstein, R., Hood, D. W., Kalia, A., Moore, C. E., Zhou, J., and Spratt, B. G. Proc. Natl. Acad. Sci. USA 98 (2001) 182–187. [57] Linz, B., Schenker, M., Zhu, P., and Achtman, M. Mol. Microbiol. 36 (2000) 1049–1058. [58] Zhu, P., Vander Ende, A., Falush, D., Brieske, N., Morelli, G., Linz, B., Popovic, T., Schuurman I. G., Adegbola, R. A., Zurth, K., Gagneux, S., Platonov, A. E., Riou, J.-Y., Caugant, D. A., Nicolas, P., and Achtman, M. Proc. Natl. Acad. Sci. USA 98 (2001) 5234–5239. [59] Bart, A., Barnabé, C., Achtman, M., Dankert, J., van der Ende, A., and Tibayrene, M. Infect. Gen. Evol. 1 (2001) 117–122. [60] Radu, S., Ling, O. W., Rusul, G., Karin, M. L. A., and Nishibuchi, M. J. Microbiol. Methods 46 (2001) 131–139. [61] Perna, N. T., Plunkett, G., Burland, V., Mau, B., Glasner, J. D., Rose, D. J., Meyhew, G. F., Evans, P. S., Gregor, J., Kirkpatrick, H. A., Posfai, G., Hackett, J., Klink, S., Boutin, A., Shao, Y., Miller, L., Grotbeck, E. J., Davis, N. W., Lim, A., Dimalanta, E. T., Poamousis, K. D., Apodaca, J., Anantharaman, T. S., Lin, J., Yen, G., Schwarz, D. C., Weich, R., and Blattner, F. R. Nature 409 (2001) 529–533. [62] Feng, P., Dey, M., Abe, A., and Takeda, T. Clin. Diagn. Lab. Immunol. 8 (2001) 711–717. [63] Galland, J. C., Hyatt, D. R., Crupper, S. S., and Acheson, D. W. Appl. Environ. Microbiol. 67 (2001) 1619–1627. [64] Zhao, S., Mitchell, S. E., Meng, J., Kresovich, S., Doyle, M. P., Dean, R. B., Casa, A. M., and Weller, J. W. Microb. Infect. 2 (2000) 107–113. [65] Johnson, J. R., and O’Brian, T. T. Clin. Diagn. Lab. Immunol. 7 (2000) 265–273.

318


[66] Roberts, P. H., Davis, K. C., Gartska, W. R., and Bhunia, A. K. J. Microbiol. Methods 43 (2001) 171–181. [67] Kudva, I. T., Evans, P. S., Perna, N. T., Barrett, T. J., Ausebel, F. M., Blattner, F. R., and Calderwood, S. B. J. Bacteriol. 184 (2002) 1873–2057. [68] Hahm, B.-K., Maldonado, Y., Schreiber, E., Bhunia, A. K., and Nakatsu, C. H. J. Microbiol. Methods 53 (2003) 387–399. [69] Liebana, E., Guns, D., Garcia-Migura, L., Woodward, M. J., Clifton-Hadley, F. A., and Davies, R. H. J. Clin. Microbiol. 39 (2001) 3609–3616. [70] Guerin, P. J., de Jong, B., Heir, E., Hassletvedt, V., Kapperud, G., Stymo, K., Gondrosen, B., Lassesn, J., and Andersson, Y. Epidemiol. Infect. 132 (2004) 889–895. [71] Saunders, G. C., Dukes, J., Parkes, H. C., and Cornett, J. H. Clin. Chem. 47 (2001) 47–55. [72] Khoodoo, M. H., Issaq, M. I., and Jauneerally-Fakim, Y. Lett. Appl. Microbiol. 35 (2002) 146–152. [73] Blixt, Y., Knutsson, R., Borch, E., and Radström, P. Int. J. Food Microbiol. 83 (2003) 15–26. [74] Alvaez, J., Porwollik, S., Laconcha, I., Gisakis, V., Vivanco, A. B., Gonzalez, I., Echenagusia, S., Zabala, N., Blackmer, F., McClelland, M., Rementeria, A., and Garaizar, J. Appl. Environ. Microbiol. 69 (2003) 7531–7534. [75] De Cesare, Manfreda, G., Dambaugh, T. R., Guerzoni, M. A., and Franchini, A. J. Appl. Mirobiol. 91 (2001) 780–785. [76] Liebana, E. Res. Vet. Sci. 72 (2002) 169–175 [77] Davies, P. R., Funk, J. A., and Morrow, W. E. Swine Health Prod. 8 (2000) 25–29. [78] Barber, D. A., Bahnson, P. B., Isaacson, R. E., Jones, C. J., and Weigel, R. M. J. Food Protect. 65 (2002) 1861–1868. [79] Johnson, J. R., Clabots, C., Azar, M., Boxrud, D. J., Besser, J. M., and Thurn, J. R. J. Clin. Microbiol. 39 (2001) 3452–3460. [80] Gurtel, V., and Mayall, B. C. Int. J. Syst. Evol. Microbiol. 51 (2001) 3–16. [81] Garaizar, J., Lopez-Molina, N., Laconcha, I., Baggesen, D. L., Rementeria, A., Vivanco, A., Audicana, A., and Perales, I. Appl. Environ. Microbiol. 66 (2000) 5273–5281. [82] Eriksson, J., Löfström, C., Aspán, A., Gunnersson, A., Karlsson, I., Borch, E., de Jong, B., and Radström, P. Int. J. Food Microbiol. 104 (2005) 93–103. [83] Weigel, R. M., Qiao, B.,Teferedegne, B., Suh, D. K., Barber, D. A., Isaacson, R. E., and White, B. A. Vet. Microbiol. 100 (2004) 205–217. [84] Smith, B. J., and Sivasithamparam, K. Mycol. Res. 104 (2000) 952–961. [85] Láday, M., Bagi, F., Mestreházy, A., and Szécsi, A. Mycol. Res. 104 (2000) 788–793. [86] Láday, M., and Szécsi, A. System. Appl. Microbiol. 24 (2001) 67–75. [87] Booth, M. C., Pence, L. M., Mahasreshti, P., Callegan, M. C., and Gilmore, M. S. Infect. Immun. 69 (2001) 345–352. [88] Duour, P., Jarraud, S., Vandenesch, F., Greenland, T., Novick, R. P., Bes, M., Etienne, J., and Lina, G. J. Bacteriol. 184 (2002) 1180–1186. [89] Gilot, P., and van Leeuwen, W. J. Clin. Microbiol. 42 (2004) 1265–1269. [90] Gilot, P., Lina, G., Cochard, T., and Poutrel, B. J. Clin. Microbiol. 40 (2002) 4060–4067. [91] Goerke, C., Kummel, M., Dietz, K., and Wolz, C. J. Infect. Dis. 188 (2003) 250–256. [92] Jarraud, S., Moguel, C., Thioulouse, J., Lina, G., Meugnier, H., Forey, F., Nesme, X., Etienne, J., and Vandenesh, F. Infect. Immun. 70 (2002) 631–641. [93] Peacock, S. J., Moore, C. E., Justice, A., Kantzanou, M., Story, L., Mackie, K., O’Neill, G., and Day, N. P. J. Immunol. 70 (2002) 4987–4996. [94] Papakyiacou, H., Vaz, D., Simor, A., Louie, M., and McGavin, M.J. J. Infect. Dis. 181 (2000) 990–1000. [95] van Leeuwen, W., van Nieuwenhuizen, W., Gijzen, C., Verbrugh, H., and van Belkum, A. J. Bacteriol. 182 (2000) 5721–5729.


319

[96] Murchan, S., Kaufmann, M. E., Deplano, A., de Ryck, R., Struelens, M., Zinn, C. E., Fussing, V., Salmenlinna, S., Vuopio-Varkila, J.,El Sohl, N., Cuny, C., Witte, W., Tassios, P. T., Legakis, N., van Leeuwen, W., van Belkum, A., Vindel, A., Laconcha, I., Garaizar, J., OlsonLiljequist, B., Ransjo, U., Coombes, G., and Cookson, B. J. Clin. Microbiol. 41 (2003) 1574– 1585. [97] Grundmann, H., Hori, S., Enright, M. C., Webster, C., Tami, A., Feil, E. J., and Pitt, T. J. Clin. Microbiol. 40 (2002) 4544–4546 [98] Enright, M. C., Day, N. P., Davies, C. E., Peacock, S. J. and Spratt, B. G. J. Clin. Mirobiol. 38 (2000) 1008–1015. [99] Peacock, S. J., de Silva, G. D., Justice, A., Cowland, A., Moore, C. E., Winearis, C. G., and Day, N. P. J. Clin. Microbiol. 40 (2002) 3764–3770. [100] Takeda, S., Yasunaka, K., Kono, K., and Arakawa, K. Int. J. Antimicob. Agents 14 (2000) 39–43. [101] Yoshida, J., Umeda, A., Ishimaru, T., and Akao, M. Int. J. Infect. Dis. 5 (2001) 205–208. [102] Goerke, C., Esser, S., Kummel, M., and Wolz, C. Int. J. Med. Microbiol. 295 (2005) 67–75. [103] Caddick, J. M., Hilton, A. C., Armstrong, R. A., Lambert, P. A., Worthington, T., and Elliott, T. S. J. J. Microbiol. Meth. 65 (2006) 87–95. [104] O’Toole, G. A., Kaplan, H. B., and Kolter, R. Annu. Rev. Mirobiol. 54 (2000) 49–79. [105] Junter, G.-A., Jouenne, T., and Vilain, S. Chim. Oggi 20 (2002) 57–62. [106] Junter, G.-A., Jouenne, T., and Vilain, S. Chim. Oggi 20 (2002) 77–83. [107] Junter, G.-A., Coquet, L., Vilain, S., and Jouenne, T. Enzyme Microbiol. Technol. 31 (2002) 201–212. [108] Stewart, P. S., and Costerton, J. W. Lancet 358 (2001) 135–138. [109] Bessems, E., and Terpstra, P. M. J. Int. Biodeter. Bioeng. 51 (2003) 229–302. [110] Donlan, R. M. Emerg. Infect. Dis. 7 (2001) 277–281. [111] Drenkart, E., and Ausubel, F. M. Nature 416 (2002) 740–743. [112] Davies, D. Nat. Rev. Drug Dis. 2 (2003) 114–122. [113] Mah, T.-F. C., and O’Toole, G. A. Trends Microbiol. 9 (2001) 34–39. [114] Hall-Studley, L., and Studley, P. Curr. Opin. Biotechnol. 13 (2002) 228–233. [115] Kuchma, S. L.,and O’Toole, G. A. Curr. Opin. Biotechnol. 11 (2000) 429–433. [116] Loo, C. Y., Corliss, D. A.,and Ganeshkumar, N. J. J. Bacteriol. 182 (2000) 1374–1382. [117] Schembri, M. A., Kjergaard K., and Klemm, P. Mol. Mircobiol. 48 (2003) 253–267. [118] Dykes, G.A., Sampathkumar, B., and Korber, D.R. Int. J.Food Microbiol. 89 (2003) 1–10. [119] Oosthuizen, M. C., Steyn, B., Lindsay, D., Brözel, V. S., and von Holy, A. FEMS Microbiol. Lett. 194 (2001) 47–51. [120] Oosthuizen, M. C., Steyn, B., Theron, J., Cosette, P., Lindsay, D., van Holy, A., and Brözel, V. S. Appl. Environ. Microbiol. 68 (2002) 2770–2780. [121] Otto, K., Norbeck, J., Larsson, T., Karlsson, K. A., and Hermansson, M. J. Bacteriol. 183 (2001) 2445–2453. [122] Sauer, K., and Camper, A. K. J. Bacteriol. 183 (2001) 6579–6589. [123] Sauer, K., Camper, A. K., Erlich, G. A., Costerton, J. W., and Davies, D. G. J. Bacteriol. 184 (2002) 1140–1154. [124] Steyn, B., Oosthuizen, M. C., MacDonald, R., Theron, J., and Brözel, V. S. Proteomics 1 (2001) 871–879. [125] Svanseter, G., Welin, J., Wilkins, J. C., Beighton, D., and Hamilton, I. R. FEMS Microbiol. Lett. 205 (2001) 139–146. [126] Trémoulet, F., Duché, O., Namane, A., Martinie, B., and Labadie, J.-C. FEMS Microbiol. Lett. 215 (2002) 7–14. [127] Perrot, F., Hébraud, M., Charlionet, R., Junter, G.-A., and Jouenne, T. Electrophoresis 21 (2000) 645–653

320


[128] Vilain, S., Cosette, P., Hubert, M., Lange, C., Junter, G.-A., and Juoenne, T. Anal. Biochem. 329 (2004) 120–130. [129] Formato, A., and Pepe, O. Riv. Ingegn. Agr. 4 (2000) 243–248. [130] Lavermicocca, P., Valerio, F., Evidente, A., Lazzaroni, S., Corsetti, A., and Goberti, M. Appl. Environ. Microbiol. 66 (2000) 4084–4090. [131] Pepe, O., Villani, F., Oliviero, D., Greco, T., and Coppola, S. Int. J. Food Microbiol. 84 (2003) 319–326. [132] Torriani, S., Felis, G. E., and Dellaglio, F. Appl. Environ. Microbiol. 67 (2001) 3450– 3454. [133] Pepe, O., Blaitta, G., Moschetti, G., Greco, T., and Villani, F. Appl. Environ. Microbiol. 69 (2003) 2321–2329. [134] Blaiotta, G., Moschetti, G., Simeoli, E., Andolfi, R., Villani, F., and Coppola, S. J. Dairy Res. 68 (2001) 139–144. [135] Pepe, O., Blaiotta, G., Anastasio, M., Moschetti, G., Ercolini, D., and Villani, F. Syst. Appl. Microbiol. 27 (2004) 443–453. [136] Abu Al-Soud, W., and Radström, P. J. Clin. Microbiol. 38 (2000) 4463–4470. [137] Clement, B. G., and Kitts, C. L. Biotechnology 28 (2000) 640–646. [138] Martin-Laurent, F., Philippot, L., Hallet, S., Chaussod, R., Germano, J. C., Soulas, G., and Catoux, G. Appl. Environ. Microbiol. 67 (2001) 2354–2359. [139] Stach, J. E. M., Bathe, S., Clapp, J. P., and Burns, R. G. FEMS Microbiol. Ecol. 36 (2001) 139–151. [140] Miler, D. N. J. Microbiol. Methods 44 (2001) 49–58. [141] McCaig, A. E., Glover, L. A., and Prosser, J. L. Appl. Environ. Microbiol. 67 (2001) 4554– 4559. [142] Osborn, A. M., Moore, E. R. B., and Timmis, K. N. Environ. Microbiol. 2 (2000) 39–50. [143] Dunbar, J., Ticknor, L. O., and Kuske C. R. Appl. Environ. Microbiol. 67 (2001) 190–197. [144] LaMontagne, M. G., Michel Jr, F. C., Holden, P. A., and Reddy, C. A. J. Microbiol. Methods 49 (2002) 255–264. [145] Johnsen, K., Jacobsen, C. S., Torsvik, V., and Sorensen, J. Biol. Fertil. Soils 33 (2001) 443–453. [146] Yang, Y.-H., Yao, J., Hu, S., and Qi, Y. Microbiol. Ecol. 39 (2000) 72–79. [147] Muller, A. K., Westergaard, K., Christensen, S., and Sorensen, S. J. FEMS Microbiol. Ecol. 36 (2001) 11–19. [148] Rasmussen, L. D., and Sorensen, S. J. FEMS Microbiol. Ecol. 36 (2001) 1–9. [149] de Lipthay, J. R., Tuxen, N., Johnsen, K., Hansen, L. H., Albrechtsen, H.-J., Bjerg, P., and Aamand, J. Appl. Environ. Microbiol. 69 (2003) 461–467. [150] Broholm, M. M., Rugge, K., Tuxen, N., Hojberg, A. L., Mosbek, H., and Bjer, P. L. Water Resour. Res. 37 (2001) 3163–3176. [151] Tuxen, N., de Lipthay, J. R., Albrechtsen, H.-J., Aamand, J., and Bjerg, P. L. Environ. Sci. Technol. 36 (2002) 2205–2212. [152] de Lipthay, J., Johnsen, K., Albrechtsen, H.-J., Rosenberg, P., and Aamand, J. FEMS Microbiol. Ecol. 49 (2004) 56–69. [153] Juck, D., Charles, T., Whyte, L. G., and Greer, C. W. FEMS Microbiol. Ecol. 33 (2000) 241– 249. [154] Hutsch, B. W. Eur. J. Agron. 14 (2001) 237–260. [155] Felske, A., Wolterink, A., Van Lis, R., De Vos, W. M., and Akkermans, A. D. L. Appl. Environ. Microbiol. 66 (2000) 3998–4003. [156] Philips, C. J., Harris, D., Dollhopf, S. L., Gross, K. L., Prosser, J. I., and Paul, E. A. Appl. Environ. Microbiol. 66 (2000) 5410–5418. [157] Chang, Y. J., Hussain, A., Stephen, J. R., Mullen, M. D., White, D. C., and Peacock, A. Environ. Toxicol. Chem. 20 (2001) 2462–2468.


321

[158] Piemé, A., and Ekelund, F. Soil Biol. Biochem. 33 (2001) 831–835. [159] Hill, G. T., Mitkowski, N. A, Alrich-Wolfe, L., Emele, L. R., Jurkonie, D. D., Ficke, A., Maldonado-Ramirez, S., Lynch, S.T., and Nelson, E.B. Appl. Soil. Ecol. 15 (2000) 25–36. [160] Smit, E., Leeflang, P., Gommns, S., van den Broek, J., van Mil, S., and Wernars, K. Appl. Environ. Microbiol. 67 (2001) 2284–2291. [161] Kozdroj, J., and van Elsas, J. D. Biol. Fertil. Soils. 31 (2000) 372–378. [162] Boon, N., De Windt, W., Verstraete, W., and Top, E. M. FEMS Microbiol. Ecol. 39 (2002) 101–112. [163] Kolb, S., Knief, C., Stubner, S., and Conrad, R. Appl. Environ. Micobiol. 69 (2003) 2423–2429. [164] Seghers, D., Verthé, K., Reheul, D., Bulcke, R., Siciliano, S. D., Verstraete, W., and Top, E. M. FEMS Microbiol. Ecol. 46 (2003) 139–146. [165] Gray, N. D., and Head, I. M. Environ. Microbiol. 3 (2001) 481–492. [166] Yin, B., Crowley, D., Sparovek, G., De Melo, W. J., and Borneman, J. Appl. Environ. Microbiol. 66 (2000) 4361–4365. [167] Fernandez, A. S., Hashsham, S. A., Dollhopf, S. L., Raskin, L., Glagoleva, O., Dazzo, F. B., Hickey, R. F., Criddle, C. S., and Tiedje, J. M. Appl. Environ. Microbiol. 66 (2000) 4058–4061. [168] Webster, G., Embley, T. M., and Prosser, J. I. Appl. Environ. Microbiol. 68 (2002) 20–30. [169] Duineveld, B. M., Kowalchuk, G. A., Keijzer, A., van Elsas, J. D., and van Veen, J. A. Environ. Microbiol. 67 (2001) 172–178. [170] Yang, C. H., and Crowley, D. E. Appl. Environ. Microbiol. 66 (2000) 345–351. [171] Gomes, N. C. M., Heuer, H., Schonfeld, J., Costa, R., Mendonca-Hagler, L., and Smalla, K. Plant Soil 232 (2001) 167–180. [172] Lottman, J., Heuer, H., de Vries, J., Mahn, A., During, K., Wackernagel, W., Smalla, K., and Berg, G. FEMS Microbiol. Ecol. 33 (2000) 41–49. [173] Smalla, K., Wieland, G., Buchner, A., Zock, A., Parzy, J., Kaiser, S., Roscot, N., Heuer, H., and Berg, G. Appl. Environ. Microbiol. 67 (2001) 4742–4751. [174] Hastings, R. C., Butler, C., Singleton, I., Saunders, J. R., and McCarthy, A. J. Lett. Appl. Microbiol. 30 (2000) 14–18. [175] Griffiths, B. S., Ritz, K., Wheatley, R., Kuan, H. L., Boag, B., Cristensen, S., Ekelund, F., Sorensen, S. J., Muller, S., and Bloem, J. Soil Biol. Biochem. 33 (2001) 1713–1722. [176] Gray, N. D., Hastings, R. C., Sheppard, S. K., Loughnane, P., Lloyd, D., McCarthy, A. J., and Head, I. M. FEMS Microbiol. Ecol. 46 (2003) 11–22. [177] Aittokallio, T., Ojala, P., Nevalainen, T. J., and Nevalainen, O. Electrophoresis 21 (2000) 2947–2956. [178] Kropf, S., Heuer, H., Gruning, M., and Smalla, K. J. Microbiol. Methods 57 (2004) 187–195. [179] Rivero, C., Chienje, T., a, L. Q. and Martinez, G. Geoderma 123 (2004) 355–361. [180] Ryckeboer, J., Mergaert, J., Vaes, K., Klammer, S., De Clercq, D., Coosemans, J., Insam, H., and Swings, J. Ann. Microbiol. 53 (2003) 349–410. [181] Peters, S., Koschinsky, S., Schwieger, F.,and Tebbe, C. C. Appl. Environ. Microbiol. 66 (2000) 930–936. [182] Alfreider, A, Peters, S., Tebbe, C. C., Rangger, A., and Insam, H. Compost Sci. Util. 10 (2002) 303–312. [183] Stubner, S. J. Microbiol. Methods 57 (2004) 219–230. [184] Hermansson, A., and Lindgren, P. Appl. Environ. Microbiol. 67 (2001) 972–976. [185] Phillips, C. J., Harris, D., Dollhopf, S. L., Gross, K. L., Prosser, K. L., Prosser, J. I., and Paul, E. A. Appl. Environ. Microbiol. 66 (2000) 5410–5418. [186] Carnol, M., Kowalchuk, G. A., and De Boer, W. Soil. Biol. Biochem. 34 (2002) 1047–1050. [187] Kowalchuk, G. A., and Stephen, J. R. Ann. Rev. Microbiol. 55 (2001) 485–529. [188] Rowan, A. K., Snape, J. R., Fearnside, D., Barer, M. R., Curtis, T. P., and Head, I. M. FEMS Microbiol. Ecol. 43 (2003) 195–206.

322


[189] Freitag, T. E., and Prosser, J. I. Appl. Environ. Microbiol. 69 (2003) 1359–1371. [190] Kurola, J., Wittmannn, C, Salkinoja-Salonen, M., Aarnio, T., and Romantschuk, M. FEMS Microbiol. Ecol. 53 (2005) 463–472. [191] Okano, Y., Hristova, K. R., Leutenegger, C. M., Jackson, L. E., Denison, R. F., Gebreyesus, B., Lebauer, D., and Scow, K. M. Appl. Environ. Microbiol. 70 (2004) 1008–1016. [192] Innerebner, G., Knapp, B., Vasara, T, Romantschuk, M., and Insam, H. Soil Biol. Biochem. 38 (2006) 1092–1100. [193] Wuchter, C., Marquardt, J., and Krumbein, W. E. Helgol. Mar. Res. 57 (2000) 13–19. [194] Pukall, R., Kramer, I., Rohde, M., and Stackebrandt, E. Syst. Appl. Microbiol. 24 (2001) 623–633. [195] Hentschel, U, Schmid, M., Wagner, M., Fieseler, L., Gernert, C., and Hacker, J. FEMS Microbiol. Ecol. 35 (2001) 305–312. [196] Thevanathan, R., Nirmala, M., and Manoharam, A. Seaweed Res. Util. 22 (2000) 189–197. [197] Gerdes, G., Kadagies, N., Kaselowsky, J., Lauer, A., and Scholz, J. Facies 50 (2005) 363–389. [198] Thakur, N. L., and Anil, A. C. J. Chem. Ecol. 26 (2000) 57–71. [199] Harder, T., Lau, S. C. K., Dobretsov, S., Fang, T. K., and Quian, P.-Y. FEMS Microbiol. Ecol. 43 (2003) 337–347. [200] Masin, M., Jezbera, J., Nedoma, J., Straskrabova, V., Hejzlar, J., and Simek, K. Hydrobiologia 504 (2003) 99–113. [201] Eiler, A., Langenheder, S., Bertilsson, S., and Tranvik, L. J. Appl. Environ. Microbiol. 69 (2003) 3701–3709. [202] Kittelman, S., and Harder, T. J. Exp. Marine Biol. Ecol. 327 (2005) 201–209. [203] Hejazi, R. F., Husain, T., and Khan, F. I. J. Hazard. Mater. B99 (2003) 287–302. [204] Juck, D., Charles, T., Whyte, L. G., and Greer, C. W. FEMS Microbiol. Ecol. 33 (2000) 241–249. [205] Bundy, J. G., Paton, G. I., and Campbell, C. D. J. Appl. Microbiol. 92 (2002) 276–288. [206] Maila, M. P., Randima, P., Dronen, K., and Cloete, T. E. Soil. Biol. Biochem. 38 (2006) 303–310. [207] Grayston, S. J., Griffith, G. S., Mawdsley, J. L., Campbell, C. D., and Bardgett, R. D. Soil Biol. Biochem. 33 (2001) 533–551. [208] Fritze, H., Pietikenian, J., and Pennanen, T. Eur. J. Soil Sci. 51 (2000) 565–573. [209] Clegg, C. D., Ritz, K., and Griffiths, B. S. Appl. Soil Ecol. 14 (2000) 125–134. [210] Clegg, C. D., Loell, R. D. L., and Hobbs, P. J. FEMS Microbiol. Ecol. 43 (2003) 263–270. [211] Heinonsalo, J., Jorgensen, K. S., Haahtela, K., and Sen, R. Can. J. Microbiol. 46 (2000) 451–464. [212] Dobler, R., Saner, M., and Bachofen, R. Biorem. J. 4 (2000) 41–56. [213] Wenderoth, D. F., Stackebrandt, E., and Reber, H. H. Soil Biol. Biochem. 33 (2001) 667–670. [214] Ellis, R. J., Neish, B., Trett, M. W., Best, J. G., Weightman, A. J., Morgan, P., and Fry, J. C. J. Microbiol. Methods 45 (2001) 171–185. [215] Gong, P., Siciliano, S. D., Srivastava, S., Greer, C. W , and Sunahara, G. I. Hum. Ecol. Risk Assess. 8 (2002) 1067–1081. [216] Gremion, F.,Chatzinotas, A., and Harms, H. Environ. Microbiol. 5 (2003) 896–907. [217] Luo, Y. M., Christie, P., and Baker, A. J. M. Chemosphere 41 (2000) 161–164. [218] Delorme, T. H., Galiardi, J. V., Anle, J. S., and Chaney, R. L. Can. J. Microbiol. 47 (2001) 773–776. [219] Whiting, S. N., de Souza, M. P., and Terry, N. Environ. Sci. Technol. 35 (2001) 3144–3150. [220] Gremion, F., Chatzinotas, A., Kaufmann, K., Von Sigler, W., Harms, H. FEMS Microbiol. Lett. 48 (2004) 273–283. [221] Ramadan, Z., Hopke, P. K., Johnson, M. J., and Scow, K. M. Chem. Intell. Lab. Syst.75 (2005) 23–30.


323

[222] Lueders, T., and Friedrich, M. Appl. Environ. Microbiol. 66 (2000) 2732–2742. [223] Mizukami, S., Takeda, K., Akata, T., and Fujita, T. Jap. J. Soil. Sci. Plant. Nutr. 72 (2001) 633–241 (in Japanese, with English summary). [224] Ramakrishnan, B., Lueders, T, Dunfiled, P. S., Conrad, R., and Friedrich, M. W. FEMS Microbiol. Ecol. 37 (2001) 175–186. [225] Kruger, M., Frenzel, P., Kemnitz, D., and Conrad, R. FEMS Microbiol. Ecol. 51 (2005) 323–331. [226] Fey, A., Chin, K. J., and Conrad, R. Environ. Microbiol. 3 (2001) 295–303. [227] Sima, S.,Sordal-Klippert, M., Briukhanov, A., Netrusov, A., Linder, D., and Thauer, R. K. Appl. Environ. Microbiol. 67 (2001) 3041–3045. [228] Seedorf, H., Dreisbach, A., Hedderich, R., Shima, S., and Thauer, R. K Arch. Microbiol. 182 (2004) 126–137. [229] Kaku, N., Ueki, A., Fuji, H., and Ueki, K. Soil Biol. Biochem. 32 (2000) 2001–2010. [230] Ikenaga, M., Asakawa, S., Muraoka, Y., and Kimura, M. Soil Sci. Plant Nutr. 50 (2004) 701– 711. [231] Hashimoto-Yasuda, T., Ikenaga, M., Asakawa, S., Kim, H.-Y., Okada, M., Kobayashi, K., and Kimura, M. Soil Sci. Plant. Nutr. 51 (2005) 91–100. [232] Imachi, H., Sekiguchi, Y., Kamagata, Y., Ohashi, A., and Harada, H. Appl. Environ. Microbiol. 66 (200) 3608–3615. [233] Cahyani, V. R., Matsuya, K., Asakawa, S., and Kimura, M. Soil Sci. Plant Nutr. 49 (2003) 619–630. [234] Watanabe, T., Kimura, M., and Asakawa, S. Biol. Biochem. 38 (2006) 1264–1274. [235] Merwin, I. A., Byard, R., Robinson, T. L., Carpenter, S, Hoying, S. A., Iungerman, K. A., and Fragione, M. New York Fruit Quarterly 9 (2001) 11–15. [236] Mazzola, M., Funnell, D. L., and Raaijmakers, J. M. Microbiol. Ecol. 48 (2004) 338–348. [237] Duniway, J. M. Phytopathology 92 (2002) 1337–1343. [238] Martin, F. N. Ann. Rev. Phytopathol. 41 (2003) 325–350. [239] Blok, W. J., Coenen, T., Puji, A. S., and Termorshuizen, A. J. Biocycle 43 (2002) 58–60. [240] Isutsa, D. K. and Merwin, I. A. HortScience 35 (2000) 262–268. [241] Leinfelder, M. MSc thesis, Department of Horticulture, Cornell University, Ithaca, NY. [242] Rumberger, A., Yao, S., Merwin, I. A., Nelson, E. B., and Thies, J. E. Plant Soil 264 (2004) 246–260. [243] Ibekwe, A. M., Papiernik, S. K., Gan, J., Yates, S. R., Yang, C. H., and Crowley, D. E. Appl. Environ. Microbiol. 67 (2001) 3245–3257. [244] Inakagi, F., Sakihama, Y., Inoue, A., Kato, C., and Horikoshi, K. Environ. Microbiol. 4 (2002) 277–286. [245] Granatstein, D., and Mazzola, M. Bull. OILB/SROP 24 (2001) 265–271. [246] Neilsen, G. H., Hogue, E. K., Neilsen, D., and Forge, T. Acta Horticult. 638 (2004) 347–356. [247] Yao, S., Merwin, I. A., Abawi, G. S., and Thies, J. E. Soil Biol. Biochem. 38 (2006) 587– 599. [248] Soltis, D., and Soltis, P. Proc. Natl. Acad. Sci. USA 97 (2000) 7051–7057. [249] Oja, T. Biochem. Syst. Ecol. 30 (2002) 433–449. [250] Trakoontivakorn, G., Nakahara, K., Shinmoto, H., Takenaka, M., Onishi-Kameyama, M., Ono, H., Yoshida, M., Nagata, T., and Tsushida, T. J. Agric. Food Chem. 49 (2001) 3046–3050. [251] Hussin, K. H., Ibrahim, H., and Ali, D. A. H. A. J. Trop. Subtrop. Bot. 9 (2001) 49–54. [252] Van der Bank, H., Van der Bank M., and Van Wyk, B.E. Biochem. Syst. Ecol. 29 (2001) 469–514. [253] Paisooksantivatana, Y., Kako, S., and Seko, H. Sci. Horticult. 88 (2001) 299–307. [254] Karpinska, B., Karlsson, M., Scinkel, H., Streller, S., Suss, K., Melzer, M., and Wingsle, G. Plant. Physiol. 126 (2001)1668–1677.

324 [255] [256] [257] [258] [259] [260] [261] [262] [263] [264] [265] [266] [267] [268] [269] [270] [271] [272] [273] [274] [275] [276] [277] [278] [279] [280] [281] [282] [283] [284] [285]

Multivariate Methods in Chromatography: A Practical Guide Inokuchi, R., and Okada, M. Plant Sci. 161 (2001) 35–43. Lange, O., and Schifino-Wittmann, M. Ann. Bot. 86 (2000) 39–345. Vanijajiva, O., Suvachittanont, W., and Sirirugsa, P. Biochem. Syst. Ecol. 31 (2003) 499–511. Mateus-Avois, L., Mangin, P., and Saugy, M. J. Chromatogr. B 791 (2003) 203–216. La, S., Cho, J., Kim, J.-H., and Kim, K.-R. Anal. Chim. Acta 486 (2003) 171–182. Fuhrman, S., Cunningham, M. J., Liang, S., Wen, X., and Somogyi, R. BioSyst. 55 (2000) 5–14. Fuhrman, S., Cunningham, M. J., Wen, X., Zweiger, G., Seilhammer, J. J., and Somogyi, R. BioSyst. 55 (2000) 15–24. Cunningham, M. J., Liang, S., Fuhrman, S., Seilhamer, J. J., and Somogyi, R. Ann. NY Acad. Sci. 919 (2000) 52–67. Brown, M. P. S., Grundy, W. N., Lin, D., Cristianini, N., Sugnet, C. W., Furey, T. S., Ares, M., and Haussler, D. Proc. Natl Acad. Sci. USA 97 (2000) 262–267. Cunningham, M. J. J. Pharmacol. Toxicol. Methods 44 (2000) 291–300. Huitao, L., Ketai, W., Hongping, X., Xingguo, C., and Zhide, H. Chromatographia 55 (2002) 579–583. Mateus, L., Cherkaoui, S., Christen, P., and Veuthey, J.-L. J. Chromatogr. A 868 (2000) 285– 294. Rauchensteiner, F., Matsumura, Y., Yamamoto, Y., Yamaji, S., and Tani, T. J. Pharm. Biomed. Anal. 38 (2005) 594–600. Miksik, I., Eckhardt, A., Cserháti, T., Forgács, E., Zicha, J., and Deyl, Z. J. Chromatogr. A 921 (2001) 81–91. Egeberg, P. K., and Bergli, S. O. J. Chromatogr. A 950 (2002) 221–231. Rodriguez-Nogales, J. M. Food Chem. 98 (2006) 782–789. Sorensen, J., and Benfeldt, C. Int. Dairy J. 11 (2001) 355–362. Liu, H.-X., Zhang, R.-S., Yao, X.-J., Liu, M.-C., Hu, Z.-D., and Fan, B.-T. Anal. Chim. Acta 525 (2004) 31–41. Yucel, Y., and Demir, C. Talanta 63 (2004) 451–459. Malavaara, P., Harjula, P., Alén, R., and Knuutinen, J. Chemom. Intell. Lab. Syst. 52 (2000) 117–122. Strasik, S., Danková, M., Molnárová, M., Olvecká, E., and Kaniansky, D. J. Chromatogr. A 990 (2003) 23–33. Fuguet, E., Ráfols, C., Bosch, E., Abraham, M. H., and Rosés, M. J. Chromatogr. A 942 (2002) 237–248. Burns, S. T., and Khaledi, M. G. Anal. Chem. 76 (2004) 5451–5458. Metting, H. J., van Zomeren, P. V., van der Ley, C. P., Coenegracht, P. M. J., and de Jong, G. J. Chromatographia 52 (2000) 607–613. Zhang, Y., Li, H., Hou, A., and Havel, J. Chemom. Intell. Lab. Syst. 82 (2006) 165–175. Zheng, Y.-F., Kong, H.-W., Xiong, J.-H., Lv, S., and Xu, G.-W. Clin. Biochem. 38 (2005) 24–30. Chen, A., Li, C., Gao, W., Hu, Z., and Chen, X. J. Pharm. Biomed. Anal. 37 (2005) 811–816. Detroyer, A., Vander Heyden, Y., Cambré, I., and Massart, D. L. J. Chromatogr. A 986 (2003) 227–238. Jiskra, J., Claessens, H. A., Cramers, C. A., and Kaliszan, R. J. Chromatogr. A 977 (2002) 193–206. Skarpeid, H.-J., Moe, R. E., and Indahl, U. G. Meat Sci. 57 (2001) 227–234. Keller, D. W., and Porter, M. D. Anal. Chem. 77 (2005) 7399–7404.

Index

Numbers in italics refer to figures, those in bold refer to tables Acidithiobacillus ferrooxidans 276, 276 acrylic fibres 103–4, 104, 105 adsorption TLC 120 affinity chromatography 167 agrobiochemistry 162–8 agrobiology 36–46 alcohols 126, 127 retention index 97–9 aldehydes 99–101, 102, 244 alfafa 162–4 alkanes 36 Alzheimer’s disease 156 amines 193–4 amino acids 124–5, 125 binding strength 131, 132, 133 in garlic 104–5 retention time 156–9 in wine 193–4 ammonia-oxidising bacteria (AOB) 290, 291–2, 291 amphetamines 35–6 analyte differentiation 153–256 biological applications 156–62 health care 153–6 analyte retention 117–36

anilides 224–7 aniline derivatives 228–31 anthocyanidines 164–7, 167 anthocyanins 187, 188, 216–18, 217 anti-depressant drugs 185, 186 anti-inflammatory drugs 184 antioxidants in tea 204 in wines 190–2 apples apple replant disease (ARD) 298 quality 59–60, 63 aromatic acids 307 arsenic 208–10 artificial neural networks (ANN) 241 gasoline samples 95–6 imperatorin 301–2 soil characteristics 295–7, 297 ascorbic acid 214, 215 Astragalus caprinus 164, 165 atropine 302 bacteria 168, 170 ammonia-oxidising 290, 291–2, 291 and ARD 298

Multivariate Methods in Chromatography: A Practical Guide Tibor Cserháti C 2008 John Wiley & Sons, Ltd

326

Index

bacteria (Continued ) on bryozoa 292–3, 292 communities 286–98 fingerprint determination 291 influential factors 289–90 electrophoretic studies 274–85 and grassland management 294–5 gut 269–70, 270, 270 and herbicides 286, 288 lactic acid 285 on marine invertebrates 292–3 metal contamination 295, 296, 297 phospholipids 168, 170 in soils 286–9, 293–4 balsamic vinegar 208 barbituric acid/barbiturates 138, 178 barley grains 67 basil 206–8, 207, 208, 209 beans (red kidney) 66, 70 benzene derivatives 232 benzenesulfonamidefluoroquinolones 137–8 benzimidazole/benztriazole derivatives 120–1 benzoic acid derivatives 256 benzothiazole derivatives 181–3 bile acids 126–7, 128, 129 lipophilicity 126–7 binding amino acids 131, 132, 133 herbicides 131, 132 organic solvents 233–4 pesticides 132, 133, 134 zein-coated compounds 232–3, 234 binding parameters 233–4 and structural descriptors 233–4 binding strength 131, 132 biology 36–46, 156–62 biopartitioning micellar chromatography (BMC) 239, 239 analyte retention behaviour 239–41 Boesembergia spp. 299 breast cancer 310–12, 312 bromegrass 298, 299 bromobenzene hepatocity 270–1 Bromus spp. 298, 299 Brucella spp. 276 bryozoa 292–3, 292 buffers 267, 306 Burkholderia pseudomallei 276

Caco-2 permeability 185 Camellia spp. 202, 203–4, 204 Candida albicans 274–5 canonical correlation analysis (CCA) 4 canonical discriminant analysis (CDA) 300, 301 capacity factors 142–3 capillary columns 9, 10 capillary electrophoresis (CE) buffer nature 267 instrumentation 265–6 principles 265 capillary gel electrophoresis (CGE) 266 capillary isoelectric focusing (CIEF) 266 capillary isotachophoresis (CITP) 266 capillary zone electrophoresis (CZE) 266, 300–9 ephedrine and derivatives 300 peptides in rats 302–6 pharmacology 300–2 capsanthin 214, 214 Capsicum annuum 212, 213, 214, 215 see also paprika; pepper Capsicum frutescens 115, 116, 216, 217 Capsicum spp. 64 carbonyl compounds (airborne) 238 carotenoids 210–16 decomposition rate 212, 214, 214 SRA 212, 213 storage 212, 214, 214, 215 carrier gas 10 casein 306–7 cattle meat 314, 315, 315 cellulose acetate electrophoresis (CAE) 280–2 central nervous system (CNS) 29, 31 chalcone derivatives 183–4 charge-transfer TLC 122 cheese 72–3, 73, 195–201 ewes’ milk 199 ripening 307, 307 organic acids 195–6, 196 salt content 196–7 starter culture 197–9 temperature 196–7 Chelonia mydas 39 chemopreventive agents 140, 274, 275 chilli powders 115, 116, 216, 217 chlorophyll 168–9 cholesterol—coprostanol conversion 269–70, 270, 270 chromatographic retention 101

Index chromatographic systems classification 149–53 cluster analysis 151–2, 151, 152, 153, 154 factor analysis 153, 155 linear solvation energy relationship 149–50 multilinear regression analysis 149 nonlinear regression analysis 149 orthogonality 151–2 principal component analysis 149–51, 150 Chuangxiong 176, 180 cider 192–3, 193 cigarette smoking 155, 155 classification and regression trees (CART) 255 cluster analysis (CA) 5, 14 Acidithiobacillus ferrooxidans 276 ammonia-oxidising bacteria 290, 291 arsenic in fish 208–10, 210 Astragalus caprinus 164, 165 Bromus spp. 298, 299 bryozoan bacterial communities 292–3, 292 chemopreventive agents 274, 275 chromatographic systems 151–2, 151, 152, 153, 154 ciders 192–3, 193 coumarins and derivatives 176, 179 fatty acids 115, 116 fungi 244, 245 Fusarium spp. 281–2, 281 health studies 155, 155 and herbicides 286, 288 heroin samples 32–4, 34 hop oils 61, 65 humic materials 167, 169 inorganic pollutants 248, 249, 250 and inverse gas chromatography 22 and isozyme profiles 299, 300 jet fuel studies 94–5, 95 marble samples 307–8, 309 meat samples 75–6, 77 mercury poisoning 244–6, 246 and MRSA 282, 283 mustard and pickles volatiles 60–1, 64 PAHs 218, 219, 221, 222 Panax notoginseng 174–6, 177 paprika 212, 213 peptides 196 Pinus spp. 42 plant oil components 58–9, 60 Pseudomonas aeruginosa cells 284–5, 284 Qingkailing injections 174, 175

327

Salmonella spp. 277–80, 278, 279, 280 shellfish poisoning 169, 172 soil contamination 286, 289, 293–4, 293, 294 stationary phase behaviour 12–14, 14, 23–6, 27, 227, 228 packing materials 149, 149 thin-layer chromatography 115, 116 thyroid cancer 300, 302 and TTGE 290, 290 VOC emissions 76–9, 79 water toxicity 88–9, 89 wheat grain 67–8, 71 wine samples 48, 50, 139, 140 ageing 190, 191 zein binding 232–3, 234 Cnidium monnieri 176 cocoa 204–5 coffee 61–3, 66, 67, 201–2 columns 9–10 community level physiological profiles (CLPP) 286, 289 compost 80, 80, 167, 291–2 conifers 36–7, 39 corrrespondence factor analysis (CFA) 20 cotinine 29, 30 coumarins and derivatives 120, 176, 178, 179 cumin 70–1, 72 cyclodextrin (CD) 123–4, 123, 124 dead time 142 denaturing gradient gel electrophoresis (DGGE) 286–9, 289, 292 detectors 10–11 diesel 93, 94 dihydroxynaphthalene 231, 232 direct thermal desorption (DTD) 43 discriminant analysis (DA) 4 cheese ripening 195–6 see also canonical discriminant analysis (CDA); linear discriminant analysis (LDA); stepwise discriminant analyis (SDA) discriminant function (DF) 62, 67, 202 distribution constant 11 disulfides molecular descriptors 14–16 PCA loadings 16, 16 double gradient method 242, 242 Douglas fir 36–7 dry deposit flux (DDF) 251, 251

328

Index

Ecstasy tablets 36 electrically driven systems 265–324 separation characteristics 266 electrochemically modulated liquid chromatography (EMLC) 315 electroosmotic flow (EOF) 266, 267 electrophoretic mobility 307, 308 enantiomers 255 environmental analyses 76–97 environmental pollutants see pollutants enzyme catalysed reactions 253 ephedrine and derivatives 300 Escherichia coli 276, 277 esters 38, 38, 99, 99, 100 saturated indices 20–2 Eucalyptus spp. 36, 37 Eurycoma longifolia 32, 33, 33 extra virgin olive oils (EVOOs) 56, 56 factor analysis (FA) 2–3, 4 chromatographic systems 153 dry deposit flux 251 PAHs 80, 83 pesticides 130–1, 131, 132 phase classification 147 and phenol-micelle partition coefficient 86–8, 88 pollutants 246–8, 248 polychlorinated biphenyls 83–6, 85 soil organic matter 89–90 total suspended particulate (TSP) 252 fatty acids 37, 38, 125–7, 127 in goat meat 75–6 lipophilicity 125–6 in plant oils 58, 58, 205–6 in Sebastes spp. 39–41, 41, 42 and species classification 41 strucutral descriptors 125–6 in thin-layer chromatography 115, 116 in trout 75 in turtles 39 fermentation 285, 285 fish 73–5, 74 arsenic content 208–10, 210 flavonoids 164, 166 food and food products 46–76 of plant origin 59–68 forensic analyses 32–6 fragmental constant approach (FCA) 309–10, 311

fruit 206 fungi component determination 38, 40 Fusarium spp. 167, 280–2, 281 fuzziness/fuzzification 267, 268, 273–4, 274 garlic 104–5, 105 gas chromatography (GC) 9–111 instrumentation 9–12 principle 9 gas chromatography combustion isotope ratio mass spectrometry (GC-C-IRMS) 36 gas chromatography-flame ionization detector (GC-FID) 35–6 terpenoid analysis 36–7 gas chromatography Fourier transform infrared (GC-FTR) spectroscopy 36 gasoline 90–7, 92 GC-FID analysis 90 sample classification 95–6, 97 SPME-GC-MS analysis 93 gel electrophoretic techniques 267–300 and lymphomas 273–4 malignant protein identification 270–3 gel image matching 267 baseline reduction 269 fuzzy alignment 267, 268 noise reduction 268–9, 269 gel permeation chromatography 167 gelatins 210, 211, 212 generalised rank annihilation method (GRAM) 224, 226 gildings 104–5 ginsenosides 176, 177 glutathione 212 Glycyrrhiza spp. 302, 303 goat meat 75–6, 77 grape skins 216–18 grassland management 294–5, 295 group contribution approach (GCA) 309–10, 311 halides 88–9 haloacetic acids (HAAs) 88–9 headspace sampling 9 headspace solid phase micro-extraction (HS-SPME) see solid phase micro-extraction (SPME) health seasonal 30, 31 studies 153–6

Index heavy metal contamination 295, 296, 297 hepatocity 270–3, 271, 272 herbal medicines 174–8 quality control 174 herbicides 115, 131, 222–3 and bacterial communities 286, 288 binding 131 heroin 32–5, 35 hierarchical ascending classification (HAC) 20, 21 high performance anion-exchange (HPAE) 66 high performance liquid chromatography (HPLC) 6–7, 120, 140–256 advantages 140–1 analyte differentiation 153–256 hydrolysis reactions 254, 255 rate constants 255 multivariate classification 144–53 parameters 6–7 capacity factors 142–3 dead time 142 peak capacity 142 plate height 141–2 phase classification 145–9 practicalities adsorption 143 particle diameter 144 permeability 144 stationary phase 143–4 surface area 144 terpenoid analysis 36–7 wine analysis 54 high performance thin-layer chromatography (HPTLC), lichen samples 138–9 honey 66, 70–2, 70, 71 amino acid content 71–2 honeybees 68–70 hop oils 61, 65, 66 humic materials 45, 47, 132, 134, 167, 169 hydrolysis reactions 254, 255 immobilized artificial membrane (IAM) chromatography 185 imperatorin 301–2 inclusion complexes 17 injectors 9 inks (ballpoint) 253 inorganic ions 194, 194 inorganic pollutants 246–9, 248, 249, 249

329

inverse gas chromatography (IGC) 22 ion exchange chromatography (IEC) 193–4 isoelectric focusing (IEF) 315 jet fuels 94, 94, 95 k-nearest neighbour (k-NN) method 34 ketones 99–101, 102 Kováts index see retention index lactate dehydrogenase 161, 163 lactic acid bacteria 285 Lactobacillus plantarum 285 Lactococcus spp. 197, 200, 200 lactone isomers 17, 19, 20 Leontodon spp. 164 lichen—sandstone interaction 138–9, 139 lignin samples 308–9 linear discriminant analysis (LDA) 187, 188, 194 and fuzzification 273–4, 274 and inorganic ions 194, 195 triacylglycerols 205, 205, 206 linear regression analysis 1, 12, 13 linear solvation energy relationship (LSER) 17, 18, 149–50, 235, 252–3 retention times 241 linear solvent strength (LSS) 235, 241 linseed oil paints 244, 244, 245 lipophilicity 136, 138 bile acids 126–7 fatty acids 125–6 nicotinic acid derivatives 118 pesticides 127, 129–30 RP-HPLC 136–7 RP-TLC 136–7 liposomes 185, 186 liquid chromatography (LC) 113–264 classification 113 defined 113 liquorice 302, 303 liver diseases 156 low-density lipoprotein (LDL) 274 lymph nodes and lymphomas 273–4 Madeira wines 49 Maillard reaction products 61 Malus domestica see apples mandarin cultivars 58, 59 marble analysis 307–8, 308, 309

330

Index

meat cattle 314, 315, 315 fatty acids 75–6 goat 75–6, 77 Medicago sativa 162–4 Mencia wines 49–52 mercury 244–6, 246, 247, 286 methanogenesis 297–8 methyl esters 126, 127 micellar electrokinetic chromatography (MEKC) 265, 266, 309–15 overlapped peaks 310 retention times 309–10, 311 micellar solid-phase microextraction (MSPME) 86–8 and phenols 86–8 microbes 38, 286–98 see also bacteria microbiology 168–73 microorganisms 274–85 see also bacteria; microbes microwave-assisted extraction (MAE) 214–16, 219–21, 222 milk 306–7 mineral oils see oils (mineral) minerals, in wine 187, 187, 194, 194, 195 molluscs 244–6, 246, 247 monoterpenes 39, 42, 43 MRSA 282, 283, 284 multilocus enzyme electrophoresis (MLEE) 274–5 multinear regression analysis (MLR) see multivariate (multiple) linear regression (MLR) multiple nonlinear regression (MNLR) 2 multivariate (multiple) linear regression (MLR) 1–2, 101, 252–3 barbituric acid derivatives 138 chromatographic systems 149, 153 fatty acids 125, 126 herbicides 115 pharmaceuticals 28–9 plant oils 56 retention indexes 20–2 retention parameters 137 rhubarb 176, 180 stationary phase behaviour 12–14, 16–17, 17, 23–4, 26 mustard 60–1, 64

natural medicines see herbal medicines natural organic matter (NOM) 306 natural product chromatograms 26–8 Neisseria meningitis 277 neuroblastoma 271–3, 272, 273, 273 nicotinic acid derivatives 118–20, 119 lipophilicity 118 retention behaviour 118–20 topological indices 118–20 nitrates 88–9 nonlinear mapping (NLM,NLMAP) 5, 123, 124, 125 peptides 303, 304, 305 stationary phases 227, 228, 229, 231 steroidal drugs 178–81, 182 nonlinear regression analysis 149 nonsteroidal anti-inflammatory drugs 121–2, 121, 122 nucleosides 300, 301, 310–12, 312 O-, N-, S-heterocyclic compounds 101–3 obesity 156 Ocimum spp. 206–8, 207, 208, 209 oils (mineral) biodegradation 95, 96 see also diesel; gasoline oils (plant) see plant oils oligostyrene isomers 253 olive oils 205–6 Australian 57 chromatographic profiles 57, 57 extra virgin 56, 56 Oncorhynchus mykiss 75, 75 organic solvents 232–3 binding parameters 233–4 organophosphorus insecticides 129, 130, 131 Origanum vulgare 43, 45 orthogonal chromatographic systems 151–2 overpressured-layer chromatography (OPLC) 139 oxo compounds 99–101, 102, 244 packed columns 9–10 paddy fields 297–8 Panax notoginseng 174–6 paprika 212, 213, 214, 215 parallel factor analysis (PARAFAC) 61 parameters 6–7 binding 233–4 fuzzification 273–4, 274

Index gas chromatography 6 HPLC 6–7, 141–3 retention 17, 136–8 solvent strength 236 stepwise regression analysis 17 thin layer chromatography 6–7 partial least squares (PLS) techniques 2 chalcone derivatives 183–4 linseed oil paints 244, 245 meat samples 315, 315 obesity 156 retention indices 16, 101, 102 peak capacity 142 peak overlap 185, 310 peak purity determination 183 peak resolution 235 pepper 64, 68, 214, 214 peptides in cheese ripening 196–9, 197, 198 profiles in rats 302–6, 304, 304, 305, 305 retention behaviour 159–61, 160, 162, 163 surfactant interactions 135, 135 permeability 144 pesticides 127–36, 132, 222, 223, 286 adjuvants 135–6 binding 132, 133 to -cyclodextrin polymer (BCDP) 132, 134 to humus extract 132, 134 lipophilicity 127, 129–30 polar surface area (PSA) 132, 133 petrol see gasoline petroleum 244 soil contamination 286 pharmaceuticals 28–32 anti-inflammatory 184 and central nervous system 29, 31 retention times 184 synthetic 178–86 phenolic compounds and antioxidants 190–2 phenolic acids 164, 166 phenols and derivatives 86–8, 88, 187, 227 bind to zein 135–6, 228–31, 230, 231 QSSR studies 255–6 polyphenols 187, 188, 188, 188 retention characteristics 234–5, 235, 236, 237 in wines 187, 187, 189–90 and ageing 190, 190

331

phospholipids (PLs) 38, 161–2, 164, 168, 244, 245 in bacteria 168, 170 composition 140 phytoplankton 169–72, 172, 173, 173 pickles 60–1, 64 pigments algal 168–9, 171, 173 phytoplankton 169–72, 173 Pinus spp. 42, 43, 44 plankton 169–72, 172, 173, 173 plant components 38, 39 plant metabolites 43, 45 plant oils 56–9, 205 components 59, 60, 61, 62 fatty acids 58, 58, 205–6 volatiles 57–8 plant tissues 298–300 plate height 141–2 polarity indicators 19–20 pollutants 218–21, 232–43 inorganic 246–9, 248, 249, 249 NLMAP 232, 233 organic solvents 232–3 retention prediction 235 toxicity prediction 239 polybrominated diphenyl esters 90, 92 polychlorinated biphenyls (PCBs) 83–6, 85, 86 from river sediments 86 polycyclic aromatic hydrocarbons (PAHs) 80–3, 81, 83, 84, 218–21 in marine sediments 218 physicochemical parameters 219 retention time 219, 220 in soil 218 in surface water 80, 82 in urban air 80–3 vehicular emissions 221 in worms 219–21 polyethylene (PE) systems 22–3, 25 poly( p-phenylenethynylene) (PPE) 103, 104 polyphenols 187, 188, 188, 188 polysaccharides 252 polystyrene 234, 253 polyurethane (PU) systems 22–3, 25 potato pathogens 64, 69 principal component analysis (PCA) 2–4 acrylic fibres 103–4, 104, 105 amines 193–4 amino acids 193–4

332

Index

principal component analysis (PCA) (Continued ) amphetamines 35 anthocyanidines 164–7, 167 anti-depressant drugs 185, 186 bacteria gut 269–70, 270, 270 metal contamination 295, 296, 297 phospholipids 168, 170 benzimidazole/benztriazole derivatives 120–1, 120, 121 breast cancer 312, 312 cheese ripening 195–6, 196, 197, 199, 199, 200 Danbo 307, 307 chromatographic systems 149–51, 150 Chuangxiong 176, 180 ciders 192–3, 193 coffee 202, 202 compost pile emissions 80, 80 coumarins and derivatives 176, 178 cumin 70–1, 72 diesel 93, 94 dihydroxynaphthalene 231, 232 eigenvalue definition 3–4 fermentation 285, 285 forensic analyses 32–4, 34, 35, 35, 36 garlic proteins 104–5, 105 gasoline 93, 93, 93, 95–6 gelatins 210, 211, 212 Glycyrrhiza spp. 302, 303 grassland management 294–5, 295 health 156 hepatocity 270–3, 271, 272 heroin samples 35 honey 66, 70 honeybees 68–70 hop oils 61, 65, 66 inorganic ions 194, 194 inverse gas chromatography 22 jet fuel analysis 94–5, 94 Leontodon spp. 164 lichen—sandstone interaction 138–9, 139 linseed oil paints 244, 244 liposomes 185, 186 marble samples 307–8, 308 meat products 75–6 mercury poisoning 244–6, 247 MRSA 282, 284 natural organic matter 306, 306

neuroblastoma 271–3, 272, 273, 273 Ocimum spp. 206–8, 207, 208, 209 PAHs 80, 82, 220–1, 220, 221 peak purity determination 183 pepper varieties 64, 68 peptides in cheese ripening 196, 197–9, 198 in rats 303–6, 304, 304, 305, 305 retention times 159–61, 160, 162, 163 pesticides 222, 223 phase classification 145–6 phenolic compounds 234–5, 235, 236, 237 in wines 187, 187 Pinus spp. 42, 44 plant components 38, 39, 58 pollutants 248–9, 249, 250 polychlorinated biphenyls 86, 86, 87 potato pathogens 64, 69 Qingkailing injections 174, 174, 175 red kidney beans 66, 70 resins 106, 106, 107 retention characteristics 17–19 of columns 243, 243 of esters 20–2, 22 of hydrocarbons 20 Salix spp. 176 snow sampling 249–51 soil contamination 293–4, 293, 294 stationary phase behaviour 12–14, 17–19, 23–4 strawberries 206, 207 surfactants 309, 310, 310 T-RFLP 286, 287 tea 202, 203–4, 204 thin-layer chromatography 115 three-dimensional 4 thyroid cancer 300, 301 toxins 167, 168 triazines 223–4, 224, 225 two-dimensional 3 uses 3 VOC emissions 76–9, 79 waxes (skiing) 105, 106 wine analysis 49, 52, 139, 139, 189–90, 189 complexity 48, 48 off-flavours 54, 55 principal component regression (PCR) 278–80 meat samples 314, 315 Prorocentrum lima isolates 169, 172

Index proteolysis 196–7 starter culture 197–9 proteomics technique 270–3, 272 Pseudomonas aeruginosa 284–5, 284 Pseudotsuga menziesii 36–7 pulsed field gel electrophoresis (PFGE) 276–7, 278–80 MRSA investigation 282 purine nucleobases 162 pyrolysis-gas chromatography (Py-GC) 36, 38, 40 acrylic fibres 103–4 canauba waxes 103, 103 and poly( p-phenylenethynylene) 103, 104 resin mixtures 105–6 sewage sludge 90 soil organic matter 89–90 wood samples 45, 47 Qingkailing injections 174, 174, 175 quantitative structure-activity relationship (QSAR) 239, 240 quantitative structure-retention relationship (QSSR) 236–8, 255–6 analyte retention behaviour 239–41 quince jam 208 redfish 39–41, 41, 42 rennet 201, 201 resin mixtures 105–6, 106, 107 trans-resveratrol 188 retention characteristics parameters 136–8 and biological activity 136–8 of pharmaceuticals 312 stationary phases coating 146 RP-HPLC 313–15, 313 retention factor 113–14 retention index 11, 14, 16, 101 alcohols 97–9 for O-, N-, S-heterocyclic compounds 101–3 prediction 99 saturated esters 99, 99, 100 sulfur compounds 99 retention times 11–12 of amino acids 156–9 BMC studies 239–41 correspondence factor analysis 20 hierarchical ascending classification 20, 21

333

hydrophobic-substraction model 242–3 IAM chromatography 185 LSER studies 241 LSS model 241 of PAHs 219, 220 partition coefficient based model 242, 243 of peptides 159–61, 160, 162, 163 of pharmaceuticals 184 QSRR studies 239–41 shift corrections 178 structural descriptors model 242, 243 reversed-phase high performance liquid chromatography (RP-HPLC) columns 227–32, 227, 228, 229, 230 coumarin derivatives 120 hydrophobic surface area 136–7 lipophilicity 136–7 pesticides 129–30 multicomponent peak resolution 221–2 phenol/aniline derivatives 229, 230 reversed-phase thin layer chromatography (RP-TLC) 115 bile acids 126–7 cyclodextrins 123–4, 124 fatty acids 125–6 hydrophobic surface area 136–7 lipophilicity 127–30, 136–7 pesticides 127–30 rhubarb 176, 180 river sediments 86 Saccharomyces cerevisiae 285 salami 76, 78 Salix spp. 176 Salmonella spp. 277–80, 278, 279, 280 Salviae miltiorrizae 312 saponins 174–6, 177 Sebastes spp. 39–41, 41, 42 selectivity map 147, 148 separation temperature 10 sequential projection pursuit (SPP) 93, 94 serum nucleoside data 156, 158 sesquiterpenoids 164, 166 sewage sludge 90, 91 shellfish poisoning 169, 172 Shenmai injection 176 silanol groups 143–4 silica and silicic acid 114, 143–4 size exclusion chromatography (SEC) 253 smoking (cigarettes) 155, 155

334

Index

snow sampling 249–51, 251 soils bacterial community 286–9, 293–4 characteristics 295–7, 297 fertility 286–9 improvement 290 organic matter 89–90, 90 and pyrolysis-gas chromatography 89–90 petroleum contamination 286 solid phase micro-extraction (SPME) 49, 52, 54, 56 and coffee varieties 63 petroleum analysis 93 potato pathogens 64, 69 solvent strength parameters 236 spectral mapping (SPM) 5, 146, 147, 147, 147, 148 spirochetes 37, 38 Staphylococcus aureus 276–7, 282 starches 218 stationary phases 144, 227–32, 227, 228, 234 classification 234 cluster analysis 149, 149 gas chromatography comparisons 12–26 ion chromatography 147–9 porous grapitised carbon (PGC) column 146–7 principal component analysis 145–6, 146 spectral mapping 146, 147, 147, 147, 148 thin-layer chromatography 114 stepwise discriminant analysis (SDA) 190, 192 stepwise regression analysis (SRA) 2 barbiturates 178, 181 on carotenoids 212, 213 electrophoretic mobility 307, 308 peptide—surfactant interaction 135, 135 and retention parameters 17, 19, 20 steroidal drugs 123, 123, 124 retention times 178–81, 182 strawberries 206, 207 sugars 66, 70 sulfonates (aromatic) 224 supercritical fluid extraction (SFE) 216 support vector machine (SVM) 307 surface active agents see surfactants surface area 144 surfactants 309, 310, 310 as column fillers 23 peptides 135, 135, 136

tannin 139–40 tea 204 black 202–4 green 204 temporal temperature gradient electrophoresis (TTGE) 269–70, 290, 290 terminal-restriction fragment length polymorphism (T-RFLP) 286, 287 terpenoids 36–7, 37 in wines 49, 52 Theobroma cacao 204–5 thermally assisted hydrolysis and methylation (THM) 38, 40 thin-layer chromatography (TLC) 113–40 advantages 113 lichen samples 138–9 pesticides, adjuvants 135 plate development 114–15 potency values 115, 117, 118 practice 114–15 retention factor 113–14 sample preparation 114 theory 113–14 Thymus spp. 58–9, 60, 61 thyroid cancer 300, 301 tocopherols 120 tomatoes 63 topological index 101, 101 for alkanes 36 for branched alkenes 101 total suspended particulate (TSP) 251, 252 toxicity data 240 prediction of pollutants 239 water 88–9, 89 toxins 167, 168, 169, 172 transcriptomics technique 270–3, 271 triacylglycerols (TGs) 205–6 triazines 222–4, 224, 225 trihalomethanes (THMs) 88–9 trimethylsilyl (TMS) derivatives 28–9, 28, 30 trout 75, 75 turtles (green) 39 two-dimensional GC (GC x GC) jet fuel analysis 94–5, 94, 95 with time-of-flight MS (TOFMS) 43, 46 Tymbra capitata 59, 62 typical-conditions model (TCM) 235–6

Index univariate calibration 224, 226 urine nucleoside data 156, 157 UV signals 185 varimax extended rotation (VER) 235 volatile organic compounds (VOCs) 32, 32 from car seats 76, 78 from landfill sites 76–9, 79 wall paintings 252 water toxicity 88–9, 89 wheat grain 67–8, 71 whey proteins 200 willow 176 wines 47–56, 187–95 ageing 189–90, 194 amino acids 193–4 antioxidant capacity 190–2 classification 50, 52

335

enzyme treatment 48–9, 51 GC-FID analysis 48–9, 51 GC-HPLC analysis 54 geographical zones 188, 188 global volatile signature 52, 53 inorganic composition 187, 194, 194, 195 loadings of analytes 49, 51 odour unit values 53 off-flavours 54, 55 optimal composition 54, 55 overpressured-layer chromatography analysis 139, 139, 140 phenolic compounds 54–6, 187, 189–90, 189, 191 variety differentiation 49, 52 volatile compounds 48, 54–6 wood tissue sampling 45, 47 zein 228–31, 232–3, 234

Multivariate Methods in Chromatography: A Practical Guide

Gas Chromatography and Lipids: A Practical Guide

Multivariate Methods in Epidemiology

A Practical Guide to Research Methods

Validating Chromatographic Methods: A Practical Guide

A practical guide to research methods

Practical Protein Chromatography

A practical guide to pseudospectral methods

Gas Chromatography and Mass Spectrometry. A Practical Guide

Gas Chromatography and Mass Spectrometry: A Practical Guide

Methods of Multivariate Analysis

Affinity Chromatography Methods Protocols

Methods of multivariate analysis

Gas Chromatography and Mass Spectrometry: A Practical Guide

Gas Chromatography and Mass Spectrometry. A Practical Guide

Methods of Multivariate Analysis

Methods of Multivariate Analysis

Thin Layer Chromatography: A Modern Practical Approach

Quantitative Thin-Layer Chromatography: A Practical Survey

Practical Methods in Ecology

Practical High-Performance Liquid Chromatography

Multivariate Statistical Methods in Quality Management

High Resolution Chromatography: A Practical Approach (Practical Approach Series)

High Resolution Chromatography: A Practical Approach (Practical Approach Series)

Practical high-performance liquid chromatography

Affinity Chromatography Methods and Protocols

Lecturing: A Practical Guide

Beekeeping: A Practical Guide

Playwriting: A Practical Guide

Chemometrics: A Practical Guide

Multivariate Methods in Chromatography: A Practical Guide

Gas Chromatography and Lipids: A Practical Guide

Multivariate Methods in Epidemiology

A Practical Guide to Research Methods

Validating Chromatographic Methods: A Practical Guide

A practical guide to research methods

Practical Protein Chromatography

A practical guide to pseudospectral methods

Gas Chromatography and Mass Spectrometry. A Practical Guide

Gas Chromatography and Mass Spectrometry: A Practical Guide

Methods of Multivariate Analysis

Affinity Chromatography Methods Protocols

Methods of multivariate analysis

Gas Chromatography and Mass Spectrometry: A Practical Guide

Gas Chromatography and Mass Spectrometry. A Practical Guide

Methods of Multivariate Analysis

Methods of Multivariate Analysis

Thin Layer Chromatography: A Modern Practical Approach

Quantitative Thin-Layer Chromatography: A Practical Survey

Practical Methods in Ecology

Practical High-Performance Liquid Chromatography

Multivariate Statistical Methods in Quality Management

High Resolution Chromatography: A Practical Approach (Practical Approach Series)

High Resolution Chromatography: A Practical Approach (Practical Approach Series)

Practical high-performance liquid chromatography

Affinity Chromatography Methods and Protocols

Lecturing: A Practical Guide

Beekeeping: A Practical Guide

Playwriting: A Practical Guide

Chemometrics: A Practical Guide

Recommend Documents