JOURNAL OF CHROMATOGRAPHYLIBRARY- volume 57
retention and selectivity in liquid chromatography prediction, standardisa...
96 downloads
631 Views
7MB 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
JOURNAL OF CHROMATOGRAPHYLIBRARY- volume 57
retention and selectivity in liquid chromatography prediction, standardisation and phase comparisons
This Page Intentionally Left Blank
JOURNAL OF CHROMATOGRAPHY LIBRARY- volume 57
retention and selectivity in liquid chromatography prediction, standardisation and phase comparisons edited by
Roger M. Smith Department of Chemistry, Loughborough University of Technology,Loughborough, Leicestershire LEI 7 3TU, UK
ELSEVIER Amsterdam
-Lausanne-New York -Oxford -Shannon
-Tokyo
1995
ELSEVIER SCIENCE B.V. Sara Burgerhartstraat 25 PO. Box 211,1000 AE Amsterdam, The Netherlands
ISBN 0-444-81539-2
0 1995 Elsevier Science B.V. 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 or otherwise, without the prior written permission of the publisher, Elsevier Science B.V., Copyright & Permissions Department, P.O. Box 521,1000 AM Amsterdam, The Netherlands. Special regulations for readers in the U.S.A. - This publication has been registered with the Copyright Clearance Center Inc. (CCC), Salem, Massachusetts. Information can be obtained from the CCC about conditions under which photocopies of parts of this publication may be made in the U.S.A. All other copyright questions, including photocopying outside of the U.S.A., should be referred to the Publisher. No responsibility is assumed by the publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein.
This book is printed on acid-free paper. Printed in The Netherlands
V
Contents .
Retention prediction based on molecular structure ....................... R.M. Smith Introduction....................................................................................................... Structure and retention ...................................................................................... 1.2.1 Chromatographic functional group contributions .................................. 1.2.2 Related prediction studies...................................................................... Functional group effects on retention indices .................................................... 1.3.1 Retention prediction based on n-alkanes and n-alkylbenzenes.............. 1.3.2 Retention prediction based on alkan-2-ones.......................................... Prediction of retention indices based on alkyl aryl ketones ............................... 1.4.1 Monofunctional compounds .................................................................. 1.4.1.1 Aromatic functional groups ..................................................... 1.4.1.2 Aliphatic functional groups ..................................................... 1.4.1.3 Relationship between substituent indices and octanol-water partition substituent increments............................................... 1.4.2 Polyfunctional compounds .................................................................... 1.4.2.1 General prediction model ........................................................ 1.4.2.2 Meta and para groups ............................................................. 1.4.2.3 Ortho-substituents ................................................................... 1.4.3 CRIPES and expert systems .................................................................. 1.4.4 Comparisons with published retention values........................................ Other retention index prediction studies............................................................ 1.5.1 Retention prediction based on polynuclear aromatic hydrocarbons ...... Conclusion......................................................................................................... Acknowledgements............................................................................................ Appendix 1.1: Coefficients of regression equations for the effect of eluent on parent, aromatic and aliphatic substituent indices ........................................ References .........................................................................................................
Chapter 1 1.1 1.2
1.3
1.4
1.5 1.6 1.7 1.8
1.9
.
Retention prediction of pharmaceutical compounds...................... K. Valk6 Introduction ....................................................................................................... Definition and determination of retention.......................................................... Dependence of retention on the column and mobile phase composition ........... Correlation of retention parameters to the molecular parameters obtained by molecular modelling .......................................................................................... Retention prediction based on topological matrix and information the0ry ........ . . Retention prediction based on the hydrophobicity of drugs ..............................
1
1 2 3 4 5 6 8 11 13 13 15 19 21 23 29 32 33 38 40 41 43 43 44 45
Chapter 2
47
2.1 2.2 2.3 2.4
47 48 50
2.5 2.6
53 60 62
VI
Contents
Retention prediction based on empirical increment values ................................ Retention prediction based on experimental retention values. thermodynamic considerations with multiparameter approaches ..................................... 2.9 Applications of retention predictions of pharmaceutical compounds ................ 2.10 Acknowledgements............................................................................................ 2.11 References ......................................................................................................... 2.7 2.8
68 76 87 90 90
.
Retention index scales used in high-performance liquid chromatography ................................................................................. R.M. Smith 3.1 Introduction ....................................................................................................... 3.1.1 Relative retention times ......................................................................... 3.1.2 Internal and external standards .............................................................. 3.1.3 Retention indices ................................................................................... 3.2 Retention index scales in chromatography ........................................................ 3.2.1 Gas chromatography.............................................................................. 3.2.2 Supercritical fluid chromatography ....................................................... 3.2.3 Liquid chromatography ......................................................................... 3.2.4 Micellar electrokinetic chromatography ................................................ 3.3 Retention index scales in high-performance liquid chromatography................. 3.3.1 n-Alkanes ............................................................................................... 3.3.2 n-Alkylbenzenes .................................................................................... 3.3.3 Alkan-2-ones ......................................................................................... 3.3.4 Alkyl aryl ketones. ................................................................................. 3.3.5. 1-Nitroalkanes....................................................................................... 3.3.6 Polynuclear aromatic hydrocarbons....................................................... 3.3.7 Miscellaneous retention index scales..................................................... 3.3.7.1 Phenolic esters......................................................................... 3.3.7.2 Aliphatic esters ........................................................................ 3.3.7.3 Other retention index scales .................................................... 3.3.8 Comparisons between retention index scales......................................... 3.4 Applications of retention index scales ............................................................... 3.4.1 Reproducibility and transferability of retention indices......................... 3.4.2 Identification ......................................................................................... 3.4.3 Characterization of separation systems .................................................. 3.4.3.1. Column hold-up volume .......................................................... 3.4.3.2 Stationary phase characterization............................................ 3.4.3.3 Mobile phase characteristics ................................................... 3.4.4 Lipophilicity and biological activity ...................................................... 3.4.5 Structure-retentionrelationships............................................................ 3.5 Conclusions ....................................................................................................... 3.6 References ......................................................................................................... Chapter 3
93 93 94 95 97 98 99 100 103 105 107 107 109 109 111 115 115 116 116 117 118 118 122 122 125 125 126 127 133 135 137 139 140
Contents
VII
Chapter 4. 4.1 4.2 4.3
4.4
4.5
4.6 4.7 4.8 4.9
Application of retention indices for identification in high performance liquid chromatography .............................................. R.M. Smith Introduction ....................................................................................................... Advantages and problems.................................................................................. Pharmaceuticals and toxicological drug samples............................................... 4.3.1 Toxicological drug analysis................................................................... 4.3.1.1 Studies based on the alkan-2-ones........................................... 4.3.1.2 Studies based on the alkyl aryl ketones ................................... 4.3.1.3 Studies based on the 1-nitroalkanes ........................................ 4.3.2 Drug metabolites ................................................................................... Natural products ................................................................................................ 4.4.1 Fungal metabolites................................................................................. 4.4.1.1 Mycotoxins.............................................................................. 4.4.1.2 Other hngal metabolites ......................................................... 4.4.2 Plant products ........................................................................................ 4.4.2.1 Spices and flavour components............................................... 4.4.2.2 Plant toxins .............................................................................. 4.4.2.3 Gliadins................................................................................... 4.4.3 Lichen constituents ................................................................................ 4.4.4 Other natural products ........................................................................... Environmental samples...................................................................................... 4.5.1 Chlorinated compounds......................................................................... 4.5.2 PAH and aromatic hydrocarbons........................................................... Miscellaneous samples ...................................................................................... Conclusions ....................................................................................................... Appendix 1: Reported retention indices in HPLC ............................................. References .........................................................................................................
145 145 145 147 147 148 149 156 156 157 157 158 159 160 161 162 163 163 164 164 164 165 165 165 166 167
Chapter 5. 5.1 5.2
5.3
Application of nitroalkanes and secondary retention index standards for the identification of drugs ......................................... M. Bogusz Introduction....................................................................................................... The use of HPLC as a standardized identification method in toxicology .......... 5.2.1 Standardization of retention using straight phase silica......................... 5.2.2 Standardization of retention using reversed phase silica ....................... 5.2.2.1. The concept of secondary standards for retention index scale 5.2.2.2 Assessment of identification potentials of a HPLC/RI system 5.2.2.3 Influence of a biological matrix on the identification potential of a HPLC/RI system ........................................................ 5.2.3 Standardization of detection for toxicological screening procedures .... References .........................................................................................................
171 171 173 173 175 183 199 203 204 205
VIII
Contents
Chapter 6.
209
6.1 6.2 6.3 6.4 6.5
209 210 213 215 217 217 219 221 221 223 228 230 230
6.6
6.7 6.8
Identification using retention indices in gradient HPLC ............... P . Kuronen Introduction ....................................................................................................... Problems ofreversed-phase columns ................................................................ Selection of a retention index standard.............................................................. Principles of gradient elution............................................................................. Chromatographic behaviour of retention index standards ................................. 6.5.1 Isocratic conditions................................................................................ 6.5.2 Gradient elution ..................................................................................... Retention indices in qualitative identification ................................................... 6.6.1 Calculation of retention indices ............................................................. 6.6.2 Reproducibility of gradient retention indices ........................................ 6.6.3 Confirmation of identification ............................................................... Conclusions ....................................................................................................... References .........................................................................................................
Chapter 7. 7.1 7.2
7.3 7.4 7.5
7.6 7.7 7.8
Characterization of retention and selectivity in reversed-phase LC using interaction indices ............................................................. P . Jandera Introduction ....................................................................................................... Interaction indices as the descriptors of retention ............................................. 7.2.1 Retention in binary mobile phases ......................................................... 7.2.2 Retention in ternary mobile phases ........................................................ Calibration of the scale of interaction indices ................................................... Prediction of the retention under changing mobile phase composition using interaction indices.............................................................................................. Interaction indices and the selectivity of separation .......................................... 7.5.1 Non-homologous compounds ................................................................ 7.5.2 Homologous and oligomeric series........................................................ Conclusions ....................................................................................................... Glossary ofthe terms ......................................................................................... References .........................................................................................................
.
Lipophilic and polar indices ............................................................. P . Jandera 8.1 Introduction ....................................................................................................... 8.2 Retention in homologous series as the basis of lipophilic and polar indices ..... 8.3 Molecular structure and lipophilic and polar indices......................................... 8.3.1 Anc and Aq indices as the descriptors of the lipophilicity and polarity of solutes .......................................................................................... 8.3.2 Structural contributions to lipophilic and polar indices ......................... 8.4 Prediction of retention using lipophilic and polar indices ................................. 8.4.1 Selection of the reference calibration homologous series...................... 8.4.2 Precision of the predicted retention data ............................................... 8.5 Characterization of selectivity using lipophilic and polar indices .....................
Chapter 8
235 235 236 238 243 247 250 253 253 255 263 264 266 269 269 269 274 274 276 279 279 279 283
Contents
8.6 8.7 8.8
8.5.1 Binary mobile phases............................................................................. 8.5.2 Ternary mobile phases ........................................................................... 8.5.3 Gradient elution ..................................................................................... Conclusions ....................................................................................................... Glossary of the terms ......................................................................................... References .........................................................................................................
Chapter 9.
Solvent selectivity .............................................................................. S.D. West ............................................................................................ 9.1 Introduction ....................................................................................................... 9.2 Experimental ..................................................................................................... 9.2.1 Chemicals and reagents ......................................................................... 9.2.2 Instrumentation...................................................................................... 9.2.3 Determination of retention data ............................................................. 9.2.4 Selection of solvents and solutes ........................................................... 9.3 Results and discussion 9.3.1 Prediction of retention and resolution with steroids .............................. 9.3.1.1 Retention indices as a function of volume fraction of strong solvent ..................................................................................... 9.3.1.2 Prediction of retention indices for steroids.............................. 9.3.1.3 Prediction of resolution of steroid mixtures ............................ 9.3.1.4 Optimization of resolution of steroid mixtures........................ 9.3.1.5 Resolution and the solvent selectivity triangle concept ........... 9.3.2 Retention and selectivity studies with benzene derivatives ................... 9.3.2.1 Retention index variation with solvent selectivity................... 9.3.2.2 Resolution and the solvent selectivity triangle ........................ 9.3.2.3 Prediction of resolution ........................................................... 9.3.2.4 HPLC resolution as a function of retention index differences. 9.3.2.5 Preadjustment of retention indices for prediction of resolution .......................................................................................... 9.3.3 Reasons for failure of the solvent selectivity triangle ............................ 9.3.4 Extension of theory to gas chromatography........................................... 9.3.4.1 Prediction of resolution ........................................................... 9.3.4.2 McReynolds constants and resolution ..................................... 9.3.4.3 Application of the selectivity triangle to the characterization of GC stationary phase selectivity ........................................... 9.3.5 Characterization of RP-HPLC selectivity with adjusted retention indices....................................................................................................... 9.3.5.1 Calculation of adjusted retention indices................................. 9.3.5.2 Probes for characterization of RP-HPLC solvent selectivity... 9.3.5.3 Quantitative prediction of resolution with any RP solvent ...... 9.4 Conclusions ....................................................................................................... 9.5 Acknowledgments ............................................................................................. 9.6 References .........................................................................................................
IX
283 287 290 292 292 294 297 297 297 299 299 299 299 300 301 301 302 303 307 309 311 311 311 314 3 14 318 320 323 323 324 326 327 327 329 331 334 335 335
X
Contents
.
Chapter 10
Retention and selectivity for polycyclic aromatic hydrocarbons in reversed-phase liquid chromatography ...................................... L.C. Sander and S.A. Wise 10.1 Introduction ....................................................................................................... 10.2 Stationary phase characteristics affecting selectivity in RPLC .......................... 10.2.1 Phase type .............................................................................................. 10.2.2 Isomer separations................................................................................. 10.2.3 Assessing column shape selectivity ....................................................... 10.2.4 Pore size effects ..................................................................................... 10.2.5 Bonding density..................................................................................... 10.2.6 Bonded phase length.............................................................................. 10.2.7 Mobile phase composition..................................................................... 10.2.8 Temperature........................................................................................... 10.3 Retention indexes .............................................................................................. 10.3.1 Retention index data .............................................................................. 10.3.2 Length-to-breadth ratio .......................................................................... 10.3.3 Planar and non-planar PAHs ................................................................. 10.3.4 Methyl-substituted PAHs....................................................................... 10.4 Summary ............................................................................................................ 10.5 References .........................................................................................................
11.1
11.2
11.3
11.4
.
Comparison of novel stationary phases ........................................... J.J. Pesek and E.J. Williamsen Introduction ....................................................................................................... 11.1.1 Characteristicsfor the ideal reversed-phase HPLC stationary phase ..... Characterizationtechniques............................................................................... 1 1.2.1 Spectroscopic techniques....................................................................... 1 1.2.1.1 IR methods .............................................................................. 11.2.1.2 NMR methods ......................................................................... 1 1.2.1.3 ESCA methods ........................................................................ 11.2.2 Thermal methods ................................................................................... 1 1.2.3 Elemental analytical methods ................................................................ 11.2.4 Chromatographic characterization......................................................... Monomeric octadecyl silica............................................................................... 1 1.3.1 Separation mechanisms for ODS phases ............................................... 11.3.2 ODS limitations and adjustments........................................................... 11.3.2.1 Different bonded phase-silica linkages.................................... 11.3.2.2 Differences in ODS columns ................................................... Novel silica phases ............................................................................................ 11.4.1 Non-C18alkane phases .......................................................................... 11.4.2 Materials containing other functional groups ........................................ 11.4.2.1 Cyanopropyl-bondedphases ................................................... 11.4.2.2 Amine and phenyl phases ........................................................ 11.4.2.3 Non-alkyl phases bonded through a reactive olefin................. 11.4.2.4 Polymer bonded phases ...........................................................
Chapter 11
337 337 338 338 341 343 346 349 350 352 353 357 358 360 364 365 368 368 371 371 371 372 372 372 374 375 376 376 377 378 378 379 381 383 383 383 385 385 385 385 388
Contents
XI
11.4.2.5 Liquid crystal phases ............................................................... 11.4.3 Chiral phases ......................................................................................... 11.4.3.1 Ligand exchange chromatography........................................... 11.4.3.2 Direct enantiomeric separation................................................ 11.4.3.3 Pirkle phases ........................................................................... 11.4.3.4 Protein phases ......................................................................... 11.4.3.5 Organometallic phases ............................................................ 11.5 Non-silica-based stationary phases .................................................................... 11.5.1 Alumina-based stationary phases........................................................... 11.5.1.1 Monomeric alumina phases ..................................................... 1 1.5.1.2 Polymeric alumina phases ....................................................... 11.5.2 Carbon-based zirconia ........................................................................... 11.5.3 Polymer-based stationary phases ........................................................... 11.6 Conclusion ......................................................................................................... 11.7 References .........................................................................................................
.
Multivariate characterization of RP-HPLC stationary phases ..... A . Bolck and A.K. Smilde Introduction ....................................................................................................... Multivariate characterization ............................................................................. 12.2.1 Some basic multivariate statistical concepts .......................................... 12.2.1.1 Data matrices; centering and scaling ....................................... 12.2.1.2 Visualization of data................................................................ 12.2.2 Principal component analysis ................................................................ 12.2.2.1 The concept of principal component analysis ......................... 12.2.2.2 The calculation of the principal components........................... 12.2.2.3 A least squares interpretation of principal component analysis .................................................................................... 12.2.3 A principal component example ............................................................ 12.2.4 Three-way analysis ................................................................................ 12.2.5 A three-way analysis example ............................................................... Marker selection ................................................................................................ 12.3.1 The choice of markers ........................................................................... 12.3.1.1 The determinant criterion ........................................................ 12.3.1.2 The induced variance criterion ................................................ 12.3.1.3 Principal component analysis .................................................. 12.3.1.4 Marker selection and three-way analysis................................. 12.3.2 A marker selection example .................................................................. Predictions......................................................................................................... 12.4.1 Ordinary least squares (OLS) ................................................................ 12.4.2 Partial least squares (PLS) ..................................................................... 12.4.2.1 PLS1 ........................................................................................ 12.4.2.2 PLS2 ........................................................................................ 12.4.2.3 PLS predictions ....................................................................... 12.4.3 PLS predictions and marker selection ...................................................
388 390 390 391 391 392 393 393 393 393 395 395 397 399 399
Chapter 12
403
12.1 12.2
403 404 405 406 407 410 410 412
12.3
12.4
414 415 417 420 422 423 423 423 424 425 425 428 429 432 432 433 434 434
XI1
Contents
12.4.4 A PLS example...................................................................................... 12.5 Practical examples ............................................................................................. 12.5.1 Calibration of octadecyl modified stationary phases of different batches ................................................................................................... 12.5.2 Calibration of stationary phases of different types ................................ 12.6 Appendix A ....................................................................................................... A.l Rank ...................................................................................................... A.2 Spectral decomposition ......................................................................... A.3 The singular value decomposition (SVD).............................................. 12.7 References ......................................................................................................... Subject Index
............................................................................................................
435 438 439 443 445 445 446 446 447 451
XI11
Preface This book grew out of a long standing interest in the ways in which retention and the selectivity of separation in liquid chromatography are dependent on the structure of the analyte and on changes in the mobile and stationary phases. These relationships are at the heart of an understanding of the operation of liquid chromatography and of the ways in which the chromatographer can manipulate the conditions of a separation to achieve the analysis of a complex sample. The factors involved in these processes are complex and even 90 years after the pioneering work of Tswett are still not fully understood. Any progress is linked to the development of an understanding of the physical chemical process of solvation and the physicochemical nature of the stationary and mobile phases. Chromatography is also a valuable practical analytical method and much can be learnt by studying relative interactions and by comparing the behaviour of analytes with different chemical structures under different separation conditions. To achieve this objective, techniques for recording relative retentions are needed so that results can be reproduced in different laboratories or by different operators. However, liquid chromatography has a notoriously poor transferability, the same high versatility which enables separations to be precisely optimized also means that small changes between systems can alter the separations. This book addresses some of the ways in which these problems have been overcome to enable retention predictions, identifications and the characterization of the properties of mobile and stationary phases, to be carried out. The work owes much to studies in gas chromatography, in particular the work of Kovhts in providing a retention index scale and of Rohrschneider and McReynolds on the comparison of stationary phases. A theme which leads through the different chapters is the value of relative measurements. Most obviously in the descriptions of the different retention index scales in liquid chromatography and their application to the identification of a wide range of analytes. The indices also form the basis of one of the studies on retention prediction, the other relating retention to the contribution of analytes to partition coefficients. Related methods have been used to compare analytes and their interaction properties. The final group of chapters investigates methods for the comparison of mobile and stationary phases not just by using a simple solvent strength parameter but by examining the comparative interaction of the phases to different types of analytes either in terms of their shapes or physical properties. Bringing these chapters together enables the different approaches to be compared and illustrates the values of each. Hopefully, this will stimulate further research or different approaches for this is by no means the full description of the mechanism of retention. Much more still needs to be done, in particular to understand how complex molecules behave. In this case, the chromatographic behaviour of the analyte under different conditions may itself provide valuable information about the physical properties of the analyte.
XIV
Preface
I would like to thank many of the contributors for useful and interesting discussion of their work and the stimulation it has provided for our own studies. I would also thank my research and project students at Loughborough University of Technology, who have contributed to our own studies in this field. In the same way that their individual contributions have together built our overall study, so I hope that the chapters of this book will contribute to an overall greater understanding of the retention process in liquid chromatography. Roger M Smith May 1994
xv
List of Contributors M. BOGUSZ
Institut fur Rechtsmedizin, Medizinische Fakultat, RheinischWestflilische Technische Hochschule Aachen, Pauwelsstrasse 30, 0-5100, Germany
A. BOLCK
Faculty of Mathematics and Natural Sciences, University Centre for Pharmacy, University of Groningen, Antonius Deusinglaan 2, 9713 A W Groningen, The Netherlands
P. JANDERA
Department of Analytical Chemistry, University of Pardubice, Faculty of Chemical Technology, Ndm. Legii 565, 532 10 Pardubice, Czech Republic
P. KURONEN
Department of Chemistry, P.O. Box 6 (Yuorikatu 20), University of Helsinki, FIN-00014 Helsinki, Finland
J.J. PESEK
Department of Chemistry, Sun Jose State University, Sun Josk, CA 95192-0101, USA
L.C. SANDER
Chemical Science and Technology Laboratory, Organic Analytical Research Division, National Institute of Standards and Technology, Gaithersburg, MD 20899-0001, USA
A.K. SMILDE
Laboratory for Analytical Chemistry, Nieuwe Achtergracht 166, I018 WY Amsterdam, The Netherlands
R.M. SMITH
Department of Chemistry, Loughborough University of Technology, Loughborough, Leicestershire, LEI I 3TU, UK
K. VALKO
Department of Physical Sciences, Wellcome Research Laboratories, Langley Court, Beckenham, Kent, BR3 3BS, UK
S.D. WEST
North American Environmental Chemistry Laboratory, DowElanco, P.O. Box 68955, 9410 Zionsville Road, Indianapolis, IN 46268-1053, USA
E.J. WILLIAMSEN
Department of Chemistry, Sun Jose State University, Sun Jost!, CA 95192-0101, USA
S.A. WISE
Chemical Science and Technology Laboratory, Organic Analytical Research Division, National Institute of Standards and Technology, Gaithersburg, MD 20899-0001, USA
This Page Intentionally Left Blank
Journal of Chromatography Library, Vol. 57: Retention and Selectivity in Liquid Chromatography R.M. Smith, editor 0 1995 Elsevier Science B. V. All rights reserved
1
CHAPTER 1
Retention prediction based on molecular structure Roger M. Smith Department of Chemistry, Loughborough University of Technology, Loughborough, Leicesfershire,LEI I 3TU UK
1.1 INTRODUCTION The retention of a particular analyte in a reversed-phase liquid chromatographic system is dependent on many factors, the structure of the analyte, the nature and chemistry of the stationary phase, the composition of the mobile phase and the temperature. Some of these factors are reasonably well understood, at least on an empirical level, and chromatographers can manipulate eluent composition and even temperature to alter retentions in a predictable manner. However, the effect of the chemical structure of the analyte on retention is probably the least well described parameter. Most chromatographers recognise the broad influence of polarity and size and their effect on hydrophobicity but not the detailed impact of the addition of a methoxyl, carboxamide or other functional group. Nor in most cases is it possible to predict the composition of the eluent required to result in a predetermined retention (capacity) factor (k). Instead the experimental conditions to achieve a particular retention are usually selected by analogy with related compounds or fiom experience of analysing a wide range of samples. Most analytical methods in liquid chromatography are then refined on a trial and error basis. However, in recent years two methods to aid the chromatographer in refining a separation have become available. The first requires no knowledge of the structure of the analyte. A computer programme, often an expert system, uses the retention factors of the components of a mixture fiom a gradient or isoeluotropic set of separations to propose an eluent mixture, which is predicted to provide optimal resolution or overall run times. These techniques include systems such as Drylab, PESOS, ICOS, and DIAMOND, which are based on prediction and mapping methods, and chemometric techniques, including iteration and Simplex optimization methods. These methods have been well reviewed in recent years [l-31. Future developments are likely to see the expert systems being supReferences pp. 45-46
2
Chapter I
plemented by neural networks, which should enable them to “learn” about the properties of a particular column and instrument, before making their predictions [75]. In most assays the structure of the analyte is known and the second approach has been to predict the retention from the molecular structure. This can be carried out directly by the summation of the retention properties of the structural components or by deriving a physical property, such as the octanol-water distribution coefficient (log P), which can then be related to retention by comparison with analytes of known value [3]. As the structures of any impurities or metabolites in a sample are often known, it should also be possible to predict the optimum conditions for their resolution from the main components. This approach has the potential for true prediction as it can propose initial chromatographic conditions, designed from the start to achieve a particular separation. Two different but closely related aspects of this approach form the subject of this and the following chapters. The recent literature also includes numerous papers on retention prediction which related retentions under one set of conditions with those using a different proportion of modifier or temperature. For example, changes in retention with mobile phase composition have been recently discussed by Valko ef al. [4]. A second closely related area has been the selection of robust methods that although they may not be optimized to give the maximum resolution, nevertheless provide methods which are less susceptible to small changes in eluent composition, temperature and or different columns [5,6]. In real life situations this may be an important consideration if the method is to form part of an official method or is required for long-term studies of the stability or quality of a product. Again computer assistance has been provided for the selection of testing conditions and the evaluation of the results.
1.2 STRUCTURE AND RETENTION
The concept that the retention of an analyte in gas or liquid partition chromatography can be expressed as the summation of factors related to its skeleton and individual functional groups was originally proposed by Martin [7].He suggested that the retention of a analyte can be expressed by the summation of contributions from each of the structural components, alkyl-chains, aromatic rings and fictional groups. These substituent values are related to their effects on other equilibria and are recognised as examples of a linear freeenergy relationship. The early work in this area on gas-liquid chromatography and thinlayer chromatography have been reviewed by Kaliszan [S,9]. These concepts have led to a wide range of studies, which have examined the effect of the different substituents on the retention of an analyte in liquid chromatography. These quantitative structure-retention relationships (QSRR) studies have encompassed physical properties, topological indices, and additive functions and have been reviewed in detail [8-121. Similar concepts have long been used for the prediction and calculation of octanol-water partition coefficients (log P) in quantitative structure-activity relationship (QSAR) studies which are important in relating biological activity to structural features. Hansch and Leo [13] have shown that the log P can be calculated by the summation of a value for a parent compound with contributions for each substituent (rc constants) and a
Retentionprediction based on molecular structure
3
similar approach based on fragmentalv) constants has been used by Rekker [14]. There is often a good correlation between the octanol-water partition coefficients and chromatographic retention and numerous studies have used HPLC techniques to measure effective log P values [10,12,15]. The technique works well if a group of analytes are structurally related but compounds of different structural types may show a poorer correlation. However, comparatively relatively little use has been made of the n or f constants to calculate log P values for retention prediction. In a series of studies, Jim0 and Kawasaki [ 16-1 81 predicted the retention factors of alkylbenzenes and substituted aromatic compounds. More recently, the relationship has been used by Valko and co-workers [19] as the basis of a retention prediction system (see Chapter 2). This work has formed the basis of a computer program, which also incorporates the ability to handle partially ionized analytes. Some of the advantages and limitations of this method have recently been evaluated by Fekete et al. [20]. An alternative approach for the prediction of retentions in liquid chromatography is to use the summation of retention increments, which have been determined by comparison of substituted and unsubstituted analytes. These can be expressed either as functional group contributions (Section 1.2.1) or retention index increments (Sections 1.3-1 S). This approach has also been examined in other branches of chromatography. Peng et al. [21] have examined the prediction of the retention indices of analytes, based on their molecu1: structure on an apolar column in gas chromatography. The number of atoms, the aromatic increment and the group retention functions (GFW)were all important. They used a combination of the number of carbon atoms, carbon atom equivalents, and group retention factors for substituents and functional groups. They took into account the effects of rings, is0 and neo-carbons and found that predicted and experimental values were within *3%. In a second paper, they examined these effects for separations on polar columns [22]. A similar approach has also been reported by Evans and co-workers [23] based on the molecular weights and selectivity indices of the analytes.
1.2.1 Chromatographic functional group contributions A frequently applied approach, to relate retention to changes in structure, has been the functional group contribution (z) to the logarithm of the retention factor. The values o f t are determined by comparison of the retention of substituted analytes with the corresponding unsubstituted analyte (Eq. 1.1).
The measurement and application of group contributions have been comprehensively reviewed by Smith [24]. The magnitude of the contributions for individual functional groups differ with the eluent composition and their magnitude usually decreases significantly with increased organic modifier. The contributions also differ with different organic modifiers in the mobile phase. However, these parameters have not been widely used for retention prediction because few studies have examined the relationship between mobile phase composition and the magnitude of the contribution. The contributions were References pp. 45-46
4
Chapter 1
frequently deliberately extrapolated to 100% water as the eluent to give composition independent values (z,), which were then compared with other physical parameters. Probably the most widely investigated functional group change in retention with structure is the methylene group contribution. Numerous studies of homologous series [24] have shown that there is a systematic change in the logarithm of the retention factor with the carbon number. This change is usually similar for all homologous series, irrespective of the other groups present. For example, Figge and co-workers [25] reported a constant change for a series of homologous analytes, n-alkanes, n-alkenes, n-alkylbenzenes, fatty acid methyl esters, alkan-3-ones, 2-n-alkyl-pyridines, 1 -n-alkanols. This relationship also forms the basis of most retention index series and is discussed further in Chapter 3. A difficulty with many of the retention studies, such as the hnctional group contributions, which are based on retention factors (k), is that the increments are very dependent of the experimental conditions, such as temperature and the eluent composition. Frequently these have not been closely controlled and the resulting retention values are often unique to that individual system of mobile phase and column. Many of these problems of reproducibility and transferability between systems can be overcome by using relative retention measurements, such as retention indices (see Chapter 3). A retention index scale effectively compares the increment for a functional group with the corresponding methylene increment in the same system. Both should be similarly affected by the small changes in the strength of the eluent and by temperature, so that retention index based group contributions should be almost independent of the eluent composition and of the make of stationary phase. Unless there are changes in the relative interactions between the methylene or other functional groups and the stationary phase, the retention increments should be largely independent of the brand of stationary phase and carbon loading of the columns, even though these differences can significantly effect retention factors (k).
1.2.2 Related prediction studies
In an extensive series of studies, Jandera has examined the description of the interaction of an analyte in terms of lipophilicity (rice) and polarity interaction indices (4).The values of these terms for a number of functional groups have been determined by comparison with the retentions of the homologous n-alkylbenzenes. This work has been recently reviewed [26] and is described in more detail in Chapters 7 and 8. Galusko proposed that it should be possible to predict the retention of a compound based on the summation of the effects of the bond dipoles and partial molar volumes of the substituents [27]. This system has now been developed into ChromDream, a computer-based prediction system [28]. Their model is based on a two-layer continuum model of reversed-phase chromatography and the differences in molecular solvation energies in the two phases. The retention of an analyte is described by Eq. (1.2), in which Vi are the increments of the partial molar volume fragments in water and Ge,si,H20are the increments of energy of interaction of bond dipoles with water. a, b and c are the parameters of the reversed-phase system and depend on the column and phase ratio and are characteristic of the stationary layer and mobile phase.
Retention prediction based on molecular structure
hk,
=~(cc)~’~
+~(C,AG~.~.J,H,O)+C
i
5
(1.2)
J
The values of a, b and c for a particular separation system have to be determined by using reference compounds. The parameters a and b are related to differences in the SUTface tension and dielectric permeability of the sorbent surface layer and mobile phase, respectively, and can be related to differences between stationary phases. The programme gave accurate predictions over a range of eluent composition for a wide range of aromatic analytes. However, the model does not take stereochemical and intramolecular interactions into account and discrepancieswere found for ortho-disubstituted analytes. A common approach for retention correlation has been to relate a range of physical parameters related to structure, such as shape and connectivity parameters, to retention factors using multivariant analysis [9,10]. The resulting regression equations can then be used for retention prediction. However, although the correlations are frequently excellent, the addition of new model compounds to the data set will often markedly change the coefficients and even cause the significant terms to change. Thus although the regressions can “predict” the retention of analytes that are included the original model data set, they frequently fail to predict accurate retentions for new compounds. A conceptual problem is that there is often no clear connection between the properties that are used as terms in the regression and structural or physical properties, which are generally accepted as being significant in liquid chromatography. The correlation may be valid but only because the parameters are also indirectly related to a parameter of relevance. This purely chemometric approach has been reviewed [9,11] but the real prediction power appears to be limited. 1.3 FUNCTIONAL GROUP EFFECTS ON RETENTION INDICES
Retention indices, which are determined by logarithmic interpolation between the retention factors of a series of homologous standards, provide a reproducible mode of retention measurement (Chapter 3). They can form the basis of reliable and transferable retention comparisons. In comparison to functional group contributions, they are more independent of the eluent composition and stationary phase. Functional group or substituent increments (Is, x) can be determined as the differences between the retention index of a parent compound (ZR-H)and those of substituted derivatives (&) (Eq. 1.3).
Is, x = IR-x- ZR-H As the functional group index increments are not related to the absolute retention times but to the methylene group contribution = 100) by definition) they should be largely independent of eluent composition. If interactions occur between multiple substituents, the differences between the simple summation of the contributions from the individual groups and the experimental retention index value, are defined as the interaction index ZI. Because of the range of polarities of analytes in HPLC, a number of different retention index scales have become established ([29], Chapter 3). The most frequently employed scales have often also been used for retention prediction. As each scale is based on the methylene increment, the results should be equivalent but small differences might be exReferences pp. 45-46
Chapter 1
6
pected from changes in separation conditions. The n-alkanes and n-alkylbenzenes are both highly retained and have been used for non-polar analytes (Section 1.3.1). The alkyl aryl ketones (Section 1.4) cover a wide range of polar and non-polar analytes and have formed the basis of a major prediction study (Section 1.4). The more rapidly eluted alkan2-ones have been used primarily for the prediction of drugs and their metabolites (Section 1.3.2). Although the 1-nitroalkanes have a similar polarity, no prediction studies have been carried out with these standards. Some prediction work has used a retention index scale based on standards with increasing numbers of aromatic rings (Section 1.5).
1.3.1 Retention prediction based on n-alkanes and n-alkylbenzenes
Morishita et al. [30] determined the retention index increments for four aryl substituents using the n-alkane scale (Table 1.1a). The values of the increments reflected the expected hydrophobicities, although the methyl group increment (dZ= 89.3) was lower then the nominal value for a methylene group (dZ= 100) and may have reflected hyperconjugation with the ring. The relative magnitudes of the increments corresponds to those obtained in later studies on the alkyl aryl ketone scale [3 11. They also studied the interactions between two substituents on the aromatic ring, by measuring the difference between the experimental indices and the calculated values obtained by the summation of substituent increments (Table I. Ib). The increments for interTABLE 1 . 1 SUBSTITUENT AND INTERACTION INCREMENTS OF RETENTION INDICES BASED ON THE nALKANE SCALE (a) Substituent increment SIX = I,+x - IA-H. For comparison, substituent index I~,A-x values are given from the regression equations based on alkyl aryl ketone retention index scale [31] Functional group
Substituent increment (SIX)
I ~ ,[311 ~ - ~
89.3 41.9 -195.7 -213.6
104 -87 -290 -302
(b) Interaction increments between substituents on an aromatic ring. Calculated as the differences between experimental retention index and summed index values for substituents and parent Interaction increment (611,~-y)
Groups X
Y
2-Y
3-Y
4-Y
CH3 CH3 CH3 CH3 NH2
CH3 OH NH2 NO2 NO2
-12.7 -11.1 -24.7 -27.4 140.8
3.1 -24.5 -24.3 -2.5 84.5
5.0 -24.1 -23.5 -19.2 48.8
Based on Morishita and co-workers [30]. Conditions: column, Partisil ODS-3; eluent, methanol-water (70:30).
Retention prediction based on molecular structure
7
actions between methyl groups were small (61 = -13 to 5 ) but between amino and nitro groups much larger changes were observed. The ortho-nitro substituent (dZ= 141) on aniline had a major effect and increased the retention by considerably more than the metuor puru-substituents (61 = 85 and 49, respectively) suggesting that hydrogen-bonding was occurring between the ortho-groups. Using the index increments for the substituents and interactions, they calculated the predicted retention indices for a number of trisubstituted aromatic compounds and found a good correspondence with experimental values. The deviations were between -14 and +10 units; for example, 3,5-dimethylaniline ZCdc= 163 and robs= 166; 2-methyl-4-nitroaniline, Zcdc = 79 and Zobs = 80. A similar approach was subsequently used in a study of sulphur compounds by MSckel [32]. He compared the retention of a series of homologous thiols and alcohols with the corresponding alkanes and determined the retention index increments for the replacement of a methylene group by a hydroxyl (OH) group (-510 to -519) or thiol (SH) groups (-1 80 to -206) (the values increased slightly with chain length) using methanol-water (70:30). Slightly confusingly, the retention indices of the parent compounds in this paper were based on the number of non-hydrogen skeletal atoms (C plus S) so that the replacement increment for the thiol group in heptylthiol (]= 599) was calculated as 599 - 800 [(C, + S) x 100 = -201)l. These replacement values correspond to substituent increments for hydroxyl of ZoH = -410 to -419 units and for thiol of ZsH = -80 to -106 units. Thioethers (R-S-R) had retentions similar to the monothiols. The increments for the hydroxyl group were similar to those found later for the aliphatic hydroxyl (Is,R-oH=-362 in methanol-buffer (50:50), see Table 1.7) using the alkyl aryl ketone scale. The retentions of the alkylpolysulphides (R-S,-R) were also examined [32]. When a rnethylene group in tetradecane (Z= 1400) was replaced by a sulphur atom to give hexylheptyl sulphide (I = 1060) the retention decreased markedly. A second replacement by a sulphur atom gave dihexyl disulphide with a similar retention but as the proportion of sulphur atoms in the chain was increased further, the polarity decreased to eventually give dipropyl octasulphide (1= 1259) and dimethyl dodecasulphide ( I = >1400). These changes were explained as the initial formation of a local polar centre and then an increase in retention as the non-polar polysulphide replaced the methylene groups. In subsequent studies, MSckel and co-workers [33] determined the coefficients, which related the retention indices of homologous analytes to the number of carbon atoms (nc) (Eq. 1.4).
They used these values to demonstrate that for most homologues the change ( B ) in the retention index on the n-alkane scale on the addition of a methylene group was close to 100 units. However, smaller values were found for Ph-(CH2),-Ph (88.99 units) and RS9-R (78.84 units). The A term indicated the effect of the functional group. A comparison of the saturated and unsaturated hydrocarbons suggested that the addition of an olefinic group increased retention by 61.85 units but an acetylene group decreased retention markedly (A = -265.96). These increments were translated into “chromatographic free energy” changes. They also found a linear relationship for each series between retention index and calculated total surface area. In a subsequent study [34] they reported the corReferences pp. 45-46
Chapter I
8 TABLE 1.2 INDEX INCREMENTS FOR SUBSTITUENTS ON HOMOLOGOUS SERIES OF ANALYTES Substituent
Correlation coefficients
B
A ~
H Br SH OH CN -04-
0 -1 12.4 -145.9 -653.1 -625.4 -211.6 -108.2
~~
100 97.5 98.4 101.3 101.3 86.1 90.9
Based on data from Mockel and co-workers [34]. Determined from the intercept of plot between 1 , and carbon number. Eluent methanol. IK = A + BNc. The intercept A is equivalent to the substituent index value.
responding values for a wider range of substituted homologues (Table 1.2). The A values reflected the effects due to the different groups. However, these values were not used for retention prediction. Subsequent studies by Aced and co-workers have examined the 1,nbi(alkylthioa1kanes) R-S-(CH2)n-S-R and reported a constant methylene increment of 88 units as n was increased [35]. An example of the application of use of multivariant analysis to correlate physical properties and retention indices has been reported by Dimov [36]. He identified terms which could be used to precalculate the retention indices of isoalkanes in HPLC. His initial equation Icalc = 41.09 + 0.92756PCI (PCI = physicochemical index) gave only a correlation of 0.9861 and was expanded to give a model equation by the addition of terms for different type of atoms (for example nq = quaternary carbons) (Eq. 1.5). Icalc= 42.17 + 0.33448PCI + 58.336n0- 1 1.629ncH3+ 5.056nL
- 3.729nd - 3.44nq- 7.55ni This equation gave a correlation of 0.99959 between experimental and calculated values. The model provided good predictions for m b e r isoalkanes but was not tested for other groups. 1.3.2 Retention prediction based on alkan-2-ones In studies of the application of the alkan-2-one retention index scale, Baker and coworkers explored two main aspects. Firstly, they developed a relationship between octanol-water partition constants and retention indices, which could be used as a predictor of retention and, secondly, they examined the systematic changes in retention indices with the addition of a functional group to a parent compound. In other studies, retention indices based on the alkan-Zones were used as a measure of lipophilicity in QSAR comparisons (Chapter 3).
Retention prediction based on molecular structure
9
The prediction studies examined groups of related pharmaceuticals either as part of QSAR studies or in the examination of drug metabolism. Baker proposed [37] that as the addition of a methylene group (whose x value is 0.50) changes the retention index value by 100 units, other functional groups should show a corresponding effect. The increment for the addition of a function group should therefore be 200 times its x constant (Eq. 1.6)
AI = 200n
(1.6)
Thus, if the retention index of a parent compound (ZR-H) was known, the predicted retention indices for substituted derivatives (ZR-x) could be calculated from the nx constants by using Eq. (1.7).
This assumption was tested using groups of propanolols, barbiturates, anthranilic acids [37], narcotic analgesics and nortropanes [38]. For the barbiturates, barbital (diethylbarbitone) was chosen as the parent compound and the calculated and measured values of the retention indices were determined. The correlation was good for the alkylated and aryl substituted barbiturates but poorer for thiopental and thioamylal. Overall the highest deviation was 43 units with a mean error of 29 units. For the anthranilates [37], which contained a wider range of bctional groups, the mean error between the measured and predicted indices was only 22 units, even though the carboxylate group in each compound would have been nearly completely ionized (Table 1.3). For the propanolol derivatives [37] the calculation was more complicated, as nxvalues were not available for all the functional groups and some had to be determined from model compounds. There was a reasonable correlation between predicted and experimental retention indices, which was attributed to uncertainties in the partition coefficients. When the same approach was applied to the narcotic analgesics and related compounds [38] using morphine as the parent compound, the comparisons showed a greater variation (Table 1.4), which was attributed to stereochemical factors such as the shielding TABLE 1.3 MEASURED AND CALCULATED RETENTION INDICES OF ANTHRANILATES DERIVED FROM CONTRIBUTIONS BASED ON HANSCH CONSTANTS Functional group (X)"
Measured
Calculated
530 565 586 606 656 678
530 (Ired 560 578 634 670 734
Based on values from Baker [37]. aCompounds, C6H4(C02H)-NH-S02-C6H~-x. bIcajc=Ir,+ 20Ozx based on the alkan-2-one scale.
References pp. 45-46
zxconstant
Retention indexb
0.15 0.24 0.52 0.70 1.02
Chapter 1
10 TABLE 1.4 COMPARISON OF MEASURED AND CALCULATED RETENTION INDICES OF NARCOTIC ANALGESICS AND RELATED COMPOUNDS Compound
Morphine Oxymorphine Oxocodeine N aloxone Codeine Dihydrocodeine Hydromorphone Ethylmorphine Hydrocodone Heroin Pentazocine Levallorphan Phenazocine Levorphanol Dextromethorphan
Retention index (0 Observed
Calculated
621 529 615 668 705 71 1 712 783 79s 805 965 1002 1005 1126 1284
621 (Ired 329 459 469 75 1 811 615 85 1 74s 925 1275 1279 1427 1145 1269
Based on Baker and co-workers [38]. Index values calculated as Table 1.3 using the alkan-2-one scale. Reference compound, morphine.
of polar substituents by the 14-hydroxy group in the oxymorphone type of drugs. These effects were used to characterise the stereochemistry of isomeric N-substituted 3-propananilidonortropane analogues. The a-isomer consistently had a higher retention index than the B-isomer, in which the polar 3-propanilido group is less shielded. During a study of metabolites, Baker [39] found that the retention index values of glucuronides were much lower (dI = -244 f 3 1 units) than the parent compounds, such as morphine. In this case the corresponding n value for the glucuronide group was not available. However, by using the shift in the retention index values, Eq. (1.6) could be used to calculate the effective constant (ng~dglucwoni~e = -1.22 f 0.16). In all these examples, a hydroxy group on the parent structure had been converted into the glucuronide but if hydroxylation also occurred as part of the metabolism then the expected change in the retention indices would be -378 units. These changes were subsequently used to study the retentions of metabolites from primaquine [40]. Most of the hydroxylated metabolites were identified by comparison of their retention indices with standards. Two polar metabolites were identified as possible glucuronides because their retention indices (A, I = 538 and B, I = 681) were smaller, by about 244 units, than primaquine (I= 733). However, they were not hydrolysed to the starting material by B-glucuronidase, even through the first metabolite was cleaved by the treatment. By using Eq. (1.7), Hufford et al. [41] were able to predict the retention indices of metabolites formed by microbial action on imipramine. The values for the hydroxylated and desmethyl derivatives closely matched the empirical values (2-hydroxy, Icalc = 789, Iexpt = 769; 10-hydroxy, Zcalc = 725, Iexpt = 720 and N-desmethyl, Icdc = 846, Iexpt = 807) and helped to confirm their identification. They were also able to obtain a tentative as-
Retention prediction based on molecular structure
11
signment of 10-hydroxydesmethylimipramine as a previously unidentified metabolite (Icalc = 529, Zexpt = 484) but no authentic sample was available for confirmation. The influence of stereochemical effects on the ability to use retention index increments to predict retention has been examined using a number of azabicycloakanes and bicycloalkanes [42]. From a regression analysis, Baker et al. were able to determine the empirical effect of selected functional groups, however, these differed from the expected values based on x constants. For example, replacing OH by OCH3should increase retention indices by 234 units but the observed increase was only 160 units and for the introduction of phenolic OH, the expected change was -134 units but the observed value was -84 units. It appeared that the increment for a substituent was dependent on the total number of polar substituents in the molecule. When this was taken into account the empirical Values for these groups became 205 and -144, respectively, much closer to the predicted values. The regression analysis also demonstrated a difference between stereochemical isomers of 57 units. A similar difference of 67 units was observed for a number of related azabicyclooctanes. These results suggest that in complex systems, the ability to obtain good predictions may be affected by interactions between groups, which depend on their relative positions in space. The prediction of retention indices has also been used to assist the identification of natural products from plant and microbial sources. During a study to identify the urushiol congeners from poison ivy and poison oak, Ma and co-workers [43] calculated the retention indices for unsaturated and acetylated derivatives of 3-pentadecylcatechol (PDC), and 3-heptadecylcatechol(HDC) by using Eq. (1.7) (PDS-triene Zexpt = 1416; ZCdc = 1393 and HDC-diacetate Zexpt = 2026 and Zcalc = 2023) (see also Table 4.9). However, they found that the value of x = -0.63 (corresponding to Zx = -125) for the double bond gave a better fit with the experimental value than the standard value (nx = -0.30). This prediction calculation provided a simple method for the characterization of the congeners, which avoided the need for derivatizationor synthesis. Magg and Ballschmiter found systematic changes in the retention indices (IRCOMe) in ergopeptines with changes in the structure [44]. The magnitudes of the changes were almost identical on three different column systems, even though the absolute values of the indices differed. However, the values were not correlated with Hansch x values, although the trends agreed.
1.4 PREDICTION OF RETENTION INDICES BASED ON ALKYL ARYL KETONES
The alkyl aryl ketones [45] have been identified as suitable retention index standards for a wide range of analytes (see Chapter 3), because of their ready detectability, easy availability and similar polarities, and hence retentions, to many aromatic analytes. It was also shown that retention indices based on this scale are highly reproducible and can be transferred between columns more readily than retention factors [46,47]. The retention indices were relatively insensitive to the exact eluent composition and would not be affected by small variations such as might occur in the preparation of an eluent mixture by different operators. References pp. 45-46
Chapter I
12
Because of this independence from the chromatographic conditions, Smith proposed [31] that the retention indices could form the basis of a retention prediction system, whose conclusions would be more generally applicable than values calculated for one specific retention system. The predicted retention index value for an analyte could then be converted into the corresponding retention factor for a column-eluent system by using Eq. (1A), whose constants A and B can be determined fiom the retention factors (k)of a series of alkyl aryl ketones, which have defined retention indices (I = carbon number nc x 100). log k = A + BI
(1.8)
The intention was to develop a system which could predict the retention of a compound fiom its structure and the mobile phase modifier. It should then be possible to use this approach to predict the conditions for a separation or the optimum conditions to achieve a particular resolution. Even if the retention index of a complex analyte could not be predicted, it should be possible to calculate relative retentions, compared to a parent compound in the same way that Baker (in the previous section) was able to estimate the retentions of metabolites and congeners. From the start, it was recognised that because the interactions between substituents are not hlly understood, particularly on heterocyclic and aromatic rings systems, it would probably be not be possible to make completely accurate predictions for complex molecules. However, any deviations between experimental and predicted index values could be used to examine the interactions between the functional groups. The basis of the prediction system was that the retention index of an analyte, in a selected eluent, could be calculated by the summation of the retention index of a parent compound, substituent index values for each substituent plus interaction index terms required to describe interactions between substituents, such as H-bonding, steric and electronic interactions) (Eq. 1.9).
where Ipis the retention index value of a parent compound, IS,R is the substituent index contribution from saturated aliphatic carbons, are the substituent index contributions for substituents on an aromatic ring, Is,R-x are the substituent index contributions for substituents on aliphatic carbons (these substituents will include unsaturated and carbonyl groups), and I1,y-Z are the interaction index contributions between substituents or groups (Y-Z) to account for H-bonding, steric and electronic effects. Although some reports had suggested that there is a nearly linear relationship between percentage composition of the eluent and retention parameters, such as log k, Schoenmakers el al. [48] found a closer correlation can usually be obtained with a quadratic relationship, particularly if a wide range of eluent compositions were being compared. Consequently, each of the terms in the prediction system (Eq. 1.9) was defined for each organic modifier as an experimentally determined quadratic equation (Eq. 1.10): I =ax2 + bx + c (x
= % of
organic modifier in the eluent).
(1.10)
Retention prediction based on molecular structure
13
The a, b, and c coefficients for each modifier derived from the different components of the prediction equation (Eq. 1.9) could then be summed to give an overall quadratic equation (Eq. 1.1 1) for the retention index of the analyte. (1.11) Benzene was selected as the parent compound, because all its substituted derivatives could be detected spectroscopically. A wide range of derivatives was also readily available, substituted both directly on the aromatic ring and on aliphatic side chains, which means that both types of substituent indices could be determined. Suitable polysubstituted standards were also readily available for the determination of the interaction terms. In order to ensure a consistent data set [49], a single batch of stationary phase was used throughout the study. The separations conditions were also standardised. The eluent was buffered to pH 7.0 using a constant ionic strength buffer prepared by weight and the column temperature was controlled at 30°C. Regular test samples were examined and the reproducibility of different columns was checked. Although there were some changes in the retention factors (k),the retention indices were consistent and a variance of less than 10 units was considered to have been reached across the study, which lasted over 2 years (see Chapter 3). All the terms generated in the project were brought together into a database, which could be interrogated using a expert system programme CFUPES (Chromatographic Retention Index Prediction Expert System) to generate predicted retention indices (see later). 1.4.1 Monofunctional compounds Although the retention indices of analytes are largely independent of the proportion of modifier in the mobile phase, there are small changes and the first stage of the project was to determine the baseline values for benzene with different modifiers [31,50]. For methanol, acetonitrile and THF the relationships were curved (Fig. 1.1) and could be matched closely in each case by a quadratic equation (Appendix 1.1). These correlations provided smoothed values, which were used as the defined retention indices of the unsubstituted parent compound (Ip). 1.4.1.I Aromaticfunctional groups
Initially, the retention factors of 16 monosubstituted model compounds and the alkyl aryl ketones (from acetophenone to heptanophenone) were measured in a range of different eluents from methanol-pH 7 buffer (40:60) to (80:20) and acetonitrile-pH 7 buffer (30:70) to (80:20) [31,51]. Although the retention indices were also determined in each case for 90% modifiers, the corresponding retention factors were so small that these values were considered unreliable and were not used in the correlations. Subsequently the retentions of these compounds were also examined using THF-pH 7 buffer (20230) to (60:40) eluents [50]. The monosubstituted model compounds covered a wide range of functional groups, however, it was not possible to examine the aryl carboxylic and sulphonic acid groups, as they were ionized at the pH of the buffer.
References pp. 45-46
14
Chapter 1
'loo
-
n.
=
1
1000
Y
c
.-0 Y
c
2! 900 Q)
a
MeOH
800
1
I
I
I
20
40
60
80
Organic modifier
(%I
Fig. 1.1, Comparison of experimental retention indices (symbols) of benzene with calculated values of parent index values (Ip, lines) derived from quadratic relationships (Table 1.7). Eluents: 0 , methanol-buffer; W, acetonitrile-buffer, THF-buffer. Values from [3 I] and [50].
+,
In each case, the retention indices were calculated by comparison with the retention factors of the alkyl aryl ketones injected in the same series of separations (e.g. for methanolic eluents see Table 1.5). The substituent increments for each functional group were then determined using Eq. (1.3) (Table 1.6). The increments for the relatively non-polar groups were almost constant with proportion of modifier but the polar groups showed greater changes. Particularly for the acetonitrile eluents the relationships were often nonlinear. However, the changes were small compared to the changes in retention factors for these compounds. These empirical substituent increments were fitted to quadratic equations to give the coefficients a, b and c, which could be used to predict the substituent index at any eluent composition. Throughout the study, if the change with mobile phase composition was small ( 4 0 units), a single c term was determined. The calculated substituent index values from the quadratic expressions gave a close fit with the experimental increment values (for example, Fig. 1.2). More recently, naphthalene as an alternative parent compound and some additional monosubstituted compounds have been examined [52] and the fill set of coefficients is listed in Appendix 1.1. The advantage of this approach is that the predicted substituents indices reflect the differences in the interaction selectivity of the modifier. For example, the calculated substituent indices, in three isoeluotropic eluents, methanol-buffer (70:30), acetonitrilebuffer (50:50) and THF-buffer (40:60) are often markedly different (Table 1.7). The changes in the nitro group ( Zs,k-x=-54 in methanol and -56 in THF) and carbomethoxyl group (Zs,k-x = -10 in methanol and -80 in THF), match the published re-
Retention prediction based on molecular structure TABLE 1.5 RETENTION INDICES OF MODEL COMPOUNDS IN METHANOL-pH 7.0BUFFER ELUENTS Compound
Acetophenonea Aniline Anisole Benzaldehyde Benzamide Benzene Benzeneb Benzonitrile Benzyl alcohol Benzyl bromide Benzyl chloride Benzyl cyanide Biphenyl Bromobenzene Chlorobenzene Methyl benzoate Nitrobenzene Phenol Toluene
Retention index (0 Methanol (%)
40
50
60
70
80
805 650 884 774 605 883 885 776 689 991 962 773 1205 1027 998 899 851 685 987
806 658 904 777 589 915 913 788 691 1004 976 766 1222 1051 1021 904 857 683 1019
803 657 917 777 578 938 938 775 698 1019 992 763 1231 1065 1036 904 864 680 1039
803 659 934 775 570 958 961 774 684 1030 992 738 1247 1088 1051 910 874 67I 1065
804 639 954 784 551 983 982 760 675 1059 994 722 1270 1110 1072 914 874 650 1095
Data from Smith and Burr [31]. aDefined value I = 800. bParent index (Ip) values for benzene derived Erom quadratic regression equation (Appendix I. 1).
versa1 in the order of elution of methyl benzoate and nitrobenzene between these two modifiers [53], which is often used as an example of eluent selectivity. The phenolic hy= -229 in methanol, -243 in acetonitrile, and -142 in THF) and acetyl droxyl group (II= -1 13 in methanol, -127 in acetonitrile and -165 in THF) also reflect the marked changes that are frequently observed between separations in isoeluotropic eluents. The way in which the substituent index values change with the proportion of modifier also differs between the modifiers (Fig. 1.3). The chloro group index is unaffected by the proportion of acetonitrile, changes slightly with methanol and markedly with THF. These changes contrast with prediction methods based on log P values or on the extrapolation from a single log ko or zo value in pure water. In those cases, the relative order of elution is predetermined by the original values and the predicted relative order of elution will be the same in each eluent. Although the magnitudes of any changes can depend on the strength of the eluent, each will behave in a corresponding manner. 1.4.1.2 Aliphatic Jirnctional groups
Although by definition, the addition of a methylene group increases the retention index of a compound by 100 units = loo), the index increments for the successive addition of a methylene group to benzene to increase the length of the alkyl side chain were not sysReferences pp. 4 5 4 6
16
Chapter 1
tematic (dIcH2= 87 - 112 units) [50,54] suggesting that an interaction was occurring with the benzene ring. In particular, there was a reduced increment (dZ= 87-92 units) for substitution directly onto the benzylic carbon, i.e. toluene to ethylbenzene. An interaction index correction term (ZI,PHCH~R= -12 for methanol and acetonitrile and -14 in TKF) was therefore defined for alkyl substitution onto a benzylic carbon ) [50,54]. A similar effect had also been observed for the effect of the Hansch n coefficients on alkyl substituents on benzene. The step between the n values for methyl and ethyl substituents of 0.46 is smaller than for the addition of a methylene group to longer chains, which range from 0.56 to 0.58 [13]. The retention indices of a number of 1, 2, and 3-substituted toluenes, ethylbenzenes and propylbenzenes were measured and the substituent index increments determined relative to the calculated retention indices for the corresponding alkylbenzenes [54]. Because they might be partially or completely ionized in the mobile phase, carboxylic and sulphonic acids and amines were not examined. The increments for the substituents differed according to their proximity of the benzene ring. It was assumed that by the 3-position of propylbenzene the effect of the ring should be negligible and these retention index increments (Table 1.8), when available, were used to determine the substituents indices. However, in some cases only substituted toluenes or ethylbenzenes were initially available TABLE 1.6 RETENTION INDEX INCREMENTS FOR SUBSTITUENTS ON AN AROMATIC RING IN METHANOL ELUENTS Substituent
CONH2 NH2 CH20H OH CHO CH2CN CN COCH3b NO2 OCH3 C02CH3 H CH2CI CH3 CI CH2Br Br Ph
Retention index increment (Soa Methanol (?h)
Hansch a constant
40
so
60
70
80
-280 -23 5 -1 96 -20s -111 -1 12 -109 -85 -34 -1 14 0 77 102 113 106 142 320
-324 -255 -222 -230 -136 -147 -133 -1 13 -56 -9 -9 0 63 106 108 91 138 309
-360 -281 -240 -258 -161 -175 -1 63 -138 -74 -20 -34 0 54 101 98 81 127 293
-391 -3 02 -277 -290 -1 86 -223 -187 -161 -87 -27 -5 1 0 31 104 90 69 127 286
4 31 -343 -307 -332 -198 -260 -222 -182 -108 -2 8 -68 0 12 113 90 77 128 288
Values from Smith and Burr [3 13. aIncrement61 = I k - x - Ip (Ip = calculated value for benzene). bBased on defined value of I = 800.
-1.49 -1.23 -1.03 -0.67 -0.65 -0.57 -0.57 -0.55 -0.28 -0.02 -0.01 0.00 0.17 0.56 0.71 0.79 0.86 1.96
Retention prediction based on molecular structure
17
0
0
0
I
I
I
0
-2004
-250-
-300-
-350
I
I
1
(Table 1.8) [54]. The correlation coefficients for changes with the proportion of modifier were calculated and were used to determine the corresponding aliphatic substituent indices (Zs,R-x, Fig. 1.4). The retention indices of the substituted toluenes and ethyl benzenes were used to derive interaction index terms for the effect of the phenyl groups but generally these values were small and except for cyano, hydroxyl and carboxamide groups on toluene could probably be ignored ZI,ph-x < 50 units)[54]. Values from additional model compounds have been derived more recently [51] and the full set of coefficients for the substituent indices are listed in Appendix 1.1. Typical values for the aliphatic substituent indices (Is,R-x) have been calculated for isoeluotropic eluents (Table 1.7) and these were often markedly different from the corresponding aromatic substituents. In particular, the aliphatic hydroxyl group = -362, -459, and -434, in methanol, acetonitrile and THF, respectively) has a greater ef= -229, -243 and -142, respectively). These fect than the phenolic hydroxyl differences corresponded to those observed in the studies described earlier (Tables 1.1 and 1.3). The addition of an aliphatic bromo group causes little change to the retention of an analyte (I=3, -13, and 2, respectively), whereas the aromatic bromo group would increase the retention markedly (Zs, r-Br = 135, 127 and 117, respectively). References pp. 45-46
18
Chapter 1
TABLE 1.7 COMPARISON OF CALCULATED SUBSTITUENT INDICES IN ISOELUOTROPIC ELUENTS Substituent
Substituent index ( I s , ~ - xor IS, -x))a MeOH-bufferb
Benzene Ip Naphthalene
4,k-x
SO2CH3 CONH2 CONHCH3 CON(CH312 NHCOCH3 NH2 OH CHO OCOCH3 CN COCH3 NO2 OCH3 C02CH3 H NHC2H5 F NH(NH312 CH3
c1
Br Ph CHCH-CH3
Hansch n constant
(5050)
MeCN-bufferb (40:60)
THF-bufferb (30:70)
913 1104
927 1101
966 1112
-374 -321 -278 -199 -240 -254 -229 -138
-342 -392 -339 -274 -293 -230 -243 -141
-134 -1 13 -54 -1 1 -10 0 -40 6 44 100 105 135 305 239
-111 -127 -54 -19 -34 0 -1 1 5 46 100 98 127 27 1 215
-299 -381 -366 -381 -278 -216 -142 -165 -127 -133 -165 -5 6 -3 0 -80 0 8 16 17 100 95 117 24 1 190
-0.97 -1.23 -0.67 -0.65 -0.64 -0.57 -0.55 -0.28 -0.02 -0.01 0.00 0.08 0.14 0.18 0.56 0.71 0.86 1.96 Fragmental constant (FJ 1131
IS.R-X
CONH2 OH (primary) OH (sec.-tertiary) CHO CN COCH3 OCH3 C02CH3 Br H CI CH3 CHCH-CH3
-1.82 -1.49 -1.27
-432 -362 -394 -324 -300 -215 -167 -158 3 0 -39 100 217
-540 -459 -494 -326 -280 -226 -190 -186 -13 0 -49 100 199
-532 -434
-2.18 -1.64
-363 -316 -181 -227 -186 2 0 -27 100 -34
-1.27 -1.13 -1.10 -0.72 0.20 0.23 0.06 0.77
a I ~ , and ~ . IS,R-X - ~ calculated from coefficients in Appendix 1.1 [31,50,52,54]. bEluents selected to give similar retention factors for acetophenone:methanol-buffer (50:50), k = 3.23; acetonitribbuffer (40:60), k = 2.91; THF-buffer (30:70), k = 3.48.
Retention prediction based on molecular structure
19
200
100 X P)
-0 C
c C
aJ
0
3
.-c u) n ¶ 4-
cn
-100
-200 0
20
40
60
Organic modifier
80
100
(%I
Fig. 1.3. Comparison of substituent indices of the aryl chloro group (Ar-CI, solid symbols) and carbomethoxyl group (Ar-COzCH3, open symbols) in THF-buffer eluent (0 and 0 )methanol-buffer eluents (0and H) and acetonitrile-buffer eluents (A and A)[50]. Reproduced with permission.
Further studies examined the effect of unsaturation and chain branching in aliphatic side chains [52,55]. The olefinic group (-CH=CH-) had much smaller effect (ZS,CH:CH = 83-122 in 3-phenyl-I-propene and Is,CH:CH = 103-145 in 1-phenyl-1-propene) = 200) in methanolic and acetonitrile modifiers. With than two methylene groups THF as the modifier, the olefinic group in I-phenyl-1-propene had an even smaller contribution (Zs,CH:CH = 47-105) and appeared to make a large negative contribution (-105 to -320) in 3-phenyl-1-propene [52]. There were differences in the retention indices of isomeric alkylbenzenes and isomeric and corresponding isomeric phenylpropanols [%I. These lead to the introduction of terms for alkyl branching (II,brmch = -12 in methanol and -20 in acetonitrile) and for secondary and tertiary hydroxyl groups, which were slightly larger (about 40-50 units more negative) than those for primary hydroxyl groups. The coefficients for these terms and for olefinic groups are included in Appendix 1.1 and typical values are given in Table 1.7.
I.4.1.3 Relationship between substituent indices and octanol-water partition substituent increments Within groups of closely related compounds, log k of analytes in reversed-phase HPLC References pp. 45-46
Chapter I
20
TABLE 1.8 RETENTION INDEX INCREMENTS FOR ALIPHATIC SUBSTITUENTS IN METHANOL-pH 7.0 BUFFER ELUENTS Retention index increment Methanol (%) 40
50
60
70
80
1013
1039
1061
1082
-195
-261
-293
-327
-300
Substituents on ethylbenzene Calculated parent I
1073
1101
1126
1149
1170
COCH; CORa OCH3 ORa
-368 -1 89 -289 -150 -250
-403 -216 -316 -167 -267
-459 -243 -343 -181 -28 1
-486 -272 -372 -194 -294
-530 -303 -403 -202 -302
Substituents on n-propylbenzene Calculated parent I
1173
1201
1226
1249
1270
CN CO2CH3 C02Ra CI Br
-333 -266 -127 -227 -26 8
-366 -302 -158 -258 -42 -5
-391 -334 -186 -286 -5 1 -12
-436 -3 83 -216 -316 -65 -20
-477 -428 -24 1 -34 1 -80 -70
Substituents on toluene Calculated parent I
985
Data from Smith and Burr [54]. alOO subtracted for the methyl group contribution. b200 subtracted for the ethyl group contribution.
have frequently been linearly correlated with the octanol-water partition coefficients (log P)[10,151which can be calculated in an additive manner from the Hansch substituent constants x and the octanol-water log P value of a parent [13]. The values for the substituent indices should be thus related to the increments reported for the prediction of octanol-water partition coefficients, although, as noted earlier, the values will differ because of the selectivity of the modifier. This relationship might provide an mechanism by which estimated Z could be obtained for substituents not determined experimentally (e.g. the non-ionized carboxylic acid group) as was demonstrated by Baker (Eq. 1.6). values and The relationships between the initially predicted substituent index the Hansch x constants (Table 1.7) were therefore determined for a range of different eluent compositions [3 1,501. The correlation coefficients (r) were good for methanolbuffer eluents and improved with increasing buffer proportion (r = 0.9607-0.9849) but poorer for acetonitrile-buffer (0.9627-0.9739) [311. Similar results were also obtained for the full set of predicted indices in the three isoeluotropic eluents (Table 1.7). The hy-
21
Retentionprediction based on molecular structure
*0°1 100
I
.
.
CHS
Br
CI c
c
-200
CHO
OR
CO2R CN COR
OH
Acetonltrlle concentratlon (%) Fig. 1.4. Comparison of experimental values of retention index increments (symbols) and calculated aliphatic substituent indices (ZS,R-X) (lines) in acetonitrils-buffer eluents. Points are experimental values and curves are calculated substituent indices [54]. Reproduced with permission.
droxyl and carboxamide groups.had more negative than expected predicted indices and the sulphonamide group was less negative than predicted (Figs. 1.5-1.7). The fragmental constants F, (Table 1.7) are usually regarded as a closer match with the aliphatic contributions to log P [ 131. However, the correlations were poorer ranging from 0.925 to 0.869 in methanol and 0.924 to 0.860 in acetonitrile with the higher correlations being obtained with low proportions of modifier [52].
1.4.2 Polyfunctional compounds It was known from the studies of Hansch constants as predictors of octanol-water partition coefficients, that for disubstituted aromatic compounds containing polar groups, simple summation of the x terms to calculate log P was not very successful [56]. This was assumed to be due to electronic, steric and hydrogen-bonding interactions between the substituents. Initially, this led to the measurement of individual sets of x y values for each parent substituent. However, this approach is impractical for a wide range of combinations of groups as each group would need a separate set of interaction terms. Clearly a References pp. 45-46
Chapter 1
22
2.0
4-
1.o
E
m
CI
In
c
0 0
l= c
0.0
0 In C
m
=
-1.0
-2.0 -400
-200
200
0
400
Substituent index (I) Fig. 1.5 Comparison of Hansch x constants and calculated substituent indices for methanol-buffer (50:SO) based on Table 1.7.
2.0
-
0
, c
1.o 1.0
Ph
,‘
,,
-
c Q
In
,
C
0 0
l= c
0.0 -
,&‘O
0 In
c m
=
,
NHCOCH,
OH0
,‘ ,,
0 ,’ -1.0 CONHCH, ,’ 0 ,’
’
0
-2.0 I --400 -400
0
SO2NH2 I
I
I
I
-200
0
200
400
Substituent index (I) Fig. 1.6. Comparison of Hansch x constants and calculated substituent indices in acetonitrile-buffer (4050) based on Table 1.7.
Retention prediction based on molecular structure
c
23
1.0 -
U‘
c
,
0’
c Q ln
.‘U
c 0
0
t r
0.0 -
0
.‘0
ln
c Q
=
-1.0 -
‘ -2.0 I -400
s
OH
NWOCH,
0 SOzNH? I
I
I
i
-200
0
200
400
Substituent index (I) Fig. 1.7. Comparison of Hansch II values for substituents and calculated substituent indices in THF-buffer (30:70) based on Table 1.7.
similar problem will occur in the prediction of retention in HPLC and it is possible to examine the systems used in QSAR as models for chromatographic prediction systems 1571. 1.4.2.1 General prediction model
In the studies of log P,it was found that, for the meta- and para-isomers, the difference between the n value of a substituent X with benzene as the parent compound and that with phenol as the parent could be described using the Hammett constant CT of the substituent [56]. This led to a more general equation, which attempted to quantify the effect of a group Y on the Hansch constant of a substituent X in terms of their “susceptibility” constants @) and Hammett constants (0)[56,58-591. nX(phY) -nX(F’bH)
‘PgX
‘PflY
(1.12)
p x andp, are the susceptibilitiesof X and Y to the modifying effects of Y and X, respectively. Values for the susceptibility constants (Table 1.9) were derived experimentally using multiple regression analysis, In a similar study Leo [60,6 11 developed a simplified model to facilitate rapid estimation of the interaction terms. He included additional terms were included to account for intramolecular hydrogen-bonding (FHB),the negative ortho effect (F,) and the presence of alkyl-aryl systems (Fa$). Linear regression analysis gave the following correlation (Eq. 1.9) with experimental partition coefficients. References pp. 45-46
24
Chapter I
TABLE 1.9 VALUES OF u AND p USED IN CALCULATIONS OF INCREMENTS Substituent
P
urneta
upara
Inducers CN NO2 Br CI
0.56 0.71 0.39 0.37
0.66 0.78 0.23 0.23
0.00
Bidirectional CHO COzCH3 COCH3 CONH2 OCH3
0.35 0.37 0.38 0.28 0.12
0.42 0.45 0.50 0.36 -0.27
0.44 0.27 0.27 0.72
0.12 -0.16 -0.07
-0.37 -0.66 -0.17 -0.01
1.06 1.08 0 0
Responders OH NH2 CH3 Phenyl
0.06
0.00 0.00
0.00
0.50
Data from Hansch and Leo [13].
Iog P = C x + Fa - 0.29 F m - 0.1SF,$
(1.13)
in which Fa = p (T +p o y. By analogy, Smith and Burr [57]proposed that a corresponding general equation can be derived for the interaction increments in HPLC in which the different components are expressed in retention index units. 4,x-Y = ( a x P; p*, F& and
-Ia* P t ) -I-G B
+ F0*
(1.14)
Fo* correspond to the terms in Eq. (1.13) but are expressed in retention
index units. It is hoped that in each case, they could be directly related through a common regression equation to the eluent composition (Eq. 1.15). p* =p(ax2+ bx + c)
(1.15)
Leo noted the (T constants are valid for up to 80% organic modifier in aqueous solutions [60] and so they should be applicable in the present eluents. In preliminary calculations, it appeared that the meta- and para-interactions differed so that instead of common (T values as suggested by Leo [60,61], published urneta and apWa values (Table 1.9) [I31 were used. The term for alkyl-aryl substitution (Fa@)was omitted as it was thought that this effect had already be covered by the interaction term for alkyl substitution on a benzylic carbon. In order to study the interactions, the retention factors of 7 3 ortho-, metu-, and parasubstituted toluenes and phenols have been examined [57] However, a number of the
Retention prediction based on molecular structure
25
compounds, such as 4-nitrophenol (pK, = 7.1) had to be excluded from the subsequent calculations because they appeared to be partially ionized in the pH 7.0 eluent. The substituted toluenes showed only minor effects, mainly due to steric or electronic interactions. The phenols were expected to demonstrate stronger electronic effects and many were capable of intramolecular hydrogen-bonding. As with the monosubstituted compounds, the retention factors changed significantly with eluent composition but the retention indices for the substituted phenols were relatively constant across the composition ranges (Table I . 10). For most of the substituents the meta and para-isomers were similar (*50 units) but frequently the ortho-isomer was significantly different (up to 400 units). These differences were also reflected in the corresponding octanol-water partition coefficients (Table 1.10). For example, the retention indices of the methyl 2-, 3-, and 4hydroxybenzoates (Z= 940, 681 and 667) and partition coefficients (log P = 2.55, 1.89 and 1.96, respectively, both reflected a significant reduction in the polarity of the ortho isomer due to hydrogen-bonding. However, when the results for the substituted toluenes and phenols were compared they gave separate but parallel relationships (Fig. I A). The separation corresponded to the differences between the comparative values of substituent indices and the n constants for the phenolic hydroxyl and methyl groups. The ortho- interactions of methyl 2-hydroxybenzoate and other 2-hydroxy-carbonyl compounds were so marked that they appeared to behave as if they were substituted toluenes. The retention indices were used to calculate the interactions increments (61 values) between the substituents, as the differences between the measured I value and the summation of the previously determined parent index values for benzene together with the substituent indices (Zs,k-x) for the individual groups [57]. For most of substituted toluenes, the interaction increments were negligible or small. The most significant changes were found for 2-methylbenzamide (for example, 61= -64, -33 and -32 for the 2-, 3-, 4isomers respectively in acetonitrile-buffer (60:40)). In earlier work, Clark and his coworkers reported that this compound was eluted more rapidly than the 3- and 4-isomers, probably due to steric interaction of the 2-methyl group causing the amide group to be less coplanar with the aromatic ring and hence more polar [62]. Many of the substituted phenols showed much larger interaction increments, which changed significantly with eluent composition (Table I . 11). The smallest effects, usually 1.5) for 56 compound pairs and negative I d values (i.e. R, < 1.5) for 10 compound pairs using methaTABLE 9.9 PREDICTION OF RESOLUTION WITH METHANOL AT q5 = 0.513 Compound pair
Predicted retention indexa
12
Id
Experimental resolution
(minimum)’
Cortisone Hydrocortisone
635.4 672.0
666.7
5.3
1.91
Prednisone Hydrocortisone
624.8 672.0
657.3
14.7
2.59
Prednisone Cortisone
624.8 635.4
657.3
-21.9
0.66
Hydrocortisone Betamethasone
672.0 744.3
699.4
44.9
5.95
Betamethasone Corticosterone
744.3 746.6
764.0
-17.4
0.19
Betamethasone Reichstein’s substance S
744.3 759.8
764.0
-4.2
1.31
Betamethasone Androstenedione
744.3 807.1
764.0
43.1
6.98
Corticosterone Reichstein’s substance S
746.6 759.8
766.0
-6.2
1.14
Dexamethasone Reichstein’s substance S
747.4 759.8
766.0
-7.0
1.16
Androstenedione Testosterone
807.1 841.8
820.1
21.7
4.45
Testosterone Spironolactone
841.8 848.4
851.1
-2.7
0.98
Ethisterone 1 7a-Acetoxyprogesteronc
850.6 906.3
858.9
47.4
4.61
aFrom Eq. (9.2). ’From Eq. (9.6),with 12 (minimum) = 0.893(/,) + 90.3.
307
Solvent selectivity
no1 at $ = 0.513. Comparisons of these predictions with experimental resolutions resulted in an accurate prediction in every case. Several examples are given in Table 9.9. 9.3.1.4Optimization of resolution of steroid mixtures
In addition to predicting if a given pair of compounds can be completely resolved under a given set of conditions, it is also possible to predict the optimum conditions for yielding baseline resolution in minimum time. To optimize resolution in this manner, it is necessary to calculate a value of $ where I d = 0 (i.e. where R, = 1.5). In Table 9.10, I d values are calculated using experimental retention indices and 1, (minimum) values for three compound pairs with methanol as the strong solvent at seven values of @. Plotting Id on the x-axis versus $ on the y-axis yielded a linear relationship described by the following equation:
where q50 is the y-axis intercept and therefore represents the volume fraction of strong solvent at which 1, = 0 (and R, = 1.5). The $o regression coefficients for the three compound pairs are listed in Table 9.10. This technique predicted baseline resolution at rp0 = 0.579 for pair number 1, at $o = 0.662 for pair number 2, and at $o = 0.734 for pair number 3. To test the accuracy of this approach, a mixture of the compounds was chromatographed isocratically at each of the three rp0 values calculated to give baseline resolution for a given pair in the mixture. The resulting chromatograms are presented in Fig. 9.1. As expected, baseline resolution was achieved for each compound pair at the predicted value of $@ However, none of the isocratic conditions that resulted in baseline resolution for a given pair resulted in the optimum separation of all of the components in the mixture. At $o = 0.579 (Fig. 9.1, chromatogram A), the retention times of the three latest eluting components were longer than necessary to achieve baseline resolution, resulting in an unnecessarily long analysis time of nearly 40 min. Increasing $, to 0.662 (Fig. 9.1, chromatogram B) or 0.734 (Fig. 9.1, chromatogram C) substantially reduced the analysis time, but at the expense of decreased resolution of compound pair number 1. Figure 9.2 demonstrates that Eq. (9.8) can be used to estimate a gradient system that should optimize the separation of the entire mixture. In this system, $ was held isocratiTABLE 9.10 Id AS A FUNCTION OF $J WITH METHANOL AS THE STRONG SOLVENT
Compound pair no.a 1
2 3
Id as a function of $J
Intercept ($0)
0.550
0.575
0.600
0.625
0.650
0.675
0.700
5.9 17.9 48.3
-3.4 13.8 43.4
-4.5 15.0 37.2
-17.3 12.3 35.4
-15.2 5.5 28.5
-28.3 -0.5 17.3
-45.7 -11.1 6.1
0.579 0.662 0.734
"Pair no. 1, prednisone and hydrocortisone; pair no. 2, androstenedione and testosterone; pair no. 3, testosterone and 17a-acetoxyprogesterone.
References pp. 335-336
308
Chapter 9
--
0
10
20 30 Minutes
40
0
10
20 30 Minutes
40
0
10
20 3 0 4 0 Minutes
Fig. 9.1 Chromatograms demonstrating the use of Eq. (9.8) for predicting isocratic values of q5 that will yield baseline resolution for given compound pairs using methanol as the strong solvent: (A) q5 = 0.579; (B) q5 = 0.662; (C) q5 = 0.734. (Reproduced from Ref. 21 with permission of Preston Publications, A Division of Preston Industries, Inc.)
-
0
10
20 30 Minutes
40
Fig. 9.2. Chromatogram demonstrating the use of Eq. (9.8) for predicting values of q5 for optimizing the gradient resolution of all five compounds in a mixture in minimum time. (Reproduced from Ref. 21 with permission of Preston Publications, A Division of Preston Industries, Inc.)
Solvent selectivity
309
cally at 0.579 for 10 min to provide resolution of pair number 1. At 10 min, cp was increased to 0.662 and held isocratically for 5 min to provide baseline resolution for pair number 2. At 15 min, $ was increased to 0.734 and held isocratically to resolve the last pair. This technique resulted in baseline resolution for all five compounds in minimum time. 9.3.1.5 Resolution and the solvent selectivity triangle concept
Several discrepancies in the solvent selectivity triangle concept were observed for the experimentally determined selectivities of the various solvents in this study. Examination of the slopes (a) describing the change in the steroid retention indices as a function of q5 showed that they varied considerably among solvents from the same selectivity group (Table 9.4). For example, the average slope was 2.3 times greater for 1-propanol than for methanol, even though both solvents are classified in Group I1 and should therefore result in similar selectivity for the steroids relative to the 2-keto alkane standards. Likewise, the slopes for the two Group V1 solvents were substantially different, averaging 1.87 times greater for 2-methoxyethyl acetate than for acetonitrile. In some instances, slopes were actually more similar for solvents in different groups than for those in the same group. For example, the average slope for 2-methoxyethyl acetate in Group V1 (-663.5) more closely resembled that of tetrahydrofuran in Group I11 (-636.0) than that of its Group VI counterpart, acetonitrile (-354.7). Similarly, the average slopes for methanol (-394.5) in Group I1 and acetonitrile (-354.7) in Group VI were more similar than those for their group counterparts, 1-propanol (-901.4) and 2methoxyethyl acetate (-663 3. Consequently, a study was undertaken in which experimental resolutions (R,) were determined by RP-HPLC for three steroid compound pairs using a total of 10 solvents from three different selectivity groups. Resolution was calculated using the following classical equation:
where tz and tl are the observed retention times and w1 and w 2are the peak widths at baseline. In order to ensure that the comparison of resolution was done at constant solvent strength, the capacity factor (k) of the earliest eluting compound in each solute pair was adjusted to 2.00 f 0.03 for each solvent in a binary mixture with water. As shown in Table 9.1 1, the results of this study confirmed that solvents in the same selectivity group frequently did not give similar resolution, even at identical solvent strength. Perhaps the most dramatic examples occurred with compound pair number 1, consisting of spironolactone and ethisterone. Within Group 111, 2-ethoxyethanol, 2-methoxyethanol, and tetrahydrofuran resulted in an experimental resolution of 0.67, 1.15, and 3.26, respectively. Likewise, the resolution of this compound pair by Group VI solvents was greater than baseline for dioxane (R,= 1.70), but less than baseline with 2-methoxyethyl acetate (R, = 0.87) and acetonitrile (R, = 0.59). Similar discrepancies occurred in the attempted separation of compound pair number 2 (prednisone and hydrocortisone) with Group VI solvents. The R, values ranged from 0.25 for acetonitrile to 1.34 for 2-methoxyethyl acetate. References pp. 335-336
310
Chapter 9
TABLE 9.11 COMPARISON OF SOLVENT SELECTIVITYFOR THE EXPERIMENTAL RESOLUTION (Rs) OF THREE STEROID PAIRS Solvent
Solvent group
Polarity
Experimental resolution
(P’) Pair no. la
Pair no. 2a
Pair no. 3a
2-Propanol I-Propanol Ethanol MethanoI
I1 I1 I1 I1
3.9 4.0 4.3 5.1
0.78 1.01 0.71 0.26
1.53 1.64 1.88 1.58
1.35 1.29 0.81 0.93
Tetrahydrofuran 2-Ethoxyethanol 2-Methoxyethanol
I11 I11 111
4.0 5.0 5.5
3.26 0.67 1.15
1.81 2.54 1.93
0.42 1.00 1.03
2-Methoxyethyl acetate Dioxane Acetonitrile
VI VI VI
4.0 4.8 5.8
0.87 1.70 0.59
1.34 0.90 0.25
2.94 2.09 2.83
aPair no, 1, spironolactone and ethisterone; pair no. 2, prednisone and hydrocortisone; pair no. 3, betamethasone and Reichstein’s substance S.
The results in Table 9.1 1 also demonstrated that actual separations were frequently more alike with solvents classified in different selectivity groups than for those within the same group. For example, the resolution of prednisone and hydrocortisone by two of the Group I11 solvents, tetrahydrofuran (R, = 1.81) and 2-methoxyethanol (R, = 1.93), much more closely resembled the resolution obtained with ethanol (R, = 1.88) from Group I1 than that with the other Group I11 solvent, 2-ethoxyethanol (R, = 2.54). Also, the resolution of spironolactone and ethisterone with a Group I11 solvent, 2-ethoxyethanol (R, = 0.67), more closely resembled those obtained with ethanol (R, = 0.71) from Group I1 and acetonitrile (R, = 0.59) from Group VI than those with the other Group I11 solvents, tetrahydrofuran (R, = 3.26) and 2-methoxyethanol (R, = 1.15). These results contradicted the theory of the solvent selectivity triangle concept, which states that solvents in the same group should result in similar selectivity, while those in different groups should yield different selectivities [26,27]. A literature search conducted on publications listed in the Science Citation Index demonstrated widespread usage of the selectivity triangle as a rationale for solvent selection. However, with the exception of a publication by Lewis et ul. [33], definitive studies on the accuracy of the solvent groupings in the selectivity triangle appeared to be lacking. Lewis et ul. studied the separation of polystyrene oligomers using a total of 27 solvents representing all eight of the selectivity groups and concluded that the solvent triangle did not accurately predict selectivity for the separations being studied. These authors reported that the degree of solute solubility in the pure mobile phase solvents was a better predictor of selectivity than were the groupings of the solvent triangle. Thus, studies with steroids and polystyrene oligomers had clearly demonstrated that solvents classified within the same group of the solvent selectivity triangle did not necessarily result in similar selectivity. Both of these studies were conducted with compounds
Solvent selectivity
311
containing similar functionalities. Consequently, it was of interest to determine if the solvent selectivity triangle approach would accurately predict selectivity for compounds containing different hctionalities.
9.3.2 Retention and selectivity studies with benzene derivatives
9.3.2. I Retention index variation with solvent selectivity HPLC retention indices (Z) were determined for the 16 compounds on a CI8 column for each of the 12 strong solvents in aqueous binary mixtures of the mobile phase. The solvents used for the study with the benzene derivatives are listed in Table 9.1. The fraction of the organic solvent in the mobile phase ($) was adjusted to yield a capacity factor (k) of 4.00 f 0.04 for benzene. The values for q5 and I are listed in Tables 9.1 and 9.12, respectively. In most cases, considerable variation in retention indices occurred for solvents classified in the same solvent group, suggesting quantitative selectivity differences for the solutes relative to the 2-keto alkane standards. In many cases, retention indices for a given compound were more similar for solvents in different groups than for those in the same group. For example, the retention index of chlorobenzene with methanol from Group I1 was 812.7, which more closely resembled that of 2-ethoxyethanol in Group I11 (809.2) and acetonitrile in Group VI (813.7) than those of the other Group I1 solvents, ethanol (871.5), 1-propanol (979.8), and 2-propanol (968.5). Likewise, the retention index for nitrobenzene with ethanol from Group I1 (654.0) most closely resembled that with acetonitrile from Group VI (653.6), and the retention index for nitrobenzene with methanol from Group I1 (635.0) most closely resembled those of 2-ethoxyethanol in Group I11 (628.0) and dioxane in Group VI (626.9). The retention index of acetophenone with tetrahydrokan (THF) from Group I11 (583.7) much more closely resembled those of the four solvents from Group I1 (568. I to 579.6) and acetonitrile from Group VI (579.1) than those for any of the other four solvents in Group I11 (513.4 to 531.4). In general, there was very little or no correlation between retention indices and the solvents grouped according to the selectivity triangle concept. 9.3.2.2 Resolution and the solvent selectivity triangle To more thoroughly evaluate the correlation of selectivity with the solvent groupings in the solvent triangle, a study was undertaken in which experimental resolution (R,) was determined by RP-HPLC for the 16 benzene derivatives using all 12 of the mobile phase solvents. In order to ensure that the comparison was done at constant solvent strength, the volume of each strong solvent (q5) in a binary mixture with water was adjusted to yield a capacity factor of 4.00 f 0.04 for benzene. The results of this study confirmed that solvents in the same selectivity group seldom give similar resolution, even at constant solvent strength. The resolution of all 15 compounds from benzene using each of the 12 solvents is shown in Table 9.13. Numerous examples of extreme variation of R, within the solvent groups are evident, with resolution frequently being more alike for solvents classified in different groups than for those within a given group. For example, the resolution of benzene and chlorobenzene with ethanol from Group I1 (6.03) most closely resembled those of dimethylformamide (DMF) in Group 111 (6.14) and acetonitrile in Group VI Referencespp. 335-336
TABLE 9.12 RETENTION INDICES OF SEVERAL AROMATIC COMPOUNDS WITH 12 RF' SOLVENTSa Compound
-H -CI -Br -CH3 -CH=CHz -NO2 -COCH3 -COOCH3 -CHO -OH -cN -COCI -333
0-NT
m-NT p-NT
Group II
Group VI
Group I11
MeOH
EtOH
1-PrOH
2-PrOH
DMF
DMSO
2-EE
2-ME
THF
ACN
Diox
2-MEA
714.7 812.7 844.2 819.1 849.4 635.0 579.6 687.6 553.2 476.8 552.3 687.1 699.5 710.8 736.8 723.8
776.5 871.5 901.3 884.3 907.9 654.0 568.1 678.0 555.9 494.0 557.1 681.9 735.3 732.8 760.4 743.3
870.2 979.8 1010.2 994.2 1024.0 675.1 579.4 691.8 572.0 435.1 572.7 693.0 784.9 767.2 792.6 776.4
870.6 968.5 1005.6 997.7 1022.0 680.0 575.0 694.1 567.1 504.9 562.6 696.4 784.3 761.8 790.8 774.0
634.0 718.4 736.7 725.4 731.8 601.0 531.4 618.6 508.7 514.2 531.1 620.7 622.5 661.8 691.0 681.9
551.2 637.9 649.7 666.9 665.0 546.2 517.9 615.7 494.9 459.9 505.5 616.0 576.7 614.1 626.3 624.0
734.7 809.2 828.0 839.5 843.3 628.0 513.4 630.5 518.1 491.9 520.9 634.1 687.0 687.0 720.3 704.9
697.4 768.1 784.6 789.7 787.2 598.7 515.7 627.0 500.0 432.0 503.8 627.8 654.9 662.5 693.4 682.4
811.9 885.2 900.2 911.3 9 18.4 705.2 583.7 677.0 595.8 598.2 620.3 678.3 755.5 774.8 803.4 786.7
720.4 813.7 840.6 815.0 836.6 653.6 579.1 674.1 567.8 460.8 593.9 673.5 698.3 728.6 750.4 740.0
722.5 801.4 816.9 816.3 823.4 626.9 552.3 664.0 547.8 471.8 560.6 661.4 693.4 698.6 721.4 705.5
757.3 833.7 853.1 848.1 857.3 664.6 569.5 672.5 573.7 499.8 595.6 672.5 719.2 732.9 752.6 739.2
aSee Table 1 for solvent abbreviations.
TABLE 9.13 EXPERIMENTALRESOLUTION OF SEVERAL AROMATIC COMPOUNDS FROM BENZENE WITH 12 RP-HPLC SOLVENTS IN AQUEOUS BINARY MOBILE PHASES Compound
4 1
-Br 4H3 -CH=CH;? -NO2 -COCH3 -COOCH3 -CHO -OH
-cN -COCl 4CH3 0-NT rn-NT P-NT
Group III
Group II
Group VI
MeOH
EtOH
I-PIOH
2-PrOH
DMF
DMSO
2-EE
2-ME
THF
ACN
Diox
2-MEA
8.29 10.87 8.77 10.81 4.78 7.04 1.64 6.81 9.94 7.19 1.71 1.oo 0.28 1.60 0.67
6.03 8.44 6.67 7.78 5.10 6.56 4.06 6.90 8.68 7.39 4.03 1.97 2.03 0.81 1.66
4.61 5.82 4.91 6.69 5.52 7.98 5.40 7.89 10.21 7.88 5.81 3.05 3.33 2.60 3.80
3.67 5.44 4.75 6.07 4.86 6.27 5.04 6.50 8.02 7.18 5.24 2.75 3.29 2.65 2.97
6.14 8.48 7.16 9.62 1.76 5.59 0.82 6.20 4.73 4.20 0.76 0.54 1.74 3.58 2.86
4.84 6.42 8.02 6.78 0.22 1.26 3.01 2.39 2.82 1.55 3.27 0.91 3.42 3.65 3.78
3.97 5.41 5.52 5.50 3.73 6.68 3.98 6.76 8.09 6.56 3.64 1.88 2.03 0.64 1.40
3.32 4.13 4.25 3.98 3.25 4.33 1.97 4.30 6.06 4.70 2.35 1.57 1.27 0.15 0.58
3.34 4.15 4.50 4.95 3.92 7.61 4.80 6.79 8.17 6.27 4.23 2.17 1.44 0.35 1.07
6.71 9.00 6.54 9.19 3.68 6.57 2.37 6.36 10.95 6.29 2.65 1.30 0.46 1.90 1.21
3.78 4.74 4.63 5.11 2.73 4.41 1.90 4.02 5.72 4.16 2.04 1.02 0.87 0.04 0.64
4.65 5.87 5.27 5.83 3.43 6.11 3.64 6.16 7.53 6.27 3.72 1.80 1.12 0.90 0.25
314
Chapter 9
(6.71). Likewise, the resolution of benzene and anisole with methanol from Group I1 (1.OO) most closely resembled those obtained with dimethylsulfoxide (DMSO) in Group I11 (0.91) and dioxane in Group V I (1.02). Numerous other examples are obvious in Table 9.13. The discrepancies in solvent selectivity within solvent groups were not limited to resolution from benzene, but rather were universal for all of the compounds studied. Other examples of extreme variation are shown in Table 9. I4 with all solvent groups for the resolution of nitrobenzene and anisole from benzoyl chloride and for the resolution of three positional isomers of nitrotoluene. Methanol was the only Group I1 solvent and DMSO was the only Group I11 solvent providing resolution greater than baseline (i.e. R, 2 1.5) for nitrobenzene and benzoyl chloride. In Group 11, only methanol failed to completely resolve anisole from benzoyl chloride, while only three of the Group I11 solvents and one of the Group VI solvents provided baseline resolution. For the resolution of 0- and m-nitrotoluene, methanol was the only Group I1 solvent and DMF and 2ethoxyethanol were the only Group 111 solvents providing complete separation. In Group 111, the resolution of nitrobenzene and benzoyl chloride ranged from 0.2 to 3.2, and the resolution of anisole and benzoyl chloride ranged from 0.1 to 2.1. 9.3.2.3 Prediction of resolution
As indicated in Section 9.3.1.3. for steroids, the ability of a solvent to resolve a given pair of compounds can be predicted using Z.(minimum) values and Eq. (9.7). Regression coefficients for the slope (3) and y-axis intercept (y) are contained in Table 9.15 for the values of @ given in Table 9.1. A positive value for indicates that the retention index of the second component is greater than that required for baseline resolution, while a negative Z, indicates that resolution cannot be achieved. This approach for predicting resolution is demonstrated to also provide accurate predictions for benzene derivatives in Table 9.16 by using the previous example of nitrobenzene and benzoyl chloride in which the selectivity triangle concept failed to predict differences in solvent selectivity. From Table 9.12, the retention indices of nitrobenzene and benzoyl chloride with methanol at @ = 0.589 are 635.0 and 687.1, respectively. From Table 9.15, the minimum retention index necessary for complete resolution from nitrobenzene with methanol as the strong solvent is calculated as:
Z.(minimum) = 0.8581(635.0) + 127.9 = 672.8
(9.10)
Z, thus equals 687.1 - 672.8 = 14.3, indicating that the retention index of benzoyl chloride is 14.3 units greater than the minimum required for baseline resolution from nitrobenzene. An experimental resolution (R,)of 2.5 confirmed that complete resolution was easily obtained. In like manner, this technique correctly predicted that DMSO was the only other solvent capable of completely resolving this pair of solutes (Table 9.15). 9.3.2.4HPLC resolution as afunction of retention index diferences A common misconception with retention index systems in both GC and HPLC is that two
liquid phases producing similar retention indices or retention index differences will result in a similar resolution of the mixture of compounds [38 and references therein]. Without
b
8 m Ti
0 3
2
3cu
TABLE 9.14 VARIATION OF EXPERIMENTAL RESOLUTION OF AROMATIC COMPOUNDS WITH SOLVENTS CLASSIFIED ACCORDING TO THE SOLVENT SELECTIVITY TRIANGLE
m
z 0,
$Compounds
cu
Y cu cu