57 Plant Metabolomics (158pages) Biotechnology in Agriculture and Forestry

Biotechnology in Agriculture and Forestry
Edited by
T. Nagata (Managing Editor)
H. Lörz
J. M. Widholm
Volumes already published and in preparation are listed at the end of this book.
Biotechnology in
Agriculture and Forestry 57
Plant Metabolomics
Edited by
K. Saito, R.A. Dixon, and L. Willmitzer
With 96 Figures, 29 in Color, and 10 Tables
123
Series Editors
Professor Dr. Toshiyuki Nagata
University of Tokyo
Graduate School of Science
Department of Biological Sciences
7-3-1 Hongo, Bunkyo-ku
Tokyo 113-0033, Japan
Professor Dr. Horst Lörz Professor Dr. Jack M. Widholm
Universität Hamburg University of Illinois
Biozentrum Klein Flottbek 285A E.R. Madigan Laboratory
Zentrum für Angewandte Molekularbiologie Department of Crop Sciences
der Pflanzen (AMP II) 1201 W. Gregory
Ohnhorststraße 18 Urbana, IL 61801, USA
22609 Hamburg, Germany
Volume Editors
Professor Dr. Kazuki Saito
Chiba University
Graduate School of Pharmaceutical Sciences
Yayoi-cho 1-33, Inage-ku
Chiba 263-8522, Japan;
RIKEN Plant Science Center
Yokohama 230-0045, Japan
Professor Dr. Richard A. Dixon Professor Dr. Lothar Willmitzer
Plant Biology Division Max Planck Institute
Samuel Roberts Noble Foundation of Molecular Plant Physiology
2510 Sam Noble Parkway Am Mühlenberg 1
Ardmore, OK 73401, USA 14476 Golm, Germany
Library of Congress Control Number: 2005936763
ISSN 0934-943X
ISBN-10 3-540-29781-2 Springer Berlin Heidelberg New York
ISBN-13 978-3-540-29781-9 Springer Berlin Heidelberg New York
This work is subject to copyright. All rights reserved, whether the whole or part of the material is concerned,
specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction
on microfilm or in any other way, and storage in data banks. Duplication of this publication or parts thereof
is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current
version, and permission for use must always be obtained from Springer. Violations are liable for prosecution
under the German Copyright Law.
Springer is a part of Springer Science + Business Media
springer.com
© Springer-Verlag Berlin Heidelberg 2006
Printed in Germany
The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply,
even in the absence of a specific statement, that such names are exempt from the relevant protective laws and
regulations and therefore free for general use.
Editor: Dr. Dieter Czeschlik, Heidelberg, Germany
Desk Editor: Dr. Andrea Schlitzberger, Heidelberg, Germany
Cover design: design&production GmbH, Heidelberg, Germany
Typesetting and production: LE-TEX Jelonek, Schmidt & Vöckler GbR, Leipzig, Germany
Printed on acid-free paper 31/3152 5 4 3 2 1 0
Preface
Metabolomics is a rapidly-emerging sector of post-genome research. The

metabolome (a set of all metabolites of an organism) represents not only
the ultimate phenotype of cells by the perturbation of gene expression and the
modulation of protein functions caused by the environment or mutations, but
the metabolome can also feed back on gene expression and protein function.
Therefore, metabolomics plays a key role for understanding cellular systems.
Metabolomics is applied to a variety of biological fields from medical science
to agriculture. Nevertheless, metabolomics research is particularly important
in the plant field because plants collectively produce a huge variety of chemical
compounds, far more than animals and even microorganisms. The number
of all metabolites in the plant kingdom is estimated at 200,000 or more. In
addition, most of the human-beneficial properties of plants, be they foods,
medicinal resources, or industrial raw materials, are ascribed to plant metabo-
lites.
This book aims to review the current status of plant metabolomics research.
Since metabolomics itself is a new field, no such comprehensive book has yet
been published. The chapters are divided into three sections: analytical tech-
nology, bioinformatics, and applications. These represent three major elements
of metabolomics research. Each chapter provides cutting-edge information
contributed by leading researchers from throughout the world.
We hope that this book will be a landmark for plant metabolomics research
into the future and will give beneficial guidance to graduate students and
researchers in academia, industry, and technology transfer organizations. Since
metabolomics is still a growing discipline, further technology development in
chemical analysis and bioinformatics will be required. We look forward to
breakthrough technology innovations in metabolomics, and yet unforeseen
findings and applications in plant science.
Finally, we would like to acknowledge our contributors who have enthusias-
tically put their efforts to ensure the high scientific quality of this volume. We
also would like to thank our colleagues at Springer.
January 2006 Kazuki Saito,

Richard A. Dixon,
and Lothar Willmitzer
Contents
Section I Analytical Technology
I.1 Gas Chromatography Mass Spectrometry . . . . . . . . . . . . . . . . . . . . 3

J. Kopka
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 GC-MS Profiling Technology in a Nutshell . . . . . . . . . . . . . . . . . . . 5
3 Short Excursion into Nomenclature and Definitions . . . . . . . . . . . . 10
4 Present Challenges of GC-MS Profiling . . . . . . . . . . . . . . . . . . . . . . 13
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
I.2 Current Status and Forward Looking Thoughts

on LC/MS Metabolomics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
L.W. Sumner
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2 Chromatography Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3 Limitations of Current Metabolic Profiling Approaches
and Proposed Solutions to Advance Metabolomics . . . . . . . . . . . . . 25
4 Future Directions and Forward-Looking Thoughts . . . . . . . . . . . . . 28
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
I.3 Plant Metabolomics Strategies Based upon Quadrupole Time

of Flight Mass Spectrometry (QTOF-MS) . . . . . . . . . . . . . . . . . . . . 33
H.A. Verhoeven, C.H. Ric de Vos, R.J. Bino, and R.D. Hall
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2 The Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3 Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
4 Application of QTOF MS-based Plant Metabolomics Analyses . . . . 38
5 Conclusions and Future Prospects . . . . . . . . . . . . . . . . . . . . . . . . . 46
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
VIII Contents
I.4 Capillary HPLC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

T. Ikegami, E. Fukusaki, and N. Tanaka
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2 Monolithic Silica Columns for Micro HPLC . . . . . . . . . . . . . . . . . . 49
3 Applications of Monolithic Silica Columns to Metabolomics . . . . . . 54
4 Two-Dimensional HPLC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5 Combination of Reversed-Phase HPLC
and Other Separation Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
6 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
I.5 Capillary HPLC Coupled to Electrospray Ionization

Quadrupole Time-of-flight Mass Spectrometry . . . . . . . . . . . . . . . . 65
S. Clemens, C. Böttcher, M. Franz, E. Willscher,
E. v. Roepenack-Lahaye, and D. Scheel
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
2 Extraction, Chromatography and Mass Spectrometry . . . . . . . . . . . 67
3 Potential and Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
4 Conclusions and Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
I.6 NMR Spectroscopy in Plant Metabolomics . . . . . . . . . . . . . . . . . . . 81

J.L. Ward and M.H. Beale
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
2 High-throughput Screening by 1D 1 H-NMR . . . . . . . . . . . . . . . . . . 82
3 Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
4 Two-dimensional NMR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
5 Stable Isotope Labelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
6 Hyphenated NMR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
7 Discussion: Applying NMR to Plant Metabolomics . . . . . . . . . . . . . 88
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
I.7 Hetero-nuclear NMR-based Metabolomics . . . . . . . . . . . . . . . . . . . 93

J. Kikuchi and T. Hirayama
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
2 Historical Aspects of NMR Studies of Plant Metabolism . . . . . . . . . 93
3 1 H-NMR-based Metabolomics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4 Use of Stable Isotope Labeling Technique to Enable Monitoring
of the Dynamic Movement of Metabolites . . . . . . . . . . . . . . . . . . . . 94
5 Approach for Hetero-nuclear NMR-based Metabolomics . . . . . . . . 95
6 Prospects for the Future . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
Contents IX
Section II Bioinformatics
II.1 Bioinformatics Approaches to Integrate Metabolomics

and Other Systems Biology Data . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
B. Mehrotra and P. Mendes
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
2 Databases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3 Data Visualization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
4 Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
II.2 Chemometrics in Metabolomics – An Introduction . . . . . . . . . . . . 117

J. Trygg, J. Gullberg, A.I. Johansson, P. Jonsson, and T. Moritz
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
2 Theory and Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
3 Example: Metabolomics Study on Arabidopsis Mutants . . . . . . . . . 125
4 Summary and Future Prospectives . . . . . . . . . . . . . . . . . . . . . . . . . 126
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
II.3 Map Editor for the Atomic Reconstruction of Metabolism (ARM) . 129
M. Arita, Y. Fujiwara, and Y. Nakanishi
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
2 Definition of Metabolic Information . . . . . . . . . . . . . . . . . . . . . . . . 131
3 Metabolic Map Editor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
4 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
II.4 AraCyc: Overview of an Arabidopsis Metabolism Database

and its Applications for Plant Research . . . . . . . . . . . . . . . . . . . . . . 141
S.Y. Rhee, P. Zhang, H. Foerster, and C. Tissier
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
2 Database Content . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
3 Search, Browse, and Analyze Functionalities . . . . . . . . . . . . . . . . . . 145
4 Applications of AraCyc . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
5 Current Issues and Future Directions . . . . . . . . . . . . . . . . . . . . . . . 152
6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
II.5 KaPPA-View: A Tool for Integrating Transcriptomic

and Metabolomic Data on Plant Metabolic Pathway Maps . . . . . . . . 155
T. Tokimatsu, N. Sakurai, H. Suzuki, and D. Shibata
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
2 General Features of the KaPPA-View Tool . . . . . . . . . . . . . . . . . . . . 155
X Contents
3 Plant Metabolic Pathway Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158

4 Integration of Transcriptomic and Metabolomic Data
on Pathway Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
5 Comparison with Other Databases and Tools . . . . . . . . . . . . . . . . . 159
6 Limitations and Future Improvements . . . . . . . . . . . . . . . . . . . . . . 160
7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163
II.6 KNApSAcK: A Comprehensive Species-Metabolite

Relationship Database . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
Y. Shinbo, Y. Nakamura, M. Altaf-Ul-Amin, H. Asahi,
K. Kurokawa, M. Arita, K. Saito, D. Ohta, D. Shibata,
and S. Kanaya
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 165
2 Search Options of the KNApSAcK Database . . . . . . . . . . . . . . . . . . 166
3 Statistics of the Database . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
4 Classification Based on Common Metabolites . . . . . . . . . . . . . . . . . 177
5 Conclusion and Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
6 Access to KNApSAcK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
Section III Applications
III.1 Systems Biology: A Renaissance of the Top-down Approach

for Plant Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
F. Carrari, N. Schauer, L. Willmitzer, and A.R. Fernie
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
2 Re-emergence of Top-down Thinking . . . . . . . . . . . . . . . . . . . . . . . 186
3 Systems Biology in Non-plant Systems . . . . . . . . . . . . . . . . . . . . . . 186
4 Systems Biology in Plant Systems . . . . . . . . . . . . . . . . . . . . . . . . . . 188
5 Dynamic Profiling in Plant Cells . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
6 Conclusions and Future Perspectives . . . . . . . . . . . . . . . . . . . . . . . 195
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
III.2 Systems-based Analysis of Plant Metabolism by Integration

of Metabolomics with Transcriptomics . . . . . . . . . . . . . . . . . . . . . . 199
M.Y. Hirai, T. Tohge, and K. Saito
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
2 Understanding Whole Plant Metabolism – Our Aims and Strategy . 199
3 Metabolome and Transcriptome Analyses . . . . . . . . . . . . . . . . . . . . 200
4 Studies on Sulfur Metabolism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201
5 Studies on Anthocyanin Metabolism . . . . . . . . . . . . . . . . . . . . . . . . 206
Contents XI
6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
III.3 Targeted Profiling of Fatty Acids and Related Metabolites . . . . . . . . 211

T.R. Larson and I.A. Graham
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
2 Metabolite Profiling Techniques Used to Study Plant Lipid
Metabolism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
3 Future Developments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
III.4 Metabolic Profiling and Quantification of Carotenoids

and Related Isoprenoids in Crop Plants . . . . . . . . . . . . . . . . . . . . . . 229
P.D. Fraser and P.M. Bramley
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
2 Analytical Methodologies Employed in the Analysis
of Carotenoids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
3 Examples of Carotenoid/isoprenoid Profiling . . . . . . . . . . . . . . . . . 237
4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
III.5 Metabolomics and Gene Identification

in Plant Natural Product Pathways . . . . . . . . . . . . . . . . . . . . . . . . . 243
R.A. Dixon, L. Achnine, B.E. Deavours, and M. Naoumkina
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
2 Gene Discovery – Past and Present Strategies . . . . . . . . . . . . . . . . . 243
3 Enzyme Promiscuity in Natural Product Pathways . . . . . . . . . . . . . 246
4 Examples of the Use of Metabolomics in the Elucidation
of Gene Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
5 Single Cell or Isolated Tissue Metabolomics . . . . . . . . . . . . . . . . . . 253
6 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
III.6 Metabolomic Analysis of Catharanthus roseus

Using NMR and Principal Component Analysis . . . . . . . . . . . . . . . 261
H.K. Kim, Y.H. Choi, and R. Verpoorte
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
2 Experimental Consideration for Metabolomics Using NMR . . . . . . 262
3 Application of NMR for Plant Metabolome . . . . . . . . . . . . . . . . . . . 266
4 Principal Component Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 273
5 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
XII Contents
III.7 Metabolomics of Plant Secondary Compounds:

Profiling of Catharanthus Cell Cultures . . . . . . . . . . . . . . . . . . . . . . 277
M. Orešic̀, H. Rischer, and K.-M. Oksman-Caldentey
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
2 Metabolomics as a Platform to Study Plant Secondary Metabolites . 278
3 Case Study: Metabolic Profiling of Catharanthus roseus Cells . . . . . 280
4 Protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
5 Perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
III.8 The Taxus Metabolome and the Elucidation

of the Taxol® Biosynthetic Pathway in Cell Suspension Cultures . . . 291
R.E.B. Ketchum and R.B. Croteau
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
2 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
3 Protocol . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
III.9 The Use of Non-targeted Metabolomics in Plant Science . . . . . . . . . 311

T. Daskalchuk, P. Ahiahonu, D. Heath, and Y. Yamazaki
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311
2 Fundamental Investigations into Plant Metabolomics . . . . . . . . . . . 313
3 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
III.10 Plant Metabolite Profiling for Industrial Applications . . . . . . . . . . . 327

R.N. Trethewey
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
2 The Metabolome . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
3 Profiling Technologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
4 High Throughput Metabolite Profiling . . . . . . . . . . . . . . . . . . . . . . 332
5 Industrial Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335
6 Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338
Subject Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341

List of Contributors
L. Achnine
Plant Biology Division, Samuel Roberts Noble Foundation, 2510 Sam Noble
Parkway, Ardmore, OK 73401, USA
P. Ahiahonu
Phenomenome Discoveries Inc., 204–407 Downey Road, Saskatoon,
Saskatchewan, Canada S7N 4L8
M. Altaf-Ul-Amin
Department of Bioinformatics and Genomics, Graduate School of
Information Science, Nara Institute of Science and Technology (NAIST),
Takayama-cho 8916–5, Ikoma, Nara 630–0101, Japan
M. Arita
Department of Computational Biology, Graduate School of Frontier Sciences,
The University of Tokyo, 5–1–5 Kashiwanoha, Kashiwa, 277–8561 Japan,
e-mail: [email protected]
H. Asahi
M.H. Beale
The National Centre for Plant and Microbial Metabolomics, Rothamsted
Research, West Common, Harpenden, Herts. AL5 2JQ, UK,
R.J. Bino
Plant Research International, P.O. Box 16, 6700 AA Wageningen, The
Netherlands
C. Böttcher
Leibniz Institute of Plant Biochemistry, Weinberg 3, 06120 Halle/Saale,
Germany
XIV List of Contributors
P.M. Bramley
School of Biological Sciences, Royal Holloway, University of London, Egham,
Surrey, TW20 0EX, UK
F. Carrari
Max Planck Institute of Molecular Plant Physiology, Am Mühlenberg 1,
14476, Golm, Germany
Y.H. Choi
Division of Pharmacognosy, Section Metabolomics, Institute of Biology,
Leiden University, P.O. Box 9502, 2300RA, Leiden, The Netherlands
S. Clemens
Germany, e-mail: [email protected]
R.B. Croteau
Institute of Biological Chemistry, Washington State University, Pullman, WA
99164, USA
T. Daskalchuk
Saskatchewan, Canada S7N 4L8, e-mail: [email protected]
B.E. Deavours
R.A. Dixon
Parkway, Ardmore, OK 73401, USA, e-mail: [email protected]
A.R. Fernie
14476, Golm, Germany, e-mail: [email protected]
H. Foerster
Carnegie Institution, Department of Plant Biology, 260 Panama Street,
Stanford, CA 94305, USA
M. Franz
Germany
List of Contributors XV
P.D. Fraser
School of Biological Sciences, Royal Holloway, University of London, Egham,
Surrey, TW20 0EX, UK, e-mail: [email protected]
Y. Fujiwara
Department of Computational Biology, Graduate School of Frontier Sciences,
The University of Tokyo, 5–1–5 Kashiwanoha, Kashiwa, 277–8561 Japan
E. Fukusaki
Department of Biotechnology, Graduate School of Engineering, Osaka Univ,
2–1 Yamadaoka, Suita, 565–0871, Japan,
I.A. Graham
CNAP, Department of Biology (Area 7), University of York, PO Box 373, York
YO10 5YW, UK
J. Gullberg
Umeå Plant Science Centre, Department of Forest Genetics and Plant
Physiology, Swedish University of Agricultural Sciences, SE-901 87 Umeå,
Sweden
R.D. Hall
Netherlands, e-mail: [email protected]
D. Heath
M.Y. Hirai
RIKEN Plant Science Center, Suehiro-cho 1–7–22, Tsurumi-ku, Yokohama,
Kanagawa 230–0045, Japan
T. Hirayama
International Graduate School of Arts and Sciences, Yokohama City
University, 1-7-29 Suehiro, Tsurumi-ku, Yokohama, 230-0045 Japan
T. Ikegami
Department of Polymer Science and Engineering, Kyoto Institute of
Technology, Matsugasaki, Sakyo-ku, Kyoto, 606–8585, Japan
A.I. Johansson
Umeå Plant Science Centre, Department of Forest Genetics and Plant Physiol-
ogy, Swedish University of Agricultural Sciences, SE-901 87 Umeå, Sweden
XVI List of Contributors
P. Jonsson
Research Group for Chemometrics; Organic Chemistry, Department of
Chemistry, Umeå University, SE-901 87 Umeå, Sweden
K.-M. Oksman-Caldentey
VTT Biotechnology, P.O. Box 1500, 02044 VTT, Finland,
e-mail: Kirsi-Marja.Oksman@vtt.fi
S. Kanaya
Takayama-cho 8916–5, Ikoma, Nara 630–0101, Japan,
R.E.B. Ketchum
Institute of Biological Chemistry, Washington State University, Pullman, WA
99164, USA, e-mail: [email protected]
J. Kikuchi
RIKEN Plant Science Center, 1-7-22 Suehiro-cho, Tsurumi-ku, Yokohama,
230-0045 Japan, e-mail: [email protected]
H.K. Kim
Leiden University, P.O. Box 9502, 2300RA, Leiden, The Netherlands,
J. Kopka
Max Planck Institute of Molecular Plant Physiology, Am Mühlenberg 1, 14476
Potsdam-Golm, Germany, e-mail: [email protected]
K. Kurokawa
T.R. Larson
CNAP, Department of Biology (Area 7), University of York, PO Box 373, York
YO10 5YW, UK, e-mail: [email protected]
B. Mehrotra
Virginia Bioinformatics Institute, Virginia Polytechnic Institute and State
University, Washington St., MC 0477, Blacksburg, Virginia 24061, USA
List of Contributors XVII
P. Mendes
Virginia Bioinformatics Institute, Virginia Polytechnic Institute and State
University, Washington St., MC 0477, Blacksburg, Virginia 24061, USA,
T. Moritz
Umeå Plant Science Centre, Department of Forest Genetics and Plant
Physiology, Swedish University of Agricultural Sciences, SE-901 87 Umeå,
Sweden, e-mail: [email protected]
Y. Nakamura
Y. Nakanishi
Intec Web and Genome Informatics Corporation
M. Naoumkina
D. Ohta
Department of Plant Genes and Physiology, Graduate School of Agriculture
and Biological Sciences, Osaka Prefecture University, Gakuen-cho 1–1, Sakai,
Osaka 599–8531, Japan
M. Orešic̀
VTT Biotechnology, P.O. Box 1500, 02044 VTT, Finland
S.Y. Rhee
Stanford, CA 94305, USA, e-mail: [email protected]
C.H. Ric de Vos

Netherlands
H. Rischer
VTT Biotechnology, P.O. Box 1500, 02044 VTT, Finland
XVIII List of Contributors
K. Saito
Chiba University, Graduate School of Pharmaceutical Sciences, Yayoi-cho
1-33, Chiba 263-8522, Japan
Kanagawa 230–0045, Japan, e-mail: [email protected]
N. Sakurai
Kazusa DNA Research Institute, 2–6–7 Kazusa-kamatari, Kisarazu, Chiba
292–0818, Japan
N. Schauer
D. Scheel
Germany
D. Shibata
292–0818, Japan, e-mail: [email protected]
Y. Shinbo
New Energy and Industrial Technology Development Organization, Toshima,
Tokyo 170–6028, Japan
L.W. Sumner
The Samuel Roberts Noble Foundation, 2510 Sam Noble Parkway, Ardmore,
OK 73401, USA, e-mail: [email protected]
H. Suzuki
292–0818, Japan
N. Tanaka
Department of Polymer Science and Engineering, Kyoto Institute of
Technology, Matsugasaki, Sakyo-ku, Kyoto, 606–8585, Japan
C. Tissier
T. Tohge
Kanagawa 230–0045, Japan
List of Contributors XIX
T. Tokimatsu
292–0818, Japan
R.N. Trethewey
metanomics GmbH and metanomics Health GmbH, Tegeler Weg 33, 10589
Berlin, Germany, e-mail: [email protected]
J. Trygg
Research Group for Chemometrics; Organic Chemistry, Department of
Chemistry, Umeå University, SE-901 87 Umeå, Sweden
E. v. Roepenack-Lahaye
Germany
H.A. Verhoeven
Netherlands
R. Verpoorte
Leiden University, P.O. Box 9502, 2300RA, Leiden, The Netherlands
J.L. Ward
The National Centre for Plant and Microbial Metabolomics, Rothamsted
Research, West Common, Harpenden, Herts. AL5 2JQ, UK,
L. Willmitzer
E. Willscher
Germany
Y. Yamazaki
P. Zhang
I.1 Gas Chromatography Mass Spectrometry
J. Kopka1
1 Introduction
GC-MS technology has been used for decades in studies which aim at the
exact quantification of metabolite pool size and metabolite flux. Exact quan-
tification has traditionally been focused on a single or small set of predefined
target metabolites. Today GC-MS is one of the most widely applied technology
platforms in modern metabolomic studies. Since early applications in unrav-
elling the mode of action of herbicides (Sauter et al. 1988) it has experienced
a renaissance (Fig. 1) in post-genomic, high-throughput fingerprinting and
metabolite profiling of genetically modified (e. g. Roessner et al. 2001a,b, 2002;
Fernie et al. 2004) or experimentally challenged plant samples (e. g. Cook et
al. 2004; Kaplan et al. 2004; Urbanczyk-Wochniak and Fernie 2005). Metabolic
phenotyping and analysis of respective phenocopies by metabolite profiling
has become an integral part of plant functional genomics (Fiehn et al. 2000b;
Roessner et al. 2002; Fernie et al. 2004). The essence of metabolite profiling,
namely the non-biased screening of biological samples for changes of metabo-
lite levels relative to control samples, has been thoroughly discussed earlier
and is clearly distinguished from fingerprinting approaches and the concept
of exact quantification (Fiehn et al. 2000b; Sumner et al. 2003; Birkemeyer et
al. 2005).
GC-MS-based metabolome profiling analysis is on the verge of becoming
a routine technology. This fact substantially contributes to the development
of metabolomics as a fourth integral part of the Rosetta stone for functional
genomics and molecular physiology (Trethewey et al. 1999; Fiehn et al. 2000b;
Trethewey 2004). Nevertheless, GC-MS technology is already challenged again
by new bottlenecks and demands for improved data sets which are optimised
for the mathematical modelling tools currently developed in the fields of bioin-
formatics and biological systems analysis.
The challenges of modern, multi-parallel, GC-MS based metabolite anal-
ysis are manifold: (i) automation of sample preparation, wet chemistry and
data processing after acquisition for increased throughput and reproducibil-
ity, (ii) extension of the analytical scope of metabolomics studies, for example
by combined analysis of single samples using multiple analytical technol-
ogy platforms, and combined analysis with the proteome and transcriptome
1 MaxPlanck Institute of Molecular Plant Physiology, Am Mühlenberg 1, 14476 Potsdam-Golm,
Germany, e-mail: [email protected]
Biotechnology in Agriculture and Forestry, Vol. 57

Plant Metabolomics (ed. by K. Saito, R.A. Dixon, and L. Willmitzer)
4 J. Kopka
Fig. 1. Literature survey of publications which associate the concepts, “metabolite”, “profiling”,
and “gas chromatography” performed on 1/2005. A total of ∼500 citations without conference
proceedings, abstracts and book chapters were found. The frequency of publications in all bio-
logical sciences (open circles) is compared to the contribution by plant metabolomics community
(closed circle)
(Weckwerth et al. 2004b), (iii) profiling of trace compounds, or signalling

molecules in the presence of bulk metabolites (Mueller et al. 2002; Birkemeyer
et al. 2003; Schmelz et al. 2003, 2004), (iv) increasing accuracy in multi-parallel
metabolite quantification (Birkemeyer et al. 2005), (v) combining profiling and
flux analyses (Roessner-Tunali et al. 2004), (vi) establishment of quantitative
repeatability, unambiguous nomenclature and comparability between analy-
ses performed in different laboratories or using different analytical technology
platforms (Schauer et al. 2005), and (vii) finally – perhaps the most important
challenge of all metabolomic investigations – the identification of the unidenti-
fied majority of metabolic components from metabolite profiling experiments
(Fiehn et al. 2000a; Schauer et al. 2005).
In agreement with the focus of this chapter the above challenges have pre-
dominantly analytic or technical motivation. The breakthrough of metabolo-
mic investigations, however, will depend on the access to hitherto unavail-
able fundamental insights into metabolic and systems interactions. Increas-
ingly integrative studies which consider the metabolome, proteome, tran-
scriptome, and genome evolution of an organism have been initiated and
are to be expected. Promising steps have been made – using GC-MS tech-
nology – towards network analysis (Fiehn 2003; Weckwerth et al. 2004a)
and correlation studies between or within metabolome and transcriptome
Gas Chromatography Mass Spectrometry 5
constituents (Urbanczyk-Wochniak et al. 2003; Steinhauser et al. 2004; Kopka

et al. 2005). A detailed discussion of these general aspects including GC-
MS studies and beyond can be found in the applications section of this
book.
2 GC-MS Profiling Technology in a Nutshell
Metabolite profiling with GC-MS involves six general steps:

1. Extraction of metabolites from the biological sample, which should be as
comprehensive as possible, and at the same time avoid degradation or mod-
ification of metabolites (e. g. Kopka et al. 2004).
2. Derivatisation of metabolites making them amenable to gas chromatogra-
phy. Metabolites which are not volatile per se require chemical modification
prior to GC analysis.
3. Separation by GC. High resolution GC can also be highly reproducible as
it involves automated sample injection robotics, highly standardised con-
ditions of gas-flow, temperature programming, and standardised capillary
column material.
4. Ionisation of compounds as they are eluted from the GC. Electron impact
(EI) ionisation is most widely used, as it is the technology which is least
susceptible to suppression effects and produces reproducible fragmentation
patterns.
5. Time resolved detection of molecular and fragment ions. Mass separation
and detection can be achieved with different mass-detection devices, in-
cluding sector field detectors, quadrupole detectors (QUAD), ion trap tech-
nology, and time-of-flight detectors (TOF). The choice of detectors depends
on the targeted analytical niche. GC-MS systems with QUAD detection are
most widely spread for routine analysis. Ion trap technology allows MS×MS
(two-dimensional MS) analysis for structural elucidation and targeted quan-
tification of trace compounds (e. g. Mueller et al. 2002). TOF detection can
either be tuned to fast scanning rates (van Deursen et al. 2000) or to high
mass precision comparable to sector field systems. Fast scanning GC-TOF-
MS enables the, today, most advanced technology in the GC-MS field, namely
two dimensional GC×GC-TOF-MS (two-dimensional GC-TOF-MS) (Ryan et
al. 2004; Sinha et al. 2004a–c).
6. Acquisition and evaluation of GC-MS data files. All GC-MS system manu-
facturers provide software which is tuned for targeted, quantitative metabo-
lite analysis. The targeted approach involves unequivocal identification of
predefined metabolites by expected chromatographic retention times and
mass-spectral fragmentation patterns and quantitative calibration by au-
thentic standard concentrations. Recent software developments support the
non-targeted analysis of GC-MS patterns, and the full evaluation of all re-
solved compounds. This feature of GC-MS allows discovery of novel hitherto
6 J. Kopka
unknown metabolites. As we are far from knowing all possible metabolites

of a given organism, non-biased, truly comprehensive data evaluation is the
most essential requirement of metabolite profiling.
2.1 Chemical Derivatisation and Chromatography
The principles of fast metabolic sample inactivation and nondestructive ex-

traction are common to all metabolome analyses. In contrast to all other
technologies GC-MS is inherently restricted to volatile and temperature-stable
compounds. The scope of GC-MS for metabolite analysis is limited by the
typical temperature range of commercial capillary columns, for example up
to 320–350 ◦ C. The lower temperature range is determined by ambient tem-
perature, but cold trapping devices and isothermal GC allow analysis of low
molecular weight gases and highly volatile metabolites. GC received a consid-
erable extension of applications through the development of a highly versatile
tool box of derivatisation reagents, which chemically transform non-volatile
metabolites into volatile analytes for GC-MS analysis (e. g. Knapp 1979; Blau
and Halket 1993; Toyo’oka 1999). To date, GC-MS profiling of metabolites in
plants has largely been confined to compounds, recovered in the methanol-
water phase after methanol-water/chloroform extraction of tissues (Fiehn et
al. 2000a; Roessner et al. 2000; Duran et al. 2003; Barsch et al. 2004; Gullberg
et al. 2004; Strelkov et al. 2004; Broeckling et al. 2005). Although not all hy-
drophilic compounds can be volatilised by derivatisation, the following classes
of compounds are detected routinely: amino-, organic-, and aromatic-acids,
amines, sugars up to trisaccharides, alcohols and polyols, and some mono-
phosphorylated metabolites.
The current limitations of metabolite preparation and derivatisation strat-
egy, namely methoxyamination with subsequent direct trimethylsilylation of
predominantly polar metabolites, call for extension. Application of other tech-
nology platforms is an obvious route and will be discussed in the following
chapters. Here a short appraisal of the potential of chemical derivatisation is
attempted. Four main types of reaction schemes will be discussed.
1. Alkoxyamination by reagents, such as methoxyamine CH3 −O−NH2 , sta-
bilises carbonyl moieties in native metabolite structures, but forms E-
and Z-isomers of the −N=C< double-bond substituents. Keto-enol tau-
tomerism is suppressed, as is the decarboxylation of unstable β-carbonyl-
carboxylic acids. In addition, the formation of acetal- or ketal-structures in
aqueous solution is inhibited. These equilibrium reactions generate mul-
tiple intramolecular and water adducts, for example the typical α- and
β-conformers of reducing sugars. Ether- and ester-conjugates are mostly
stable when exposed to methoxyamine reagent and maintain conformation.
So far other alkoxy-reagents – for example hydroxylamine, ethyloxyamine,
or benzyloxyamine – have not been exploited for systematic discovery of
metabolites with carbonyl moieties:
2. Silylation reagents classify into those which introduce either a trimethyl-

silyl (TMS) moiety, −Si(CH3 )3 , or a dimethyl-(tert-butyl)-silyl (TBS) moiety,
−Si(CH3 )2 −C(CH3 )3 . TMS reagents have been well investigated and are
known to have the widest derivatisation spectrum (Little 1999; Halket et al.
2005). TMS has the potential to substitute all exchangeable, “acidic” protons
of a metabolite. Steric hindrance of TMS substitution is rare but common
with the bulkier TBS reagent. The benefit of the TBS reagent is higher
tolerance for the presence of water and clear mass spectral fragmentation.
However, vicinal diols, which typically occur in sugars, are only partially
derivatised.
3. Alkylation reactions, mostly methylation, are widely used to derivatise car-
boxylic acids and alcohols. The enormous reactivity of available reagents –
some allow for flash derivatisation during hot GC injection – leads to
transalkylation of ester-bonds und consequently breaks down complex
metabolites, such as glycero- and phospholipids. Alkylation of sugars leads
to derivatives which are more volatile than the TMS derivatives and therefore
allow analysis of higher sugar oligomers.
4. Acylation reactions, mostly acetylation or trifluoro-acetylation, are less re-
active than transalkylation. Reagents usually form stable ester and amide
bonds and break down only activated metabolic intermediates, e. g. thio-
esters.
In conclusion further developments of alternate GC-MS profiling techniques
need to employ more selective combinations of metabolite fractionation and
derivatisation schemes. Solid phase extraction can be explored to partition
and concentrate metabolites amenable to alternate subsequent derivatisation.
On the other hand, vapour phase extraction (VPE) for the separation and
concentration of volatile derivatisation products prior to GC injection may
prove promising (Schmelz et al. 2003, 2004). VPE has the potential to be
a robust technique and was shown to operate with a range of commonly used
reagents.
2.2 Mass Detection and Quantitative Calibration Techniques
One of the major criticisms and pitfalls of metabolome analyses is best ex-
plained by so-called matrix effects. This well-known effect describes unex-
pected losses or increased recovery of metabolites in complex extracts com-
pared to pure authentic preparations. Matrix effects on one hand are caused
by the presence of compounds which either specifically inhibit extraction or
chemical analysis of metabolites. Positive matrix effects can stabilise other-
wise labile compounds in the presence of suitable chemicals. Typical exam-
ples are suppression effects of soft ionization techniques, for example electro-
spray ionization (ESI) or matrix assisted laser desorption ionization (MALDI).
Electron-impact ionization (EI) typically used in GC-MS profiling is not sus-
ceptible to suppression. Instead GC injection is the crucial step which may
8 J. Kopka
cause discriminations, especially in view of the complex and rather crude

extracts which are typically injected.
So far, only exemplary – albeit time demanding – thorough tests for un-
expected matrix effects have been performed with selections of chemically
Fig. 2. Mass spectra of deuterated and 13 C labeled MSTs help structural elucidation and recov-
ery analysis of metabolites. Labeled and non-labeled MSTs of Glycine N, N-di(trimethylsilyl)-,
trimethylsilyl ester are shown. Oryza sativa L. cv. Nipponbare was labelled in vivo using deuter-
ated water or 13 CO2 . MSTs representing the fully labeled mass isotopomers demonstrate presence
of two carbon atoms (left panel) and two non-exchangeable hydrogen atoms (right panel). Mass
fragments which exhibited a mass shift of 1 amu (red) or 2 amu (blue) are indicated
Fig. 3a–c. Mass spectral deconvolution of deuterated mass isotopomers. Succinic acid
di(trimethylsilyl) ester was partially labelled in vivo by exposing Oryza sativa L. cv. Nippon-
bare to deuterated water. Metabolite profiles were performed on a Pegasus II GC-TOF-MS system
(LECO, St. Joseph, MI, USA) with 20 scans s−1 . Mass spectra were deconvoluted using Chro-
maTOF software version 1.00, with baseline offset just above noise, smoothing and peak width
set to 10 and 2 scans, respectively: a selected ion traces of non-deuterated (D0 , m/z = 247) and
deuterated (D1−4 , m/z = 248 − 251) M+ − 15 mass fragments. Mass fragments at 252 and 253 amu
are carbon mass isotopomers of D4 ; b peak area compared to deconvoluted peak height. Peak
area integration does not allow differentiation of contributions by carbon mass isotopomers;
c deconvoluted mass spectra of D0−4 . Inset shows partial deconvolution of D0−4 carbon mass
isotopomers and missing carbon mass isotopomers of D1−3
diverse, representative metabolites (e. g. Roessner-Tunali et al. 2003; Gullberg

et al. 2004). Therefore technologies are required to improve quantitative stan-
dardisation for the comparison of increasingly diverse biological samples and
experimental conditions.
10 J. Kopka
For this purpose, full saturating 13 C in vivo labelling was developed using
yeast which is one of the most important organisms in systems biology (e. g.
Stephanopoulos et al. 2004). Metabolites of yeast were demonstrated to be fully
labelled when provided with an exclusive carbon source, such as U-13 C-glucose
(Mashego et al. 2004; Birkemeyer et al. 2005). Refer to Birkemeyer et al. (2005)
for detailed discussion of potential applications for 13 C-labelled metabolomes.
Similar approaches are possible in plants (Figs. 2 and 3).
In short, standardised in vivo labelled extracts of yeast or other microor-
ganisms can substitute the rather small number of chemically synthesised
mass isotopomers used in earlier studies (Fiehn et al. 2000a; Gullberg et al.
2004). Typically a standardised labelled reference sample is combined in equal
amounts with non-labelled experimentally challenged samples. The advan-
tages of this approach are (i) the presence of a mass isotopomer for all iden-
tified but also all hitherto non-identified metabolites, (ii) the concentration
of each mass isotopomer is inherently adjusted to the endogenous metabolite
concentration, (iii) metabolic components can easily be distinguished from
laboratory contaminations, and (iv) recovery of all metabolic components can
be determined with the appropriate mass isotopomer.
Thus metabolite profiling will achieve the same level of transcriptome and
proteome experiments, which utilize differential fluorescent probes or differ-
ential isotope coded tagging, respectively. In conclusion, comprehensive in
vivo isotope labelling will help to establish quantitative between laboratory
comparability of GC-MS based metabolome experiments. More importantly,
we expect metabolome experiments with full mass isotopomer standardisa-
tion to be also independent of the mass spectrometric platform, e. g. CE-MS,
LC-MS, or possibly even MALDI-TOF-MS.
3 Short Excursion into Nomenclature and Definitions
Concise and unambiguous description of GC-MS metabolite profiling results

requires clear definitions. The definitions suggested within this section are
biased towards the specifics of GC-MS technology but may also be applied to
other technology platforms. This section is intended as a contribution to the
ongoing process of unifying data formats and concepts within the field of plant
metabolomics (e. g. Fiehn 2002; Bino et al. 2004; Jenkins et al. 2004).
3.1 Metabolite and Analyte
Routine GC-MS profiling analysis (Fiehn et al. 2000b; Roessner et al. 2000)
has an upper size exclusion limit which is roughly equivalent to a persily-
lated trisaccharide derivative (MW:1296), hexatriacontane (MW:506), or hen-
triacontanoic acid trimethylsilylester (MW:523). Even though it may appear
tempting, metabolite and analyte are best not defined by molecular weight.
A metabolite may be described as a compound which is internalised, chem-

ically converted or secreted by an organism, but is not synthesised by DNA
replication, transcription, or translation. Post-processing events of DNA, RNA
and proteins, such as DNA methylation, RNA splicing, sequence specific
protease cleavage or post-translational modification are not attributed to the
metabolome. The origin of a metabolite is not exclusively dependent on the
biosynthetic capacity of an organism or delimited by the genomic inventory.
Metabolites may readily be exchanged between organisms, for example in
plant microbe interactions, and – like drugs or pesticides – can today be of
anthropogenic/xenobiotic origin.
In contrast to LC- or CE-MS, GC-MS analysis requires clear distinction
between metabolite and analyte, because – depending on choice of chemical
derivatisation – metabolites may be chemically transformed before quantifi-
cation. The term analyte may be used to address the chemical structure and
compound which is submitted to GC-MS and finally detected and quantified.
An analyte can be identical with the metabolite, if the metabolite is not chem-
ically derivatised. Single metabolites may have more than one analyte, if the
chosen derivatisation reaction generates more than one derivative, for example
methoxyamination (see above). In these cases preferred and alternate analytes
exist for quantification. Analytes of one metabolite may differ in abundance,
i. e. a major and one, even multiple, minor analytes may exist. Standardisation
by stable mass isotopomers corrects the quantification errors which may arise
from unforeseen matrix effects on analyte ratios during chemical derivatisa-
tion of GC injection.
Different metabolites may be chemically transformed into the same ana-
lyte structure. In addition a single analyte may arise from inadequate chro-
matographic separation of isomers. For example, the biochemically distinct
stereoisomeric structures of dl-amino acids are only separated by specialised
chiroselective chromatography. These analytes have composite properties in
contrast to absolutely specific analytes.
These concepts are not unique to GC-MS technology. Analyte sensitivity,
accuracy, and potentially composite analyte properties need to be thoroughly
considered in MS-MS applications, non-chiroselective capillary electrophore-
sis or liquid chromatography, and in cases of adduct-formation or multiply
charged ions.
3.2 Mass Spectral Tag (MST) and Mass Fragment
GC-MS metabolite profiles resolve hundreds of analytes, which represent

metabolites, but also internal standard substances and laboratory contam-
inations. Typical GC-MS profiles may contain approximately 100 identified
analytes of metabolites. The chemical structure of the majority of GC-MS an-
alytes, however, is still unknown. Each new biological object or experimental
condition still gives rise to new, hitherto unidentified, chemical components.
12 J. Kopka
Because in non-biased analysis of GC-MS profiles identified and unidenti-

fied components are equally important, we created the term mass spectral tag
(MST), i. e. a mass spectrum which is characterised by a specific chromato-
graphic retention and by repeated occurrence in a single or multiple types of
biological samples (Colebatch et al. 2004; Desbrosses et al. 2005). MSTs repre-
sent analytes. MSTs can be identified, in other words, unequivocally linked to
a chemical structure. The use of MSTs allows uncoupling of metabolite profil-
ing experiments from the time consuming process of chemical identification.
MSTs can be used to track analytes in different experiments or laboratories
(Schauer et al. 2005). Thus MST identification can be performed even years
after the first discovery.
MSTs of GC-EI-MS profiles are composed of multiple characteristic mass
fragments in constant relative abundances. In most cases residual, non-frag-
mented molecular ions are rare or even absent. In consequence GC-MS allows
selection of multiple mass fragments which all represent the same MST and ex-
hibit the same quantitative changes. Typically one quantifying mass fragment
(QM) and a set of specific, supporting qualifying mass fragments are selected
in GC-MS analysis (Halket et al. 2005). The criteria for the proper choice of QMs
are equal to the choice of a preferred analyte. QMs need to be selective, i. e. not
composite, in the context of the complexity of co-eluting MSTs. Therefore, the
best QM is the most abundant among the available selective mass fragments.
3.3 Response and Relative Quantification of Metabolite Pools
GC-MS metabolite profiling studies monitor relative changes in metabolite

pool sizes and but also allow insight into flux, i. e. the dynamic turnover of
metabolite pools or metabolite substructures (e. g. Fischer and Sauer 2003;
Sauer 2004; Roessner-Tunali et al. 2004). Flux experiments are easily distin-
guished from above mentioned saturating in vivo labelling experiments. Flux
experiments monitor the initial kinetics of labelling and thus stable isotopes
are only partially incorporated into metabolite pools. In contrast, saturating
in vivo labelling reaches the endpoint of a completely stable isotope labelled
metabolome.
MSTs are quantified by ion currents of QMs which are recorded after an-
alyte ionization, fragmentation and mass separation. Ion currents in GC-MS
are monitored either by peak area or peak height. Both measurements need
to be baseline corrected for electronic and chemical noise. The resulting cor-
rected values are defined to be what we call responses, i. e. XQM of fragment
QM (Colebatch et al. 2004; Desbrosses et al. 2005). The fragment response is
routinely normalised to the amount of the sample, for example fresh or dry
weight. In addition each response is corrected for recovery effects, which may
occur at any step of the analytical process between metabolic inactivation of
the sample and final recording of ion currents. Different levels of recovery
correction exist: (i) correction by extract and sample volume, (ii) correction
by addition of a constant amount of a representative internal standard com-

pound (IS), and (iii) normalization by chemically identical, but stable-isotope
labelled mass isotopomers of each metabolite. The normalised response (NQM )
−1
is, consequently, NQM = XQM × X−1 IS × sample weight , where XIS ideally rep-
resents a mass isotopomer response of QM. In a further step, the normalised
response of a fragment, NQM , is divided by the average relative response of
QM as determined in a set of reference samples, avgNQM(ref) . The resulting
quotient, Ri = NQM × avgN−1 QM(ref) , is called response ratio Ri . Ri describes
the x-fold changes in metabolite pools sizes relative to the reference samples.
Typical reference samples are taken at the start of a time series experiment or
are mock-treated biological controls.
In GC-MS profiling analyses the standard deviation of normalised responses
is dependent on the chemical nature of metabolite and analyte. Average relative
standard deviations (RSD) of 10% (Weckwerth et al. 2004b) or 13.8% (5.5–
33.4%; Gullberg et al. 2004) were reported for replicate GC-MS analyses. These
analyses included extraction as well as derivatisation and were performed using
representative analytes. Use of isotope labelled standardisation was reported to
reduce RSD further to approximately 6.9–9.7% residual experimental variance
(Gullberg et al. 2004).
4 Present Challenges of GC-MS Profiling
4.1 Standardisation of GC-MS Systems
GC-MS profiles, with the exception of GC×GC-TOF-MS data, are in essence

three-dimensional and comprise a chromatographic time-resolved axis, a sec-
ond coordinate axis which represents the mass to charge ratio (m/z, z = 1
in GC-MS with only rare exceptions), and an intensity axis which monitors
the ion current (IC) and thus the abundance of molecules or mass frag-
ments. A substantial breakthrough for GC-MS analyses was the early establish-
ment of generally accepted calibration substances and procedures, so-called
tuning routines, which allowed comparison of mass spectra from GC-MS
systems of virtually all manufacturers and from different hyphenated mass
detection technologies. In addition the widely used electron-impact ionisa-
tion technique (EI) ensured stable fragmentation ratios, which are in first
approximation independent of analyte concentration. However, comparability
was only achieved by restriction to 1 amu precision.
The chromatography axis is less standardised, not least because of multi-
ple types of available capillary GC-columns which have different chromato-
graphic properties and thus serve different separation problems. In addition
slight changes in temperature program, pressure and flow settings of both
carrier gas and injection technique, as well as slight production differences
of column manufacturers cause minor but perceptible changes in retention
14 J. Kopka
time. Retention time indices (RI), based on homologous series of internal ref-
erence substances, such as n-alkanes, have been introduced early to aid GC
analyses (Kovàts 1958). Use of an n-alkane RI system in GC-MS metabolite pro-
filing substantially improves the reproducibility of the chromatography axis.
The currently achievable accuracy of RI prediction was recently investigated
in three different profiling laboratories which use the same type of capillary
column but different GC-MS systems (Schauer et al. 2005). In this investiga-
tion the possibility of predicting RIs of more than 100 identified analytes was
tested. Mathematical regression resulted in an average accuracy of ±5.4 RI
units.
The IC intensity axis in GC-MS is standardised for high vs low mass discrim-
ination. The GC-MS tuning includes processes which ensure constant ratios
of high vs low mass intensities. However, mass spectra which are recorded by
either QUAD-MS or fast scanning GC-TOF-MS detection may differ in this
respect. Fast scanning GC-TOF-MS systems (e. g. Pegasus II MS system, LECO,
St. Joseph, MI, USA) have increased sensitivity of small mass fragments and
reduced sensitivity in the high mass range.
4.2 Deconvolution and Alignment of Mass Spectral Tags
The principal challenge in GC-MS profiling analysis is the automated unrav-

elling of the multiple partially co-eluting MSTs which comprises a GC-MS
chromatogram. One of the fundamental advances in GC-MS technology has
been the development of algorithms and software for the so-called deconvo-
lution of mass spectra from GC-MS chromatograms (Halket et al. 1999; Stein
1999; Shao et al. 2004; AnalyzerPro, http://www.spectralworks.com), specific
software for the deconvolution of fast scanning GC-TOF-MS data files, e. g.
ChromaTOF software used by Vreuls et al. (1999), Veriotti and Sacks (2001) or
Jonsson et al. (2005), and ongoing developments for automated processing of
GC×GC-TOF-MS chromatograms (Ryan et al. 2004; Sinha et al. 2004a,b). The
inherent steps of deconvolution are (i) mass resolved baseline subtraction of
electronic and chemical noise, (ii) assignment of retention time and/or reten-
tion time indices (RI) to chromatographic peak apices and respective MSTs,
and (iii) accurate separation of MSTs from closely co-eluting analytes, the most
challenging and advanced but still error-prone part (Fig. 3).
Even though automated mass spectral deconvolution has fundamentally
facilitated GC-MS analyses of complex mixtures, accuracy and limitations of
respective software have so far not been systematically compared and assessed.
Typical errors of mass spectral deconvolution are (i) accidental generation of
MSTs due to noise fluctuations, (ii) deconvolution of multiple MSTs from
a single component, (iii) incomplete MSTs which lack one or multiple mass
fragments (Fig. 3c), and (iv) chimeric MSTs, i. e. composite mass spectra of
co-eluting compounds. The co-elution problem of complex mixtures has been
fundamentally improved by introduction of fast scanning GC-TOF-MS and is
today technically best solved by GC×GC-TOF-MS, using a set of two capillary

GC columns with alternate phase-polarity (Sinha et al. 2004c).
Reliable alignment of identical MSTs in sets of consecutive GC-MS chro-
matograms is required for rapid, repeatable and automated comparative high-
throughput analysis of large samples sets. So far, software solutions and
novel algorithm developments for the alignment of complex mixtures depend
on close to constant chromatographic retention within series of consecutive
GC-MS chromatograms (Duran et al. 2003; Jonsson et al. 2004; metAlign,
http://www.metalign.nl). Indeed, consistent run-to-run retention times are
considered to be crucial to the application of chemometrics on complex mix-
tures, especially in the field of two-dimensional separations (Sinha et al. 2004c).
In conclusion, automated mass spectral deconvolution of GC-MS profiles
appears to be in principal solved by both GC×GC technology and deconvolution
algorithm, but the optimum solution still has to be found (Halket et al. 2005).
In contrast prediction of chromatographic shifts in complex mixtures with
highly dynamic range of concentrations has not been satisfyingly solved. As
there is currently no solution – other than recalibration with pure standard
substances – addressing the problem of RI shifts will be crucial for future
GC-MS based metabolite profiling and identification of MSTs.
4.3 Identification of Mass Spectral Tags
Identification of MSTs requires chromatographic separation as well as mass

spectrometric information (Wagner et al. 2003), mainly because plants like
microorganisms contain a multitude of isomeric metabolites (e. g. Barsch et al.
2004; Stephanopoulos et al. 2004; Strelkov et al. 2004). These isomers give rise
to MSTs, which can be chromatographically resolved but have almost identi-
cal mass spectra. Today, GC-MS appears to have found a generally accepted
standard for mass spectral comparison. The NIST mass spectral search and
comparison software (Ausloos et al. 1999; Stein 1999) has been integrated into
the customised operating software of most GC-MS manufacturers. The GC-MS
technology is in this respect more advanced than LC-MS (Halket et al. 2005).
However, mass spectral search and comparison software, which harbours in-
formation on chromatographic retention in what we suggested to call MSRI
libraries (Wagner et al. 2003; Kopka et al. 2005) and the automated utilization
of this information for probability based matching, would be highly desirable.
The new version NIST05 (National Institute of Standards and Technology,
Gaithersburg, MD, USA) of a mass spectral search and comparison software
now makes RI information available but currently does not utilize RI for auto-
mated matching. The result of a hitherto manual inventory of Oryza sativa L.
cv. Nipponbare leave profiles is shown in Fig. 4.
Two different approaches exist for the identification of unknown MSTs from
GC-MS profiles: (i) the “bottom up” approach in which metabolites of interest
to a particular researcher are analysed by the purchase of authentic standard
16 J. Kopka
Fig. 4. Synthetic and representative GC-MS profiles of Oryza sativa L. cv. Nipponbare leaves:
A – 132 identified MSTs representing 109 known metabolites; B – 12 added internal stan-
dard substances; C – 148 unidentified MSTs which match previous MSRI library entries;
D – all previously observed MSTs present in the MSRI library at GMD (http://csbdb.mpimp-
golm.mpg.de/gmd.html)
substances, which are subsequently mapped by standard addition experiments

onto established standardised GC-MS systems, and (ii) the “top down” ap-
proach whereby structural elucidation is performed on a hitherto unknown,
but important target MST. The work of “top down” structural identification
is highly time-demanding and involves preparative enrichment, purification,

spectroscopic, mass spectral and NMR analyses of the preparation and finally
chemical synthesis of the predicted structure. As a consequence the “bottom
up” approach prevails in most laboratories and “top down” identification is
currently restricted to potentially novel signalling compounds or marker sub-
stances of specific biological samples and experimental conditions.
In order to avoid unnecessary “top down” investigations reliable identi-
fication by prior standard addition experiments is essential. MSRI library
collections of mass spectra (Kopka et al. 2005), which comprise frequently
observed identified and non-identified MSTs, appear to represent the most
effective means to pool the identification efforts currently performed in many
laboratories around the world (Schauer et al. 2005). Identified and yet uniden-
tified, MSTs can efficiently be shared by public resources such as GMD@CSBDB
(http://csbdb.mpimp-golm.mpg.de/gmd.html). In addition mass spectral iden-
tifications and chromatographic sequence of analyte elution can be transferred
between laboratories. “Bottom up” identifications performed in parallel may
be used for inter-laboratory confirmation of identifications and reduce the risk
of unnecessary structural elucidation projects.
Currently the MSRI libraries available from GMD@CSBDB include in total
more than 2000 evaluated mass spectral data sets obtained using GC-QUAD-
and GC-TOF-MS technology platforms with 1089 non-redundant and 360 iden-
tified MSTs. Future efforts at GMD aim to refine mass spectral quality and
annotation, and will add stable isotope labelled variants of MSTs (e. g. Fig. 2)
for improved mass spectral interpretation of unidentified MSTs. The num-
ber of identified analytes and metabolites will continuously be extended and
annotations updated.
Acknowledgements. I would like to thank A.R. Fernie, A. Erban, Max Planck Institute of Molecular
Plant Physiology, Potsdam-Golm, Germany for critically reading and discussing my manuscript.
My thanks extend to Prof. Dr. Le Tran Binh, Institute of Biotechnology (IBT), Hanoi, Vietnam,
for sharing his expertise in rice cultivation. This work was supported by the Max-Planck society,
and the Bundesministerium für Bildung und Forschung (BMBF), grant PTJ-BIO/0312854.
References
Ausloos P, Clifton CL, Lias SG, Mikaya AI, Stein SE, Tchekhovskoi DV, Sparkman OD, Zaikin V,
Zhu D (1999) The critical evaluation of a comprehensive mass spectral library. J Am Soc Mass
Spectrom 10:287–299
Barsch A, Patschkowski T, Niehaus K (2004) Comprehensive metabolite profiling of Sinorhizobium
meliloti using gas chromatography-mass spectrometry. Funct Integrat Genomics 4:219–230
Bino RJ, Hall RD, Fiehn O, Kopka J, Saito K, Draper J, Nikolau BJ, Mendes P, Roessner-
Tunali U, Beale MH, Trethewey RN, Lange BM, Wurtele ES, Sumner LW (2004) Potential
of metabolomics as a functional genomics tool. Trends Plant Sci 9:418–425
Birkemeyer C, Kolasa A, Kopka J (2003) Comprehensive chemical derivatization for gas
chromatography-mass spectrometry-based multi-targeted profiling of the major phytohor-
mones. J Chromatogr A 993:89–102
18 J. Kopka
Birkemeyer C, Luedemann A, Wagner C, Erban A, Kopka J (2005) Metabolome analysis: the

potential of in vivo labeling with stable isotopes for metabolite profiling. Trends Biotechnol
23:28–33
Blau K, Halket JM (1993) Handbook of derivatives for chromatography, 2nd edn. Wiley, New York
Broeckling CD, Huhman DV, Farag MA, Smith JT,May GD, Mendes P, Dixon RA, Sumner LW
(2005) Metabolic profiling of Medicago truncatula cell cultures reveals the effects of biotic
and abiotic elicitors on metabolism. J Exp Bot 56:323–336
Colebatch G, Desbrosses G, Ott T, Krusell L, Kloska S, Kopka J, Udvardi MK (2004) Global changes
in transcription orchestrate metabolic differentiation during symbiotic nitrogen fixation in
Lotus japonicus. Plant J 39:487–512
Cook D, Fowler S, Fiehn O, Thomashow MF (2004) A prominent role for the CBF cold response
pathway in configuring the low-temperature metabolome of Arabidopsis. Proc Natl Acad
Sci USA 101:15243–15248
Desbrosses G, Kopka J, Udvardi MK (2005) Legume metabolomics: development of GC-MS
resources for functional genomics of plant-microbe interactions. Plant Physiol 137:1302–
1318
Duran AL, Yang J, Wang L, Sumner LW (2003) Metabolomics spectral formatting, alignment and
conversion tools (MSFACTs). Bioinformatics 19:2283–2293
Fernie AR, Trethewey RN, Krotzky AJ, Willmitzer L (2004) Metabolite profiling: from diagnostics
to systems biology. Nat Rev Mol Cell Biol 5:763–769
Fiehn O (2002) Metabolomics – the link between genotypes and phenotypes. Plant Mol Biol
48:155–171
Fiehn O (2003) Metabolic networks of Cucurbita maxima phloem. Phytochem 62:875–86
Fiehn O, Kopka J, Trethewey RN, Willmitzer L (2000a) Identification of uncommon plant
metabolites based on calculation of elemental compositions using gas chromatography and
quadrupole mass spectrometry. Anal Chem 72:3573–3580
Fiehn O, Kopka J, Dörmann P, Altmann T, Trethewey RN, Willmitzer L (2000b) Metabolite profiling
for plant functional genomics. Nat Biotechnol 18:1157–1161
Fischer E, Sauer U (2003) Metabolic flux profiling of Escherichia coli mutants in central carbon
metabolism using GC-MS. Eur J Biochem 270:880–891
Gullberg J, Jonsson P, Nordström A, Sjöström M, Moritz T (2004) Design of experiments: an
efficient strategy to identify factors influencing extraction and derivatization of Arabidopsis
thaliana samples in metabolomic studies with gas chromatography/mass spectrometry. Anal
Biochem 331:283–295
Halket JM, Przyborowska A, Stein SE, Mallard WG, Down S, Chalmers RA (1999) Deconvolution
gas chromatography mass spectrometry of urinary organic acids – potential for pattern recog-
nition and automated identification of metabolic disorders. Rapid Commun Mass Spectrom
13:279–284
Halket JM, Waterman D, Przyborowska AM, Patel RKP, Fraser PD, Bramley PM (2005) Chemical
derivatization and mass spectral libraries in metabolic profiling by GC/MS and LC/MS/MS.
J Exp Bot 56:219–243
Jenkins H, Hardy N, Beckmann M, Draper J, Smith AR, Taylor J, Fiehn O, Goodacre R, Bino RJ,
Hall RD, Kopka J, Lane GA, Lange BM, Liu JR, Mendes P, Nikolau BJ, Oliver SG, Paton NW,
Rhee S, Roessner-Tunali U, Saito K, Smedsgaard J, Sumner LW, Wang T, Walsh S, Wurtele ES,
Kell DB (2004) A proposed framework for the description of plant metabolomics experiments
and their results. Nature Biotechnol 22:1601–1606
Jonsson P, Gullberg J, Nordström A, Kusano M, Kowalczyk M, Sjöström M, Moritz T (2004)
A strategy for identifying differences in large series of metabolomic samples analyzed by
GC/MS. Anal Chem 76:1738–1745
Jonsson P, Johansson AI, Gullberg J, Trygg J, A J, Grung B, Marklund S, Sjostrom M, Antti H,
Moritz T (2005) High-throughput data analysis for detecting and identifying differences
between samples in GC/MS-based metabolomic analyses. Anal Chem 77: 5635–5642
Kaplan F, Kopka J, Haskell DW, Zhao W, Schiller KC, Gatzke N, Sung DY, Guy CL (2004) Exploring
the temperature-stress metabolome of Arabidopsis. Plant Physiol 136:4159–4168
Knapp DR (1979) Handbook of analytical derivatization reactions. Wiley, New York

Kopka J, Fernie AF, Weckwerth W, Gibon Y, Stitt M (2004) Metabolite profiling in Plant Biology:
Platforms and Destinations. Genome Biol 5(6):109–117
Kopka J, Schauer N, Krueger S, Birkemeyer C, Usadel B, Bergmüller E, Dörmann P, Gibon Y,
Stitt M, Willmitzer L, Fernie AR, Steinhauser D (2005) GMD@CSBDB: The Golm Metabolome
Database. Bioinformatics 21:1635–1638
Kovàts ES (1958) Gas-chromatographische Charakterisierung organischer Verbindungen: Teil 1.
Retentionsindices aliphatischer Halogenide, Alkohole, Aldehyde und Ketone. Helv Chim Acta
41:1915–1932
Little JL (1999) Artifacts in trimethylsilyl derivatization reactions and ways to avoid them. J Chro-
matorgr A 844:1–22
Mashego MR, Wu L, van Dam JC, Ras C, Vinke JL, van Winden WA, van Gulik WM, Heijnen JJ
(2004) MIRACLE: mass isotopomer ratio analysis of U-13 C-labeled extracts. A new method
for accurate quantification of changes in concentrations of intracellular metabolites. Biotech
Bioeng 85:620–628
Mueller A, Duechting P, Weiler EW (2002) A multiplex GC-MS/MS technique for the sensitive
and quantitative single-run analysis of acidic phytohormones and related compounds, and
its application to Arabidopsis thaliana. Planta 216:44–56
Roessner U, Wagner C, Kopka J, Trethewey RN, Willmitzer L (2000) Simultaneous analysis of
metabolites in potato tuber by gas chromatography-mass spectrometry. Plant J 23:131–142
Roessner U, Luedemann A, Brust D, Fiehn O, Linke T, Willmitzer L, Fernie AR (2001a) Metabolic
profiling allows comprehensive phenotyping of genetically or environmentally modified plant
systems. Plant Cell 13:11–29
Roessner U, Willmitzer L, Fernie AR (2001b) High-resolution metabolic phenotyping of geneti-
cally and environmentally diverse plant systems – identification of phenocopies. Plant Physiol
127:749–764
Roessner U, Willmitzer L, Fernie AR (2002) Metabolic profiling and biochemical phenotyping of
plant systems. Plant Cell Rep 21:189–196
Roessner-Tunali U, Hegemann B, Lytovchenko A, Carrari F, Bruedigam C, Granot D, Fernie AR
(2003) Metabolic profiling of transgenic tomato plants overexpressing hexokinase reveals
that the influence of hexose phosphorylation diminishes during fruit development. Plant
Physiol 133:84–99
Roessner-Tunali U, Lui J, Leisse A, Balbo I, Perez-Melis A, Willmitzer L, Fernie AR (2004) Flux
analysis of organic and amino acid metabolism in potato tubers by gas chromatography-mass
spectrometry following incubation in 13 C labelled isotopes. Plant J 39:668–679
Ryan D, Shellie R, Tranchida P, Casilli A, Mondello L, Marriott P (2004) Analysis of roasted coffee
bean volatiles by using comprehensive two-dimensional gas chromatography-time-of-flight
mass spectrometry. J Chromatogr A 1054:57–65
Sauer U (2004) High-throughput phenomics: experimental methods for mapping fluxomes. Curr
Opin Biotechnol 15:58–63
Sauter H, Lauer M, Fritsch H (1988) Metabolite profiling of plants – a new diagnostic technique.
Abstr Pap Am Chem Soc 195:129
Schauer N, Steinhauser D, Strelkov S, Schomburg D, Allison G, Moritz T, Lundgren K, Roessner-
Tunali U, Forbes MG, Willmitzer L, Fernie AR, Kopka J (2005) GC-MS libraries for the
rapid identification of metabolites in complex biological samples. FEBS Lett 579:1332–
1337
Schmelz EA, Engelberth J, Alborn HT, O’Donnell P, Sammons M, Toshima H, Tumlinson JH (2003)
Simultaneous analysis of phytohormones, phytotoxins, and volatile organic compounds in
plants. Proc Natl Acad Sci USA 100:10552–10557
Schmelz EA, Engelberth J, Tumlinson JH, Block A, Alborn HT (2004) The use of vapor phase
extraction in metabolic profiling of phytohormones and other metabolites. Plant J 39:790–808
Shao XG, Wang GQ, Wang SF, Su QD (2004) Extraction of mass spectra and chromatographic
profiles from overlapping GC/MS signal with background. Anal Chem 76:5143–5148
20 J. Kopka
Sinha AE, Fraga CG, Prazen BJ, Synovec RE (2004a) Trilinear chemometric analysis of twodimen-
sional comprehensive gas chromatography-time-of-flight mass spectrometry data. J Chro-
matogr A 1027:269–277
Sinha AE, Hope JL, Prazen BJ, Nilsson EJ, Jack RM, Synovec RE (2004b) Algorithm for locating
analytes of interest based on mass spectral similarity in GC×GC-TOF-MS data: analysis of
metabolites in human infant urine. J Chromatogr. A 1058:209–215
Sinha AE, Prazen BJ, Synovec RE (2004c) Trends in chemometric analysis of comprehensive
two-dimensional separations. Anal Bioanal Chem 378:1948–1951
Stein SE (1999) An integrated method for spectrum extraction and compound identification from
gas chromatography/mass spectrometry data. J Am Soc Mass Spectrom 10:770–781
Steinhauser D, Usadel B, Luedemann A, Thimm O, Kopka J (2004) CSB.DB: a comprehensive
systems-biology database. Bioinformatics 20:3647–3651
Stephanopoulos G, Alper H, Moxley J (2004) Exploiting biological complexity for strain improve-
ment through systems biology. Nat Biotechnol 22:1261–1267
Strelkov S, von Elstermann M, Schomburg D (2004) Comprehensive analysis of metabolites
in Corynebacterium glutamicum by gas chromatography/mass spectrometry. Biol Chem
385:853–861
Sumner LW, Mendes P, Dixon RA (2003) Plant metabolomics: large-scale phytochemistry in the
functional genomics era. Phytochemistry 62:817–836
Toyo’oka T (1999) Modern derivatization methods for separation science. Wiley, New York
Trethewey RN (2004) Metabolite profiling as an aid to metabolic engineering in plants. Curr Opin
Plant Biol 7:196–201
Trethewey RN, Krotzky AJ, Willmitzer L (1999) Metabolic profiling: a Rosetta stone for genomics?
Curr Opin Plant Biol 2:83–85
Urbanczyk-Wochniak E, Fernie AR (2005) Metabolic profiling reveals altered nitrogen nutrient
regimes have diverse effects on the metabolism of hydroponically-grown tomato (Solanum
lycopersicum) plants. J Exp Bot 56:309–321
Urbanczyk-Wochniak E, Luedemann A, Kopka J, Selbig J, Roessner-Tunali U, Willmitzer L,
Fernie AR (2003) Parallel analysis of transcript and metabolic profiles: a new approach in
systems biology. EMBO Reports 4:989–993
Van Deursen MM, Beens J, Janssen HG, Leclercq PA, Cramers CA (2000) Evaluation of time-of-
flight mass spectrometric detection for fast gas chromatography. J Chromatogr A 878:205–213
Veriotti T, Sacks R (2001) High-speed GC and GC/time-of-flight MS of lemon and lime oil samples.
Anal Chem 73:4395–4402
Vreuls RJJ, Dallüge J, Brinkman UAT (1999) Gas chromatography – time-of-flight mass spec-
trometry for sensitive determination of organic microcontaminants. J Microcolumn Sep
11:663–675
Wagner C, Sefkow M, Kopka J (2003) Construction and application of a mass spectral and retention
time index database generated from plant GC/EI-TOF-MS metabolite profiles. Phytochem
62:887–900
Weckwerth W, Loureiro ME, Wenzel K, Fiehn O (2004a) Differential metabolic networks unravel
the effects of silent plant phenotypes. Proc Natl Acad Sci USA 18:7809–7814
Weckwerth W, Wenzel K, Fiehn O (2004b) Process for the integrated extraction, identification and
quantification of metabolites, proteins and RNA to reveal their co-regulation in biochemical
networks. Proteomics 4:78–83
I.2 Current Status and Forward Looking Thoughts
on LC/MS Metabolomics
L.W. Sumner1
1 Introduction
The metabolome can be viewed as the consequential end products of gene

expression and the goal of metabolomics includes the comprehensive evalu-
ation of the metabolome (Trethewey et al. 1999; Fiehn et al. 2000; Trethewey
2001; Oliver et al. 2002; Sumner et al. 2003). Quantitative and qualitative mea-
surements of large numbers of cellular metabolites provide a high-resolution
biochemical phenotype of an organism which can be used to monitor and
assess gene function (Fiehn et al. 2000) or a system’s response (Weckwerth
2003). Although mRNA/transcripts represent a mechanism for information
transmission from the genome to the cellular machinery for protein synthesis,
mRNA levels do not always correlate well with protein levels (Gygi et al. 1999).
Furthermore, once translated a protein may or may not be enzymatically
active as post translational modifications, protein sorting, protein–protein
interactions, and controlled proteolysis all contribute to the regulation of ac-
tive enzyme levels. Due to these factors, changes in the transcriptome or the
proteome may not always lead to alterations in the metabolic phenotype. In
addition, the majority of transcript and protein annotations are currently
inferred based on sequence or structural similarity. It is estimated that less
than 10% of annotated genes have experimental evidence supporting assigned
function and thus, the accuracy of these annotations are of some uncertainty
(Somerville and Somerville 1999; Somerville and Dangl 2000). In the absence
of functionally annotated database information, transcript or protein profiling
often yields limited information. For example, transcriptomics or proteomics
often reveal the differential accumulation of a hypothetical or unannotated
protein; however, without annotation it is very difficult to infer biological con-
text. Microarray or proteomics experiments may also yield putative or generic
protein identifications such as a putative peroxidase or peroxidase-like pro-
tein. These generic annotations have limited information as many of these
enzymes are promiscuous and/or involved in a large number of different re-
actions. However, metabolomics has the ability to reveal that the accumulated
peroxidase/enzyme is more specifically related to lignification or to another
specific biochemistry. Thus, profiling the metabolome may actually provide
the most direct and “functional” information of the “omics” technologies.
1 TheSamuel Roberts Noble Foundation, 2510 Sam Noble Parkway, Ardmore, OK 73401, USA,

22 L.W. Sumner
The plant metabolome is quite complex with current estimates on the order
of 15,000 metabolites within a given species and over 200,000 different metabo-
lites within the plant kingdom (Dixon 2001; Hartman et al. 2005). Due to the
chemical complexity of the plant metabolome, it is generally accepted that
a single analytical technique will not provide comprehensive visualization of
the metabolome, and therefore, multiple technologies are generally employed.
The selection of the most suitable technology is generally a compromise be-
tween speed, chemical selectivity and instrumental sensitivity. Tools such as
nuclear magnetic resonance spectroscopy (NMR) are rapid, highly selective,
and non-destructive, but have relatively lower sensitivity. Other methods such
as capillary electrophoresis (CE) coupled to laser induced fluorescence (LIF)
detection are highly sensitive, but have limited chemical selectivity. Chromato-
graphically coupled mass spectrometry methods such as gas chromatography
(GC)/mass spectrometry (MS) and liquid chromatography (LC)/MS offer the
best combination of sensitivity and selectivity, and therefore are central to
most metabolomics approaches. Mass selective detection provides highly spe-
cific chemical information including molecular mass and/or characteristic
fragment ion(s) information that are directly related to chemical structure.
This information can be utilized for compound identification through spectral
matching with data compiled in libraries for authentic compounds or used for
de novo structural elucidation. Further, chemically selective MS information
can be obtained from extremely small metabolite quantities with limits of de-
tection in the pmole and fmole level for many primary and secondary plant
metabolites.
GC/MS has proven capability for profiling large numbers of metabolites
with reports covering several hundred to slightly more than a thousand var-
ious components (Fiehn et al. 2000; Roessner et al. 2000, 2001; Birkemeyer
et al. 2003; Wagner et al. 2003; Broeckling et al. 2005; Schauer et al. 2005;
Welthagen et al. 2005). The term component is used because a large num-
ber of metabolites often yield more than one derivatized component which
are observed in the GC/MS analysis. The achievable range and number of
metabolites profiled by GC/MS can be attributed to the high separation effi-
ciencies of long (30−60 m) capillary GC columns (i. e. N ≥ 250,000 for 60 m).
These high efficiencies enable the separation of very complex mixtures, and
with mass selective detection, qualitative identification of a significant pro-
portion of these compounds is achievable. This makes GC/MS a very effi-
cient and cost effective metabolomics tool. A major prerequisite for GC/MS
is sample volatility which is necessary to enable separation in the gas phase.
Analytes may be innately volatile or chemically derivatized to yield volatile
compounds. Unfortunately, there exist a large number of metabolites which
are not amenable to GC/MS even following derivatization. These include com-
pounds such as phenylpropanoid and other natural product glycosides whose
labile glycosidic bonds degrade during heating and vaporization. Thus, alter-
native techniques are necessary and especially so for the study of secondary
metabolism.
Current Status and Forward Looking Thoughts on LC/MS Metabolomics 23
Liquid introduction techniques for mass spectrometry such as electrospray

ionization, atmospheric chemical ionization, and photo ionization remove the
necessity for chemical derivatization. Thus, aqueous samples can be analyzed
with minimal sample processing or even directly from the tissue source (Takats
et al. 2004). Further, these techniques allow for the analyses of more labile
and larger metabolites, and for the coupling of liquid separation technologies
to mass spectrometry. Therefore high performance liquid chromatography
(HPLC) and CE are readily coupled to mass spectrometry to yield powerful
tools for targeted metabolic profiling and non-targeted metabolomics.
The utility of LC/MS emanates from the coupling of a ‘universal’ separa-
tion technology to a selective and sensitive mass analyzer detector. HPLC is
commonly considered a universal separation technique because of its appli-
cability to a broad range of chemical classes with a diversity of physical and
chemical properties. For example, HPLC has been utilized for the analysis
of ionic compounds, inorganics, volatile organics, polar organics, non-polar
organics, lipids, amino acids, carbohydrates, nucleotides, carotenoids, phenyl-
propanoids, hormones, peptides, proteins, and the list goes on. The major point
is that HPLC can be used for many of those compounds commonly analyzed by
GC and many more. LC/MS also removes the need for derivatization and thus,
complex samples can be analyzed directly or with minimal sample processing.
As a result of these favorable properties, it is not surprising that LC/MS and
LC coupled to tandem mass spectrometry (LC/MS/MS) have become popular
tools for metabolism investigations.
HPLC is performed on various scales utilizing different column sizes. Gen-
eral values are provided in Table 1 for preparative, analytical, micro, capillary
and nano-scale modes of HPLC. Generally, preparative scale HPLC is used
for compound(s) purification and analytical scale is traditionally used for
the quantitative analyses of plant extracts. However, smaller scale technolo-
gies (micro, capillary, nano) are now commercially available for quantitative
analyses. These smaller scale separations offer significant sensitivity enhance-
ments, and thus reduce the amount of material necessary for analysis. Further,
capillary and nano HPLC often offer increased chromatographic resolution.
Unfortunately as the separation scale gets smaller it becomes more difficult to
reproducibly generate mobile phase gradients and the retention time variance
increases. However, this problem is continually decreasing as novel instrumen-
tation and approaches become available.
Table 1. General liquid chromatographic scales
Scale Column internal diameter Flow rate
Preparative 2.1–>200 mm 10 mL/min,

Analytical (conventional) 2.1−4.6 mm 1.0 mL/min
Micro 1.0 mm 200 μL/min
Capillary 300 μm–1 mm id 4 μL/min
Nano 25−300 μm id 200 nL/min
24 L.W. Sumner
2 Chromatography Theory
Currently, the chromatographic performance of HPLC, relative to GC and CE,
is lower, and there is a significant need for improvement. However, to dis-
cuss this issue and possible improvements in detail, several terms must be
defined. A number of quantifiers are used to assess chromatographic perfor-
mance. These include resolution (Rs ), selectivity (α), efficiency (N), and peak
capacity (n) which are defined below:
1. Resolution (Rs ) is a quantifier of the degree of separation between mixture
components, i. e. two peaks ta and tb with peak widths at the base wa
and wb . A resolution of 1 indicates that two adjacent peaks are baseline
resolved. Resolution can also be expressed as a function of the theoretical
plate number (N) and selectivity (α) as defined below in Eq. (1):
√
2(tb − ta ) 2ΔtR N α−1 k2
Rs = = Rs = (1)
wa + wb wa + wb 4 α 1 + k2
2. Selectivity (α), which is also referred to as the separation factor, is a ratio of
the retention or capacity factor (k ) of two peaks. The capacity factor is a rel-
ative retention parameter that has been normalized using the void elution
time (tv ) or volume (Vv ) and is therefore independent of column geometry –
see Eq. (2). The void value is the volume or time of an unretained component.
The selectivity parameter provides a quantifier of the relative separation of
two components. Selectivity can be altered based on the chemical composi-
tion of the stationary phase, stationary phase manufacturer, mobile phase,
and pH:
k2 t1 − tv V1 − V2
α= k1 = = (2)
k1 tv Vv
3. Column efficiency is usually quantified based upon a column’s theoretical
plate number (N) which is unitless and a measure of band broadening per
unit time – see Eq. (3). This can be practically quantified using retention
time (tR ) and peak width. Peak width can be defined at the base (Wb ) or at
half height (w1/2 ) as they are directly related if one assumes a Gaussian peak
shape, i. e. Wb = 1.698 w1/2 = 4σ where σ equals the standard deviation of
the peak. Alternatively, plate number can be calculated using the column
resolution (R) and selectivity (α).
4. Separation efficiency is also quantified using a normalized theoretical plate
number based on column length, i. e. (N/L) with units of plates/m. The
theoretical plate number can be dramatically increased by decreasing the
peak width. Plate number and efficiency are also related to particle size (dp )
and column length (L) as described below:
2 2
tR tR tR 16R2 L
N= = 16 = 5.54 = = (3)
σ Wb w1/2 (1 − α)2 dp
5. Peak capacity (n) is a measure of the maximum number of theoretical peaks

resolvable by the chromatographic system based on optimum performance
and equal variation in the partitioning of all components in the mixture –
see Eq. (4). The peak capacity is a good parameter for estimating the maxi-
mum number of compounds resolvable by a given chromatographic system.
Ideally this value should approach or exceed the number of compounds that
need to be separated, i. e. the number of metabolites:
√
N t2
n= ln +1 (4)
4R t1
3 Limitations of Current Metabolic Profiling Approaches

and Proposed Solutions to Advance Metabolomics
Currently, the major limitation of metabolomics is its inability to comprehen-

sively profile all of the metabolome. This inability is directly related to the
chemical complexity of the metabolome, the biological variance inherent in
most living organisms, and the dynamic range limitations of most instrumen-
tal approaches (Sumner et al. 2003). Many biological responses to altered gene
expression or to environmental stimuli result in both quantitative and qual-
itative changes in metabolite pools. Understanding these responses is most
dependent upon the qualitative identification of the altered metabolite. Quan-
titative measurements are also important, as both temporal and spatial changes
in metabolite concentrations are expected; however this information is of little
use if it cannot be assigned to a specific metabolite or biological process. Thus,
comprehensive qualitative and quantitative analysis of all metabolites within
a cell, tissue or organ is the ultimate goal of metabolomics; however, this is still
a very ambitious goal and far from a reality for any system. Bino and colleagues
(Bino et al. 2004) proposed two major objectives to increase the comprehensive
nature of metabolomics. They were:
1. Increase the current capacity for metabolite separation and differentiation
(i. e. the number of resolvable components within the complex metabolome
mixture) using multi-dimensional separations.
2. Increase the number of identifiable metabolites through the generation of
spectral libraries, high resolution accurate mass measurements, and tandem
mass spectrometry.
Unfortunately, the separation of complex metabolome mixtures is still quite
challenging. Currently, analytical scale HPLC (4.6 × 250 mm) is most com-
monly used for natural product analyses; however, the upper peak capacities
(i. e. theoretical number of maximum peaks resolvable based on optimum per-
formance) of these systems is approximately 300 (Tanaka et al. 2004). Based
on this estimate, a maximum of 300 components could be resolved in a best
26 L.W. Sumner
case scenario; however in practice this value is seldom achieved and more
realistic peak capacities are between 100 and 200. Thus, current HPLC tech-
nologies are limiting the comprehensive scope of metabolomics. Separation
efficiencies can be improved by altering selectivity, increasing column lengths,
reducing particle sizes, increasing temperature, and/or alternative column
materials. Alternatively, the utilization of multidimensional chromatography
offers increased HPLC peak capacities of greater than 1000 to provide more
comprehensive coverage of plant natural products (Tanaka et al. 2004). Each
of methods to increase HPLC efficiency is discussed below.
Typically, improving selectivity is the best approach to improving chro-
matographic resolution. Selectivity is based upon the chemical or physical
interaction properties that are fundamental to the separation process. More
precisely, the separation selectivity of specific components can be optimized
by the appropriate choice of column materials, mobile phases, and/or man-
ufacturer. Various generic and proprietary materials are available for vari-
ous chromatographic modes for HPLC. Example modes include ion-exchange,
normal-phase, reverse-phase, hydrophilic interaction, and size exclusion chro-
matography. All HPLC columns are not equal, and different particles, particle
sizes, surface modification chemistries, surface coverage, and packing pro-
cesses vary significantly from manufacturer to manufacturer. These parame-
ters dramatically influence chromatographic performance.
Often selectivity is optimized for a targeted set of analytes as a means of
increasing resolution. However, in more complex mixtures associated with
global metabolomics-based approaches, improved selectivity for one class of
compounds often results in decreased selectivity for others. Thus, techniques
(e. g. reverse-phase chromatography) with a broad range of selectivity are most
likely to be the best choices for metabolomics.
One of the simplest means of increasing resolution is to increase the number
of theoretical plates. Since the plate number is directly proportional to the
column length (Eq. (3)), one needs only to increase the column length to
increase resolution. However, Eq. (1) tells us that R is proportional to the
square root of N. Thus, to achieve a 2× increase in resolution, we would
have to square the column length. For example a 250 mm long column would
need to be extended to 625 cm (i. e. 25 × 25 cm) for a twofold increase in
resolution. Unfortunately, this is not a practical solution as the operating
pressure is directly proportional to the column length. Equation (5) defines
the relationship between pressure (ΔP), column length (L), analyte diffusion
coefficient (Dm ), particle size (dp ), mobile-phase viscosity (η), and column
permeability (K o ):

LvDm η
ΔP = (5)
dp Ko
If a typical column of 25 cm has an operational pressure of 3000 pounds per

square inch (p.s.i.), then a twofold resolution increase obtained by squaring the
column length (25 cm)2 would require an operational pressure of 75,000 p.s.i.
(i. e. 3,000 p.s.i. × 25). Although this illustrates the advantage of very high
pressure liquid chromatography which has been achieved by select groups
using custom apparatuses (MacNair et al. 1997, 1999; Tolley et al. 2001; Patel et
al. 2004; Shen et al. 2005), commercial pumps do not operate at these pressures
(most commercial HPLC pumps have a 5,000-p.s.i. limit). Therefore, significant
resolution enhancements achieved through longer columns is limited for most
researchers. With that said, several companies (i. e. Waters and JASCO) have
recently introduced 15,000-p.s.i. HPLC pumps.
Equation (5) reveals that the pressure differential is proportional to the
mobile phase viscosity (η). Thus, lowering of the mobile phase viscosity (η)
by increasing the temperature can lower the operational pressure and allow
the use of longer columns for resolution enhancement (Djordjevic et al. 1998,
1999, 2000). Selectivity is also affected by temperature and additional efficiency
can be achieved by heating alone. However, one must ensure analyte thermal
stability if elevated temperature separations are to be employed.
Equation (5) also shows that the pressure is a function of the column per-
meability (K o ). New monolithic columns offer greater permeability and lower
pressures, thus allowing for the use of longer columns. The continuous bed
stationary phases of these columns consist of porous polymeric materials gen-
erated from silica or organic materials such as acrylamide, styrene, acrylate,
or methacrylate monomers which result in lower back-pressure than packed
particles. The lower back-pressure allows for the use of longer columns and
hence greater efficiencies. Several groups have reported on the use of up to
1 m capillary columns (Que and Novotny 2002; Legido-Quigley et al. 2003;
Tolstikov et al. 2003; Tanaka et al. 2004) and this technology looks promis-
ing.
Plate number and efficiency are also related to particle size (dp ) and column
length (L) as shown in Eq. (3). This equation shows that decreasing the particle
size increases the theoretical plate number/efficiency (MacNair et al. 1997,
1999; Tolley et al. 2001; Shen et al. 2005). However, Eq. (5) shows again that
pressure increases with smaller particle size. Fortunately, new commercial
ultra-high pressure liquid chromatography pumps (UPLC) are now available
from multiple manufacturers that allow the use of smaller particles in the range
of 1−2 μm. These instruments offer substantial resolution enhancements with
plate numbers on the order of several hundred thousand and peak capacities in
excess of 400 (Wilson et al. 2005). In addition to increased resolution, UPLC also
offers higher speed separations as the optimum flow velocity has a significantly
broader range which allows for increased flow rates without significant loss of
resolution (Wilson et al. 2005). Estimates of up to ninefold increases in flow
rates without significant loss of resolution have been suggested (Wilson et al.
2005). It is important to note that ultra-high pressure separations result in
increased frictional heating; however this can be reduced by down-scaling the
chromatography dimensions with the heating being negligible in columns of
less than 1 mm (MacNair et al. 1997).
28 L.W. Sumner
4 Future Directions and Forward-Looking Thoughts

Although several of the above principles can be used to achieve enhanced
chromatographic resolution, the resolution enhancements are still far from
that which is needed for very complex metabolomics mixtures. To separate
these mixtures, peak capacities of thousands to tens of thousands are neces-
sary. Currently, only multidimensional chromatographic methods offer peak
capacities of this magnitude (Mondello et al. 2002; Evans and Jorgenson 2004).
Multidimensional chromatography utilizes combinations of two or more sep-
aration mechanisms with different selectivity, e. g. ion-exchange and reverse-
phase or capillary electrophoresis and reverse-phase LC. These systems offer
enhanced resolution due to the utilization of multiple columns with inde-
pendent chemistries which expands the selectivity of the method. Recall that
selectivity improvements can dramatically improve resolution. The maximum
peak capacity of a multidimensional system is the product of the two or more
individual separation dimensions. For example, a realistic system that has
a peak capacity in the first dimension (ny ) of 150 and the peak capacity in
the second dimension (nz ) of 50, then the total maximum peak capacity of the
multidimensional system is ny × nz = 150 × 50 = 7500. If one considers that
an individual metabolome consists of 15,000 metabolites, then one recognizes
that this is a considerable increase in comprehensive coverage.
Multidimensional LC-LC separations have been capitalized upon in the area
of proteomics and are often referred to as multidimensional protein identifi-
cation technology (i. e. MUDPIT; Washburn et al. 2001; Wolters et al. 2001);
however multidimensional separations have only recently been pursued for
metabolomics using GC×GC/time-of-flight (TOF)-MS (Welthagen et al. 2005).
Unfortunately, these complex separations will come with increased analysis
times, but I believe they will be worth the additional temporal costs.
The above discussion focuses on homogenous multidimensional separa-
tions (i. e. LC×LC/MS or GC×GC/MS, but multidimensional LC×GC separa-
tions are possible. In fact, the combination of these technologies is commonly
referred to as unified chromatography (Chester and Parcher 2001; Chester and
Pinkston 2002; Wells et al. 2002, 2003; Luo et al. 2003) and often associated
with supercritical fluid chromatography (Chester and Parcher 2001; Chester
and Pinkston 2002; Mondello et al. 2002; Wells et al. 2002, 2003; Luo et al.
2003). Although this technology is conceptually exciting, it is still somewhat
empirically limited. Another possible LC×GC approach would be to couple
HPLC with ion mobility mass (IMS) TOF-MS spectrometry (Verbeck et al.
2002; Guevremont 2004; Liu et al. 2004; Shvartsburg et al. 2005). In this config-
uration, analytes are ionized as they elute from the HPLC and an electrostatic
field propels the analyte ions through a gas field maintained at elevated, atmo-
spheric, or sub ambient pressures. Ions of different size and geometric structure
traverse the gas field at different rates dependent upon their charge and col-
lisional cross section therefore allowing separation. The LC-IMS method has
been demonstrated for proteomics (Lee et al. 2002; Matz et al. 2002; Liu et al.
2004) and more recently applied to metabolite analyses (Kapron et al. 2005).
Extension of this concept to metabolomics will surely occur.
The above text discusses multidimensional chromatographic approaches in
an on-line context. However, multidimensional approaches can also be pur-
sued in an off-line, multiplexed, or parallel approach. For example, fractions
can be collected off-line using a separate HPLC. The fractions can then be con-
centrated and reinjected onto an on-line LC/MS system. Alternately, fractions
of the same samples could be injected onto a series of parallel systems using
different methods (i. e. GC/MS, LC/MS, or various selective modes of each
performed with different column selectivities). This is our current approach.
For example, samples are fractionated and/or enriched and then the polar and
lipophilic fractions are analyzed by GC/MS. In addition methanolic extracts
are analyzed for phenolic/saponin content. An interesting concept would be to
design a multiplexed system, with multiple chromatographic-mass spectrom-
etry systems operating in an integrated manner. For example, a multiplexed
chip system with each chip having a slightly different selectivity and indepen-
dent mass analyzer could be designed to increase the comprehensive coverage.
Such a system with on-line enrichment could also be used to address dynamic
range limitations that currently exist for specific compound classes such as
phytohormones.
If higher resolution chromatography is obtained, mass analyzers must also
be employed with compatible scans speeds to record data for compounds
eluting in very short temporal periods. It is expected that LC peak widths of
1−5 s will be routine in the very near future. For accurate quantification, it is
commonly accepted that the sampling rate should be sufficient to capture 10
data points across the eluting peak with higher sampling rates being beneficial.
Thus, sampling rates should be less than 0.1 s or greater than 10 Hz. This
is achievable with current TOF-MS analyzers. It is worth mentioning that
quadrupole based mass analyzers, including traps, can approach these speeds;
however, TOF mass spectrometers equipped with delayed extraction and ion-
reflectrons also offer improved mass accuracy over quadrupoles.
Improvements in the accuracy of the mass analyzer can further enhance
metabolite differentiation, elemental composition determination, identifica-
tion, and allow for the profiling of greater numbers of metabolites. Mass
accuracy is directly related to the mass resolution or the ability of the mass
analyzer to resolve compounds of different m/z values. Mass resolution is de-
fined in Eq. (6) and is a function of mass (M) divided by the peak width (ΔM)
which is most commonly defined at half-height:
M
Rm = (6)
ΔM
Often, LC/MS is performed with ion-traps or quadrupole mass analyzers that
yield mass accuracies in the range of 1.0−0.1 Da. Unfortunately, many metabo-
lites have similar nominal masses which can not be differentiated at this level
of mass accuracy. For example, the important natural products genistein and
30 L.W. Sumner
medicarpin have similar nominal masses of 270, but have different accurate
masses of 270.2390 (C15 H10 O5 ) and 270.2830 (C16 H14 O4 ) respectively due to
different chemical compositions. If one could measure their mass with suf-
ficient accuracy, then one could differentiate these compounds in the mass
domain even if they could not be physically separated in the chromatographic
domain. This mass differentiation can be achieved at a mass resolution (M/ΔM)
greater than 6136. Compounds with closer accurate masses such as rutin
(C27 H30 O16 = 610.5180) and hesperidin (C28 H34 O15 = 610.5620) would re-
quire a higher mass resolution of 13,864 for their differentiation. Mass resolu-
tions on the order of 10,000 can be achieved with modern TOF-MS analyzers,
and resolutions in excess of 100,000 with sub-part-per-million mass accuracies
(i. e. less than 0.001 at m/z of 1,000 Da) are achievable with Fourier transform
ion cyclotron mass spectrometry (FTMS). Newer technologies, such as Thermo
Electron Corporation’s Orbitraps are currently surfacing that also offer high-
resolution solutions. Although high resolution accurate mass measurements
have great advantages, this technology is still rather costly.
Interestingly, a significant argument can be made that accurate mass mea-
surements significantly reduce the need for ultra-high resolution separations
due to the enhanced separation in the mass domain. However if the chro-
matography step is omitted or compressed significantly, then ion suppression,
competitive ionization, and other matrix affects become increasingly more
problematic. I personally believe that both improved chromatographic reso-
lution and accurate mass measurements offer the best solution and that the
combination of these techniques will provide greater comprehension and con-
fidence in our ability to profile the metabolome. Further, I also believe that the
needed magnitude of enhancements in chromatographic resolution can only
be achieved with multidimensional approaches.
References
Birkemeyer C, Kolasa A, Kopka J (2003) Comprehensive chemical derivatization for gas
chromatography-mass spectrometry-based multi-targeted profiling of the major phytohor-
mones. J Chromatogr A 993:89–102
Broeckling CD, Huhman DV, Farag MA, Smith JT, May GD, Mendes P, Dixon RA, Sumner LW
(2005) Metabolic profiling of Medicago truncatula cell cultures reveals the effects of biotic
and abiotic elicitors on metabolism. J Exp Bot 56:323–336
Chester T, Parcher JF (2001) Blurring the Boundaries. Science 291:502–503
Chester T, Pinkston J (2002) Supercritical fluid and unified chromatography. Anal Chem 74:2801–
2811
Dixon RA (2001) Phytochemistry in the genomics and post-genomics eras. Phytochemistry
57:145–148
Djordjevic N, Houdiere F, Fowler P (1998) High temperature and temperature programming in
capillary HPLC. Biomed Chromatogr 12:153–154
Djordjevic N, Fitzpatrick F, Houdiere F, Lerch G, Rozing G (2000) High temperature and temper-
ature programming in capillary electrochromatography. J Chromatogr A 887:245–252
Djordjevic NM, Fowler PWJ, Houdiere F (1999) High temperature and temperature programming
in high-performance liquid chromatography: Instrumental considerations. J Microcolumn
Separ 11:403–413
Evans C, Jorgenson J (2004) Multidimensional LC-LC and LC-CE for high-resolution separations
of biological molecules. Anal Bioanal Chem 378:1952–1961
Fiehn O, Kopka J, Dormann P, Altmann T, Trethewey RN, Willmitzer L (2000) Metabolite profiling
for plant fuctional genomics. Nat Biotechnol 18:1142–1161
Guevremont R (2004) High-field asymmetric waveform ion mobility spectrometry: a new tool
for mass spectrometry. J Chromatogr A 1058:3–19
Gygi SP, Rochon Y, Franza BR, Aebersold R (1999) Correlation between protein and mRNA
abundance in yeast. Mol Cell Biol 19:1720–1730
Hartman T, Kutchan TM, Strack D (2005) Evolution of metabolic diversity. Phytochemistry
66:1198–1199
Kapron J, Jemal M, Duncan G, Kolakowski B, Purves R (2005) Removal of metabolite interfer-
ence during liquid chromatography/tandem mass spectrometry using high-field asymmetric
waveform ion mobility spectrometry. Rapid Commun Mass Spectrom 19:1979–1983
Lee Y, Hoaglund-Hyzera C, Srebalus Barnes C, Hilderbrand A, Valentine S, Clemmer D (2002)
Development of high-throughput liquid chromatography injected ion mobility quadrupole
time-of-flight techniques for analysis of complex peptide mixtures. J Chromatogr B Anal
Technol Biomed Life Sci 782:343–351
Legido-Quigley C, Marlin N, Melin V, Manz A, Smith N (2003) Advances in capillary electrochro-
matography and micro-high performance liquid chromatography monolithic columns for
separation science. Electrophoresis 24:917–944
Liu X, Plasencia M, Ragg S, Valentine S, Clemmer D (2004) Development of high throughput
dispersive LC-ion mobility-TOFMS techniques for analysing the human plasma proteome.
Brief Funct Genomic Proteomic 3:177–186
Luo Z, Xiong Y, Parcher J (2003) Chromatography with dynamically created liquid “stationary”
phases: methanol and carbon dioxide. Anal Chem 75:3557–3562
MacNair J, Lewis K, Jorgenson J (1997) Ultrahigh-pressure reversed-phase liquid chromatography
in packed capillary columns. Anal Chem 69:983–989
MacNair J, Patel K, Jorgenson J (1999) Ultrahigh-pressure reversed-phase capillary liquid chro-
matography: isocratic and gradient elution using columns packed with 1.0-micron particles.
Anal Chem 71:700–708
Matz L, Dion H, Hill H (2002) Evaluation of capillary liquid chromatography-electrospray ioniza-
tion ion mobility spectrometry with mass spectrometry detection. J Chromatogr A 946:59–68
Mondello L, Lewis AC, Bartle KD (2002) Multidimensional Chromatography. Wiley, Chichester,
UK
Oliver DJ, Nikolau B, Wurtele ES (2002) Functional genomics: high-throughput mRNA, protein,
and metabolite analyses. Metab Eng 4:98–106
Patel K, Jerkovich A, Link J, Jorgenson J (2004) In-depth characterization of slurry packed capillary
columns with 1.0-microm nonporous particles using reversed-phase isocratic ultrahigh-
pressure liquid chromatography. Anal Chem 76:5777–5786
Que A, Novotny M (2002) Separation of neutral saccharide mixtures with capillary electrochro-
matography using hydrophilic monolithic columns. Anal Chem 74:5184–5191
Roessner U, Wagner C, Kopka J, Trethewey RN, Willmitzer L (2000) Simultaneous analysis of
metabolites in potato tuber by gas chromatography-mass spectrometry. Plant J 23:131–142
Roessner U, Luedemann A, Brust D, Fiehn O, Linke T, Willmitzer L, Fernie AR (2001) Metabolic
Tunali U, Forbes M, Willmitzer L, Fernie A, Kopka J (2005) GC-MS libraries for the rapid
identification of metabolites in complex biological samples. FEBS Lett 579:1332–1337
32 L.W. Sumner
Shen Y, Zhang R, Moore R, Kim J, Metz T, Hixson K, Zhao R, Livesay E, Udseth H, Smith R
(2005) Automated 20 kpsi RPLC-MS and MS/MS with chromatographic peak capacities of
1000–1500 and capabilities in proteomics and metabolomics. Anal Chem 77:3090–3100
Shvartsburg A, Tang K, Smith R (2005) Optimization of the design and operation of FAIMS
analyzers. J Am Soc Mass Spectrom 16:2–12
Somerville C, Dangl J (2000) Plant biology in 2010. Science 290:2077–2078
Somerville C, Somerville S (1999) Plant functional genomics. Science 285:380–383
Takats Z, Wiseman JM, Gologan B, Cooks RG (2004) Mass spectrometry sampling under ambient
conditions with desorption electrospray ionization. Science 306:471–473
Tanaka N, Kimura H, Tokuda D, Hosoya K, Ikegami T, Ishizuka N, Minakuchi H, Nakanishi K, Shin-
tani Y, Furuno M, Cabrera K (2004) Simple and comprehensive two-dimensional reversed-
phase HPLC using monolithic silica columns. Anal Chem 76:1273–1281
Tolley L, Jorgenson J, Moseley M (2001) Very high pressure gradient LC/MS/MS. Anal Chem
73:2985–2991
Tolstikov VV, Lommen A, Nakanishi K, Tanaka N, Fiehn O (2003) Monolitichic silica-based
capillary reversed-phase liquid chromatography/electrospray mass spectrometry for plant
metabolomics. Anal Chem 75:6737–6740
Trethewey RN (2001) Gene discovery via metabolic profiling. Curr Opin Biotechnol 12:135–138
Trethewey RN, Krotzky AJ, Willmitzer L (1999) Metabolic profiling: a rossetta stone for genomics?
Verbeck G, Ruotolo B, Sawyer H, Gillig K, Russell D (2002) A fundamental introduction to ion
mobility mass spectrometry applied to the analysis of biomolecules. J Biomol Tech 13:56–61
Wagner C, Sefkow M, Kopka J (2003) Construction and application of a mass spectral and
retention time index database generated from plant GC/EI-TOF-MS metabolite profiles.
Phytochemistry 62:887–900
Washburn M, Wolters D, Yates J (2001) Large-scale analysis of the yeast proteome by multidi-
mensional protein identification technology. Nat Biotechnol 19:242–247
Weckwerth W (2003) Metabolomics in systems biology. Annu Rev Plant Biol 54:669–689
Wells P, Zhou S, Parcher J (2002) Gas-liquid chromatography with a volatile “stationary” liquid
phase. Anal Chem 74:2103–2111
Wells P, Zhou S, Parcher J (2003) Unified chromatography with CO2-based binary mobile phases.
Anal Chem 75:18A–24A
Welthagen W, Shellie RA, Spranger J, Ristow M, Zimmermannn R, Fiehn O (2005) Comprehensive
two-dimensional gas chromatography-time-of-flight mass spectrometry (GC×GC-TOF) for
high resolution metabolomics: biomarker discovery on spleen tissue extracts of obese NZO
compared to lean C57BL/6 mice. Metabolomics 1:65–73
Wilson I, Nicholson J, Castro-Perez J, Granger J, Johnson K, Smith B, Plumb R (2005) High reso-
lution “ultra performance” liquid chromatography coupled to oa-TOF mass spectrometry as
a tool for differential metabolic pathway profiling in functional genomic studies. J Proteome
Res 4:591–598
Wolters D, Washburn M, Yates J (2001) An automated multidimensional protein identification
technology for shotgun proteomics. Anal Chem 73:5683–5690
I.3 Plant Metabolomics Strategies Based upon
Quadrupole Time of Flight Mass Spectrometry
(QTOF-MS)
H.A. Verhoeven1,2 , C.H. Ric de Vos1,2 , R.J. Bino1,2 , and R.D. Hall1,2
1 Introduction
The growing interest in the use of metabolomics technologies in plant research

has come about both due to the broad value of such approaches in almost every
field of plant science and also through the improvements in instrumentation
and bioinformatics tools which has been realised in recent years. A compre-
hensive overview of all the various technologies available is beyond the scope
of this chapter but the reader is referred to other chapters in this volume or to
a number of recent reviews for information on the different approaches (Fiehn
2001, 2002; Sumner et al. 2003; Goodacre et al. 2004). However, MS-based
strategies, and in particular in combination with GC or LC separation technol-
ogy, are proving most popular as these combine very high analytical precision
with an equally high detection sensitivity. This enables reliable measurements
to be made down to the femtomolar range (Fernie 2003). Furthermore, recent
advances in electronics and computing have given rise to the development
of yet a new generation of mass spectrometers to supplement the traditional
magnetic sector and scanning quadrupole instruments that have been around
for several decades now. In this new generation, instruments based on ion
traps and time-of-flight (generally referred as TOF) are the most prominent.
In particular, the TOF instruments have become popular due to their rela-
tively simple construction and their capacity to be combined with a number
of other technologies to enable multi-dimensional analysis. This has resulted
in an unprecedented expansion of our metabolomic capabilities. For exam-
ple, the fast spectral acquisition capacity of TOF instruments has resulted in
approximately 1000 components being detected in leaf extracts and an analyt-
ical capacity of 1000 samples per month has been achieved (Weckwerth et al.
2001). Such sample numbers and breadth of metabolite detection represent the
arrival of true metabolomics research in the true sense of the word. Since that
time, our capacity for metabolomic analyses has continued to improve. The
high mass accuracy, high resolution, good dynamic range and the large diver-
sity of detectible masses possible with TOF instruments, in association with
their intrinsic high sensitivity, are therefore the main reasons behind the many
applications, first in the field of proteomics, and now also in metabolomics.
1 Plant ResearchInternational, P.O. Box 16, 6700 AA Wageningen, The Netherlands
2 Centrefor BioSystems Genomics, P.O. Box 98, 6700 AB Wageningen, The Netherlands, e-mail:
[email protected]

34 H.A. Verhoeven, C.H. Ric de Vos, R.J. Bino, and R.D. Hall
2 The Technology
Time of flight was already introduced in the early 1960s but was quickly re-
placed by other approaches. This was due to the lack of sufficiently fast elec-
tronics needed to process data on a nanosecond scale. Thirty years later in
the early 1990s, the development of high megahertz and even gigahertz digital
circuits led to the dramatic increase in the application of TOF technology. This,
combined with new developments in the area of sample introduction and ion-
isation of (macro)molecules, has subsequently led to many new applications
of (TOF-based) mass spectrometry in the fields of biology and pharmacy.
A TOF instrument serves as the main mass analyser, and its principle is
based on ions with different mass/charge ratios having different flight times in
a field-free drift zone once they have been accelerated by a very short electric
pulse from the electrodes of an accelerator: lighter ions travel faster through
the measurement chamber than the heavier ones. A thorough discussion on
the physical principles can be found in, for example, Guilhaus (1995). In most
TOF instrument designs, ions are detected using Micro Channel Plates (MCP),
which, on capturing the ions, generate a cascade of electrons to amplify the
signal so that it can be detected by the associated electronics (see Fig. 1 for
a schematic representation of the various parts). Several ion recorders have
been used with the various designs of hybrid TOF mass spectrometer. The two
most widely used are the time-to-digital converter (TDC) and the transient
recorder or analogue to digital converter (ADC) (Chernushevich et al. 2001).
The type of detector affects both the dynamic range of the signals that can be
measured and also the mass accuracy. In a TDC, every individual ion generates
a pulse. This pulse is shaped into a digital signal, of which the rising flank is
used for timing. The time passed since the start of the ion accelerating pulse
and its arrival at the MCP is stored in the memory. This system is very accurate
over the entire mass range and is optimally suited for the accurate timing of
low ion counts. It is, however, less suitable or even unsuitable for the detection
of ions arriving simultaneously at the MCP since these will be recorded as
being single events and this will thus lead to an underestimation of the signal.
TDCs also suffer from an additional limitation concerning the detection of
ions. During the time required to process one pulse, the detector is ‘blind’ to
new incoming pulses. This so-called dead time, not only leads to a further
underestimation of the signal, but also it causes a shift in the observed m/z
value towards lower values. This can lead to serious deviations from the true
accurate mass at high to very high signal intensities. These problems occur to
a lesser extent in instruments equipped with an ADC, since these machines can
sample the analog output of the MCP at very high frequencies, thus providing
multiple data points per observed m/z value. In this way, multiple ions arriving
at the same time will lead to a linear increase in peak area. In some designs, TDC
and ADC are both used to combine the high mass accuracy of the TDC at low
ion rates, with the high dynamic range and accurate m/z value measurements
of an ADC at high ion rates.
(Q)TOF-MS in plant metabolics 35
Fig. 1. Schematic diagram showing the main components of a typical quadrupole/time of flight
configuration. Ions enter the instrument on the left, and pass through the first quadrupole.
This can be operated either in ion transfer mode, which allows all ions to pass, or in selective
mode, which is used for precursor scanning and alignment of the different quadrupoles. The
ion transfer is a special quadrupole, intended to separate the operating pressures in the different
compartments of the quadruple section. In the collision cell, a collision gas can be present in
order to induce fragmentation of the incoming ions. When no gas is present molecular ions will
be detected. Subsequently, molecular ions and/or charged fragments enter the TOF tube, where
they are collected and pushed into the drift zone. During this transition, ions are accelerated in
an electric field in the accelerator assembly, consisting of several ion lenses, which determines
their kinetic energy. The ions now all follow a trajectory towards the reflectron, consisting of
a pile of cylindrical ion lenses at different potentials, which causes them to be repelled towards
the detector, the multi channel plate. Here the ions strike the surface of the detector, which finally
converts the arrival of every single ion into a measurable electric current. Additional electronics
is required to process the electrical signals and the timing between pushing the ions into the drift
zone and their arrival at the detector
Regardless of the high mass accuracy, resolution and sensitivity, the applica-
tion of TOF instruments in structure elucidation is quite limited due to the ab-
sence of filtering and scanning capabilities. Consequently, hybrid instruments
have since been designed to cope with these shortcomings. These machines
include the addition of ion trap(s), quadrupoles or combinations thereof, to
the basic TOF analyser. One key example is the now increasingly well-known
and widely used QTOF system. These instruments rely on the combination
of two or more quadrupoles with a TOF analyser. The first quadrupole (Q1)
serves as a mass filter or ion tunnel, depending on the operational mode, with
the second quadrupole (Q2) serving as the collision cell for the fragmentation
of the ions which have passed through Q1. This fragmentation is achieved
using an electric field to accelerate the ions, in combination with a collision
gas such as nitrogen or argon. Fragmentation can be controlled by varying the
(very low) pressure of the collision gas and/or by varying the collision energy
through altering the acceleration voltage of the cell. Collisions with the gas
molecules also result in a cooling of the ions, which incurs that their kinetic
energy is transferred. This results in a more homogeneous energy distribution
of the individual ions, which in turn improves the mass accuracy capacity of
the instrument. The ions and/or ion fragments are subsequently collected in
the accelerator part of the TOF instrument where a very short pulse is applied
to the electrodes of the chamber to eject the ions. In the case of orthogonal
ejection, the differences in kinetic energy in the z-axis will be less than in case
of forward ejection. The differences in kinetic energy are also further reduced
in the reflectron lens, which repels the ions towards the detector. Here, ions
with higher energy will travel further than lower energy ions, thus reducing
the difference. A number of variations on this basic design have been created.
These include, for example, the modification of the second quadrupole into
a linear ion trap with axial ejection through the addition of a number of ex-
tra ion lenses (Hager 2002). As a result, new possibilities are created, such as
the ability to store specific ions, which can be selectively ejected for complex
MS/MS or MSn analyses (Hager 2002). The high mass accuracy can be further
improved by using an internal (reference) standard that is sampled at regular
intervals throughout the entire analysis period. This reference is then used
to correct the instrument calibration on-the-fly (lock mass correction). Such
a capacity for continuous (re)calibration is particularly useful, if not essential,
in the case of long series of chromatographic runs where excellent, long-term
stability of the mass accuracy can be continuously achieved down to ±5 ppm.
This is significant as mass accuracy at or below this level allows us to predict
the chemical composition of a given ion by using the small known differences
in atomic masses of the various atomic elements. In this way, a first predic-
tion can be made about the nature and identity of the molecular component.
Combined with other data (retention time, N rule, stable isotope distribution
of 13 C etc.) this can then enable the list of possible molecular identities to
be reduced even further and thus come closer to translating MS output into
named metabolites.
Combining the results obtained from several biological samples into a sin-
gle comparative analysis is an arduous task that requires the precise align-
ment and matching of peaks representing the same compound over all chro-
matograms. Due to its relatively robust chromatography and compound sepa-
ration efficiency, GC-(TOF)-MS of derivatized extracts is at present generally
preferred over LC-MS in metabolomic studies (Fiehn et al. 2000; Roessner et al.
2001a,b; Fernie et al. 2004). Nevertheless, GC-MS is less suitable for semi-polar
compounds among which are key classes of plant (secondary) metabolites
including flavonoids, (glyco-)alkaloids, glucosinolates and saponins. Recent
advances in techniques for improving resolution in LC by using capillary elec-
trophoresis (Soga et al. 2002), hydrophilic interaction columns (Tolstikov and
Fiehn 2002) and monolithic columns (Tolstikov et al. 2003) demonstrate the
high potential which TOF technology has for LC-MS to complement GC-MS in
unravelling metabolic profiles.
3 Data Analysis
Data analysis is perhaps the most crucial step in any metabolomics strategy
and the importance of bioinformatics tools should not be underestimated. In
a standard (ideal) approach, a whole range of standards would be used to assist
in identifying through simple linkage, which peaks in an MS output represent
which metabolites. However, as the vast majority of metabolites present in
complex plant extracts are as yet unknown and are not commercially available,
as is especially true for the secondary plant metabolites, this approach is
unfeasible at present for a true untargeted metabolomics approach. Another
strategy is therefore required which enables the automated and essentially
blind direct comparison of large numbers of spectra. Since most datasets are
very complicated, dedicated metabolomics software is needed for this purpose.
Some of this software is already available but more still needs to be developed
and this represents a major task for the next five years.
Data manipulation is essential for reliable metabolomics analyses and spe-
cial attention has to be paid to aspects such as baseline correction and noise
elimination. In addition, in the case of LC-MS, particular attention also needs
to be given to reliable correction of local drifts in retention time and accurate
mass. Different compositions of eluant can cause significant variation in base-
line especially when using steep LC gradients. For the successful correction of
such baseline fluctuations, the chromatogram has to contain a region without
strong peaks. Digital filtering will enable the elimination of excess noise which
would otherwise lead to the generation of erroneous (false) peaks. Some re-
cent software packages are able to deal with a number of these problems in
TOF data analysis. Another key element is the need to correct for retention
time fluctuations. Unlike capillary gas chromatography which is generally very
stable, liquid chromatography often suffers from relatively large, non-linear
(localised) fluctuations in retention time. This can be due to small differences
in pH, temperature, or the co-elution of components which interact differently
with the stationary phase. Consequently, this problem prevents a simple direct
comparison of different samples. A number of algorithms have been designed
to correct for this phenomenon. One such approach, based on photodiode
array type data, uses correlation optimised warping of the chromatograms to
achieve alignment of shifted peaks in the chromatograms (Nielsen et al. 1998).
For MS data, MetAlign™ software, in contrast, uses specific mass peaks with
strong local maxima throughout the chromatogram as ‘landmark peaks’ with
which to correct for chromatographic shifts over the entire series of analyses
(Vorst et al. 2005). After correction, unbiased, direct spectral comparisons,
based on mass peak intensities are possible and contrasting mass signals can
be reliably identified and extracted. Differential chromatograms are produced
from which all unchanging peaks have been removed to reveal the true extent
of the differences between two (groups of) samples in one or both directions.
This dedicated software can automatically handle hundreds of full scan MS
datasets obtained by either LC or GC, and is independent of type of mass
spectrometer. Another package, Markerlynx™, which is dedicated to Waters

instruments, exploits the high mass accuracy and resolution of the (Q)TOF
technology. Here, the distribution of specific ions in a predefined mass win-
dow, usually in a 20−50 ppm range, over the chromatogram is analysed and
retention shifts are corrected within a predetermined retention time window.
In this way, metabolites can be compared over many samples using high mass
resolution. Both approaches have their own advantages and limitations. In our
lab we more or less routinely use metAlign™ to process LC-QTOF and GC-TOF
data. However, in cases where there are insufficient landmark peaks, metAl-
ign™ is unable to perform a thorough alignment. When spectra are noisy or
highly complex, the Markerlynx™ approach will likely give misalignments.
How to proceed further with the processing of corrected chromatograms is
dependent upon the type of metabolomics analysis required. Within
metAlign™ there is a tool to extract user-defined significant differences be-
tween two groups of samples based on the Student t-test. In many cases,
where large sample numbers are being compared, multivariate analysis will
be desirable. For this purpose, software originally developed for the analy-
sis of microarray datasets can be advantageously applied, since metabolomic
data share a number of problems similar to those which were encountered
with microarray data. We use, for example, the GeneMaths software package
(Vorst et al. 2005), which enables rapid statistical data analysis and provides
clear graphic outputs of the results in the form of histograms and principle
components plots. Other software packages should be equally valuable.
4 Application of QTOF MS-based Plant Metabolomics Analyses
4.1 Rapid Sample Profiling by Direct Flow Injection Analysis (DFI)
In plant metabolomics there is often an initial desire for a rapid pre-screening

of the samples. This is especially the case when dealing with large sample
numbers where only a limited number of individuals might be expected to
be different. One can think here for example, of natural populations, potential
mutants or the progeny from a breeding cross (Bino et al. 2005; Hall et al. 2005).
Direct Flow Injection can effectively be used to get a rapid, overall impression
of the composition of a biological extract. It is an unbiased analysis, and seeks
to cover as many metabolites as possible in a single short run. The only se-
lective property is the type of ionisation used, i. e. positive vs negative, ESI
(ElectroSpray ionisation) vs APCI (Atmospheric Pressure Chemical Ionisa-
tion). The advantage of this approach is time-saving. When using chromato-
graphic separation to prevent excessive component interaction, run times of
30−90 min are usually required (but see as an exception; Jander et al. 2004). In
the case of DFI, run times of only 30 s to a few minutes (Goodacre et al. 2003)
may be required. For DFI, a few microlitres of extract is introduced by the
autosampler directly into the ion source and all ions with the corresponding
charge are then analysed by the MS. In this case, TOF instruments have a clear
advantage over scanning quadrupole instruments because their significantly
higher resolution allows for the simultaneous detection of many ion species
and, because no scanning is required, every individual ion can theoretically
be captured. This inevitably results in a very rich mass spectrum which is
further complicated by the many interactions which can occur between the
different components of the sample during ionization. Furthermore, unstable
ions can cause additional extensive ion vapour phase interactions. For these
reasons, there was initially considerable scepticism of the potential value of
DFI approaches for reliable metabolic analysis and these phenomena are ex-
tensively discussed elsewhere (Kebarle 2000; King et al. 2000). However, recent
publications have shown that the mass spectrum data obtained in this way is
actually highly reproducible and can effectively be used for a fast screening of
complex extracts (Aharoni et al. 2002; Goodacre et al. 2002, 2003; Castrillo et
al. 2003; Verhoeven et al., in preparation).
Data processing is in many cases the bottleneck for the successful deploy-
ment of this technology, and many applications rely on a dedicated approach to
data processing. This was clearly demonstrated for example, by the MS analysis
of unfractionated plant extracts of Pharbitis leaf sap (Goodacre et al. 2003).
Correct data processing of the complex mass spectra was found crucial for
reliable discrimination between the different physiological treatments used.
Experiments performed in our laboratory resulted in similar conclusions. Five
commercially available extracts from Salix were analysed using DFI in a QTOF
MS in positive mode. A single total ion count (TIC) injection peak was ob-
served, and all the masses obtained were combined into a single mass spectrum
per sample. The aligned spectra were processed for noise elimination, baseline
correction and then centroided to obtain the accurate masses of each m/z peak.
These were then aligned in the m/z dimension using exact masses of known
metabolites to correct for small fluctuations in exact mass due unavoidable
minor (thermal) drift in the TOF tube. Intensities of the m/z peaks were log
transformed, and exported to GeneMaths™ for multivariate analysis. Principle
Components Analysis (PCA) revealed first (Fig. 2a) that the sample replicates
(Samples 1 and 2) cluster close together reflecting the high reproducibility of
the extraction and mass profiling techniques. Sample 3 is also clearly simi-
lar in overall composition to Sample 5, whereas Sample 4 is clearly distinct
from all others. Sample 4 was found to have come from a different supplier.
Differences in sample composition were readily detected by selecting the m/z
values which were responsible for the separation of the samples in the PCA
(Fig. 2b). This example indicates the usefulness of rapid screening for quality
control of complex extracts without the need for more dedicated but time-
consuming LC separation. In a similar manner Goodacre also used a rapid DFI
approach to compare olive oil samples and to test for adulteration (Goodacre
et al. 2002).
Fig. 2. a PCA plot of the entire set of detected mass peaks of 5 Salix samples. Samples 1 and 2
were experimental replicates taken from plant extracts of the same origin, but with different
batch numbers. Samples 3, 4 and 5 were samples of unknown, but different origin. This figure
shows that experimental variation (Samples 1 and 2) is low, Samples 3 and 5 are highly similar
with respect to their overall composition while Sample 4 is distinctly different from the rest being
placed on the other side of the PCA plot. b Detailed PCA of all mass peaks in the Samples 1 to 5.
The area responsible for the grouping and positioning of Samples 3 and 5 in the bottom right
quadrant is highlighted, and the corresponding mass peaks are shown as their logarithmic ratio
on the right together with the m/z values of each. Light grey: low abundant mass peaks, dark grey:
highly abundant mass peaks
4.2 QTOF MS Coupled to HPLC
As outlined above, direct infusion methods for a (Q)TOF-MS approach are

relatively fast and simple approaches for obtaining metabolic composition
fingerprints of multiple samples which can be used to get a preliminary esti-
mation of the extent of similarities and differences between complex extracts.
However, ion suppression phenomena may result in decreased detection sen-
sitivity of some compounds, especially of those which ionize relatively poorly
(discussed in Kebarle 2000). Moreover, the unavoidable consequences of direct
infusion such as matrix-dependent ion suppression, adduct formation and un-
intended in-source fragmentation, may severely hamper the further detailed
interpretation of the origins of the differential mass signals detected. This will
thus limit possibilities for the subsequent metabolite identification involved. In
addition, with DFI analyses it is, per definition, impossible to discriminate be-
tween molecular isomers or between a quasi-molecular ion and an ion having
identical mass but which resulted from unintended in-source fragmentation.
When such problems arise, or when a more detailed analysis of interesting
samples (preselected, e. g. using a DFI approach) is required, LC separation
can be used to reduce or solve some of these problems.
Separation of metabolites in complex extracts by liquid chromatography,
prior to MS analysis, takes more time but nevertheless has a number of clear
advantages. Sensitivity of detection for most compounds will be increased, the
formation of adducts at the ionization source will be reduced and the detection
of isomeric compounds will be improved. Isomer discrimination is especially
important in plant metabolomics as plants are well-known to contain many
(secondary) metabolites that may have identical accurate mass but different
molecular structures. This is especially true for the large group of flavonoids,
within which many compounds have the same elemental composition (and
consequently the same accurate mass) whereas the chemical structures are
quite distinct, e. g. kaempferol and scutellarein both of which have a neutral
accurate mass of 286.04721. Furthermore, when using chromatographic sepa-
ration, it is also possible to collect additional valuable structural information
by applying, e. g. on-line tandem MS and/or by making use of other molecu-
lar characteristics such as UV-Vis absorbance and fluorescence which can be
detected on-line prior to the molecules entering the MS. It is this combination
of technologies which has made QTOF-based MS analysis a popular choice.
4.2.1 HPLC-PDA-ESI-QTOF-MS
The key to successful, full scale metabolomics analysis is the establishment

of a technology platform which generates the maximum amount of reliable
information in a single analytical run. For example, the LC-TOF MS based
metabolomics system used in our laboratory incorporates a Waters Alliance
2795 HT autoinjector and HPLC pump system fitted with a column oven,
a Waters 2996 photodiode array detector (PDA to give absorbance spectra in
the 190−700 nm range) and a QTOF Ultima API mass spectrometer with MS/MS
capability. In this system, four sets of data are therefore obtained simultane-
ously: UV/vis spectra, retention time, accurate mass and, when applied, MS/MS
fragment information. A lock mass™ spray module is routinely connected to
the ESI source in order to correct, on-the-fly, for any small measurement de-
viations from the exact mass (e. g. Wolff et al. 2001). We routinely use the syn-
thetic peptide leucine enkephalin, which is continuously supplied by a separate
low-flow HPLC pump as a reference lock mass in both positive and negative
ESI measurement modes. By making combined use of accurate mass, MS/MS
fragmentation information and absorbance characteristics, the number of pos-
sible elemental compositions and isomers can be narrowed down and essential
structural information about a specific metabolite can be derived. For instance,
most plant extracts contain multiple (poly)phenolic compounds, among which
many isomers exist. Upon MS/MS fragmentation, many isomeric forms can
already be distinguished, e. g. quercetin-rhamnoside (m/z 449.1079 in ESI
positive mode) provides a fragment of 303.0505 while kaempferol-glucoside
(also m/z 449.1079) provides a positively-charged fragment of 287.0556. How-
ever, with a QTOF, fragmentation experiments are not always conclusive. For
instance, the glycosylated flavonoids kaempferol-3-O-glucoside and cyanidin-
3-O-glucoside have identical mass and elemental composition of C21 H20 O11 ,
and show more or less similar MS/MS fragmentation patterns in ESI-positive
mode with the loss of the glucosidic group leaving C15 H10 O6 as the major
fragment. However, in contrast, their UV-Vis absorbance characteristics are
markedly different, with only the red-coloured cyanidin-glycoside having sig-
nificant absorbance at wavelengths between 500 and 520 nm. This additional
PDA information is therefore key to rapid isomer discrimination in this case.
A typical chromatogram obtained by reversed phase HPLC separation of
a crude plant extract and subsequent on-line detection of eluting compounds
by both PDA and QTOF MS is shown in Fig. 3. The observed mass of the metabo-
lite eluting at retention time 23.48 min was 611.1596. Taking into account the
uneven mass (indicating an even number of nitrogen atoms) and the isotopic
distribution (indicating the absence of sulphur atoms), about 36 different el-
emental compositions are possible at 5 ppm accuracy (9 at 1 ppm accuracy)
Fig. 3. Typical LC-PDA-QTOF MS chromatograms (base peak intensities) obtained by injection

of 5 μl of an aqueous-methanol extract of tomato peel: a photodiode array signal (240−600 nm)
b QTOF-ESI+ -MS signal (m/z 100–1500). Indicated are retention times of the most intense peaks
Fig. 4. a Absorbance spectrum of chromatographic peak at retention time 23.48 min. b QTOF-
ESI-MS/MS spectrum of chromatographic peak at retention time 23.48 min. Observed accurate
mass of parent ion [M+H] + = 611.1596 corresponds to an elemental composition of C27 H31 O21
(−2.6 ppm) and its fragments obtained correspond to C21 H21 O16 (+1.9 ppm) and C15 H11 O7
(−3.9 ppm)
with parameter settings of C ≤ 75, H ≤ 100, O ≤ 75, N ≤ 10 and P ≤ 4.

Subsequent on-line LC-MS/MS fragmentation experiments (Fig. 4b) showed
neutral losses of 146 and 162, and accurate mass fragments ([M+H] + ) of
465.1042 and 303.0493. The accurate mass and MS/MS fragmentation pattern
correspond to a metabolite having an elemental composition of C27 H30 O21 ,
e. g. a diglycosylated anthraquinone, an (iso)flavonoid or a benzoyl-benzoic
acid. The PDA spectrum of the same chromatographic peak (Fig. 4a) showed
two absorbance maxima at around 255 and 360 nm, indicative of a flavonol-
type flavonoid with at least two hydroxyl groups on the B ring (Markham
1989). The combination of accurate mass, MS/MS fragments and UV/vis ab-
sorbance spectrum indicates that the most likely candidate is quercetin-3-O-
rutinoside. This is also supported by the knowledge that this flavonoid has
been reported to be the major flavonol in tomato fruits (Muir et al. 2001;
Le Gall et al. 2003). Subsequent comparison with an authentic standard in-
deed revealed identical chromatography, accurate mass, MS/MS fragmen-
tation pattern as well as absorbance spectrum thus confirming peak iden-
tity.
4.2.2 Metabolomics to Characterize Tomato Mutants
The LC-PDA-QTOF-MS-based platform approach has been shown to be an

effective, reproducible and sensitive method for non-targeted metabolomic
profiling. Sample preparation, chromatographic system and accurate mass
measurements have been optimized in order to screen hundreds of extracts
in an unsupervised stable manner. After unbiased alignment of the chro-
matograms, the data are imported into multivariate analyses software to eluci-
date the biological variables underlying the data structure (Vorst et al. 2005).
As an example of the power of non-targeted metabolomics approaches using
LC-QTOF MS, we recently reported on the effect of a single mutation on the
metabolic profile of ripe fruits of tomato (Bino et al. 2005). In tomato, several
natural photomorphogenic mutants are known and these have been the sub-
ject of detailed physiological investigations. One mutant, carrying one of the
high pigment (hp-1, hp-1w , hp-2, hp-2j , and hp-2dg ) mutations, is characterized
by its exaggerated light responsiveness. Generally, these mutants have higher
pigmentation levels in their hypocotyls, leaves and fruit in comparison to their
semi-isogenic, wild-type counterparts (Levin et al. 2003; van Tuinen et al.
2005). The more intense colour of the fruits is a clear indication that these mu-
tants accumulate more all-trans lycopene in their ripe fruits. However, by using
a metabolomics approach it became clear that the metabolic perturbations in
these fruit were much more extensive than just involving lycopene (Bino et
al. 2005). The hp-2dg mutant and wild-type tomato plants (cv. Manapal) were
grown simultaneously under controlled environmental conditions and fruit
samples were pooled per plant. Different, complementary metabolic profiling
techniques, including GC-MS and LC-MS, were applied to measure as many
compounds as possible in ripe fruits. For non targeted profiling of non-volatile

(semi-polar) compounds, aqueous methanol extracts were prepared and sub-
jected to reversed phase HPLC using both PDA (240−600 nm) and QTOF-MS
(m/z 100–1500; ESI positive and negative modes). A lock mass spray (sampled
every 10 s) was used to enable accurate mass measurements. Raw data were
processed by the metAlign™-software and mass traces were extracted (6168
in negative mode and 5401 in positive mode with a ratio of > 3) and aligned
across all samples. Pair-wise comparisons (Student t-test) could then be used
to determine statistically significant differences between hp-2dg and wild-type
fruit extracts. Differential chromatograms were produced from the original
LC-MS software, and from these (Fig. 5) it was evident that the hp-2dg muta-
tion resulted in a significant increase in many compounds (246 mass signals in
negative mode and 137 in positive mode were > twofold higher) and a decrease
in only a small number of other compounds (57 mass signals being a factor 2
or more lower in negative mode and 5 mass signals twofold lower in positive
mode). The metabolites corresponding to the differential masses were identi-
fied using accurate mass, MS/MS fragmentation experiments and absorbance
spectra (PDA) information. In this way, it was possible to identify a number
of phenolic compounds, flavonoids and alkaloids that were significantly in-
creased in the hp-2dg mutant (Bino et al. 2005) pointing clearly to a pleiotropic
Fig. 5a,b. Metalign™-processed LC-QTOF MS chromatograms (recorded in ESI-negative mode)

showing metabolites that are significantly different (Student t-test, p < 0.01; n = 5) between hp-
2dg and wild-type tomato fruits: a metabolites at least twofold higher in hp-2dg than in wild-type;
b metabolites at least twofold higher in wild-type than in hp-2dg . Retention times and nominal
masses of metabolites are indicated. 100% scale of y-axis (TIC) is 25,000 in a and 500 in b
effect of photomorphogenic mutations on tomato fruit metabolism which was

much greater than was initially visible.
5 Conclusions and Future Prospects
The examples presented in this chapter clearly underline the versatility of

hybrid TOF mass spectrometers, and their capabilities with regard to metabolic
profiling, structure elucidation and compound identification, using accurate
mass determinations and MS/MS fragmentation. The high sensitivity and mass
resolution allows the rapid screening of complex plant extracts by DFI, suitable
for semi quantitative high throughput (pre)screening. More detailed analysis
is possible when MS detection is preceded by applying separation technologies
such as LC. Data processing and efficient data handling are becoming more and
more the bottleneck in the process, especially when high throughput screening
is required and the currently available bioinformatics tools are inadequate.
Another bottleneck is the low number of available reference compounds needed
for definitive identification of differentially accumulating components. Key
developments for the near future will therefore have to be made in these areas if
true plant metabolomics strategies are to become routine. With better software
and more easily mined databases we will be best equipped for the identification
of the large numbers of the highly chemically diverse components typically
present in complex plant extracts.
References
Aharoni A, de Vos CHR, Verhoeven HA, Maliepaard CA, Kruppa G, Bino RJ, Goodenowe DB
(2002) Nontargeted metabolome analysis by use of Fourier Transform Ion Cyclon Mass
Spectrometry. OMICS 6:217–234
Bino RJ, de Vos CHR, Lieberman M, Hall RD, Bovy A, Jonker HH, Tikunov Y, Lommen A, Moco S,
Levin I (2005) The light-hyperresponsive high pigment-2dg mutation of tomato: alterations
in the fruit metabolome. New Phtyologist 166:427–438
Castrillo JI, Hayes A, Mohammed S, Gaskell SJ, Oliver SG (2003) An optimized protocol for
metabolome analysis in yeast using direct infusion electrospray mass spectrometry. Phyto-
chemistry 62:929–937
Chernushevich IV, Loboda AV, Thomson BA (2001) An introduction to quadrupole-time-of-flight
mass spectrometry. J Mass Spectrom 36:849–865
Fernie AR (2003) Metabolome characterization in plant system analysis. Funct Plant Biol
30:111–120
Fernie AR, Trethewey RW, Krotzky AJ, Willmitzer L (2004) Metabolic profiling: from diagnostics
to systems biology. Nature Rev Mol Cell Biol 5:763–769
Fiehn O (2001) Combining genomics, metabolome analysis and biochemical modelling to un-
derstand metabolic networks. Comp Funct Genom 2:155–168
Fiehn O (2002) Metabolomics-the link between genotypes and phenotypes. Plant Mol Biol
48:115–171
Fiehn O, Kopka J, Dörmann P, Altmann T, Trethewey RN, Willmitzer L (2000) Metabolic profiling
for plant functional genomics. Nat Biotechnol 18:1157–1161
Goodacre R, Vaidyanathan S, Bianchi G, Kell DB (2002) Metabolic profiling using direct infusion
electrospray ionisation mass spectrometry for the characterisation of olive oils. Analyst
127:1457–1462
Goodacre R, York EV, Heald JK, Scott IM (2003) Chemometric discrimination of unfractionated
plant extracts analyzed by electrospray mass spectrometry. Phytochemistry 62:859–863
Goodacre R, Vaidyanathan S, Dunn WB, Harrigan GG, Kell DB (2004) Metabolomics by
numbers: acquiring and understanding global metabolomics data. Trends Biotechnol
22:245–252
Guilhaus M (1995) Principles and instrumentation in Time-of-flight mass spectrometry. J Mass
Spectrom 30:1519–1532
Hager JW (2002) A new linear ion trap mass spectrometer. Rapid Commun Mass Spectrom
16:512–526
Hall RD, de Vos CHR, Verhoeven HA, Bino RJ (2005) Metabolomics for the assessment of
functional diversity and quality traits in plants. In: Harrigan G, Vaidyanathan S, Goodacre R
(eds) Metabolic profiling. Kluwer Acad Publ, Dordrecht, Netherlands pp 31–44
Jander G, Norris SR, Joshi V, Fraga M, Rugg A, Yu S, Li L, Last RL (2004) Application of a high-
throughput HPLC-MS/MS assay to Arabidopsis mutant screening; evidence that threonine
aldolase plays a role in seed nutritional quality. Plant J 39:465–475
Kebarle P (2000) A brief overview of the present status of the mechanisms involved in electrospray
King R, Bonfiglio R, Fernandez-Metzler C, Miller-Stein C, Olah T (2000) Mechanistic investigation
of ionization supression in electrospray ionization. J A Soc Mass Spectrom 11:942–950
LeGall G, DuPont MS, Mellon FA, Davis AL, Collins GJ, Verhoeyen ME, Colquhoun IJ (2003) Char-
acterization and content of flavonoid glycosides in genetically-modified tomato (Lycopersicon
esculentum) fruits. J Agric Food Chem 51:2438–2446
Levin I, Frankel P, Gilboa N, Tanny S, Lalazar A (2003) The tomato dark green mutation is a novel
allele of the tomato homolog of the DEFOLIATED 1 gene. TAG 106:454–460
Markham KR (1989) Flavones, flavonols and their glycosides. In: Dey PM, Harborne JB (eds)
Methods in plant biochemistry, vol 1. Academic Press, San Diego, USA, pp 197–235
Muir S, Collins GJ, Robinson S, Hughes S, Bovy A, de Vos CHR, van Tunen AJ, Verhoeyen ME
(2001) Overexpression of petunia chalcone isomerase in tomato results in fruit containing
increased levels of flavonoids. Nat Biotechnol 19:470–474
Nielsen N-PV, Carstensen JM, Smedsgaard J (1998) Aligning of single and multiple wavelength
chromatographic profiles for chemometric data analysis using correlation optimised warp-
ing. J Chromatogr A 805:17–35
Roessner U, Willmitzer L, Fernie AR (2001a) High-resolution metabolic phenotyping of geneti-
cally and environmentally diverse potato tuber systems. Identification of phenocopies. Plant
Physiol 127:746–764
Roessner U, Luedemann A, Brust D, Fiehn O, Linke T, Willmitzer L, Fernie A (2001b) Metabolic
Soga T, Ueno Y, Naraoka H, Matsuda K, Tomita M, Nishioka T (2002) Pressure-assisted capillary
electrophoresis electrospray ionization mass spectrometry for analysis of multivalent anions.
Anal Chem 74:6224–6229
Tolstikov VV, Fiehn O (2002) Analysis of highly polar compounds of plant origin: combining
of hydrophilic interaction chromatography and electrospray ion trap spectrometry. Anal
Biochem 301:298–307
Tolstikov VV, Lommen A, Nakanishi K, Tanaka N, Fiehn O (2003) Monolithic silica-based cap-
illary reversed phase liquid chromatography / electrospray mass spectrometry for plant
Van Tuinen A, de Vos CHR, Hall RD, van der Plas LHW, Bowler C, Bino RJ (2005) Use of
metabolomics for identification of tomato genotypes with enhanced nutritional value derived
from natural light-hyperresponsive mutants. In: Jaiwal PK (ed) Improving the nutritional and
therapeutic qualities of plants. (Plant Metabolic Engineering & Molecular pharming.) SciTech
Publishers, Raleigh, USA (in press)
Vorst OF, de Vos CHR, Lommen A, Staps RV, Visser RGF, Bino RJ, Hall RD (2005) A non
directed approach to the differential analysis of multiple LC/MS-derived metabolic profiles.
Metabolomics 1:169–180
Weckwerth W, Tolstikov V, Fiehn O (2001) Metabolomic characterization of transgenic potato
plants using GC/TOF and LC/MS analysis reveals silent metabolic phenotypes. Abstract:
Proceedings of the 49th ASMS Conference on Mass spectrometry and Allied Topics (1–2)
Wolff JC, Eckers C, Sage AB, Giles K, Bateman R (2001) Accurate mass liquid chromatography /
mass spectrometry on quadrupole orthogonal acceleration time-of flight mass analyzers
using switching between separate sample and reference sprays. 2 Applications using the
dual-electrospray ion source. Anal Chem 73:2605–2612
I.4 Capillary HPLC
T. Ikegami1 , E. Fukusaki2 , and N. Tanaka1
1 Introduction
Among many techniques employed for the separation and identification of

metabolites, HPLC (high performance liquid chromatography)-MS (mass spec-
trometry) is most widely applicable to metabolomics, although the chromato-
graphic efficiency is generally lower than that of the other separation tech-
niques, GC (gas chromatography)-MS or CE (capillary electrophoresis)-MS.
Recently, significant improvement was made to increase the separation capa-
bility of HPLC, which will help the analysis of complex metabolite samples.
In the field of metabolomics, because of the importance of separation and
detection of thousands of small molecules, micro HPLC techniques will be-
come a common method of separation in the near future (Tomita and Nishioka
2003). In this article, the use of long capillary columns that give high separa-
tion efficiencies in micro HPLC system, and multidimensional HPLC that can
provide even higher peak capacity will be described. Special attention will be
paid to the examples of high efficiency HPLC separations made possible by
monolithic silica columns composed of network type silica skeletons.
2 Monolithic Silica Columns for Micro HPLC
Micro HPLC systems with a monolithic silica capillary column possess the
following advantages:
1. Small consumption of stationary and mobile phases
2. High detection sensitivity for a certain amount of samples
3. High speed separation with low pressure drop
4. The possible use of a long column with 1 ∼ 2 m that can provide around
100,000 ∼ 200,000 theoretical plates
along with some disadvantages:
1 Department of Polymer Science and Engineering, Kyoto Institute of Technology, Matsugasaki,

Sakyo-ku, Kyoto, 606–8585, Japan, e-mail: [email protected], [email protected]
2 Department of Biotechnology, Graduate School of Engineering, Osaka Univ, 2–1 Yamadaoka,
Suita, 565–0871, Japan, e-mail: [email protected]

50 T. Ikegami, E. Fukusaki, and N. Tanaka
Table 1. Column sizes, flow rates, linear velocities, and degrees of sample dilution
Column type Inner Column Flow Solvent Relative

t0a
diameter volumea rate linear velocity degree
[mm (μm)] [μl] [μl/min] [s/10 cm] [mm/s] of dilutionb
Conventional 4.6 1660 1000 70 1.4 2100

Semi-micro 2.0 314 200 66 1.5 400
Micro 1.0 78 50 66 1.5 100
0.5 (500) 20 12.5 66 1.5 25
Micro-capillary 0.3 (300) 7.1 5 59 1.7 9
0.2 (200) 3.1 2 66 1.5 4
0.1 (100) 0.78 0.5 66 1.5 1
0.05 (50) 0.20 0.12 69 1.5 0.25
0.025 (25) 0.05 0.03 69 1.5 0.06
a Column lengths were 10 cm, total porosity was estimated as 0.70
b Column of id 100 μm is taken as a standard
1. Smaller sample capacities of a monolithic silica column than particle-packed

columns
2. Necessity of skill and knowledge to operate a capillary HPLC system to
obtain high separation efficiency, and insufficient supply of good columns
and instruments for capillary HPLC
Particle-packed capillary columns have been employed for separations of an-

alytes with or without the assistance of electroosmotic flow. It is possible to
pack silica particles into a fused silica capillary equipped with a frit, but it is
difficult to produce high efficiency and long-lasting columns using 1 ∼ 2 μm
particles (Novotony 1988; Knox and Grant 1991; Schmeer et al. 1995).
Recently, monolithic silica capillary columns have been reported to show
higher separation efficiencies than particle packed columns (Ishizuka et al.
2000; Tanaka et al. 2000). They consist of network silica skeletons that can
be prepared in capillaries by a sol-gel method. Monolithic silica columns
of 4.6 mm ID (inner diameter), 0.2 mm ID, 0.1 mm ID and 0.05 mm ID are
commercially available at present. Column sizes and flow rates to be employed
are listed in Table 1.
2.1 Characteristics of Monolithic Silica Columns
Here, the features of monolithic silica capillary columns and the optimization
of separation conditions will be described. The use of monolithic silica columns
consisting of network silica skeletons and through-pores for micro HPLC was
reported recently (Minakuchi et al. 1996; Tanaka et al. 2001). Monolithic silica
capillary columns were reported to provide better separation efficiencies than
particle-packed columns, and the use of these columns for proteomics and
Capillary HPLC 51
Fig. 1. Scanning electron microscope images of monolithic silica prepared from sol-gel methods:
a monolithic silica prepared in a test tube; b,c monolithic silica prepared in 50 μm ID fused silica
capillary; d monolithic silica prepared in 100 μm ID fused silica capillary; e monolithic silica
prepared in 200 μm ID fused silica capillary tube
metabolomics seems to be attractive (Cabrera 2004). Monolithic silica columns

are prepared by acid-catalyzed hydrolytic polymerization of alkoxysilanes in
the presence of water-soluble polymers such as poly(ethylene glycol) (Tanaka
et al. 2001; Cabrera 2004). Figure 1a shows a scanning electron microscope
(SEM) image of monolithic silica prepared in a test tube, while Fig. 1b–e shows
SEM images of monolithic silica columns prepared in fused silica capillaries
with 50 ∼ 200 μm internal diameter (Motokawa et al. 2002).
Currently available monolithic capillary columns include organic polymer
columns (Svec 2004) and chemically modified silica columns, and they have
the following features. Monolithic polymer columns generally show higher
permeability than particle-packed columns, and high efficiency for the sep-
aration of macromolecules (Svec et al. 2000). In the case of monolithic silica
capillary columns, silica skeletons are covalently bonded to capillary walls.
Thus, frits are not necessary to hold the skeletons in a column, and column
length can be varied in the range of 5 ∼ 200 cm after preparation. Generally,
the silica skeleton sizes are in the range of 1 ∼ 2 μm. As shown in Fig. 1a,
monolithic columns have 3 ∼ 10 times bigger (through-pore size/skeleton size)
ratio, 1 ∼ 3, than particle-packed columns with (through-pore size/particle
size) ratio, 0.25 ∼ 0.4. Monolithic silica columns produce similar separation ef-
ficiencies to particle-packed columns at much lower pressure drop. At the same
pressure drop, monolithic columns can provide higher separation efficiencies
than particle-packed columns. Moreover, due to the small silica skeleton sizes,
relatively high separation efficiencies can be expected at higher linear veloc-

ities (Minakuchi et al. 1997, 1998). In terms of separation impedance, total
performance of columns (E), monolithic silica capillary columns can produce
higher separation efficiencies, nearly 10 times greater than that of a particle-
packed column (Motokawa et al. 2002). Separation impedance is given by
Eq. (1) where N, ΔP, t0 and η stand for number of theoretical plates, column
back pressure, the elution time of an unretained solute, and viscosity of mobile
phase, respectively (Bristow and Knox 1977):
E = t0 ΔP/N 2 η = (ΔP/N)(t0 /N)(1/η) (1)
Figure 2 shows chromatograms produced by monolithic silica capillary

columns of 25 ∼ 130 cm lengths modified with C18 stationary phase (Ishizuka
et al. 2002).
The dilution factors for analytes are proportional to the internal diameters of
columns squared assuming that band broadening (peak width) and resolutions
are similar for various HPLC systems. Sample concentrations after the separa-
tion are higher in micro columns with smaller diameters, and the higher sample
concentrations can lead to higher detection sensitivity (Table 1). Because lower
flow rates can lead to higher ionization efficiencies and higher detection sen-
sitivity in LC-ESI (electrospray ionization) MS system, development of high
Fig. 2. Chromatograms obtained for alkylbenzenes (C6 H5 (CH2 )n H, n = 0 − 6) by: a–d C18 mono-
lithic silica capillary columns; e particle packed column (5 mm silica-C18 particles, Mightysil
RP18)
Capillary HPLC 53
efficiency micro HPLC system is an important issue for metabolomics studies

(Schmidt et al. 2003).
2.2 Column Efficiencies and the Optimization

of Separation Conditions
The number of theoretical plate N is a measure of the quality of a column and

elution conditions, and is given by Eq. (2) from the retention time of a peak (tR)
and peak width at half height (tw1/2 = 2.35σ , σ being the standard deviation of
a Gaussian peak). Resolution Rs is given by Eq. (4), that includes N, α (Eq. (5),
selectivity, the ratio of retention factors of two adjacent peaks), and k (Eq. (3),
a retention factor, distribution coefficient of a solute between stationary and
mobile phases, i. e. the ratio of times (tR –t0 ) to t0 , the former stands for time
the solute exists in mobile phase, and the latter stands for time the solute exists
in stationary phase). For convenient separation and detection, the k values
should be in a range of 2 ∼ 5:
N = (tR /σ )2 = 5.54(tR /tw1/2 )2 = 16(tR /tw )2 (2)

k = (tR − t0 )/t0 (3)
Rs = (N 1/2 /4)[(α − 1)/α][k/(1 + k)] (4)
α = k2 /k1 (5)
ΔP = φηuL/dp2 (u = L/t0 ) (6)
ΔP is proportional to η, u (linear velocity of the mobile phase), and L (column

length) while it is inversely proportional to dp2 , where dp stands for diameter of
particle. Thus, a column packed with particles of small diameter leads to high
separation efficiency, (greater N) at the expense of high column backpressure.
Due to the drawback, an approach to get high efficiencies by reducing diameter
of particles has a limit: since the pressure limit of a pump system is around
300 ∼ 400 bar with a normal operational pressure 100 ∼ 200 bar, the limit in
particle sizes is in a range of 1 ∼ 3 μm. The flow resistance parameter φ in
Eq. (6), is usually ca. 2000 for particle-packed columns, while φ values reach
to 200 ∼ 400 in the case of monolithic silica columns (Giddings 1965; Bristow
and Knox 1977).
A solute band is broadened when it travels outside a column due to parabolic
flow profile in a tube as well as due to slow diffusion in the stagnant mobile phase
existing in an injector, a detector, or connection tubing. Especially, for solutes
of small retention factors which elutes in early part of a chromatogram, sample
injection into a capillary column of 1 ∼ 5% of column volume has significant
influence on band spreading, mainly caused by sample diffusion at orifice in
an injector or by dead volume in all connection parts (Ikegami at al. 2004).
The split-flow injection technique is practical and useful for micro HPLC with
monolithic silica capillary columns in order to avoid the peak spreading during
injection (Taniguchi and Murata 2002). Moreover, the use of weak eluents for
sample injection is also effective to increase the separation efficiency: in the
case of reversed-phase HPLC, sample solution can be prepared with water-rich
solvent (Ikegami at al. 2004).
3 Applications of Monolithic Silica Columns to Metabolomics
Figure 3 shows chromatograms of leaf extracts of Arabidopsis thaliana by LC-

ESI-MS using 30 ∼ 90 cm monolithic silica capillary columns modified with
C18 stationary phase under gradient conditions, from aqueous ammonium
acetate buffer (pH 5.5) to acetonitrile, MeCN (Tolstikov et al. 2003). A shallow
gradient (large tG , gradient time) with a long column has lead to better sepa-
ration. The results indicate that improvement of separation by the use of the
longer columns caused the reduction of ion suppression effect by introducing
the solute bands separately into ES ionization interface. In the case of Fig. 3, the
peak capacity provided by the long monolithic silica column is still not enough
for complete separation, but it shows a feasible approach of using longer mono-
lithic silica capillary columns to achieve higher separation efficiency avoiding
ion suppression effect in the LC-ESI-MS system. This approach will result in
longer separation time, but the amount and quality of information after the
analysis of metabolites would be better than conventional LC-MS systems us-
ing particle packed columns. Connected monolithic columns (conventional
size) in series showed good separation of polyprenol homologues (Bamba et
al. 2004).
Mass spectrometry would often be used in metabolomics research due to
its superiority in both quantification and qualification. However, mass spec-
trometry has a serious drawback named ‘ionization suppression’. Ionization
suppression is a phenomenon that presence of impurity at ionization might
cause a serious impairment in qualitative accuracy (King et al. 2000; Müller
et al. 2002). Coelution in chromatography might cause ionization suppres-
sion. Even the technology of capillary monolithic chromatography might not
provide a perfect time separation that is one of the ideal solutions against
ionization suppression. Recently, stable isotope dilution technology tends to
be used as a practical tool to reduce an ionization suppression negative ef-
fect. A stable isotope dilution method employs isotopologues as an internal
standard that would be separated not by chromatography but by mass spec-
trometry to provide an accurate comparative quantification. This principle is
used in the proteomics research tool ‘isotope coded affinity tags (ICAT)’ (Han
et al. 2001). A metabolic profiling of sulfur metabolite using 34 S was reported
(Mougous et al. 2002). 13 C and 15 N stable isotope labeling techniques could be
available in some case. In addition, post sampling stable isotope labeling would
be also applicable, although D-labeling may face some difficulties (Zhang et
al. 2001; Fukusaki et al. 2005). In future a combination of monolithic capillary
Capillary HPLC 55
chromatography and stable isotope diluted comparative quantification would

be one of the de facto standard methods in metabolomics.
4 Two-Dimensional HPLC
Peak capacity (PC) given by Eq. (7) indicates the separation ability regarding
how many solutes can be potentially separated by a chromatographic system.
Retention times of the first solute and the last solute are given as t1 and tR
respectively in Eq. (7). Separation methods such as ultrahigh-pressure liquid
chromatography (UHPLC) and supercritical fluid chromatography (SFC) can
produce a PC of ca. 300/h (Shen and Lee 1998; MacNair et al. 1999), while
a conventional HPLC system gives a PC of 100 ∼ 200/h. In order to achieve
far larger PC using conventional HPLC systems, multidimensional separation
systems were shown to be effective. When two chromatographic systems with
PCx and PCy are combined to form a two-dimensional (2D) chromatography
system, PC for the total system can be theoretically estimated as a product of
two PC values as Eq. (8) (Giddings 1991):
PC = 1 + (N 1/2 /4)ln(tR /t1 ) (7)

PC2D = PCx × PCy (8)
In comprehensive 2D-HPLC separations every fraction obtained from 1st-D

separation is to be separated in 2nd-D HPLC, while the next fraction is eluted
from 1st-D. Therefore the 2nd-D column should ideally be eluted at very high
speed to meet the rate of fractionation at the 1st-D separation. The 2nd-D col-
umn should possess low-pressure drop and reasonable efficiency at high flow
rate. In addition to high efficiency and high permeability, the 1st-D and 2nd-D
columns must possess adequate difference in selectivity to effect 2D separa-
tions. Ideally the 1st-D and 2nd-D should have orthogonal selectivity or differ-
ent separation mechanisms (Bushey and Jorgenson 1990; Köhne and Welsch
1999; Wagner et al. 2002; Venkataramani and Zelechonok 2003). Ion-exchange
mode and reversed-phase mode, or size-exclusion mode and reversed-phase
mode have often been combined to effect 2D separations for peptide mixtures
in proteomics. Because a particle-packed column cannot be operated at ad-
equately high flow rate, various approaches were taken in the past: (i) small
columns were employed at 1st-D compared to 2nd-D, (ii) the first column was
eluted slowly or intermittently, or (iii) two or more sets of chromatographs were
used at the 2nd-D. Even with these methods, however, truly “two-dimensional”
HPLC is hard to achieve due to the mixing of separation modes.
Figure 4 shows a scheme of 2D-HPLC and its working principle, in which
the outlet tubing of the 1st-D column was connected to a loop of the 2nd-D
injector to couple particle the packed 1st-D column and 2nd-D monolithic
silica column run at higher linear velocity (Tanaka et al. 2004). In this case, the
Capillary HPLC
Fig. 3. Replicate injections of an Arabidopsis leaf methanol extract on capillary monolithic C18 columns in positive ionization fullscan MS, given as base peak
chromatograms. Upper panel 0.2 mm ID, 300 mm long; middle panel 0.2 mm ID, 600 mm long; lower panel 0.2 mm ID, 900 mm long column; t0 , void volume
57
Fig. 4. a Tubing connection at 2nd-D injector of simple 2D-HPLC. b Tubing connection of two
six-port valves used as 2nd-D injector
fraction from the 1st-D column is loaded and temporarily kept in a loop of the
2nd-D injector that results in mixing of separated peaks, but the flow rates of
two HPLC systems can be controlled independently. The 2nd-D separation can
be carried out at very high flow rate (for example, 10 ml/min for a 4.6 mm ID
column) throughout the separation. The simplest 2D-HPLC in Fig. 4a produced
PC = 1000 in reversed phase mode. When two six-port valves or a ten-port
valve is used at the 2nd-D HPLC in Fig. 4b, all fractions can be subjected to the
separation at the 2nd-D column to provide a comprehensive 2D-HPLC system
resulting in so-called group separation, solutes of similar structural features
appear as a group. Because of fast flow rate in the 2nd-D separation using
a 4.6 mm ID column, the 2D-HPLC system consumed a lot of mobile phase sol-
vent. In order to reduce the consumption of mobile phases, the sufficiently fast,
simple 2D-HPLC using capillary columns has been examined (Kimura et al.
2004). The use of capillary column at 2nd-D leads to less solvent consumption
and better MS detectability compared to a larger-sized column. Figure 5a shows
a 2D chromatogram for the tryptic digest of BSA (Bovine serum albumin) ob-
tained from total ion monitoring by ESI-TOF (Time of flight)-MS. From the
1st-D (2.1 mm ID, 5.0 cm long), 18 fractions were injected at 2-min intervals
into the 2nd-D reversed-phase system (4.6 mm ID, 2.5 cm long), generating
18 chromatograms that were used to produce a 2D chromatogram. Figure 5b
shows a 2D chromatogram obtained for the separation of tryptic digest of BSA
using a capillary column (100 μm ID, 10 cm long) in the 2nd D separation. The
number of spots distinguishable in vertical direction in Fig. 5b was greater than
that in Fig. 5a. This is due to the higher column efficiency and longer gradient
time in 2nd-D, along with greater MS detection sensitivity based on nearly
optimum flow rate (3 μL/min) on the capillary column, the greater amount of
sample introduced to the 2nd-D column because of the longer fractionation
interval, and the smaller extent of dilution due to the use of small diameter
column (Kimura et al. 2004).
Capillary HPLC 59
Fig. 5. Two-dimensional separation of tryptic digest of BSA in simple 2D-HPLC, 1st-D; MCI
CQK-31S column (2.1 mm ID, 50 mm long), flow rate; 50 μl/min: a 2nd-D; monolithic silica-C18
column (4.6 mm ID, 25 mm long), flow rate; 5.0 ml/min; b 2nd D; C18 monolithic column (0.1 mm
ID, 100 mm long), flow rate in a capillary column; 3.0 μl/min with a split flow/injection; linear
velocity in the column; 7.7 mm/s. ESI-TOF-MS detection, total ion chromatogram for a mass
range 400–2000
5 Combination of Reversed-Phase HPLC

and Other Separation Modes
Since many compounds of similar properties are to be separated in proteomics,

2D-HPLC hyphenated to an MS system has been employed combining ion ex-
change mode and reversed phase mode, or size-exclusion mode and reversed-
phase mode. In the case of metabolomics, combination of several different sep-
aration modes is preferable to separate a variety of substances. Reversed-phase
mode is most often employed in HPLC, where chemically bonded stationary
phases (C8, C18, and C30, etc.) have advantages in rapid equilibration with
mobile phase, high separation efficiency, and high reproducibility in gradi-
ent.
Recently, hydrophilic LC (HILIC LC) (Alpert et al. 1994; Yoshida 1997)
was shown to be effective for the separation of metabolomes utilizing the
interaction between solutes and hydrophilic functional groups on the sta-
tionary phases. The selectivities of HILIC columns are similar to those of
a conventional silica column, but HILIC columns have advantages over silica
columns in the recovery of samples, and compatibility with mobile phases
used in reversed-phase mode. Figure 6 shows a comparison of elution pat-
terns of HILIC mode and reversed-phase mode in the separation of an ex-
tract from Arabidopsis thaliana (Tolstikov and Fiehn 2002). Since the solvent
type for HILIC and reversed-phase mode are common, it is possible to com-
bine the two separation modes to form multidimensional HPLC, although
60
Fig. 6. Comparison of chromatograms of an Arabidopsis thaliana leaf methanol extract, obtained by HILIC-LC mode (top panel) and reversed-phase mode
(bottom panel): Conditions (top panel) TSK Gel Amide 80, 4.6 mm ID, 150 mm long, gradient elution from MeCN to ammonium acetate buffer (6.5 mmol/l,
pH 5.5), MeCN content (%) (time, min) 100 → 100(5) → 90(8) → 60(75) → 0(80), (bottom panel) C18 column, 4.6 mm ID, 150 mm long, gradient elution
T. Ikegami, E. Fukusaki, and N. Tanaka
from ammonium acetate buffer (6.5 mmol/l, pH 5.5) to MeCN, MeCN content (%) (time, min) 0→ 0(15) → 95(40) → 100(60) → 100(80)
Capillary HPLC 61
the compositions of mobile phases that controls the retention order are total
opposites to each other. Capillary columns for HILIC LC are under develop-
ment.
6 Outlook
Routine use of micro HPLC will need development of several important con-
stituents; the reproducible preparation of high performance columns, small-
volume pumps and gradient systems, and improvement of an injection system.
Subjects to be studied are the development of high performance monolithic
silica columns for variety of separation modes, multidimensional microLC
systems, and optimization of an interface between LC and MS instruments.
Large peak capacities realized by highly efficient microHPLC systems or mul-
tidimensional HPLC will greatly contribute to metabolomics studies when
coupled with MS instruments and stable isotope dilution methodology.
References
Alpert AJ, Shukla M, Shukla AK, Zieske LR, Yuen SW, Ferguson MAJ, Mehlert A, Pauly M, Orlando
R (1994) Hydrophilic-interaction chromatography of complex carbohydrates. J Chromatogr
A 676:191–202
Bamba T, Fukusaki E, Nakazawa Y, Kobayashi A (2004) Rapid and high-resolution analysis of
geometric polyprenol homologues by connected octadecylsilylated monolithic silica columns
in high-performance liquid chromatography. J Sep Sci 27:293–296
Bristow PA, Knox JH (1977) Standardization of test conditions for high performance liquid
chromatography columns. Chromatographia 10:279–289
Bushey MM, Jorgenson JW (1990) Automated instrumentation for comprehensive two-
dimensional high-performance liquid chromatography of proteins. Anal Chem 62:161–167
Cabrera K (2004) Application of silica-based monolithic HPLC columns. J Sep Sci 27:843–852
Fukusaki E, Harada K, Bamba T, Kobayashi A (2005) An isotope effect on the comparative
quantification of flavonoids by means of methylation-based stable isotope dilution coupled
with capillary liquid chromatograph/mass spectrometry. J Biosci Bioeng 99:75–77
Giddings JC (1965) Dynamics of chromatography, part 1. Principles and theory. Dekker, New
York
Giddings JC (1991) Unified separation science. Wiley-Interscience, New York, pp 126–128
Han DK, Eng J, Zhou H, Aebersold R (2001) Quantitative profiling of differentiation-induced
microsomal proteins using isotope-coded affinity tags and mass spectrometry. Nat Biotechnol
19:946–951
Ikegami T, Dicks E, Kobayashi H, Morisaka H, Tokuda D, Cabrera K, Hosoya H, Tanaka N (2004)
How to utilize the true performance of monolithic silica columns. J Sep Sci 27:1292–1302
Ishizuka N, Minakuchi H, Nakanishi K, Soga N, Nagayama H, Hosoya K, Tanaka N (2000) Perfor-
mance of a monolithic silica column in a capillary under pressure-driven and electrodriven
conditions. Anal Chem 72:1275–1280
Ishizuka N, Kobayashi H, Minakuchi H, Nakanishi K, Hirao K, Hosoya K, Ikegami T, Tanaka N
(2002) Monolithic silica columns for high-efficiency separations by high-performance liquid
chromatography. J Chromatogr A 960:85–96
Kimura H, Tanigawa T, Morisaka H, Ikegami T, Hosoya K, Ishizuka N, Minakuchi H, Nakanishi K,

Ueda M, Cabrera K, Tanaka N (2004) Simple 2D-HPLC using a monolithic silica column for
peptide separation. J Sep Sci 27:897–904
King R, Bonfiglio R, Fernandez-Metzler C, Miller-Stein C, Olah T (2000) Mechanistic investigation
of ionization suppression in electrospray ionization. J Am Soc Mass Spectrom 11:942–950
Knox JH, Grant IH (1991) Electrochromatography in packed tubes using 1.5 to 50 μm silica gels
and ODS bonded silica gels. Chromatographia 32:317–328
Köhne AP, Welsch T (1999) Coupling of a microbore column with a column packed with non-
porous particles for fast comprehensive two-dimensional high-performance liquid chro-
matography. J Chromatogr A 845:463–469
MacNair JE, Patel KD, Jorgenson JW (1999) Ultrahigh-pressure reversed-phase capillary liquid
chromatography: isocratic and gradient elution using columns packed with 1.0 mm particles.
Minakuchi H, Nakanishi K, Soga N, Ishizuka N, Tanaka N (1996) Octadecylsilylated porous silica
rods as separation media for reversed-phase liquid chromatography. Anal Chem 68:3498–
3501
Minakuchi H, Nakanishi K, Soga N, Ishizuka N, Tanaka N (1997) Effect of skeleton size on the
performance of octadecylsilylated continuous porous silica columns in reversed-phase liquid
Minakuchi H, Nakanishi K, Soga N, Ishizuka N, Tanaka N (1998) Effect of domain size on the
performance of octadecylsilylated continuous porous silica columns in reversed-phase liquid
Motokawa M, Kobayashi H, Ishizuka N, Minakuchi H, Nakanishi K, Jinnai H, Hosoya K, Ikegami T,
Tanaka N (2002) Monolithic silica columns with various skeleton sizes and through-pore sizes
for capillary liquid chromatography. J Chromatogr A 961:53–63
Mougous JD, Leavell MD, Senaratne RH, Leigh CD, Williams SJ, Riley LW, Leary JA, Bertozzi CR
(2002) Discovery of sulfated metabolites in mycobacteria with a genetic and mass spectro-
metric approach. Proc Natl Acad Sci USA 99:17037–17042
Müller C, Schäfer P, Störtzel M, Vogt S, Weinmann W (2002) Ion suppression effects in liq-
uid chromatography-electrospray-ionisation transport-region collision induced dissociation
mass spectrometry with different serum extraction methods for systematic toxicological
analysis with mass spectra libraries. J Chromatogr B 773:47–52
Novotny M (1988) Recent advances in microcolumn liquid chromatography. Anal Chem 60:500A–
510A
Schmeer K, Behnke B, Bayer E (1995) Capillary electrochromatography – electrospray mass
spectrometry: a microanalysis technique. Anal Chem 67:3656–3658
Schmidt A, Karas M, Dülcks T (2003) Effect of different solution flow rates on analyte ion signals
in nano-ESI MS, or: when does ESI turn into nano-ESI. J Am Soc Mass Spectr 14:492–500
Shen Y, Lee ML (1998) General equation for peak capacity in column chromatography. Anal Chem
70:3853–3856
Svec F (2004) Preparation and HPLC applications of rigid macroporous organic polymer mono-
liths. J Sep Sci 27:747–766
Svec F, Peters EC, Sýkora D, Yu C, Fréchet JMJ (2000) Monolithic stationary phases for capil-
lary electrochromatography based on synthetic polymers: designs and applications. J High
Resolut Chromatogr 23:3–18
Tanaka N, Nagayama H, Kobayashi H, Ikegami T, Hosoya K, Ishizuka N, Minakuchi H, Nakan-
ishi K, Cabrera K, Lubda D (2000) Monolithic silica columns for HPLC, micro-HPLC, and
CEC. J High Resolut Chromatogr 23:111–116
Tanaka N, Kobayashi H, Nakanishi K, Minakuchi H, Ishizuka N (2001) Monolithic LC columns.
Anal Chem 73:420A–429A
Tanaka N, Kimura H, Tokuda D, Hosoya K, Ikegami T, Ishizuka N, Minakuchi H, Nakanishi K, Shin-
tani Y, Furuno M, Cabrera K (2004) Simple and comprehensive two-dimensional reversed-
phase HPLC using monolithic silica columns. Anal Chem 76:1273–1281
Capillary HPLC 63
Taniguchi H, Murata Y (2002) The newest protocol of proteomics 9, Capillary HPLC. Cell Tech
21:1332–1343
Tolstikov VV, Fiehn O (2002) Analysis of highly polar compounds of plant origin: combination of
hydrophilic interaction chromatography and electrospray ion trap mass spectrometry. Anal
Biochem 301:298–307
Tolstikov VV, Lommen A, Nakanishi K, Tanaka N, Fiehn O (2003) Monolithic silica-based
Tomita M, Nishioka T (2003) Frontier of metabolomics. Springer, Berlin Heidelberg New York
Venkatramani CJ, Zelechonok Y (2003) An automated orthogonal two-dimensional liquid chro-
matograph. Anal Chem 75:3484–3494
Wagner K, Miliotis T, Marko-Varga G, Bischoff R, Unger KK (2002) An automated on-line multidi-
mensional HPLC system for protein and peptide mapping with integrated sample preparation.
Yoshida T (1997) Peptide separation in normal phase liquid chromatography. Anal Chem 69:3038–
3043
Zhang R, Sioma CS, Wang S, Regnier FE (2001) Fractionation of isotopically labeled peptides in
quantitative proteomics. Anal Chem 73:5142–5149
I.5 Capillary HPLC Coupled to Electrospray Ionization
Quadrupole Time-of-flight Mass Spectrometry
S. Clemens, C. Böttcher, M. Franz, E. Willscher,
E. v. Roepenack-Lahaye, and D. Scheel1
1 Introduction
Metabolite profiling in the pre-metabolomics era of the early 1970s to the late
1990s as well as the pioneering metabolomics projects since the late 1990s
have been predominantly GC-MS based. GC-MS techniques are robust and
well-established. Many primary metabolites (e. g. organic acids, sugars, amino
acids, sugar alcohols) can easily be derivatized and are therefore amenable to
GC-MS analysis. Also, spectral databases and deconvolution algorithms are
available, which help extracting meaningful information. Early on, however, it
was obvious that no single analytical technique would be sufficient to achieve
comprehensive coverage of the metabolome (Sumner et al. 2003). As stated
from the beginning and reiterated since, the chemical diversity of metabolites
makes it virtually impossible to detect all compound classes in one “catch”
(Goodacre et al. 2004; Dunn et al. 2005). That is why already the first reports
describing GC-MS-based metabolomics platforms emphasized the need to
develop complementing LC-MS platforms (Roessner et al. 2000). LC-MS covers
in principle a much wider mass range and should allow one to target many
compound classes not detectable by GC-MS. Furthermore, there is usually
no need for derivatization and LC-MS offers superior options to elucidate
unknown metabolites structurally. Particular fractions can easily be collected
for NMR analysis and metabolites/molecular ions can be further analyzed by
tandem-MS or even MSn . Hampering the adoption of LC-MS approaches for
metabolomics, however, was the fact that LC-MS has only rather recently (i. e.
in the 1990s) developed into a routine technology (Niessen 1999a).
One might argue that the need for LC-MS-based profiling is even more
pressing in plant science. A highly rich and diverse secondary metabolism is
a hallmark of plant biology. Lacking the ability to avoid or to retreat from
unfavorable conditions or potential foes, plants have evolved an enormous
metabolic plasticity, which allows them to respond dynamically to environ-
mental changes through the synthesis and/or degradation of particular com-
pounds. This is complemented by the accumulation of various pre-formed
defenses against microbial attack and other threats (Dixon 2001). Further-
more, many so-called secondary metabolites also apparently play major roles
in primary developmental processes and as signaling molecules. Flavonoids
1 Leibniz
Institute of Plant Biochemistry, Weinberg 3, 06120 Halle/Saale, Germany, e-mail:
[email protected]

66 S. Clemens et al.
and their biosynthesis, for instance, have long been investigated because of
their role in flower pigmentation, UV protection, or pathogen defense (Winkel-
Shirley 2001). More recent work demonstrated that flavonoids negatively reg-
ulate auxin transport and are required for pollen germination (Taylor and
Grotewold 2005).
A large fraction of plant secondary metabolites has been classically analyzed
by LC techniques, predominantly through separation on reversed phase mate-
rial. Thus, it is a straightforward concept to combine this with state-of-the-art
mass spectrometry in order to develop powerful metabolomics platforms that
cover important compound classes such as phenylpropanoids or alkaloids.
A look at Arabidopsis thaliana, the most important plant model species, can
illustrate the need for and the potential of LC-MS profiling. Because A. thaliana
has no history of use as a medicinal plant, it initially did not attract the atten-
tion of too many natural product chemists. As a consequence, few secondary
metabolites were identified 10 years ago. In the course of the genome sequenc-
ing, however, it became increasingly clear, that A. thaliana should produce
thousands of different compounds. The Arabidopsis genome encodes a myriad
of proteins likely to be involved in secondary metabolism (d’Auria and Ger-
shenzon 2005). There are more than 270 cytochrome P450 genes, more than
100 glycosyl transferase genes, about 50 glutathione S-transferase genes, to
name a few. For most of the encoded enzymes we do not know substrates or
products.
The first major challenge for metabolomics is the huge chemical diversity
of the metabolome. The second lies in the fact that – as indicated above for
Arabidopsis thaliana – most of the metabolites in any given higher eukaryote
are unknown. Current estimates are in the range of 4000–20,000 metabolites
for a given species (Fernie et al. 2004). Unlike for proteins, genome sequences
do not allow one to deduce the structure of the metabolites. Instead, the
structure has to be elucidated because for only a very minor portion of the
metabolites are standards available. Thus, the future success of metabolomics
will also be determined by the ability to identify reliably metabolites and
to establish the metabolomes of the important model species. Again, this is
a particularly daunting task for plants and filamentous fungi, organisms that
synthesize huge numbers of secondary metabolites, many of which might only
be synthesized in certain cell types or at particular developmental stages. LC-
MS, especially in the combination of quadrupole and time-of-flight analysis in
modern hybrid instruments, holds the promise to meet this challenge as well.
Structural information can in principal be obtained in three different ways:
(i) by determining the elemental composition through the accurate mass, (ii)
by exploiting the information provided by in-source fragmentation, and (iii)
by performing targeted CID-MS (collision-induced dissociation). In contrast,
GC-MS-based profiling faces severe limitations when it comes to de novo
identification of unknown compounds (Fiehn 2002). Molecular ions are rarely
detected because most analytes are derivatized and molecules are fragmented
by the electron impact ionization.
Capillary LC-ESI-QTOF-MS-based profiling 67
We will in the following discuss the principles of capillary LC-MS-based pro-

filing, describe the current state and present new data from our own laboratory
on the optimization and the potential of capillary LC coupled to electrospray
ionization quadrupole time-of-flight mass spectrometry (CapLC-ESI-QTOF-
MS) (von Roepenack-Lahaye et al. 2004).
2 Extraction, Chromatography and Mass Spectrometry
When optimizing extraction and chromatographic separation of low molecu-

lar weight compounds, there are various considerations which are common-
place in analytical chemistry (Niessen 1999b) and which will therefore only be
touched upon very briefly. The extraction of biological material with aqueous
methanol has so far been the most widely used option for GC-MS as well as
LC-MS metabolite profiling schemes (Roessner et al. 2000; Fiehn et al. 2000;
Tolstikov et al. 2003). For the sake of stability of compounds and reproducibility
of the analysis, cold extraction is preferred in most cases. Obviously, the choice
of solvent greatly influences scope and range of the profiling. We tested, for
instance, acetonitrile-water and methanol-water mixtures for the extraction
of Arabidopsis thaliana seeds and counted simply the number of mass signals
with a signal to noise ratio > 5 by analyzing subsets of the resulting LC-MS
chromatograms with MetAlign (www.metalign.nl) (see below). We detected
1680 mass peaks in an 80% methanolic seed extract and 1771 signals in a 50%
methanolic seed extract. Of these signals, 973 were found in both extracts.
Utilization of acetonitrile-water gave comparable results: upon extraction with
80% and 50% acetonitrile, 2070 and 1771 mass peaks, respectively, were de-
tected and 1029 were found in both extracts. Only 532 mass peaks were detected
in all extracts.
There are several classical analytical options to enrich selectively certain
compound classes by modifying the extraction. Solid-phase extraction can be
used to remove problematic compounds such as lipids or to concentrate others
that are of interest but give low signal intensity. These different options and
their effect on the metabolome coverage of LC-MS approaches have not been
systematically evaluated yet. A way of selectively targeting specific classes
of molecules is derivatization. This can permit analysis of compounds with
inadequate stability and results in better chromatographic behaviour as well
as enhanced signal intensity. Also, derivatization has been proposed as a means
to make the ionization of diverse analytes more uniform by adding a particular
chemical group (Halket et al. 2005).
A major obstacle in the development of LC-MS was the general incompat-
ibility of flow rates between LC and MS, i. e. the need to introduce a column
effluent of about 1 mL/min into a high vacuum (Niessen 1999a). One solution
to this problem was to reduce the flow rate by miniaturization of the LC col-
umn, a second to split the column effluent so that only a fraction reaches the
mass spectrometer. Often these two options are combined. In capillary liquid
chromatography the flow rate is reduced to meet the optimum flow rate range
characteristic for many ESI interfaces. Splitting occurs – if at all – prior to
chromatography between the pump and the column. Chromatography is per-
formed at low flow rates of 2−20 μL/min (Abian et al. 1999). Column diameters
are typically between 80 and 800 μm. In principle, MS is a mass flow sensi-
tive detection because the response is proportional to the actual number of
molecules reaching the detector. However, at a constant flow rate under atmo-
spheric pressure ionization conditions, MS acts as a concentration sensitive
detector, i. e. the signal is proportional to the analyte concentration in the elu-
ent (Niessen 1999a). The smaller diameter of a capillary column as compared
to a regular 4.6-mm analytical column combined with a lower flow rate allows
the use of much smaller sample volumes and lower sample concentrations.
Furthermore, depending on the design of the ESI interface a reduced flow-rate
can result in higher sensitivity due to the enhanced ionization yield of the
smaller primary droplet formation (Wilm and Mann 1994). Thus, since the
mid 1990s there has been a trend towards miniaturization of the LC (Abian et
al. 1999), although the better sensitivity – i. e. lower concentration detection
limits – is partly offset by the need to reduce the injection volume and by the
lower capacity of the column.
It is advisable to inject as small a volume as possible (and reproducible)
in a solvent of low eluotropic strength. Otherwise, retention on the station-
ary phase is incomplete and many compounds will elute partly in the flow-
through. Furthermore, separation could be seriously disturbed, which results
in unsymmetrical peak shapes and altered retention times. Figure 1 shows
the extracted ion chromatograms, which correspond to the molecular ion
of 4-glucopyranosyloxybenzoyl choline, a secondary metabolite identified in
methanolic seed extracts, injected in either 2 μL 80% methanol (a) or 2 μL 10%
Fig. 1. Influence of solvent on the retention and separation. Extracted ion chromatograms (XIC
412.0–412.5) showing the altered retention behaviour of 4-glucopyranosyloxybenzoyl choline
from a seed extract upon injection in different injection solvent mixtures: a 80% methanol –
a fraction of the metabolite elutes in the flow-through (tR = 6.60 min); b 10% methanol –
diastereomers are retained on the column and baseline-separated (tR = 16.80, tR = 20.73 min)
methanol (b) following separation on C18 phase with hydrophilic end-capping.

In case of an injection in 80% methanol most of the compound is eluted with-
out any retention, whereas upon injection in 10% methanol the compound is
retained on the stationary phase and both diastereomeres (probably cis/trans-
isomers) are baseline separated.
A second major problem for the coupling of LC to MS was initially the incom-
patibility of commonly used mobile phases with MS. Therefore, non-volatiles
such as phosphate ions or the frequently used ion-pairing agent trifluoroacetic
acid (TFA), which causes ion suppression due to the extreme ionization capac-
ity of the mother ion, had to be replaced with formate or acetate buffers. The
organic component of the mobile phases is most frequently acetonitrile, some-
times methanol. Using classical reversed phase material (RP-18, 3 or 4 μm) as
stationary phase, acceptable peak shapes for most of the compounds in leaf and
root extracts could be achieved. In general peak widths of about 0.20−0.35 min
for a 15-cm column with 3 μm particle size were observed. Application of a C18
phase with hydrophilic end-capping provides better separation for early elut-
ing analytes. In particular, aromatic amino acids and biogenic amines show
a considerably improved retention behaviour.
Concerns about the feasibility and reliability of LC-MS-based metabolite
profiling have been raised repeatedly (Fiehn 2002; Fernie et al. 2004; Kell 2004).
These concerns are mostly referring to the fact that electrospray ionization
is prone to matrix effects. This term summarizes two phenomena potentially
compromising quantification: (i) reduction or enhancement of ion signals
caused by the sample matrix, and (ii) interferences from co-eluting molecules
(Matuszewski et al. 2003). Matrix components that are non-volatile can have
dramatic effects on the ion signal of an analyte. Mechanistically this effect is
not fully understood. It is likely caused by competition between an analyte
and non-volatile matrix components for access to the droplet surface in the
spray and for reaction with ions formed during the ionization process (Niessen
1999a; Matuszewski et al. 2003; Manini et al. 2004). Thus, reproducibility of
a quantification can be compromised, a potential problem that is further aggra-
vated when diverse samples ( = matrices) are analyzed. It is important to note,
however, that matrix effects have predominantly been observed in cases where
there was little chromatographic separation. Run times and column lengths
were reduced because the coupling to MS/MS supposedly guaranteed highly
selective detection (Matuszewski et al. 2003). It is obvious, however, that good
separation prior to ionization is essential to reduce the impact of matrix effects
and to minimize ion suppression – especially when highly complex metabolite
mixtures are analyzed. The fewer analytes elute from the column simulta-
neously, the better the chances are of efficient and reproducible electrospray
ionization and detection of a particular analyte. Thus, optimal separation is
of paramount importance and, consequently, the use of very long monolithic
columns for the liquid chromatography has been proposed (Tolstikov et al.
2003). In capillary LC, particle size and column diameter severely restrict the
length of the column because of the backpressure build-up. At the same time,
a certain minimum flow rate has to be maintained in order to obtain a stable

electrospray. In our experience, therefore, it is not feasible to use columns much
longer than 20 cm unless larger particles are used. We found, however, that the
3 μm–15 cm design is sufficient to achieve very good separation. Figure 1b
shows as a selected example for the CapLC resolution the base-line separation
of the two diastereoisomers of 4-glucopyranosyloxybenzoyl choline (below we
will present and discuss an assessment of matrix effects in our metabolite
profiling scheme).
A great advantage of ESI is its ability to provide soft ionization. Nevertheless,
fragmentation can easily be induced in one of the higher-pressure regions of
the ion passageway from the source into the mass analyzer (Fig. 2a). Three po-
tentials which determine the opposite processes of declustering and focusing
vs collision induced dissociation (Fig. 2b) can be varied on a QSTAR Pulsar
and have to be optimized for the profiling in terms of mass signal yield and
appropriate distribution. First we analyzed two model compounds, namely
hirsutin (Fig. 2c) and rutin (Fig. 2d), which are prone to give in-source frag-
ments and tried to optimize the intensities of the quasi-molecular ions by
systematically ramping both declustering potentials and the focusing poten-
tial. For both hirsutin and rutin the quasi-molecular ions [M+H]+ reached
their maximum between 40 and 50 V for DP1 and 10 and 15 V for DP2. The
effects of the focusing potential on the maxima of the breakdown curves were
of minor importance. However, optimization was necessary (FP = 220 V, data
not shown). It should be clearly stated that, in principle, for every analyte, such
an optimization has to be done to get the full sensitivity, but since a highly
complex mixture of mostly unknown compounds is analyzed, compromises
have to be made. To get the optimal value of DP1 for a profiling experiment,
we measured the same methanolic leaf extract (n = 4) with different DP1
values between 15 and 60 V and analyzed the mass signals with regard to its
mass-to-charge ratio and signal-to-noise ratio distribution (Fig. 3, left panel
and right panel, respectively). We found that in such a simplified analysis the
density functions between 30 and 60 V show only minor differences and, thus,
a value of about DP1 = 45 V appears to yield the best results concerning high
signal-to-noise ratios as well as high mass-to-charge ratios.
In our metabolite profiling platform (von Roepenack-Lahaye et al. 2004) ions
are injected into a QTOF system, a hybrid mass spectrometer. Practically, the
third quadrupole in a triple quad instrument is replaced in these instruments
with a time-of-flight mass analyzer (Chernushevich et al. 2001). Other mass
analysis options for coupling to LC are discussed in detail in another chapter
(Sumner et al.; this book, Chap. I.2). Likewise, the QTOF system is dealt with
by Bino et al. (this book, Chap. I.3). Thus, we will only briefly summarize
our experience with QTOF-MS. We routinely use an acquisition period (“scan
time”) of 2 s. For the deconvolution of data, which is even more vital for LC
due to the lower resolution as compared to GC, there are currently two options
available to us. The software MetaboliteID (Applied Biosystems) allows one to
extract the mass spectra and to generate an output that lists mass peaks with
Fig. 2. Effects of ion source potentials on sensitivity and degree of in-source fragmentation:
a schematic overview of the differentially pumped (evacuated) interface between ion source and
mass spectrometer of an API QSTAR Pulsar Hybrid LC/MS system: curtain plate (CP), curtain
gas (CGs), orifice (OR), ring (RNG), skimmer (SK); b definition of electrical potentials applied in
the interfacial region: declustering potential (DP), focusing potential (FP); c,d breakdown curves
for hirsutin (c) and rutin (d) obtained in DP1 and DP2 ramping experiments
retention times, accurate mass and intensity. Self-made macros are then needed
to normalize, to align peak list and to compare intensities (von Roepenack-
Lahaye et al. 2004). These latter steps are covered by the software MetAlign,
developed by Arjen Lommen (www.metalign.nl; Tolstikov et al. 2003), which
aligns and compares sets of chromatograms to identify differentially abundant
mass signals. Reproducibility of the retention times of capillary LC is, in our
experience, high enough to allow accurate alignment of chromatograms.
Fig. 3. Effect of declustering potential modulation (DP1 variable, DP2 = 15 V, FP = 220 V) on

mass-to-charge ratio (left panel) and signal-to-noise ratio (right panel) distribution. A methanolic
leaf extract was analyzed by CapLC-ESI(+)-QTOF-MS with varying declustering potential 1
settings between 15 and 60 V. Frequencies for different mass-to-charge ratio (m/z) and signal-
to-noise ratio (S/N) classes were plotted against DP1 values (n = 4)
3 Potential and Limitations
3.1 Scope of the Analysis
Given the idealistic goal of metabolomics to achieve comprehensive coverage

of the metabolome (Oliver et al. 1998), the number of detectable metabolites is
an important feature of a metabolite profiling platform. CapLC-ESI-QTOF-MS
has great potential because of its sensitivity (Chernushevich et al. 2001). In
a single CapLC-MS run analyzed with the deconvolution software Metabo-
liteID we routinely detect about 1000–2000 mass signals, depending on the
extracted material. Running the data through the MetAlign software and apply-
ing a signal-to-noise ratio cutoff of 5 gives comparable figures. Similar, albeit
somewhat lower numbers (around 700) have been obtained for methanolic
Arabidopsis thaliana leaf extracts in pilot experiments with monolithic sil-
ica columns, coupling of the LC to ion trap MS and deconvolution through
MetAlign (Tolstikov et al. 2003).
A mass signal or m/z, however, does not necessarily represent a metabolite
in all cases. Though isotope peaks should be eliminated through the decon-
volution, many of the signals are likely to be fragments or adducts so that
any given metabolite can in theory give rise to several mass signals. Thus,
the number of detectable metabolites is certainly smaller than the number of
mass signals and cannot reliably be estimated at this point in time. It is safe
to state, however, that several hundred metabolites are routinely detectable in
a methanolic extract using the combination of capillary LC and QTOF mass
spectrometry. Formation of sodium or potassium adducts, for instance, is not
too frequent because of the use of formic acid in the mobile phase.
How large a fraction of the Arabidopsis metabolome is covered using this
technique? The range of secondary metabolite compound classes detectable
by CapLC-ESI-QTOF-MS in its current state can be assessed by searching in

the profiles for members of the various groups of metabolites known to occur
in Arabidopsis thaliana. A recent compilation listed six biosynthetic classes
(d’Auria and Gershenzon 2005): nitrogen-containing compounds, phenyl-
propanoids, benzenoids, polyketides such as flavonoids, terpenes and fatty
acid derivatives. Metabolites of five of these classes can clearly be detected
Fig. 4. Most of the known biosynthetic classes of A. thaliana secondary metabolites are de-
tectable by CapLC-ESI-QTOF-MS. CapLC-ESI(+/-)-CID-MS spectra of representative metabolites
detected in methanolic extracts of different A. thaliana tissues such as leaves, seeds and roots
(for details on the biosynthetic classes and representative metabolites see text)
by CapLC-ESI-QTOF-MS. Figure 4 displays CID-MS spectra of representa-

tives that we identified in Arabidopsis thaliana extracts. Intact glucosino-
lates can easily be detected by ESI in negative ion mode in nearly all tis-
sues of Arabidopsis thaliana. Typical hydrolysis products of glucosinolates
like isothiocyanates and nitriles can be detected by ESI(+). Examples are 8-
methylsulfinyloctylisothiocyanate (hirsutin) and 8-methylthiononanitrile as
well as several homologs. Furthermore, biosynthetic precursors of glucosino-
lates, like desulfoglucosinolates and thiohydroxamic acids, have been identi-
fied in certain cases. Indole-derived secondary metabolites such as the phy-
toalexin camalexin represent further nitrogen containing compounds that can
be detected. Notably, methanolic root extracts contain a huge variety of in-
dole derivatives. Furthermore, ascorbigens and glutathione-indole conjugates,
which result from trapping reactions of the hydrolysis products of indole glu-
cosinolates with several nucleophiles, could also be detected. As representa-
tives of the phenylpropanoids, typical esters such as sinapoyl malate in leaves
and sinapoyl choline in seeds could be specified. In particular, methanolic
seed extracts show a wealth of different choline esters (unpublished obser-
vations). Besides the corresponding substituted cinnamoyl cholines various
hydroxylated/methoxylated benzoyl cholines could be detected. Other choline
containing compounds like differentially substituted phosphatidyl cholines
have been identified in seed extracts, too. From the class of flavonoids the ma-
jor flavonols kaempferol and quercetin and their glycosides could be detected
either in positive or negative ion mode. Saccharide composition and aglycon
structure can be determined by means of MS2 and pseudo-MS3 (product ion
spectra derived from in-source CID fragments) experiments.
In conclusion, of the biosynthetic classes known to occur in Arabidopsis
thaliana, all but one (terpenes) can be detected. Judging from this comparison,
CapLC-ESI-QTOF-MS achieves a very good coverage of secondary metabolism.
In addition, many primary metabolites such as amino a cids and oligopeptides
can be analyzed. This assessment is based on data obtained in positive-ion
mode. There is a potential to improve further the reach by also measuring
in the negative-ion mode. We found that it is also possible to acquire reliable
data in the negative mode without changing the mobile phase, albeit with
a significantly reduced mass signal yield.
3.2 Quantification
Proper quantification of an analyte requires optimization of extraction, sam-

ple preparation, chromatography and detection, as well as the availability of
a pure standard – which would ideally be isotopically labeled – that can be used
to calibrate the signal. A standard also allows one to determine the dynamic
range of a metabolite in question and to search for possible matrix effects
by performing recovery experiments (Birkemeyer et al. 2005). Obviously it is
extremely difficult for an unbiased profiling of mostly unknown compounds
to meet the criteria for accurate and reproducible quantification. Standards

are available for only a subset of the detected metabolites and this subset is
particularly small for capillary LC-MS which targets hundreds of low abun-
dance and species-specific metabolites. At the same time, it is inconceivable
that all the required standards can be synthesized. Thus, any metabolomics
approach comes at a cost of reduced precision (Trethewey et al. 1999) and has
to be assessed critically and improved continuously with respect to accuracy
of quantification.
For electrospray ionization there is inherently no correlation between signal
strength and abundance when comparing different analytes because ioniza-
tion efficiency is molecule-dependent. Decisive for quantitative profiling is,
however, whether such a correlation exists for any given analyte and what
the boundary minimum and maximum detector signals are, i. e. how wide
the dynamic range is. In the absence of appropriate standards, serial dilution
experiments are a way to assess linearity at least for the mass signals that are
sufficiently strong. Our initial results demonstrated for a subset analyzed in
detail that there is indeed a good correlation over the almost two orders of
magnitude that were tested (von Roepenack-Lahaye et al. 2004). Still, because
of the analyte dependency, there is a need to gather information continuously
for more and more metabolites and to use these data for the quantification.
As a first approximation an extensive set of reference compounds for different
compound classes – if at all available – should be used. Eventually, the fasci-
nating concept of mass isotopomer ratio analysis through in vivo labeling with
13 C (Birkemeyer et al. 2005) might at some point in the future allow internal
standardization of profiling even with multicellular organisms.
The above-mentioned potential of LC-MS for severe matrix effects makes
rigorous validation a necessity (Fernie et al. 2004). Matrix effects are again
dependent on the analyte and the extract or the nature and origin of the
biological sample. Ways to assess these indirectly for an analyte in question are
spiking experiments with different matrices (Matuszewski et al. 2003). In order
to obtain an estimate of matrix effects at the profiling scale, we performed, for
instance, a series of mixing experiments. From previous data we know that
leaf and root extracts are fundamentally different in their composition (von
Roepenack-Lahaye et al. 2004). Leaf/root extracts were either diluted with 80%
methanol or with equal amounts of root/leaf extracts and analyzed (n = 4).
We focused attention on ions with a strong signal so that they would also
be reliably detectable in dilutions. Also, we made sure that signal intensities
were in the dynamic range; 45 m/z values eluting between 15 and 45 min were
selected. Figure 5 shows the ratios of “signal after dilution with methanol” to
“signal after dilution with root/leaf extract”. A ratio above 1 is indicative of
signal enhancement through the other extract, a ratio below 1 shows signal
suppression. One can see in Fig. 5 that ion suppression occurs more frequently
than enhancement. Of the 90 m/z measured in different dilutions, 76 showed
ratios of 1 ± 0.4 (equals about two times the technical variation) and are
therefore considered to be reliably quantifiable. In most experimental set-ups
Fig. 5. Evaluation of matrix effects through mixing of extracts of different origin. Signal ratios for
mass signals in crosswise matrix-diluted and solvent-diluted leaf and root extracts obtained by
CapLC-ESI(+)-TOF-MS measurements. A total of 45 mass signals were analyzed in two different
mixtures each. A ratio of “signal after dilution with methanol” to “signal after dilution with
root/leaf extract” above 1 is indicative of signal enhancement through the other extract, a ratio
below 1 of signal suppression. Most of the mass signals showed a ratio of 1 ± 0.4 (equals about
two times the technical variation). This threshold of ±40% is indicated by vertical lines
matrix effects will probably be smaller than in this pilot experiment because
matrices will be less diverse than a root and leaf extract. Information obtained
by analyses such as these can be used to weigh the data obtained in a profiling
experiment and to add “confidence tags” to each metabolite. Factored into
such “confidence tags” should also be the results on the dynamic range and
the degree of variability observed over many experiments. We conclude that
the sensitivity and range of CapLC-ESI-QTOF-MS clearly make it a powerful
approach for the identification of qualitative differences between samples but
that – as predicted by Chernushevich et al. (2001) – quantitative analysis is also
feasible provided analyte-dependent effects can be detected and corrected for.
All available data suggest that overall the analytical variation is smaller than
the biological variation – as is the case for the more established metabolomics
approaches (Dunn et al. 2005).
3.3 Identification of Unknown Metabolites
As stated in the introduction, a major potential advantage of LC-MS-based pro-

filing is the multitude of options to obtain structural information on unknown
compounds. This is of paramount importance given, for instance, the conser-
vatively estimated 5000 metabolites in Arabidopsis thaliana of which maybe
500 are annotated today (Bino et al. 2004). The first piece of information on
unknowns is the accurate mass that can be obtained by TOF-MS with a devi-
ation of only 5−10 ppm even in complex matrices (von Roepenack-Lahaye et
al. 2004; Ibanez et al. 2005). Based on this, potential elemental compositions
can be calculated. As recently proposed by Ibanez et al. (2005), the usually

large number of possibilities can be reduced by calculating theoretical iso-
topic percentages for all possible elemental compositions and by comparing
these to the experimental data. A further significant reduction of formulae can
then be achieved through the second and third layer of structural information,
in-source fragments and CID-MS spectra (see Fig. 4). The very high mass
accuracy of QTOF instruments also applies to product ion scans (Chernushe-
vich et al. 2001). With the information on accurate masses of precursor and
product ions, databases can be searched. Obviously, the success rate of this
approach is determined not only by the performance of the analysis but also
by the availability of databases. In this respect the current situation is far from
being satisfactory and future joint efforts will hopefully result in a significant
improvement (Bino et al. 2004). One should add, however, that in the end iden-
tification will in most cases be tentative and further validation will be required
(Bino et al. 2004). Also, discrimination between isomers is not possible without
standards.
4 Conclusions and Outlook
Our experience with respect to the potential of this technique for metabolomics
can in part be validated by taking a look at recent applications of LC-MS in
general and of CapLC-ESI-QTOF-MS in particular. In occupational toxicology
the superiority of LC-MS-MS with respect to sensitivity is now being exploited
for the determination of trace and ultra-trace amounts of biomarkers of expo-
sure (Manini et al. 2004). Quantification of low-abundance molecules in highly
variable complex matrices is considered feasible, provided that precautions
such as those outlined above are taken. Also, many novel metabolites have
been identified and minor metabolic routes for well-known occupational haz-
ards have been uncovered (Manini et al. 2004). Similarly, the mass accuracy
and sensitivity of QTOF-MS coupled to liquid chromatography is now being
applied to the elucidation of unknown environmental micro-contaminants in,
for instance, water samples (Ibanez et al. 2005). Studies of this kind face chal-
lenges similar to those of metabolomics experiments. The emerging picture is
that CapLC-ESI-QTOF-MS can be routinely applied (von Roepenack-Lahaye
et al. 2004; Bino et al. 2005) and has high potential not only for the iden-
tification of selected molecules but for a highly sensitive, robust metabolite
profiling that achieves very good coverage of the metabolome. Obviously this
technology will be undergoing continuous validation and improvement. De-
veloping the profiling is an iterative process. Any progress made with respect
to the availability of standards or reference compounds, the identification of
metabolites, the linear range or the possible matrix effects for a particular mass
signal has to be used to increase further the accuracy of quantification. Also,
protocols for extraction, selective enrichment of metabolites and chromato-
graphic separation have to be tailored for specific questions so that a toolbox

of different profiling schemes becomes available to fully exploit the power of
CapLC-ESI-QTOF-MS.
Despite the current limitations – comparatively high cost and the lack of
LC-MS spectral databases – this profiling approach most likely will contribute
substantially to cataloguing the metabolome of Arabidopsis thaliana and other
systems that are under investigation as models or economically important
species. Also, it will help to elucidate biological functions of metabolites and
will greatly facilitate the identification of enzyme substrates and products
through the systematic analysis of mutants and the correlation with tran-
script and protein data (Hirai et al. 2005). Extensive data matrices will allow
one to unravel metabolic and regulatory networks, especially in secondary
metabolism.
References
Abian J, Oosterkamp AJ, Gelpi E (1999) Comparison of conventional, narrow-bore and capillary
liquid chromatography mass spectrometry for electrospray ionization mass spectrometry:
Practical considerations. J Mass Spectrometry 34:244–254
Bino RJ, Ric de Vos CH, Lieberman M, Hall RD, Bovy A, Jonker HH, Tikunov Y, Lommen A,
Moco S, Levin I (2005) The light hyperresponsive high pigment-2dg mutation of tomato:
alterations in the fruit metabolome. New Phytol 166:427–438
Birkemeyer C, Luedemann A, Wagner C, Erban A, Kopka J (2005) Metabolome analysis: the
potential of in vivo labeling with stable isotopes for metabolite profiling. Trends Biotechnol
23:28–33
Chernushevich IV, Loboda AV, Thomson BA (2001) An introduction to quadrupole-time-of-flight
D’Auria JC, Gershenzon J (2005) The secondary metabolism of Arabidopsis thaliana: growing
like a weed. Curr Opin Plant Biol 8:308–316
Dixon RA (2001) Natural products and plant disease resistance. Nature 411:843–847
Dunn WB, Bailey NJ, Johnson HE (2005) Measuring the metabolome: current analytical tech-
nologies. Analyst 130:606–625
48:155–171
Fiehn O, Kopka J, Dormann P, Altmann T, Trethewey R, Willmitzer L (2000) Metabolite profiling
for plant functional genomics. Nature Biotechnol 18:1157–1161
Goodacre R, Vaidyanathan S, Dunn WB, Harrigan GG, Kell DB (2004) Metabolomics by numbers:
acquiring and understanding global metabolite data. Trends Biotechnol 22:245–252
Halket JM, Waterman D, Przyborowska AM, Patel RK, Fraser PD, Bramley PM (2005) Chemical
derivatization and mass spectral libraries in metabolic profiling by GC/MS and LC/MS/MS.
J Exp Bot 56:219–243
Hirai MY, Klein M, Fujikawa Y, Yano M, Goodenowe DB, Yamazaki Y, Kanaya S, Nakamura Y,
Kitayama M, Suzuki H, Sakurai N, Shibata D, Tokuhisa J, Reichelt M, Gershenzon J, Pa-
penbrock J, Saito K (2005) Elucidation of gene-to-gene and metabolite-to-gene networks in
Arabidopsis by integration of metabolomics and transcriptomics. J Biol Chem (epub ahead

of print)
Ibanez M, Sancho JV, Pozo OJ, Niessen W, Hernandez F (2005) Use of quadrupole time-of-flight
mass spectrometry in the elucidation of unknown compounds present in environmental
water. Rapid Commun Mass Spectrom 19:169–178
Kell DB (2004) Metabolomics and systems biology: making sense of the soup. Curr Opin Microbiol
7:296–307
Manini P, Andreoli R, Niessen WM (2004) Liquid chromatography-mass spectrometry in occupa-
tional toxicology: a novel approach to the study of biotransformation of industrial chemicals.
J Chromatogr A 1058:21–37
Matuszewski BK, Constanzer ML, Chavez-Eng CM (2003) Strategies for the assessment of matrix
effect in quantitative bioanalytical methods based on HPLC-MS/MS. Anal Chem 75:3019–
3030
Niessen WM (1999a) State-of-the-art in liquid chromatography-mass spectrometry. J Chromatogr
A 856:179–197
Niessen WM (1999b) Liquid chromatography-mass spectrometry, 2nd edn. Dekker, New York
Oliver SG, Winson MK, Kell DB, Baganz F (1998) Systematic functional analysis of the yeast
genome. Trends Biotechnol 16:373–378
Roessner U, Wagner C, Kopka J, Trethewey RN, Willmitzer L (2000) Technical advance: simul-
taneous analysis of metabolites in potato tuber by gas chromatography-mass spectrometry.
Plant J 23:131–142
Taylor LP, Grotewold E (2005) Flavonoids as developmental regulators. Curr Opin Plant Biol
8:317–323
Tolstikov VV, Lommen A, Nakanishi K, Tanaka N, Fiehn O (2003) Monolithic silica-based
Trethewey RN, Krotzky AJ, Willmitzer L (1999) Metabolic profiling: a Rosetta Stone for genomics?
Von Roepenack-Lahaye E, Degenkolb T, Zerjeski M, Franz M, Roth U, Wessjohann L, Schmidt J,
Scheel D, Clemens S (2004) Profiling of Arabidopsis secondary metabolites by capillary
liquid chromatography coupled to electrospray ionization quadrupole time-of-flight mass
spectrometry. Plant Physiol 134:548–559
Wilm MS, Mann M (1994) Electrospray and taylor-cone theory, Dole’s beam of macro-molecules
at last? Int J Mass Spectrom Ion Processes 136:167–180
Winkel-Shirley B (2001) Flavonoid biosynthesis. A colorful model for genetics, biochemistry, cell
biology, and biotechnology. Plant Physiol 126:485–493
I.6 NMR Spectroscopy in Plant Metabolomics
J.L. Ward and M.H. Beale1
1 Introduction
Nuclear magnetic resonance (NMR) spectroscopy is one of the most power-

ful and widely used structural analysis techniques available to the analytical
phytochemist and continues to be the technique of choice for unknown struc-
ture determination. As a technique for plant metabolomics it benefits from
the fact that it is non-compound class selective and non-sample destructive.
NMR spectra contain a wealth of accurate qualitative and quantitative infor-
mation regarding the components of a sample. Whilst measurements of the
1 Hs are the most commonly used for metabolomic studies, analysis of the 13 Cs
has also been employed. However, the low sensitivity of 13 C-NMR (due to its
lower natural abundance and magnetogyric ratio) prevents its routine use for
large numbers of complex extracts. General disadvantages of the sensitivity of
NMR (relative to mass spectroscopy) and overlapping signals can be largely
overcome, for single compounds and partially fractionated mixtures, by use
of instruments with higher field strength magnets (600 MHz or greater) or
by the use of modern cryoprobes. In very complex mixtures, such as crude
plant extracts, that contain compounds at widely differing concentrations,
sensitivity and overlapping signals are problematic for traditional 1D-NMR
spectral interpretation. Techniques to deal with overlapping signals, such as
2D-J-resolved spectroscopy, can provide reconstructed 1D spectra that are sim-
plified by the absence of proton–proton coupling (Viant 2003). However, in this
review we concentrate on high-throughput 1D 1 H-NMR and demonstrate how
the technique in combination with chemometrics has become well-established
as a plant metabolomic screen. We also discuss how application of 2D- and hy-
phenated NMR techniques can provide further solutions to the sensitivity and
deconvolution problems associated with analysis of complex natural product
mixtures. For more general reviews in the use of NMR in plant sciences the
reader is directed to Roberts (2000), Ratcliffe et al. (2001) and Ratcliffe and
Shachar-Hill (2001, 2005).
1 The
National Centre for Plant and Microbial Metabolomics, Rothamsted Research, West Com-
mon, Harpenden, Herts. AL5 2JQ, UK, e-mail: [email protected], [email protected]

82 J.L. Ward and M.H. Beale
2 High-throughput Screening by 1D 1 H-NMR

A key advantage in the use of 1 H-NMR spectroscopy in metabolomic screens is
the robustness of the technique such that any compound that is soluble in the
solvent of choice will be detected, providing that it contains hydrogen atoms.
Furthermore, integration of signals from different compounds is absolutely
quantitative and is truly representative of the relative concentrations of those
compounds. The fact that NMR reliably detects most compounds present gives
the technique a clear advantage in screening and fingerprinting applications
over mass spectroscopic techniques, which are beset by problems caused by
variable ionisation of different types of compounds.
Methodologies for metabolomic screening of plant extracts by NMR spec-
troscopy are based on the large body of work carried out in the biomedical
area, particularly on plasma and urine in relation to disease biomarkers and
drug metabolism. Much of this work was done by the prolific research group
of Nicholson, Lindon and Holmes at Imperial College, London (Nicholson et
al. 1999; Lindon et al. 2000, 2001, 2004; Bollard et al. 2005).
In plant metabolomics, solvent extraction of metabolites from tissue is nec-
essary. Apart from the experimental design aspects where decisions on the
plant numbers, growth, and tissue type have to be made, key choices influenc-
ing the range of metabolites detected must be made. These are (i) whether to
use fresh or freeze-dried tissue and (ii) the polarity of the solvent to be used.
On the whole, better quality NMR spectra are obtained by directly extract-
ing freeze-dried tissue with deuterated NMR solvents. Most published work
utilises polar solvents (aqueous buffers, perchloric acid, or methanol-water
mixtures), although chloroform has also been utilised (Choi et al 2004a). Fig-
ure 1a depicts the 1 H-NMR spectrum of a deuterated water-methanol extract
of Arabidopsis thaliana. The spectrum is typical for most green tissue polar
extracts (Charlton et al. 2003; Ward et al. 2003; Choi et al. 2004b,c), and is
dominated by signals arising from carbohydrates, amino-acids and organic
acids. Polar extracts of other plant tissues such as potato tubers (Defernez et
al. 2004), wheat flour (Lewis et al. 2003) and tomato fruits (Le Gall et al. 2003)
have similar NMR spectra but contain other features reflecting the concentra-
tion of certain metabolites associated with these storage tissues. Exudates from
plant roots also contain distinctive metabolites (Fan et al. 1997) and represent
an area that has been neglected in the recent upsurge in metabolomic studies.
Classic interpretation of complex NMR spectra such as Fig. 1a is difficult
although some 30 or 40 metabolites can be definitively identified by virtue of
signals that appear in non-overlapped regions of the spectrum and quantified
by integration against the internal reference standard (usually trimethylsilyl-
d4-propionate). Use of libraries of standard spectra run in the same solvent on
the same instrument (or at least on an instrument of the same field strength) is
an aid to this interpretation, as is ‘spiking’ of extracts with pure compounds. In
the example shown in Fig. 1b, clear differences in the anomeric proton region
in the spectrum of wild-type Arabidopsis thaliana and that of two different
NMR Spectroscopy in Plant Metabolomics 83
Fig. 1. a 600 MHz 1 H NMR spectrum of 4:1 (D2 O:CD3 OD) extract of freeze-dried Arabidopsis
thaliana Col-0 tissue. b Expanded portion of the spectrum featuring the carbohydrate anomeric
proton region highlighting differences in carbohydrate concentrations between Col-0 and the
starch biosynthesis mutants pgm-1 and adg-1. Labelled peaks: 1-sucrose, 2-maltose, 3-glucose.
c PCA scores plot illustrating differences observed between Col-0 and the adg-1/pgm-1 mutants
(filled squares – Col-0, open triangles – pgm-1, filled circles – adg-1). d Loadings plot of PC1
depicting the ‘spectra’ of compounds responsible for differences between Col-0 and the mutants.
e Loadings plot of PC2 depicting the differences between adg-1 and pgm-1
mutants in the starch biosynthesis pathway (pgm-1 and adg-1) can be seen.
These differences can be interpreted in relation to the known function of
the enzymes missing in the mutants. However, in most plant metabolomic
applications many hundreds of similar spectra are collected. Simultaneous
classical interpretation of these numbers of individual spectra is not possible,
but the use of chemometrics (see next section) allows the spectroscopist to
focus on compounds that are responsible for differences between individual

plants, or populations of plants, and target more detailed analysis to particular
compounds or biochemical pathways.
3 Data Analysis
NMR-based metabolomic datasets are very large, both in terms of the number
of datapoints per sample (typically 32 k or 64 k), and also the number of samples
and resulting spectra acquired (from dozens to thousands, often including
replicates). In order to draw conclusions and make comparisons between large
numbers of spectra, automated strategies must be employed for the analysis
and interpretation of such data once they have been acquired. The literature
on data analysis is extensive and will only be briefly discussed here. Interested
readers are directed to the review on pattern recognition methods (Lindon et
al. 2001) for further coverage of some of the issues.
Data manipulation typically starts with some form of ‘bucketing’ or ‘bin-
ning’ whereby the spectrum is split into discrete regions (typically between
0.01 and 0.04 ppm in width), which are then integrated to return a list of inte-
gral values for each spectrum. Whilst this reduces the resolution of the data, it
has the advantage of removing small chemical shift changes due to slight pH
variation between samples. Increasingly, work is being carried out using all of
the datapoints in the spectrum by employing an algorithm to align the peaks,
eliminating any unwanted variation (Stoyanova et al. 2004). NMR data is usu-
ally analysed initially using multivariate statistical methods such as Principal
Component Analysis (PCA). PCA is a data visualisation method, useful for ob-
serving groupings within large datasets. There are a number of commercially
available software products that carry out PCA and other related multivariate
analyses. One that has been widely used for NMR data is SIMCA-P (Umetrics,
Sweden). A PCA model can be displayed in a graphical fashion as a “scores”
plot as shown in Fig. 1c. This example compares 1 H-spectra collected from
polar extracts of wild-type Arabidopsis thaliana Col-0 with those from the two
mutants in starch biosynthesis (pgm-1 and adg-1). This plot is useful for ob-
serving any groupings in the data set and in addition will highlight outliers that
may be due to errors in sample preparation or instrumentation parameters etc.
Coefficients by which the original variables must be multiplied to obtain the
score are called “loadings”. Thus, “loading plots” [e. g. Fig. 1d,e] can be used to
detect and display the spectral areas responsible for the separation in the data,
and can be interpreted as positive and negative NMR spectra of the compounds
responsible for the differences between the clusters. The numerical value of the
loading of a given variable on a PC indicates how much the variable has in com-
mon with that component (Massart et al. 1988). In Fig. 1d, PC1 represents the
NMR spectra of compounds differing between wild-types and both mutants,
whilst PC2, Fig. 1e, represents the (smaller) difference between the mutants.
When carrying out PCA it is necessary to apply scaling methods to the

bucketed data matrices. In NMR spectroscopic data, although the integral
values across a spectrum are proportional to concentration and the number
of resonances present, the largest resonances would, without scaling, have
a dominant effect in multivariate analysis. Before PCA the data can be scaled
in different ways. In the covariance matrix method the data are just mean-
centred. In the correlation matrix method the data are mean-centred and
then the columns (variables) of the data matrix are scaled to unit variance.
Covariance matrices are most widely used for NMR data because they have
the advantage that the loadings plots retain the scale of the original data
and can be compared back to libraries of spectra for assignment. Variable
stability scaling (VAST) has recently been described and offers advantages over
previously employed scaling methods in terms of the downstream multivariate
modelling (Keun et al. 2003). This method weights each variable according
to a metric of its stability and can unearth subtle differences between lines
against backgrounds of biological variation. However, in the plant research
arena, where growth of many identical clones under controlled environment is
easily achieved, biological variation can be minimised, for example by pooling
many individuals, and thus presents less of a problem.
An alternative method of highlighting differences between sample sets
against backgrounds of biological or experimental variation is Orthogonal
Signal Correction (OSC) (Gavaghan et al. 2002). This data filtering method
can be applied to the scaled data matrix before multivariate analysis, and can,
if used carefully, yield insights that are not evident from PCA. Such deeper
mining of large data sets for differences against a background of noise can also
be obtained by supervised modelling methods such as Partial Least Squares-
Discriminant Analysis (Lindon et al. 2001). The use of neural networks to
classify spectra has been applied to the study of the herbicide mode of action
on maize seedlings (Aranìbar et al. 2001).
4 Two-dimensional NMR
Two-dimensional (2D) NMR experiments, which make use of interactions be-

tween NMR-detectable nuclei within a molecule, can be used to increase the
spectral resolution and highlight which peaks belong to the same molecule.
These experiments generally have much longer acquisition times, posing prob-
lems to those researchers wanting to carry out high throughput data collection.
Nevertheless, these experiments can be run in automation and generate data
useful to the metabolomics researcher and are particularly useful in the as-
signment of identities to unknown peaks.
The 2D experiments can be split into homonuclear and heteronuclear exper-
iments. Homonuclear experiments examine the correlations between nuclei of
the same type (commonly 1 H). The TOCSY experiment (Total Correlation
Spectroscopy) describes all interactions in a spin system and therefore is one

of the most informative experiments available. The experiment has been used
to examine complex matrices such as root exudates and cell extracts (Fan et
al. 1997). Heteronuclear correlation experiments – as the name suggests – ex-
amine the correlation between two different types of nuclei within a molecule.
Indirect experiments such as HMQC (heteronuclear multiple quantum coher-
ence) and HSQC (heteronuclear single quantum coherence) spectroscopy are
particularly useful in structure assignment as is HMBC (heteronuclear multi-
ple bond coherence) which can examine long range correlations. A relatively
new technique DOSY (diffusion ordered spectroscopy), that is not dependent
on analysis of spin–spin coupling, has been applied to analysis of complex
mixtures such as liquid foods (Gil et al. 2004). Here resonances are separated
in the second dimension by virtue of their diffusion coefficient. This coefficient
is governed by molecular size and thus DOSY represents a novel tool to decon-
volute overlapping signals and may be particularly useful in plant spectra to
assign carbohydrate signals to mono-, di- or higher saccharides.
5 Stable Isotope Labelling
NMR spectroscopy is not restricted to the analysis of 1 H signals. 2 H, 13 C

and 15 N isotopes, however, have a very low natural abundance, making the
detection of these signals difficult. Stable isotope labelling leads to the selective
enhancement of some of these signals, providing a powerful method for the
scrutiny of metabolic pathways in many organisms including plants (Roberts
2000; Roscher et al. 2000). Stable isotope labelling in plants has been extensively
reviewed in recent years (Ratcliffe et al. 2001; Ratcliffe and Shachar-Hill 2001,
2005). 13 C is the most useful isotope and there are many suitable precursors
including acetate, amino acids, carbohydrates and carbon dioxide. 15 N labelling
can be achieved using labelled nitrate or ammonium ions whilst D2 O can be
used to supply plant tissue with 2H. In the majority of cases, 13 C labelling studies
involve the use of singularly labelled precursors although multiply labelled
precursors can also be used. One of the first applications of this technique
in plant metabolism concerned the measurement of 13 C–15 N bonds using
a solid state cross polarisation technique (Schaefer et al. 1981). An important
example that demonstrates the use of isotope labelling in plant metabolonics
has been published by Kikuchi et al. (2004). Using wild type and ethanol-
insensitive mutants of Arabidopsis thaliana, labelled with 13 C, they were able,
using subtractive 2D-HSQC, to isolate and assign only those metabolites that
were affected by ethanol treatment.
Isotopic labelling techniques are often used for measuring the fluxes of
metabolites through metabolic pathways. The information can be used to
establish the identity of the biochemical pathways involved. This type of study
has been carried out to investigate metabolic pathways in plants using 2 H,
13 C and 15 N-labelling (Fox et al. 1995; Prabhu et al. 1996; Schleucher et al.
1998). The analysis of stable isotope labelling in pulse-chase and time-course
experiments can also provide quantitative information on metabolic fluxes
although this is restricted to simple linear pathways that are reasonably close
to the entry point of the label into metabolism. For complicated pathways it
may be useful to examine the distribution of the label once the system has
reached a steady state.
6 Hyphenated NMR
Liquid chromatography-NMR-mass spectrometry (LC-NMR-MS) is arguably

the most powerful of the hyphenated techniques available to the phytochemical
researcher (Hostettmann and Wolfender 2001; Wolfender et al. 2001). Metabo-
lite profiling using such hyphenated techniques can help to provide clean spec-
tral information on components of a mixture of unknown metabolites in an
extract or fraction, leading to a partial or a complete structure determination
in a single online experiment. One of the disadvantages of LC-NMR is its lack
of sensitivity, which hampers the on-flow measurement of minor metabolites.
Another problem is the need to suppress solvent signals from the mobile phase
which if left unsuppressed would dominate the spectrum. Signals residing near
these solvent peaks may be suppressed together with the solvent signal. This
can be a major drawback when dealing with unknown constituents. In these
cases the use of sequential analysis using different solvent systems for the LC
is necessary.
LC-NMR hyphenation has been a reality for over 20 years (Buddrus and Her-
zog 1980). However it is only since the improvements in solvent suppression,
NMR sensitivity and the use of shielded magnets that the technique has re-
ceived more widespread recognition. The technique has been successful when
applied to plant extracts rich in natural products of relatively low molecular
mass. Recent studies have emphasised the value of LC-NMR as a technique for
obtaining detailed chemical profiles of species for taxonomic work. For exam-
ple, using LC-NMR in both on-flow and stop-flow modes, flavones, xanthones
and secoiridoids of several Gentianaceae taxa have been identified (Wolfender
et al. 1997). Vogler et al. (1998) also used LC-NMR in the on-flow mode to
identify nine anti-bacterial sesquiterpene lactones from a partially purified
extract of Vernonia fastigiata (Asteraceae) without the need for isolation of
the individual compounds.
LC-SPE-NMR is a relatively new concept with incorporation of an online SPE
(solid phase extraction) cartridge to trap an analyte peak prior to introduction
into the NMR flow probe. This can be done in automation without interruption
of the column flow. An additional advantage of this system is that after drying
the cartridge, analytes can be eluted in fully deuterated solvents, reducing
the need for solvent suppression. Furthermore multiple trapping on the same
cartridge concentrates the analytes. There are still relatively few publications
using this technology although, an application of LC-SPE-NMR to the detection
of compounds from oregano was recently reported (Exarchou et al. 2003). Very
recently the technology was used for the rapid identification of antioxidants in
complex commercial rosemary extracts (Pukalskas et al. 2005). In this work,
all major compounds present in the extract were collected on SPE cartridges
after their separation and analysed by both NMR and ESI-MS. LC-SPE-NMR
using post column solid-phase extraction was also applied to the direct analysis
of phenolic compounds in the polar fraction of olive oil (Christophoridou et
al. 2005). As well as the identification of simple phenolic acids, lignans and
flavonoids the technique enabled the identification of several new phenolic
compounds not previously reported as constituents of olive oil.
7 Discussion: Applying NMR to Plant Metabolomics
NMR has proven to be an exceptionally useful tool in animal metabolomics.

In plant metabolomics 1D-NMR coupled with multivariate pattern matching
serves as an excellent screen to cluster plant lines/treatments by global analysis
of the total extractable metabolome. This type of analysis serves as a first pass
screen that also gives quantitative data on important abundant metabolites.
The clustering information and preliminary metabolite data can then be used
to guide more detailed analysis by other techniques. These more targeted tech-
niques include the GC-, LC- and CE-mass spectroscopic techniques discussed
elsewhere in this volume, but also include the 2D-NMR, hyphenated-NMR and
isotope-labelling techniques described above.
Examples of 1D-NMR-PCA application can be found in several areas, for
example, in functional genomics (Cornah et al. 2004; Le Gall et al. 2005), in
analysis of ecotypic and cultivar variation (Ward et al. 2003; Frederich et al.
2004), in safety evaluation of GM crops (Noteborn et al. 2000; Charlton et al.
2003; Le Gall et al. 2003; Lewis et al. 2003; Defernez et al. 2004; Manetti et al.
2004), in analysis of the affects of infection (Choi et al. 2004b,c), in classifying
mode of action of chemicals (Aranibar et al. 2001) and of course in quality
control of food and herbal products (Vogels et al. 1996; Charlton et al. 2002).
The scale of this type of screening will increase and new challenges facing
researchers in this area concern the construction of databases of fingerprints
from mutants and the interfacing of these with similar databases of spectra of
pure standards, such that automated interpretation of complex spectra can be
performed. Generic plant metabolomic problems such as temporal batch to
batch variation caused by machine or chromatographic drift are not generally
seen for NMR as instrument drift is minimal. Furthermore, for Arabidopsis
thaliana at least, biological batch to batch variation in NMR spectra has been
eliminated by careful control of growth and experimental procedures (Lewis
et al. 2003). Thus the functional genomic goal of a database of electronically
comparable profiles of large collections of gene knockout mutants and other

genetic resources may now be achievable.
The potential of 2D- and hyphenated NMR to increase the number of
metabolites that can be observed and quantified is yet to be realised. Although
throughput will be decreased, technology platforms where selected samples
from first pass 1D-NMR-MS-PCA screens are selected for further fractionation
by SPE-NMR-MS and LC-SPE-NMR-MS are being put in place. Success with
these will inevitably broaden the metabolome coverage, especially when used
in parallel with GC-, CE- and LC-MSn methods.
References
Aranìbar N, Singh BJ, Stockton GW, Ott KH (2001) Automated mode-of-action detection by
metabolic profiling. Biochem Biophysl Res Commun 286:150–155
Bollard ME, Stanley EG, Lindon JC, Nicholson JK, Holmes E (2005) NMR-based metabonomic
approaches for evaluating physiological influences on biofluid composition. NMR Biomed
18:143–162
Buddrus J, Herzog H (1980) Coupling of HPLC and NMR.1. Analysis of flowing liquid-
chromatographic fractions by proton magnetic-resonance. Org Magn Reson 13:153–155
Charlton AJ, Farrington WHH, Brereton P (2002) Application of 1 H-NMR and multivariate
statistics for screening complex mixtures: quality control and authenticity of instant coffee.
J Agric Food Chem 50:3098–3103
Charlton A, Allnutt T, Holmes S, Chisholm J, Bean S, Ellis N, Mullineaux P, Oehlschlager S (2003)
NMR profiling of transgenic peas. Plant Biotech J 2:27–35
Choi HY, Kim HK, Hazekamp A, Erkelens C, Lefeber AWM, Verpoorte R (2004a) Metabolomic Dif-
ferentation of Cannabis sativa Cultivars using 1 NMR spectroscopy and principal component
analysis. J Nat Prod 67:953–957
Choi H, Choi HY, Verberne M, Lefeber AMW, Erkelens C, Verpoorte R (2004b) Metabolic finger-
printing of wild type and transgenic tobacco plants by 1 H NMR and multivariate analysis
technique. Phytochemistry 65:857–864
Choi HY, Tapias Casas E, Kim KH, Lefeber AMW, Erkelens C, Verhoeven JTH, Brzin J, Zel J,
Verpoorte R (2004c) Metabolic discrimination of Catharanthus roseus leaves infected by Phy-
toplasma using 1 H-NMR spectroscopy and multivariate data analysis. Plant Physiol 135:398–
2410
Christophoridou S, Dais P, Tseng L-H, Spraul M (2005) Separation and identification of phenolic
compounds in olive oil by coupling high-performance liquid chromatorgraphy with post-
column solid-phase extraction to nuclear magnetic resonance spectroscopy (LC-SPE-NMR).
J Agric Food Chem 53:4667–4679
Cornah JE, Germain V, Ward JL, Beale MH, Smith SM (2004) Lipid utilisation, gluconeogenesis
and seedling growth in Arabidopsis mutants lacking the glyoxylate cycle enzyme malate
synthase. J Biol Chem 279:42916–42923
Defernez M, Gunning YM, Parr AJ, Shepherd LVT, Davies HV, Colquhoun IJ (2004) NMR and
HPLC-UV profiling of potatoes with genetic modifications to metabolic pathways. J Agric
Food Chem 52:6075–6085
Exarchou V, Godejohann M, van Beek TA, Gerothanassis IP, Vervoort J (2003) LC-UV-solid-
phase extraction-NMR-MS combined with a cryogenic flow probe and its application to the
identification of compounds present in Greek oregano. Anal Chem 75:6288–6294
Fan TMW, Lane AN, Pedler J, Crowley D, Higashi RM (1997) Comprehensive analysis of organic
ligands in whole root exudates using nuclear magnetic resonance and gas chromatography-
mass spectrometry. Anal Biochem 251:57–68
Fox GG, Ratcliffe RG, Robinson SA, Stewart GR (1995) Evidence for deamination by glutamate-
dehydrogenase in higher-plants – commentary. Can J Bot 73:1112–1115
Frederich M, Choiu YH, Angenot L, Harnischfeger G, Lefeber AWM, Verpoorte R (2004)
Metabolomic analysis of Strychnos nux-vomica, Strychnos icaja and Strychnos ignatii ex-
tracts by 1 H nuclear magnetic resonance spectrometry and multivariate analysis techniques.
Gavaghan CL, Wilson ID, Nicholson JK (2002) Physiological variation in metabolic phenotyping
and functional genomic studies: use of orthogonal signal correction and PLS-DA. FEBS Lett
530:191–196
Gil AM, Duarte I, Cabrita E, Goodfellow BJ, Spraul M, Kerssebaum R (2004) Exploratory applica-
tions of diffusion ordered spectroscopy to liquid foods: an aid towards spectral assignment.
Anal Chim Acta 506:215–223
Hostettmann K, Wolfender JL (2001) Applications of liquid chromatography/UV/MS and liq-
uid chromatography/NMR for the online identification of plant metabolites. In: Tringali C
(ed) Bioactive compounds from natural products-isolation, characterisation and biological
properties. Taylor and Francis, London, pp 31–68
Keun HC, Ebbels TMD, Antti H, Bollard ME, Beckonert O, Holmes E, Lindon JC, Nicholson JK
(2003) Improved analysis of multivariate data by variable stability scaling: application to
NMR-based metabolic profiling. Anal Chim Acta 490:265–276
Kikuchi J, Shinozaki K, Hirayama T (2004) Stable isotope labelling of Arabidopsis thaliana for an
NMR-based metabolomics approach. Plant Cell Physiol 45:1099–1104
Le Gall G, Colquhoun IJ, Davis AL, Collins GJ, Verhoeyen ME (2003) Metabolite profiling of
tomato (Lycopersicon esculentum) using 1 H NMR spectroscopy as a tool to detect potential
unintended effects following a genetic modification. J Agric Food Chem 51:2447–2456
Le Gall G, Metzdorff SB, Pedersen J, Bennett RN, Colquhoun IJ (2005) Metabolite profiling of
Arabidopsis thaliana (L.) plants transformed with an antisense chalcone synthase gene.
Lewis J, Baker JM, Beale MH, Ward JL (2003) Metabolite profiling of GM plants: the impor-
tance of robust experimental design and execution. In: Nap JP, Atanassov A, Stiekema WJ
(eds) Genomics for biosafety in plant biotechnology. NATO science series I, 359. IOS Press,
Amsterdam, pp 47–57
Lindon JC, Nicholson JK, Holmes E, Everett JR (2000) Metabonomics: metabolic processes studied
by NMR spectroscopy. Concepts Magn Reson 12:289–320
Lindon JC, Holmes E, Nicholson JK (2001) Pattern recognition methods and applications in
biomedical magnetic resonance. Prog Nucl Magn Reson Spectrosc 39:1–40
Lindon JC, Holmes E, Nicholson JK (2004) Toxological applications of magnetic resonance. Prog
Nucl Magn Reson Spectrosc 45:109–143
Manetti C, Bianchetti C, Bizzari M, Casciani L, Castro C, d’Ascenzo G, Delfini M, di Cocco ME,
Lagana A, Miccheli A, Motto M, Conti F (2004) NMR-based metabonomic study of transgenic
maize. Phytochemistry 65:3187–3198
Massart DL, Vandeginste BGM, Deming SN, Michotte Y, Kauffman L (1988) Chemometrics:
a textbook. Elsevier, Amsterdam
Nicholson JK, Lindon JC, Holmes E (1999) ‘Metabonomics’: understanding the metabolic re-
sponses of living systems to pathophysiological stimuli via multivariate statistical analysis of
biological NMR spectroscopic data. Xenobiotica 29:1181–1189
Noteborn HPJM, Lommen A, van der Jagt RC, Weseman JM (2000) Chemical fingerprinting for
the evaluation of unintended secondaty metabolic changes in transgenic food crops. J Biotech
77:103–114
Prabhu V, Chatson KB, Abrams GD, King J (1996) C-13 nuclear magnetic resonance detection
of interactions of serine hydroxymethyltransferase with C1-tetrahydrofolate synthase and
glycine decarboxylase complex activities in Arabidopsis. Plant Physiol 112:207–216
Pukalskas A, van Beek TA, de Waard P (2005) Development of a triple hyphenated HPLC-radical
scavenging detection-DAD-SPE-NMR system for the rapid identification of antioxidants in
complex plant extracts. J Chromatography A 1074:81–88
Ratcliffe RG, Shachar-Hill Y (2001) Probing plant metabolism with NMR. Annu Rev Physiol Plant
Mol Biol 52:499–526
Ratcliffe RG, Shachar-Hill Y (2005) Revealing metabolic phenotypes in plants: inputs from NMR
analysis. Biol Rev 80:27–43
Ratcliffe RG, Roscher A, Shachar-Hill Y (2001) Plant NMR spectroscopy. Prog Nucl Magn Reson
Spectrosc 39:267–300
Roberts JKM (2000) NMR adventures in the metabolic labyrinth within plants. Trends Plant Sci
5:30–34
Roscher NJ, Kruger NJ, Ratcliffe RG (2000) Strategies for metabolic flux analysis in plants using
isotope labelling. J Biotechnol 77:81–102
Schaefer J, Skokut TA, Stejskal EO, McKay RA, Varner JE (1981) Estimation of protein-turnover
in soybean leaves using magic angle double cross-polarization N-15 nuclear magnetic-
resonance. J Biol Chem 256:1574–1579
Schleucher J, Vanderveer PJ, Sharkey TD (1998) Export of carbon from chloroplasts at night.
Plant Physiol 118:1439–1445
Stoyanova R, Nicholls AW, Nicholson JK, Lindon JC, Brown TR (2004) Automatic alignment of
individual peaks in large high-resolution of spectral data sets. J Magn Reson 170:329–335
Viant MR (2003) Improved mthods for the acquisition and interpretation of NMR metabolomic
data. Biochem Biophys Res Commun 310:943–948
Vogels JTWE, Terwel L, Tas AC, van den Berg F, Dukel F, van der Greef J (1996) Detection of
adulteration in orange juices by a new screening method using proton NMR spectroscopy in
combination with pattern recognition techniques. J Agric Food Chem 44:175–180
Vogler B, Klaiber I, Roos G, Walter CU, Hiller W, Sandor P, Kraus W (1998) Combination of
LC-MS and LC-NMR as a tool for the structure determination of natural products. J Nat Prod
61:175–178
Ward JL, Harris C, Lewis J, Beale MH (2003) Assessment of 1 H NMR spectroscopy and multivariate
analysis as a technique for metabolite fingerprinting of Arabidopsis thaliana. Phytochemistry
62:949–957
Wolfender JL, Rodriguez S, Hostettmann K, Hiller W (1997) Liquid chromatography/ultra-
violet/mass spectrometric and liquid chromatography/nuclear magnetic resonance spec-
troscopic analysis of crude extracts of Gentianaceae species. Phytochem Anal 8:97–104
Wolfender JL, Ndjoko K, Hostettmann K (2001) The potential of LC-NMR in phytochemical
analysis. Phytochem Anal 12:2–22
I.7 Hetero-nuclear NMR-based Metabolomics
J. Kikuchi1,2,3 and T. Hirayama2,3,4,5
1 Introduction
Novel methods for measurement of living systems are making new break-
throughs in life science. In the era of the metabolome (analysis of all mea-
surable metabolites), a mass spectrometry (MS)-based approach is considered
to be the major technology (Aharoni et al. 2002; Fiehn 2002; Sumner et al.
2003), whereas a nuclear magnetic resonance (NMR )-based method is fre-
quently regarded as a minor technology due to its low sensitivity. However,
we intend to strengthen the NMR-based approach, using advantages of NMR
measurement, such as high quantification, non-invasive measurements, local-
ized in vivo spectroscopy, selectivity of nuclear environments, and validity of
structure analysis of diverse biomolecules including stereo-isomers. Attrac-
tive NMR-based metabolic analyses can be achieved by uniform stable isotope
labeling of organisms allowing the application of multi-dimensional NMR ex-
periments that have been used in protein structure determination (Kikuchi et
al. 2004; Kikuchi and Hirayama 2005). Using these novel methods, the dynamic
molecular networks inside cells and tissues will be dissected.
2 Historical Aspects of NMR Studies of Plant Metabolism
The history of NMR has been sharpened by a succession of major technological

and methodological advances, including greatly enhanced sensitivity due to
improvements in electronic devices, probe design, high-field superconducting
magnets, the field/frequency stability to allow multi-scan averaging, and also
the development of pulse Fourier-transform methods, significant progress
in data handling facilities, and the development of multi-dimensional NMR
1 RIKEN Plant Science Center, 1-7-22 Suehiro-cho, Tsurumi-ku, Yokohama, 230-0045 Japan,
2 International Graduate School of Arts and Sciences, Yokohama City University, 1-7-29 Suehiro,
Tsurumi-ku, Yokohama, 230-0045 Japan

3 CREST, Japan Science and Technology Agency, 4-1-8 Hon-cho, Kawaguchi, 332-0012 Japan
4 Genomic Sciences Center, RIKEN Yokohama Institute, 1-7-22 Suehiro, Tsurumi-ku, Yokohama,
230-0045 Japan
5 Laboratory of Plant Molecular Biology, RIKEN Tsukuba Institute, 3-1-1 Koyadai, Tsukuba, 305-
0074 Japan

94 J. Kikuchi and T. Hirayama
(Ernst 1992; Claridge 1999). NMR spectroscopy provides many new insights
into the physiology of higher plants. The evolution of this particular application
of NMR can be traced back to the ground-breaking 13 CNMR studies using
magic-angle spinning methods (Schaefer and Stejskal 1976). The subsequent
developments of the technique and its applications have been charted at regular
intervals in the review literature, and although not as widely exploited as its
proponents might wish, NMR is now becoming an established technique in the
armory of plant biochemists.
3 1 H-NMR-based Metabolomics
NMR signals are highly reproducible, and quantitative assessment of each
metabolite in a sample is therefore guaranteed. In contrast, MS-signals are
sometimes less quantitative due to problems of matrix effects (“ion suppres-
sion” or “ion enhancement”) (Mei et al. 2003; Mallet et al. 2004). Because
NMR is a nondestructive technique, it is easy to combine NMR analysis with
a complementary technique such as gas chromatography/MS or liquid chro-
matography/MS (Corcoran and Spraul 2003; Ott et al. 2003). In contrast to
these applications in which numerous specific metabolites can be identified in
complex mixtures, other investigators have addressed the question of whether
computer-aided comparisons of the 1 H NMR spectra of partially fractionated
extracts can yield statistically meaningful metabolic fingerprints of the ex-
tracted tissue. Using this approach, it was possible to show that there were min-
imal compositional differences between certain transgenic and non-transgenic
tobacco varieties, but only after accounting for the substantial effects of exter-
nal factors (Choi et al. 2004).
4 Use of Stable Isotope Labeling Technique to Enable

Monitoring of the Dynamic Movement of Metabolites
NMR signals can be detected from the nuclei of many isotopes; 1 H, 13 C, 15 N and
31 P are the most widely used for biological NMR spectroscopy (Ratcliffe et al.
2001). For carbon the relevant magnetic isotope is 13 C. Its natural abundance is
only 1.11%, contributing to the considerably lower sensitivity for 13C NMR than
for 1 H NMR. Accordingly, the application of 13 C NMR in unlabeled systems is
largely confined to the detection of the most abundant metabolites, such as the
organic solutes that accumulate in response to salt stress or certain secondary
metabolites (Ratcliffe and Sachar-Hill 2001). Indirect detection techniques
such as 13 C-hetero-nuclear single quantum coherence (HSQC) pulse sequence
increases the sensitivity of the experiments (Vuister and Bax 1992). An example
of this approach can be found in an analysis of alkaloid biosynthesis in vivo
(Hinse et al. 2003).
Hetero-nuclear NMR-based Metabolomics 95
The nitrogen atom has two magnetic isotopes, 14 N and 15 N, and both can be
useful for the detection of metabolites in vivo and in extracts. The practicality of
detecting the naturally abundant (99.63%) 14 N isotope was first demonstrated
in root tissues and subsequently in vivo 14 N NMR has mainly been used for the
analysis of ammonium and nitrate. The extremely low natural abundance of
the 15 N isotope (0.037%) rules out the detection of unlabeled metabolites, but
after labeling with [15 N]ammonium or [15 N]nitrate it is possible to use in vivo
15 N NMR to detect amino acids, as well as certain secondary products. NMR
methods are relatively insensitive, so only signals from compounds present at
relatively high levels (concentrations of at least 10 μmol/L) can be detected in
spectra (Krishnan et al. 2005). Since metabolic engineering often results in the
accumulation of relatively high concentrations of metabolites, this insensitivity
is often not as restrictive for compound detection and identification as it is in
other areas of biochemistry.
5 Approach for Hetero-nuclear NMR-based Metabolomics
In recent years, hetero-nuclear NMR methods and their spectral editing tech-
nologies have developed rapidly. For example, careful selection of window
functions and base-line corrections of two dimensional (2D)-spectra yielded
improved signal dispersion and line shapes of cross peaks permitting clear
subtraction 2D-spectra, a technically difficult and time-consuming procedure
using conventional 1D-NMR technology (Deferenz and Colquhoun 2003). With
the methodology used in recent protein NMR studies, differences in the molec-
ular composition between wild type and mutant strains can be easily quan-
tified. Therefore, we think advanced technologies in NMR analysis combined
with stable isotope labeling are useful tool for metabolomic analysis. We report
here stable isotope labeling experiments in Arabidopsis using carbon or nitro-
gen, two of the largest components of all organic compounds (Kikuchi et al.
2004; Kikuchi and Hirayama 2005). Figure 1 shows the basic concept of NMR-
based plant metabolomics proposed in this study. The NMR-based approach
has an advantage when comparing different samples. Spectral subtraction be-
tween different mutants or stimuli enables metabolite levels between different
samples to be quantified.
5.1 Example of Hetero-nuclear NMR Experiments In Vitro
To demonstrate the usefulness of the NMR method (metabolomics), the meta-

bolite profile of an ethanol-hypersensitive mutant of Arabidopsis, gek1 (Hi-
rayama et al. 2004), was analyzed (Fig. 2a). Arabidopsis seedlings were grown
for two weeks on agar plates (see above) and treated with 0.5% ethanol or
water for 10 h. Figure 2b shows the 1 H–13 C HSQC spectra of ethanol or water-
treated wild type Arabidopsis extracts. The subtraction spectra were obtained
Fig. 1. Comparison of ordinal metabolomics approach (left: PCA-based) and our hetero-nuclear
NMR metabolomics approach (right: multi-dimensional NMR-based)
by subtracting the spectrum of an ethanol-treated sample from that of a water-

treated sample. Figure 2c shows the subtraction spectra from the wild-type
(WT) (left) or the gek1 (right) samples. The subtraction of measured spectra
generates virtual NMR spectra that highlight compounds that are different
between samples. In the present case, the subtraction spectra clearly show that
upon ethanol treatment, glutamic acid is synthesized de novo in both the WT
and the mutant, consistent with the previous observation that ethanol is con-
verted into amino acids and lipids rapidly in plant tissues (Mellema et al. 2002;
Rawyler et al. 2002). In addition, the ethanol-hypersensitive gek1 mutant syn-
thesized proline and γ -amino butylic acid (GABA) de novo, two compounds
that have been reported to accumulate in cells under abiotic stresses such
as drought and salinity (for reviews, Hare et al. 1998; Shelp et al. 1999). The
assignments of these compounds were possible by comparing both 1 H and
13 C chemical shifts independently obtained with corresponding commercial
reagents. Since 13 C-NMR chemical shifts are sensitive to differences in chem-
ical structure but insensitive to the surrounding environment such as solvent
effects (Kikuchi and Asakura 1999), the 2D-HSQC type spectra offer exception-
ally useful information for assignment of individual chemical groups. From
this point of view, construction of a database of 2D-HSQC spectra of main
metabolites will enhance the NMR metabolomics.
Fig. 2. a Example of how spectral subtraction can be used to differentiate environmental stress
responses between WT and gek1 (Hirayama et al. 2004) mutant. b For NMR spectroscopy,
5 mg of freshly frozen samples were heated with 0.5 mL H2 O and centrifuged at 15,000 g for
5 min to remove insoluble fractions. After adding 50 μL of 2 H2 O for NMR lock, supernatants
were transferred into 5-mm NMR tubes. The spectra were measured on a Bruker DRX-500
spectrometer equipped with a 1 H inverse probe with triple axis gradient. A total of 200 complex
f1 (13 C) and 1024 complex f2 (1 H) points were recorded with 64 scans per f1 increment. The
spectral widths were 12,000 Hz and 8400 Hz for f1 and f2, respectively. c To quantify the signal
intensities, a Lorentzian-to-Gaussian window with a Lorentzian line width of 10 Hz and a Gaussian
line width of 15 Hz was applied in both dimensions, prior to Fourier transformation. A fifth order
polynomial baseline correction was subsequently applied in the f1 dimension (Kikuchi et al.
2002). The indirect dimension was zero-filled to 2048 points in the final data matrix. NMR
spectra were processed using NMRPipe software (Delaglio et al. 1995). Quantitative 2D-spectral
subtraction was accomplished by editing a macro program of the NMRPipe software. Signal
assignments are highlighted next to the corresponding cross peaks
5.2 Example of Hetero-nuclear NMR Experiments In Vivo

15 N uniformly labeled Arabidopsis seeds can be obtained from plants fed
with a nutrient solution containing 15 NO3 as the sole nitrogen source. Using
such seeds, the first 1 H–15 N HSQC-type NMR (Bodenhausen and Ruben 1980;
Grzesiek and Bax 1993) in vivo experiments in plants were performed (Kikuchi
et al. 2004). Figure 3 shows the development of the 1 H–15 N HSQC spectrum
measured in living 15 N-labeled seeds that was induced by soaking the dry
Fig. 3. Development of the 1 H–15 N HSQC spectrum measured in living 15 N-labeled seeds that
was induced by soaking the dry seeds in water (pictures shown at the bottom). A total of 128
complex f1 (15 N) and 1024 complex f2 (1 H) points were recorded with 96 scans per f1 increment.
The spectral widths were 4500 Hz and 8400 Hz for f1 and f2, respectively. Two spectra: a dark
at 0 h; b light at 78 h are shown for comparison. Signal assignments are highlighted next to the
corresponding cross peaks
seeds in water. Using in vivo measurement, dynamic movement of metabolites

can be observed. In this case, at the initial stage just after water absorption
into dried 15 N seeds, all cross peaks (especially those corresponding to peptide
backbones) were very broad due to slow molecular motion in the dried seeds
(Fig. 3a). After 12 h of imbibition, the line shapes of cross peaks, especially
those corresponding to the glycine backbone and the side-chains of glutamic
acid and asparatic acid, started to sharpen due to the enhanced molecular
motion caused by the increasing water content. The temperature shift from 4
to 22 ◦ C at 72 h of imbibition, and light illumination which started at 78 h of
imbibition, both of which accelerate germination, further sharpened all cross
peaks and enhanced the peptide backbone signal dramatically (Fig. 3b).
6 Prospects for the Future
As described above, NMR techniques possess an advantage over common

analytical methods because they simultaneously provide information on the
concentrations of numerous compounds as well as their spatial distribution.

Therefore, NMR offers useful methodology for metabolomics. At this moment,
however, there are several issues to be resolved before we can utilize the full
power of NMR measurement in metabolomics. First, the sensitivity of NMR is
rather lower. This disadvantage is being overcome with the progress in NMR
technology. The sensitivity of a spectrometer scales as the 7/4th power of the
static magnetic-field, and our group has developed the highest magnetic field
(21.2 Tesla) used in biomolecular studies (Kiyoshi et al. 2004). In addition,
NMR signal-to-noise (S/N) ratios can be substantially improved by cooling the
NMR radio frequency detector and preamplifier . We are currently develop-
ing a 4.5-K cryogenetically cooled probe for the 920-MHz NMR spectrometer
(Yokota et al. 2004). The increase of S/N gain is expected to be 8-fold, cor-
responding to a 64-fold reduction of the NMR acquisition time. The 1 H-13 C
HSQC spectrum (shown in Fig. 1) recorded by the 500-MHz spectrometer
exhibited 477 cross peaks identified by the NMRPipe software (Delaglio et
al. 1995), corresponding to 100–200 metabolites at concentrations over 10−6
mol/L. However, theoretically the 920-MHz spectrometer equipped with a 4.5-K
cryogenically cooled probe will be able to detect metabolites at concentrations
as low as 10−8 to 10−9 mol/L. Furthermore, S/N gain by the cryogenetically
cooled probe is significantly enhanced in low dielectric solvents (Horiuchi
et al. 2005). In other words, the 1 H–13 C HSQC spectra recorded by 64 scans
with the 500-MHz spectrometer (shown in Fig. 2) will be taken by only one
scan with equivalent S/N ratio for the same sample but with higher resolu-
tion due to the higher magnetic field. Second, to facilitate the identification of
metabolites in samples, a database of 2D HSQC spectra of known metabolites
is required. We have just started to construct such a database. Once devel-
oped, metabolite analyses will be conducted with simultaneous quantification
and metabolite identification. Furthermore, recent solid-state NMR methods
will facilitate the study of insoluble metabolites such as starch, cell-wall com-
ponents, and biomembranes (Kikuchi et al. 2000). Thus, the use of isotope
labeling together with newly developed NMR technologies open a new avenue
for plant metabolomics.
Acknowledgements. This work was supported in part by RIKEN GSC Internal Collaborations
(No. 830-56625), by CREST (No. A88-54366), Japan Science and Technology Agency to J.K., T.H.
We also acknowledge Grants-in-Aid for Scientific Research (No. 15710171, to J.K.; No. 15570045,
to T.H.) from the Ministry of Education, Science, Sports and Culture of Japan.
References
Aharoni A, Ric de Vos CH, Verhoeven HA, Maliepaard CA, Kruppa G, Bino R, Goodenowe DB
(2002) Nontargeted metabolome analysis by use of Fourier Transform Ion Cyclotron Mass
Spectrometry. OMICS 6:217–234
Bodenhausen G, Ruben DJ (1980) Natural abundance nitrogen-15 NMR by enhanced hetero-
nuclear spectroscopy. Chem Phys Lett 69:185–189
Choi H-K, Choi YH, Verberne M, Lefeber AWM, Erkelens C, Verpoorte R (2004) Metabolic
fingerprinting of wild type and transgenic tobacco plants by 1 H NMR and multivariate
analysis technique. Phytochemistry 65:857–864
Claridge TDW (1999) High-resolution NMR techniques in organic chemistry. Elsevier Science,
London, UK
Corcoran O, Spraul M (2003) LC-NMR-MS in drug discovery. Drug Discov Today 8:624–631
Defernez M, Colquhoun IJ (2003) Factors affecting the robustness of metabolite fingerprinting
using 1 H NMR spectra. Phytochemistry 62:1009–1017
Delaglio F, Grzesiek S, Vuister G W, Zhu G, Pfeifer J, Bax A (1995) NMRPipe: a multidimensional
spectral processing system based on UNIX pipes. J Biomol NMR 6:277–293
Ernst RR (1992) Nuclear magnetic resonance Fourier transform spectroscopy. Angew Chem Int
Ed Engl 31:805–823
48:155–171
Grzesiek S, Bax A (1993) The importance of not saturating H2 O in protein NMR. Application to
sensitivity enhancement and NOE measurements. J Am Chem Soc 115:12593–12594
Hare PD, Cress WA, van Staden J (1998) Dissecting the roles of osmolyte accumulation during
stress. Plant Cell Environ 21:535–553
Hinse C, Richter A, Provenzani J, Stöckgt J (2003) In vivo monitoring of alkaloid metabolism
in hybrid plant cell cultures by 2D cryo-probe NMR without labeling. Bioorg Med Chem
11:3913–3919
Hirayama T, Fujishige N, Kunii N, Iuchi S, Shinozaki K (2004) A novel ethanol hypersensitive
mutant of Arabidopsis. Plant Cell Physiol 45:703–711
Horiuchi T, Takahashi M, Kikuchi J, Yokoyama S, Maeda H (2005) Effect of dielectric properties
of solvents on the quality factor for a beyond 900 MHz cryogenic probe model. J Magn Reson
174:34–42
Kikuchi J, Asakura T (1999) Use of 13 C conformation-dependent chemical shifts to elucidate
the local structure of a large protein with homologous domains in solution and solid state.
J Biochem Biophys Method 38:203–208
Kikuchi J, Hirayama T (2005) Novel methods for uniform stable isotope labeling in plant and
animal systems for a hetero-nuclear NMR based metabomics. 1st Int Metabol Meeting
Kikuchi J, Williamson MP, Shimada K, Asakura T (2000) Structure and dynamics of photosyn-
thetic membrane-bound proteins in Rhodobacter sphaeroides, studied with solid-state NMR
spectroscopy. Photosyn Res 63:259–267
Kikuchi J, Iwahara J, Kigawa T, Murakami T, Okazaki T, Yokoyama S (2002) Solution structure
determination of the two DNA-binding domains in the Shizosaccharomyces pombe Abp1 pro-
tein by a combination of dipolar coupling and diffusion anisotropy restraints. J Biomol NMR
22:333–347
Kikuchi J, Shinozaki K, Hirayama T (2004) Stable isotope labeling of Arabidopsis thaliana for
a hetero-nuclear NMR-based metabolomics approach. Plant Cell Physiol 45:1099–1104
Kiyoshi T, Maeda H, Kikuchi J, Ito Y, Hirota H, Yokoyama S, Ito S, Miki T, Hamada M, Ozaki O,
Hayashi S, Kurihara N, Suematsu H, Yoshikawa M, Matsumoto S, Sato A, Wada H (2004)
Present status of 920 MHz high-resolution NMR spectrometers. IEEE Trans Appl Supercond
14:1608–1612
Krishnan P, Kruger NJ, Ratcliffe RJ (2005) Metabolic finger printing and profiling in plants by
NMR. J Exp Bot 56:255–265
Mallet CR, Lu A, Mazzeo JR (2004) A study of ion suppression effects in electrospray ioniza-
tion from mobile phase additives and solid-phase extracts. Rapid Commun Mass Spectrom
18:49–58
Mei H, Hsieh Y, Nardo C, Xu X, Wang S, Ng K, Korfmacher WA (2003) Investigation of matrix ef-
fects in bioanalytical high-performance liquid chromatography/tandem mass spectrometric
assays: application to drug discovery. Rapid Commun Mass Spectrom 17:97–103
Mellema S, Eichenberger W, Rawyler A, Suter M, Tadege M, Kuhlemeier C (2002) The ethanolic

fermentation pathway supports respiration and lipid biosynthesis in tobacco pollen. Plant J
30:329–336
Ott K-H, Aranibar N, Singh B, Stockton GW (2003) Metabolomics classifies pathways affected
by bioactive compounds. Artificial neural network classification of NMR spectra of plant
extracts. Phytochemistry 62:971–985
Ratcliffe RJ, Shachar-Hill Y (2001) Probing plant metabolism with NMR. Annu Rev Plant Physiol
Plant Mol Biol 52:499–526
Ratcliffe RJ, Roscher A, Shachar-Hill Y (2001) Plant NMR spectroscopy. Prog NMR Spect 39:267–
300
Rawyler A, Arpagaus S, Braendle R (2002) Impact of oxygen stress and energy availability on
membrane stability of plant cells. Ann Bot 90:499–507
Schaefer J, Stejskal EO (1976) C-13 Nuclear magnetic resonance of polymers spinning at magic
angle. J Am Chem Soc 98:1031–1032
Shelp BJ, Bown AW, McLean MD (1999) Metabolism and functions of gamma-aminobutyric acid.
Trends Plant Sci 4:446–452
Vuister GW, Bax A (1992) Resolution enhancement and spectral editing of uniformly 13 C enriched
proteins by homonuclear broadband 13 C–13 C decoupling. J Magn Reson 98:428–435
Yokota H, Okamura T, Ohtani Y, Kuriyama T, Takahashi M, Horiuchi T, Kikuchi J, Yokoyama S,
Maeda H (2004) 4.5 K cooling system for a cryogenically cooled probe of a 920 MHz NMR.
Adv Cryo Eng 49:1826–1833
II.1 Bioinformatics Approaches to Integrate
Metabolomics and Other Systems Biology Data
B. Mehrotra and P. Mendes1
1 Introduction
To understand the functioning of cells fully it is important to unravel the roles of

genes and their products. The study of gene transcripts (transcriptomics) and
proteins (proteomics) is progressing rapidly through the use of microarrays
and mass spectrometry. Additionally, cells contain numerous other organic
molecules not directly encoded in the DNA, the metabolites, which are critical
for cell function. Knowledge about metabolites is crucial for an understanding
of most cellular phenomena (Weckwerth 2003; Fernie et al. 2004; Kell 2004).
Metabolomics is an emerging field consisting of the study of metabolites at
a systems scale. It is similar in objectives to transcriptomics and proteomics;
two major goals of metabolomics are the identification of all metabolites in each
organism (their metabolomes) and measurements of their dynamics under
many different challenges. Integrated approaches combining metabolomics
with transcriptomics and proteomics are now underway (e. g. Verhoeckx et al.
2004; Broeckling et al. 2005) and are expected to result in much deeper insights
than any of these techniques alone.
Metabolomics shares several characteristics with proteomics and transcrip-
tomics. Like these, it is a technique where large numbers of molecules are
profiled simultaneously (though current methods identify only hundreds of
metabolites, vs thousands of proteins, and tens of thousands of transcripts).
Metabolite profiles are, like transcript and protein profiles, snapshots of the
state of a biological sample. Experimental data of all three types are usually
dominated by a number of variables (molecules) much larger than samples,
posing a hard challenge for data analysis and interpretation. Like proteomics,
most metabolomics methods rely on spectroscopies to identify molecules; how-
ever, this is done through comparison against standards, rather than by mass
fingerprints or sequence. The lack of a concept of “sequence” for metabolites
is a major difference between metabolomics and the other two methodologies.
Sequence is the key for identification of proteins and nucleic acids in large-
scale profiles, but alternative methods must be used for metabolite profiles.
De novo identification of metabolites can be done with 2D NMR, but requires
considerable amounts of highly purified material, a major obstacle. Tandem
mass spectrometry requires smaller amounts of material, but alone is often
1 Virginia Bioinformatics Institute, Virginia Polytechnic Institute and State University, Washing-
ton St., MC 0477, Blacksburg, Virginia 24061, USA, e-mail: [email protected]

106 B. Mehrotra and P. Mendes
insufficient to identify completely unknown metabolites. The alternative to

de novo metabolite identification is to construct a library of standard profiles,
created with purified metabolites (e. g. Wagner et al. 2003). While large spectral
libraries for NMR, IR and mass spectrometry have existed for decades, these
are very incomplete compared to the nearly 200,000 known natural products
(Buckingham 1994). Additionally, these published spectral libraries are often
not accurate enough for clear identifications. In the case of chromatography-
mass spectrometry techniques, much better results are obtained if one has
constructed a library with one’s own equipment, which requires a large invest-
ment of time and finance. Thus, metabolite identification is one of the fun-
damental differences between metabolomics and proteomics/transcriptomics,
and it is common to find two thirds or more unidentified metabolites in non-
targeted profiles. The second major difference between these techniques is that
metabolites have widely different chemical properties, such as polarity, volatil-
ity, molecular mass, and chemical reactivity. Comparatively, nucleic acids are
very uniform in their properties; proteins, while more diverse than the former,
are still approachable by their common properties (such as the amide bonds
that can be used to sequence them). Because metabolites vary in their com-
position and structure, they require many different methods for extraction
and separation, and no single existing technique is able to profile all metabo-
lites in a biological sample (Sumner et al. 2003). Comprehensive coverage of
the metabolome requires parallel analyses carried out with several different
techniques.
Since the turn of the century, systems approaches have regained popularity
with biologists. Perhaps this is because the analysis of purified molecules is
rapidly approaching its limit, or simply because global experimental analyses
have become possible. Either way, it is now recognized that systems biology
studies of complex cellular phenomena are sorely needed (Kitano 2002). An
increasingly appealing approach consists of experiments that simultaneously
monitor the levels of transcripts, proteins, and metabolites, and combine their
data to make inferences about the structure and dynamics of the underlying
biochemical networks (Mendes 2001; Mendes et al. 2002; Oliver et al. 2002).
In order to integrate the diverse and large amount of data generated by such
experiments, several statistical and computational methods are required. In
particular, it is important that all data be managed in a single database which
also keeps track of the intricate details about how the experiments have been
designed and the data generated. Ultimately, the data must be traceable back-
wards to samples and experiments, allowing not only for their interpretation
but also for enabling others to replicate the experiments. These data about data
are usually known as metadata and there have been several attempts at stan-
dardizing them. The MIAME standard was proposed for transcriptomics data
(Brazma et al. 2001) and received widespread support, including support by
prominent journals requiring data to conform to that standard. Similar propos-
als for proteomics data have been put forward, e. g. PEDRo (Taylor et al. 2003)
and MIAPI (Orchard et al. 2004), but these are still under development and have
Bioinformatics Approaches to Integrate Metabolomics and Other Systems Biology Data 107
not yet received the crucial support from the publishing world. Metabolomics
is no different, and recently two proposals have been published to define stan-
dards for plant metabolomic data and metadata, ArMet (Jenkins et al. 2004)
and MIAMET (Bino et al. 2004). Invariably, these attempts build upon the
MIAME standard and hopefully will soon allow unequivocal specification of
systems biology data. (The existing Systems Biology Markup Language, SBML
(Hucka et al. 2003), is a standard for specifying systems biology models, rather
than data.)
In the remainder of this chapter we will delineate ongoing efforts in our lab-
oratory pertaining to integrating metabolomics and other functional genomics
data. These efforts have arisen from our participation in plant systems biology
studies of Medicago truncatula and Vitis vinifera, and also of the yeast Saccha-
romyces cerevisiae. All of these are large team efforts and we acknowledge all
our collaborators for their vital role in these projects (see below).
2 Databases
We are developing a database system, DOME (database for OMEs), which

stores functional genomics data originated from microarray measurements of
transcripts, 2D-PAGE-QTOFMS protein assays, and GC-MS, LC-MS, CE-MS,
and CE-LIF assays of metabolites. Early on it became evident that, in order to
analyze all these data sets in an integrated way, they should be stored in a single
database. This avoids, by design, the infamous data integration problem of
bioinformatics (Davidson et al. 1995), as all data reside in the same schema,
and queries can be made across all of them irrespective of their nature. This
integration on a single schema was only possible by assuring that all necessary
metadata was included and structured in an appropriate way. Figure 1 depicts
a high-level overview of the DOME schema.
The main skeleton of the metadata schema consists of a hierarchy of ex-
periment sets, experiments, sampling points, samples, and extracts, following
exactly the way in which the biological material is manipulated in the experi-
ments. This metadata schema also allows for easy export of transcript data in
a format compatible with MIAME, and hopefully with the future proteomics
and metabolomics standards as they stabilize (it is already compatible with PE-
DRo and ArMet). Linking through metadata is one of the main ways in which
one can put together data from metabolomics with microarray and proteomics.
For example, one single sampling point (a sample collection time) is attached
to a set of perturbed and control replicate transcript levels, protein levels, and
metabolite levels, as well as to their respective average values, comparisons
and statistical significance. Thus one can relate the metabolite levels at one
time in the experiment with the transcript and protein levels and, because
the sampling points reflect experiment time, also to the molecular levels of
previous or subsequent sampling points.
Fig. 1. High level schema of the integrative functional genomics database DOME. Metadata
tables are used to provide context to the actual experimental data. Raw data from microarray,
2D-PAGE-MS, and various metabolomics technologies are stored in separate tables. These data
are transformed by their appropriate normalization methods, keeping intermediate values, and
finally arriving at numerical summaries of each sample (such as means and standard deviations),
which are then comparable across all technologies and stored in the sp summary table. It is the
data in sp summary that is then processed with higher-level statistical analyses or visualizations.
It is also at this level that background information about the known molecular biology of the
system (B-Net) can be integrated
Another feature that makes the comparison of metabolite with transcript

and protein data possible is that each of these data types is first processed in
appropriate ways, resulting in data of a similar type. Presently these data are
reduced to ratios of levels (usually the level of a molecule in the perturbed
state over its level in the control). Before data are available in this state they
need to be processed through methods that are different for each technique.
For example, the microarray data goes through a series of standardization
and normalization procedures that take into account the technical details of
microarray technology (Quackenbush 2001). The metabolite GC-MS data are
first corrected for sample size, then deconvoluted into a series of peaks, which
are identified by the software AMDIS (Davies 1998) relative to an internal
standard that was included in the samples. An important issue is that the data
should all be represented either in linear or in logarithmic space, not a mixture
of the two. It is also important to preserve raw data, because there is always the
possibility that in the future better methods to process it will appear; however,
because the raw data are not commonly used in the analysis, it is enough to store
these offline as long as the database keeps appropriate track of their location.
A third way to relate metabolite data with transcript and protein data is
through the use of existing biochemical knowledge. This is, of course, the
traditional way to analyze such comparisons; however, it is usually done by

experts in an ad-hoc way. In order for this to be automated into computational
procedures, it is necessary to represent this domain knowledge in appropriate
schemas. Several biochemical databases exist that partially fulfill our needs
(Kanehisa et al. 2004; Krieger et al. 2004), though they fall short in a number
of ways (Wittig and de Beuckelaer 2001; Xing Li et al. 2002). In particular, they
are rather poor in terms of their coverage of plant secondary metabolism, even
AraCyc (Mueller et al. 2003), which is specific for Arabidopsis. Thus we created
a sub-schema in our database to represent the existing knowledge about the
biochemistry of the species in question. Because this can be useful on its own
(i. e. independent of the experimental data), it was designed in a manner that
allows it to be an autonomous database, and has been named B-Net. B-Net has
been populated with gene/transcript information from TIGR’s gene indices
(Lee et al. 2005) and SGD (Dwight et al. 2004), with protein information from
UniProt (Bairoch et al. 2005), metabolite information from LIGAND (Kanehisa
et al. 2004) and AraCyc (Mueller et al. 2003), and supplemented with data
collected directly from the literature by a team of curators in our laboratory.
All of the facts in B-Net are documented for the type of evidence that supports
them, using a method generalized from the Gene Ontology’s evidence codes
(Ashburner et al. 2000). Note that the information imported from external
databases is filtered to remove entries that do not represent specific molecular
entities. This was particularly important in the case of LIGAND, where groups
of molecules are represented alongside individual molecules (e. g. “amino-
acids”, instead of the specific ones); however only the latter were imported to
B-Net. B-Net also classifies entries with gene ontology terms wherever possible.
It is relevant here to highlight two problems that pervade metabolomics
databases. The first is the issue of metabolites that are detected but not clearly
identified, already mentioned above. These metabolites are sometimes referred
to as “unknowns”. Despite their identity not being known, a database must
distinguish between them, and so these are usually named by the analytical
chemists through some ad-hoc scheme. Such names are often attributed in
ways that prevent comparison of data between different labs, for example
by choosing identifiers that are used in different contexts meaning different
things. A negative consequence would be that in two separate experiments
two unknown metabolites might receive a common name, even though no one
had intended to mean that they were the same molecular entity. In order to
overcome this problem, and because our database contains data originated
from several laboratories, we have developed a naming scheme that assures
that unknown metabolites from different experiments and laboratories are not
accidentally named the same thing. This naming convention has been proposed
to the community in a recent Opinion article (Bino et al. 2004) and we hope
that it becomes adopted by many laboratories, as this is the only way in which
it would become useful. Essentially, the name for each unknown metabolite is
composed of an identifier for the laboratory, one for the extraction method,
another one for the type of analysis carried out, and at least two coordinates
from the analysis. These coordinates are context specific, and could be retention
time of a separation, mass-to-charge ratio, chemical shift, wavenumber, etc.
Another objective with this naming convention is to allow for future analyses
that attempt to establish identity between these unknown metabolites. It is
expected that many of these are observed in different studies in separate
laboratories, and through the inclusion of the analysis coordinates in their
names it becomes easier to recognize that two unknowns may actually be the
same molecule. For example, if several studies consistently identified a peak
in GC-MS (using the same extraction and analysis parameters) with the same
retention time and main ion mass, then it may be that the two are the same
metabolite. By assigning names derived from this scheme, it then also becomes
possible to create lists of molecular entities that have been observed and not
yet identified (a kind of “orphan” list for metabolomics).
Another unresolved issue that is being encountered in our projects is that
the same metabolites in a sample might have been observed by more than one
technique. The problem that is posed then is which quantification should one
chose if they do not agree. This is complicated by the fact that when metabolites
appear in an analysis, they may not have been present in the original sample,
but instead result from an artifact of the extraction method. Another reason
could be that the same metabolite might be present in different locations in
the cell, leading to the metabolite being isolated in two separate pools. In the
latter case the two pools should both be represented in the database, while
in the case of artifacts, one should use only the more accurate quantification.
This issue results in a need for careful annotation of metabolomics results,
but also requires special structures in the database schema that are capable of
representing several pools of a single metabolite.
3 Data Visualization
Scientific data visualization is the activity of displaying properties of a data set

that help the human scientist to identify quickly its most important charac-
teristics. This is not a simple problem because it is very hard to identify what
would be important for each scientist, and it is as much an issue of the scientific
domain as it is of psychology. Nevertheless, there are data properties that are
generally sought by a large class of researchers, and visualization software is
focused on them. For metabolomics, a frequent way in which biochemists like
to visualize data is through the use of maps that depict portions of the bio-
chemical network. Several packages exist that allow for this map-based data
visualization (Luyf et al. 2002; Shannon et al. 2003; Thimm et al. 2004; Lange
and Ghassemian 2005), and we have also developed one (BROME, BRowser
for OMEs) that is coupled to our database system DOME for visualization of
metabolomics data with transcriptomics and/or proteomics data. This allows
our database to select only a small set of metabolites, enzymes and genes that
are present in a certain map, such that the researcher can then quickly ob-
serve their levels organized according to how we believe the biochemistry is
organized. This could help in understanding how changes in mRNA or protein
levels affect the level of metabolites in a certain pathway or network. However
this is not as straightforward as it may seem: the changes in level of mRNA are
likely very different from changes in protein levels or changes in metabolite
levels. Cells cannot tolerate large changes of many metabolites, while mRNA
levels can change widely without much toxicity. Thus, in order to visualize
the expression of metabolites, mRNA, and proteins in the same biochemical
map, they need to be expressed on different scales, or otherwise normalized to
some comparable scale. A problem with thinking about data as part of some
biochemical map (“pathway”) is that it is likely that molecules in the map are
also involved in other interactions not depicted there. Therefore, looking at
a particular slice of a network could be highly misleading. It has been shown
that the concentrations of metabolites next to each other in a metabolic map do
not necessarily have high correlation (Steuer et al. 2003; Camacho et al. 2005),
strengthening this point. In order to understand a change in the level of a par-
ticular metabolite, it may be more useful to view the expression changes of all
enzymes (i. e. their protein and mRNA levels) linked with that metabolite. For
this we have developed the concept of metabolite neighborhood maps (Xing Li
et al. 2002), which are local views of the biochemical network and consist of all
the reactions that affect the metabolite of interest, including all the metabolites
and enzymes that take part in those reactions. BROME has a large number of
maps available, from these neighborhood maps to the nice pathway maps of
the KEGG system (Kanehisa et al. 2004).
4 Data Analysis
Analysis of metabolomic data can use the same multivariate statistical meth-
ods that are widely used in microarray data analysis. These methods can be
either supervised, where each sample or variable (molecule) is associated to
an already known class, or unsupervised, where there is no pre-classification
of the data (Mendes 2002; Sumner et al. 2003; Goodacre et al. 2004). Unsuper-
vised methods are widely popular, and the most used are principal component
analysis (PCA), hierarchical clustering (HCA), k-means clustering, and self-
organizing maps (SOM). Unsupervised analyses are mostly guided by the vari-
ance and covariance (or correlation) in the data sets, so they are good at finding
patterns therein; however nothing guarantees, other than a careful experimen-
tal design, that the largest variance is indeed a result of the perturbation rather
than other unwanted effects. On the other hand, supervised analyses are guided
by the pre-existing knowledge provided by the researcher and so are usually
based on discrimination, a property that is more related to the consistency
of the members in a class, and the differences between classes. Supervised
methods already demonstrated for metabolomics data are linear discriminant

analysis (Raamsdonk et al. 2001; Bundy et al. 2005), discriminant partial least
squares (PLS-DA) (Gavaghan et al. 2002; Jonsson et al. 2004), genetic algo-
rithms (Johnson et al. 2003; Goodacre 2005), genetic programming (Allen et
al. 2003; Goodacre 2005) and other methods (Goodacre et al. 2000; Shi et al.
2004).
Obviously, much more information could be extracted from systems biology
experiments if metabolomics data were analyzed coupled with transcriptomics
and proteomics data, but this requires much attention to ensure that the data
be comparable, as has been discussed in the previous section (see also Purohit
et al. 2004). For example, multidimensional scaling should be preferred to
principal component analysis since the former takes into account the different
scales of each variable, but the latter does not (at least in its original incarnation
that is based on covariance).
We have recently tried to understand how to interpret correlation between
metabolites in metabolomics data sets (Camacho et al. 2005) following results
from an analysis by Steuer et al. (2003). Briefly, we have found that strong
correlation arises when two or more metabolites are in chemical equilibrium,
when they conserve a common chemical moiety, when their concentrations are
mostly controlled by a single enzyme, or when one of the enzymes that affects
their concentration changes with greater magnitude than the other enzymes
that also affect them (Camacho et al. 2005). We confirmed that two metabolites
next to each other in a biochemical network (e. g. substrate and product of
a single enzyme) are not expected to be strongly correlated, in general. These
conclusions put the results that have been obtained demonstrating high cor-
relations between metabolite pairs into a systems biology context; often such
correlations are only explained at a global level, and through the action of
proteins.
Since metabolomics data are best understood when accompanied by other
systems biology data, there seems to be a great need for methods that specif-
ically address data integration. As demonstrated by the study of metabolite
correlations, methods that take into consideration pre-existing biochemical
knowledge are likely to be more effective. Since many metabolomics experi-
ments are carried out through time intervals, methods that consider the time
dimension explicitly are likely to be more productive than those that do not.
5 Conclusion
In this chapter we have addressed some of the bioinformatic issues related

to metabolomics and its integration within the systems biology framework.
We believe that metabolite, transcript, and protein analyses are much more
powerful combined than individually. In order to extract maximal benefit from
such combined studies, specific bioinformatics support is necessary in the form
of databases, visualization, and data analysis. Ultimately, a full understanding

of the underlying phenomena will require an additional layer of computational
and theoretical tools, supporting the formulation and evaluation of dynamic
models that attempt to represent the biological system. Such models will need
to be predictive, but we believe that, much more than that, they need to be
explanatory. Within our laboratory we are pursuing several projects in this
direction and have a strong interest in combining that approach with the data
and informatics systems described here, as have others. This will be a topic of
much discussion in the near future and we await it with excitement.
Acknowledgements. We thank our collaborators John Cushman, Grant Cramer, Rick Dixon, Greg
May, David Schooley, Vladimir Shulaev, and Lloyd Sumner for the excellent experimental and
analytical data collection work in their laboratories. We also thank our colleagues Xing Li and
Aejaaz Kamal for their work on the DOME and BROME systems, and Diogo Camacho and Alberto
de la Fuente for the metabolite correlation analysis. We acknowledge the generous support of
the National Science Foundation’s Plant Genome Research Program (awards DBI–0109732 and
DBI–0217653).
References
Allen J, Davey HM, Broadhurst D et al. (2003) High-throughput classification of yeast mutants
for functional genomics using metabolic footprinting. Nat Biotechnol 21:692–696
Ashburner M, Ball CA, Blake JA et al. (2000) Gene ontology: tool for the unification of biology.
The Gene Ontology Consortium. Nat Genet 25:25–29
Bairoch A, Apweiler R, Wu CH et al. (2005) The Universal Protein Resource (UniProt). Nucleic
Acids Res 33:D154–D159
Bino RJ, Hall RD, Fiehn O et al. (2004) Potential of metabolomics as a functional genomics tool.
Trends Plant Sci 9:418–425
Brazma A, Hingamp P, Quackenbush J et al. (2001) Minimum information about a microarray
experiment (MIAME)-toward standards for microarray data. Nature Genet 29:365–371
Broeckling CD, Huhman DV, Farag MA et al. (2005) Metabolic profiling of Medicago truncatula
cell cultures reveals the effects of biotic and abiotic elicitors on metabolism. J Exp Bot
56:323–336
Buckingham J (1994) Dictionary of natural products. Chapman and Hall/CRC, London
Bundy JG, Willey TL, Castell RS, Ellar DJ, Brindle KM (2005) Discrimination of pathogenic clinical
isolates and laboratory strains of Bacillus cereus by NMR-based metabolomic profiling. FEMS
Microbiol Lett 242:127–136
Camacho D, de la Fuente A, Mendes P (2005) The origin of correlations in metabolomics data.
Davidson SB, Overton C, Buneman P (1995) Challenges in integrating biological data sources.
J Comput Biol 2:557–572
Davies T (1998) The new Automated Mass Spectrometry Deconvolution and Identification System
(AMDIS). Spectroscopy, Europe 10:24–27
Dwight SS, Balakrishnan R, Christie KR et al. (2004) Saccharomyces genome database: underlying
principles and organisation. Brief Bioinform 5:9–22
Gavaghan CL, Wilson ID, Nicholson JK (2002) Physiological variation in metabolic phenotyping
and functional genomic studies: use of orthogonal signal correction and PLS-DA. FEBS Lett
530:191–196
Goodacre R (2005) Making sense of the metabolome using evolutionary computation: seeing the
wood with the trees. J Exp Bot 56:245–254
Goodacre R, Shann B, Gilbert RJ et al. (2000) Detection of the dipicolinic acid biomarker in Bacil-
lus spores using Curie-point pyrolysis mass spectrometry and Fourier transform infrared
spectroscopy. Anal Chem 72:119–127
Goodacre R, Vaidyanathan S, Dunn WB, Harrigan GG, Kell DB (2004) Metabolomics by numbers:
acquiring and understanding global metabolite data. Trends Biotechnol 22:245–252
Hucka M, Beale M, Fiehn O et al. (2003) The systems biology markup language (SBML): a medium
for representation and exchange of biochemical network models. Bioinformatics 19:524–531
Jenkins H, Hardy N, Beckmann M et al. (2004) A proposed framework for the description of plant
metabolomics experiments and their results. Nat Biotechnol 22:1601–1606
Johnson HE, Broadhurst D, Goodacre R, Smith AR (2003) Metabolic fingerprinting of salt-stressed
tomatoes. Phytochem 62:919–928.
Jonsson P, Broadhurst D, Goodacre R et al. (2004) A strategy for identifying differences in large
series of metabolomic samples analyzed by GC/MS. Anal Chem 76:1738–1745
Kanehisa M, Goto S, Kawashima S, Okuno Y, Hattori M (2004) The KEGG resource for deciphering
the genome. Nucleic Acids Res 32:D277–D280
Kell DB (2004) Metabolomics and systems biology: making sense of the soup. Curr Opin Microbiol
7:296–307
Kitano H (2002) Systems biology: a brief overview. Science 295:1662–1664
Krieger CJ, Zhang P, Mueller LA et al. (2004) MetaCyc: a multiorganism database of metabolic
pathways and enzymes. Nucleic Acids Res 32:D438–D442
Lange BM, Ghassemian M (2005) Comprehensive post-genomic data analysis approaches inte-
grating biochemical pathway maps. Phytochemistry 66:413–451
Lee Y, Tsai J, Sunkara S et al. (2005) The TIGR Gene Indices: clustering and assembling EST and
known genes and integration with eukaryotic genomes. Nucleic Acids Res 33:D71–D74
Luyf AC, de Gast J, van Kampen AH (2002) Visualizing metabolic activity on a genome-wide
scale. Bioinformatics 18:813–818
Mendes P (2001) Modeling large scale biological systems from functional genomic data: parameter
estimation. In: Kitano H (ed) Foundations of systems biology. MIT Press, Cambridge, MA,
pp 163–186
Mendes P (2002) Emerging bioinformatics for the metabolome. Brief Bioinform 3:134–145
Mendes P, de la Fuente A, Hoops S (2002) Bioinformatics and computational biology for plant
functional genomics. Rec Adv Phytochem 36:1–13
Mueller LA, Zhang P, Rhee SY (2003) AraCyc: a biochemical pathway database for Arabidopsis.
Plant Physiol 132:453–460
Oliver DJ, Nikolau B, Wurtele ES (2002) Functional genomics: high-throughput mRNA, protein,
and metabolite analyses. Metabolic Eng 4:98–106
Orchard S, Hermjakob H, Julian RK et al. (2004) Common interchange standards for proteomics
data: Public availability of tools and schema. Proteomics 4:490–491
Purohit PV, Rocke DM, Viant MR, Woodruff DL (2004) Discrimination models using variance-
stabilizing transformation of metabolomic NMR data. Omics 8:118–130
Quackenbush J (2001) Computational analysis of microarray data. Nat Rev Genet 2:418–427
Raamsdonk LM, Teusink B, Broadhurst D et al. (2001) A functional genomics strategy that uses
metabolome data to reveal the phenotype of silent mutations. Nature Biotechnol 19:45–50
Shannon P, Markiel A, Ozier O et al. (2003) Cytoscape: a software environment for integrated
models of biomolecular interaction networks. Genome Res 13:2498–2504
Shi H, Paolucci U, Vigneau-Callahan KE, Milbury PE, Matson WR, Kristal BS (2004) Development
of biomarkers based on diet-dependent metabolic serotypes: practical issues in development
of expert system-based classification models in metabolomic studies. Omics 8:197–208
Steuer R, Kurths J, Fiehn O, Weckwerth W (2003) Observing and interpreting correlations in
metabolomic networks. Bioinformatics 19:1019–1026
Taylor CF, Paton NW, Garwood KL et al. (2003) A systematic approach to modeling, capturing,
and disseminating proteomics experimental data. Nat Biotechnol 21:247–254
Thimm O, Blasing O, Gibon Y et al. (2004) MAPMAN: a user-driven tool to display genomics data
sets onto diagrams of metabolic pathways and other biological processes. Plant J 37:914–939
Verhoeckx KC, Bijlsma S, Jespersen S et al. (2004) Characterization of anti-inflammatory com-
pounds using transcriptomics, proteomics, and metabolomics in combination with multi-
variate data analysis. Int Immunopharmacol 4:1499–1514
Wagner C, Sefkow M, Kopka J (2003) Construction and application of a mass spectral and
retention time index database generated from plant GC/EI-TOF-MS metabolite profiles.
Weckwerth W (2003) Metabolomics in systems biology. Annu Rev Plant Biol 54:669–689
Wittig U, de Beuckelaer A (2001) Analysis and comparison of metabolic pathway databases. Brief
Bioinform 2:126–142
Xing Li X, Brazhnik O, Kamal A et al. (2002) Databases and visualization for metabolomics. In:
Harrigan GG, Goodacre R (eds) Metabolic profiling: its role in biomarker discovery and gene
function analysis. Kluwer Academic Publ, Boston, pp 293–309
II.2 Chemometrics in Metabolomics – An Introduction
J. Trygg1 , J. Gullberg2 , A.I. Johansson2 , P. Jonsson1 , and T. Moritz2
1 Introduction
In the post-genomics era, the use of methodologies that enable transcriptomic,

proteomic and metabolomic data to be analysed in detail have revolutionized
biological investigations. One of the major advantages with metabolomics in-
vestigations compared to traditional target metabolite analysis is that metabol-
omics data can give an unbiased view of changes in metabolism during envi-
ronmental, genetic or developmental changes. Instead of tracking only a few
metabolites, changes in relative amounts in 300 to 1000 or even more metabo-
lites can be recorded and analysed, covering all major metabolic pathways. This
development has accentuated the need to apply and further develop multivari-
ate methodology. Chemometrics (see Eriksson et al. 2001) provides tools to
make good use of measured data, enabling practitioners to make sense of mea-
surements and to model quantitatively and produce visual representations of
information. Today, chemometrics has grown into a well established data analy-
sis tool in areas such as multivariate calibration, quantitative structure-activity
modeling, pattern recognition and multivariate statistical process monitoring
and control. Although seemingly diverse disciplines, the common denomina-
tors in these applications are that high complexity data tables are generated and
that these data tables can be analysed and interpreted by means of chemometric
methods.
In chemometrics, there are three basic categories of analysis (Fig. 1):
1. Exploratory analysis (Fig. 1A). This gives an overview of all the data in order
to detect trends, patterns or clusters.
2. Classification analysis and discriminant analysis (Fig. 1B), which classifies
samples into categories or classes, for example wild-type and mutant.
3. Regression analysis and prediction models (Fig. 1C) are used when a quanti-
tative relationship between two blocks of data is sought. For example, when
prediction of growth or fiber properties from mass spectrometry data.
However, in biology, chemometric methodology has still been largely over-
looked in favour of traditional statistics. It is not until recently that the
1 Research Group for Chemometrics; Organic Chemistry, Department of Chemistry, Umeå Uni-
versity, SE-901 87 Umeå, Sweden, e-mail: [email protected]
2 Umeå Plant Science Centre, Department of Forest Genetics and Plant Physiology, Swedish Uni-
versity of Agricultural Sciences, SE-901 87 Umeå, Sweden, e-mail: [email protected]

118 J. Trygg et al.
Fig. 1. Overview of the basic categories of chemometrics analysis: A overview of data structure;
B classification and discriminant analysis; C regression analysis
overwhelming size and complexity of the ‘omics’ technologies has driven

biology towards the adoption of chemometric methods. Here we will give
an introduction to chemometrics and also give examples of why and when
chemometrical methodologies should be used.
2 Theory and Methods
2.1 Making Data Contain Information – Design of Experiments
In experimental biology, e. g. when investigating how a number of different

environmental factors (e. g. temperature, day length, nutrition) affect differ-
ent responses such as growth, transcript profiles and metabolite profiles in
plants, there is a need to carry out experiments in a systematic way. One way
to investigate how the factors affect the plant’s responses is to Change One
Factor at a Time, i. e. the COST approach. This approach has severe problems:
(1) finding optimal conditions for experiments (e. g. method development),
(2) unnecessarily many experiments are needed (inefficiency), (3) ignores in-
teraction among variables (lost information) and (4) provides no map over the
experimental space.
Design of Experiments (DOE) (Lundstedt et al. 1998) is the methodology of
how to conduct and plan experiments in order to extract the maximum amount
of information in the fewest number of runs. The basic idea is to devise a small
set of experiments, in which all pertinent factors are varied systematically. It is
a fundamental tool for planning experiments and making data informative by
simultaneously, albeit in a structured way, varying controllable factors (e. g. en-
vironmental conditions, instrument settings, experimental procedures) of the
studied system. Today they comprise a tool box for virtually any experimental
problem.
2.1.1 Stages in the DOE Process
Most of us can only grasp the effect of one factor at a time in our minds, and
that often leads us into the inefficient COST approach. We need the mathe-
matics (and the computer) to keep track of the factors and their combinations.
Chemometrics in Metabolomics – An Introduction 119
In summary, (1) all factors are varied together over a set of experimental
runs, (2) noise is decreased by means of averaging, (3) the functional space is
efficiently mapped, interactions and synergisms are seen.
1. What do I want? – formulate question(s) stating the objectives and goals of
the investigation. For example identify factors (e. g. temperature, day length,
nutrition) and factor ranges (e. g. 15−25 ◦ C, 6−12 h, 1−10 mmol N/L) that
affects flowering time.
2. Screening design – finding out a little about many factors. Which factors
are the dominating ones in controlling flowering time? Screening designs
provide simple models with information about dominating variables, and
information about ranges. Pareto’s principle states that 20% of the data
(factors) account for 80% of the information. Different types of screening
designs exist – which one to choose depends on the problem. The most com-
mon one is the fractional factorials design (Fig. 2). The full factorial design
is a set of experimental runs where every level of a factor is investigated at
both levels of all the other factors. It requires N = 2k number of runs for
k factors. Investigating more than five factors with the full factorial design
can in some cases become time consuming, i. e. 25 = 32, 26 = 64, 27 = 128
experiments, etc. Instead, performing a fractional factorial design reduces
Fig. 2. Example of a full factorial design of experiments (DOE) for investigating how three factors
(temperature, day length and nutrition) control flowering time. Varying the three factors at two
levels (coded as +/-) requires 23 = 8 experiments + center points. Each experiment according to
the design set of experiments is marked with a circle in the figure. Evaluating the results from
such an experimental design reveals the influence of each of the different factors separately and
also any interactions between them. DOE is the only feasible approach to separate cause and
effect from each other
120 J. Trygg et al.
that number quickly without the loss of too much information regarding the
estimation of factors involved. Fractional factorial design takes advantage
of the fact that three-way and higher interactions are seldom significant. It
requires only N = 2k−p number of runs for k factors, where p is set manually.
For example five factors can be run in only 25−2 = 8 experiments instead
of 25 = 32 experiments compared to the full factorial design. Fractional
factorial design takes advantage of the fact that three-way and higher inter-
actions are seldom significant. The downside, of course, for not performing
all experiments, is that confounding patterns are present. In other words,
the estimated effects are not “pure” but instead mixed with higher degree
interaction effects. This loss of information is the prize we need to pay for
the reduction of the number of experiments. The degree of confounding is
determined by the choice of p.
3. Response surface modeling (RSM) and optimization (few factors) – after
screening the factors involved in, e. g. determination of flowering time or
derivatization of metabolites, the goal of the investigation is usually to
create a valid map of the experimental domain (local space) given by the
significant factors and their ranges. This is done with a quadratic polynomial
model. The higher order models have an increased complexity, and therefore
also require more experiments/factors than screening designs. Different
types of RSM designs include Central composite designs, Box Behnken
designs and D-optimal designs (see, e. g. Lundstedt et al. 1998 for more
information).
4. Robustness testing – in robustness testing of, for instance, an analytical
method, the aim is to explore how sensitive the responses are to small
changes in the factor settings, e. g. temperature. Ideally, a robustness test
should show that the responses are not sensitive to small fluctuations in
the factors, that is, the results are the same for all experiments. Robust-
ness testing is usually applied as the last test just before the release of
a product or a method. The fractional factorial design is usually applied
here.
Plant metabolomic studies typically constitute a set of samples from Ara-

bidopsis wild types and mutants. Assume that these have been subjected to
different external conditions such as variation in day length and temperature.
Design of Experiments can then be used to select representative samples, re-
lated to the biological question we are investigating (how flowering time is
affected by temperature, day length, nutrition). An experimental design in
three factors can be setup, with factor 1 (temperature), factor 2 (day length),
and factor 3 (nutrition). In total, only eight different experiments equal 2k
where k = 3 factors are required to explore the experimental space. In addi-
tion, a number of replicates, typically three experiments, are added to estimate
the noise level. By adding extra experiments, one can investigate more thor-
oughly the day length and temperature dependence (increase the number of
different day lengths and temperatures).
2.2 The Data Table, X-matrix
In plant metabolomics studies, typically a set of samples are characterised us-

ing modern instrumentation such as GC/MS, LC/MS or H1 -NMR spectroscopy.
The choice of instrument (see Sumner et al. 2003; Dunn et al. 2005) and exper-
imental procedure (Gullberg et al. 2004) are important and largely determined
by the biological system and the scientific question. Design of Experiments
can here be used to optimize the experimental protocol.
In contrast to a 1 H-NMR spectrum, GC/MS and LC/MS data must be pro-
cessed before multivariate analysis. The reason is the two-dimensional nature
(chromatogram/mass spectra) of the data for each sample. For GC/MS data,
curve resolution or deconvolution methods are mainly applied for data pro-
cessing (see, e. g. Halket et al. 1999; Jonsson et al. 2005a). This gives a resolved
spectral and chromatographic profile for each detected compound. The 1D
multivariate profile used to characterize each sample is made up of the inte-
grated areas of all detected chromatographic peaks. The corresponding mass
spectrum and retention index are used for identification purposes (Schauer
et al. 2005). For LC/MS data, curve resolution can be applied (e. g. Idborg-
Björkman et al. 2003) or a peak detection algorithm that identifies all chro-
matographic peaks and uses their integrated areas as the multivariate profile
characterizing that sample (e. g. Andreev et al. 2003). Another alternative is
to sum the chromatographic direction to create a 1D multivariate profile pro-
duced by the total intensity over all mass spectral channels (e. g. Allen et al.
2004). Recently, partly alternative methodologies have been applied to GC/MS
data (Jonsson et al. 2004, 2005a) and LC/MS data (Jonsson et al. 2005b) where
all samples are processed simultaneously and a common set of descriptor
variables are extracted.
After, e. g. the GC/MS analysis, we now have a multivariate profile (300–1000 s
of variables) for each sample that is a fingerprint of the inherent properties
(e. g. phenotype) for each sample. For multiple samples we can therefore con-
struct a two-dimensional data table, an X matrix, by stacking each sample
on top of each other. The question is then, how do we go about analysing
this multivariate, highly collinear and complex data set? The univariate ap-
proach (e. g. student’s t-test [Jackson 1991]) is not recommended. It assumes
independent variables in X (i. e. more samples than variables) and this creates
problems with interpretation, spurious correlations (so called Type I, II errors)
and the evident risk of missing information in combinations of variables. Tra-
ditional statistical methods (e. g. multiple linear regression, MLR) are also not
recommended. They also assume independent variables and have difficulties
with noisy data (Eriksson et al. 2001). Instead, multivariate analyses based on
projection methods represent a number of efficient and useful methods for
the analysis and modeling of these complex data. Projection methods convert
the multi-dimensional data table into a low-dimensional model plane, usu-
ally consisting of two to five dimensions. Principal component analysis (PCA)
(Jackson 1991) and partial least squares (PLS) (Wold et al. 1984) methods are
122 J. Trygg et al.
two widely used methods that can handle incomplete, noisy and collinear data
structures.
2.3 Geometrical Interpretation of a Data Table
An easy way to understand and appreciate projection based methods is to

translate the data table into a swarm of points in a multi-dimensional space.
For a data table or matrix X, with N rows (biological samples) and K columns
(e. g. relative amounts of different metabolites), each row (individual sample)
can be represented as a point in a K-dimensional space. Its position in this
space is given by its coordinates, i. e. its values in each of the K columns. Re-
peating this for all N rows in a matrix, we have produced a swarm of points
in K-dimensional space. Points (samples) that lie close to each other in this
multi-dimensional space are more biologically similar to each other than points
that lie far apart (dissimilar). Projection methods find a model hyperplanes
of much lower dimensionality that closely approximates X, i. e. the swarm of
points. Figure 3 gives an overview of how multivariate projection methods
work.
2.4 Principal Component Analysis
Principal Component Analysis (PCA) is the workhorse in chemometrics. It is

a multivariate projection method designed to extract and display the systematic
variation in a data matrix X. The first two principal components define a plane,
a window into the K-dimensional space. By projecting each of the sample
points (in K-dimensional space) onto this two-dimensional sub-space, it is
possible to visualize all the samples. The coordinates of each of these samples
projected onto this plane are called scores T, and they are weighted averages
of all X-variables (e. g. metabolites). Hence the visualization of these scores T
is called a score plot. The score plot is very informative because it gives an
overview of all samples in X and how they relate to each other. It may reveal
groupings of samples (clusters), trends and outliers (deviating samples). e. g.
two genotypes (wild type and mutant) would show up as two distinct clusters of
samples, representing wild type and mutant samples respectively. In addition,
an experiment that suffered from a broken GC-vial would translate into an
unique point in the score plot, i. e. an outlier (Fig. 3).
The score plot allows us to investigate the relation among the samples, but
once interesting patterns are found (groupings, outliers etc.), it is possible
to understand the reason for this, i. e. what variables (e. g. metabolites) are
responsible for this pattern found in the score plot. Hence, there also exists
a corresponding plot related to the measured variables (metabolites), i. e. the
columns in the X matrix. This plot is known as the loading plot P and describes
the influence (weight) of the X-variables (metabolites) in the model. An impor-
tant feature is that directions in the score plot correspond to directions in the
Fig. 3. (1) Each row (representing one biological sample) in a data table with K = 3 variables
can be represented as one point in a K = 3 dimensional space. The position of that point is
given by the coordinates given by the values in each of the K = 3 variables. (2) Repeating this
for all rows (samples) in a data table produces a swarm of points in K = 3 dimensional space.
Points (samples) that are close to each other have more similar biological properties than points
that are far apart. (3) Projection methods such as PCA, finds a representative low-dimensional
plane (here two-dimensional) that is a good summary of the variation in the X data table (swarm
of points). (4) This model plane can then be visualised in scatter plots (A) and provides an
overview, e. g. if there are any groupings, trends or outliers in the data. For example in the figure
(A) there is a clear separation between the Arabidopsis wild type and mutant. It is also possible
to understand the reason for this separation by looking at the direction of the model plane
with respect to the original axes (original variables). These are summarized in the PCA model
loadings, P (B)
loading plot (Fig. 3). This is a powerful tool for understanding the underlying
patterns in the data.
The PCA model can be expressed as
Model of X: X = TPT + E
where T are the scores, P defines the loadings, and E represent the residual
matrix. The residual matrix E contains the residuals for each sample between
124 J. Trygg et al.
Fig. 4. PCA summarise all variation in X into a few new variables called scores T. These new
variables are linearly weighted combinations of the original X-variables. The loadings P contain
the weights used for each X-variable and thus reveal the influence of individual X-variables
its point in K-dimensional space and its point on the model plane. The residuals
are important for detection of outliers and for defining the model boundaries
(see Fig. 4).
2.5 Partial Least Squares Projections to Latent Structures (PLS)
The PLS method is used instead of the PCA method when additional knowledge
about each sample exists, the Y matrix, e. g. genotype of each sample (wild
type/mutant). The sample information according to the design matrix from
the Design of Experiments (see Sect. 2.1) is often used as a Y matrix. Hence,
PLS represents the regression analogy of PCA working with two matrices,
X and Y (Wold et al. 1984). It is one of the most common methods when
a quantitative relationship between a descriptor matrix X and a response
matrix Y is sought. The Y matrix can contain both quantitative (e. g. glucose
concentration) and qualitative (genotype) information. This additional sample
information in Y is used by the PLS method to focus the model plane to capture
the Y-related variation in X, e. g. separation between genotypes, rather than
providing an overall view of all variation in the data as done by the PCA
model. In addition, the PLS method can also be used to predict the properties
(Y-values) of new unknown samples, e. g. predict the glucose concentration or
genotype.
The Y matrix consists of the same number of rows as the X matrix. Each
column in Y indicate a certain property, e. g. glucose concentration or genotype
for each sample. When Y contains qualitative information such as genotype, the
number of columns in Y equals the number of classes. Each row in Y describes
the group membership for that sample where “1” indicates class belonging for
that sample and “0” does not. When Y is qualitative, the PLS method is called
PLS Discriminant Analysis (PLS-DA), to distinguish it from the situation when
Y is quantitative.
3 Example: Metabolomics Study on Arabidopsis Mutants
We will work through a metabolomics example using GC/MS data from the
analysis of Arabidopsis extracts. Shoots of higher plants are characterized by
axillary branching, where the shoot branches develop from shoot meristems
located between a leaf and the shoot stem. The control of axillary shoot growth
(branching) is not well understood, but it is known that several internal fac-
tors such as the plant hormones IAA and cytokinins are involved (McSteen
and Leyser 2005). Mutations screens in Arabidopsis have identified four loci
involved in the repression of axillary bud growth, MAX1–4. Based on the mu-
tants, it is now suggested that an unknown transmittable substance might be
involved in controlling branching (see McSteen and Leyser 2005). The biosyn-
thesis of this compound in Arabidopsis is catalyzed by a number of MAX
(more-axillary growth) proteins.
We have used a metabolomics approach to classify and identify the metabolic
differences between the MAX-mutants. Root samples from WT, max3 and max4
mutants were analysed by GC/TOFMS as described by Gullberg et al. (2004).
The GC/MS data was processed by hierarchical multivariate curve resolution
(Jonsson et al. 2005a), and the obtained X-matrix was thereafter subjected to
PCA and PLS-DA analysis. The GC/MS processing resulted in 514 resolved peak
areas. Log transformation, column centering and scaling to unit variance was
done on the resolved peak areas (X-matrix) prior to modeling and two dummy
Y-variables were constructed based on the class belonging of each sample to the
Fig. 5. A PLS-DA score-plot from the analysis of metabolite profiles in roots of Arabidopsis WT,
max3 and max4. The PLS-DA model is based on WT and max3. The X-matrix was centered and
scaled to unit variance. The explained variation in the X-matrix (R2 X) is 0.74, the explained
variation in the Y matrix (R2 Y) is 0.99 and the predictive ability according to sevenfold cross-
validation (Q2) is 0.84. R2 X is the cumulative modelled variation in X, R2 Y is the cumulative
modelled variation in Y and Q2 Y is the cumulative predicted variation in Y, according to cross-
validation. The range of these parameters is 0–1, where 1 indicates a perfect fit. B Based on the
model max4 samples were predicted into the model showing that max3 and max4 are very similar
regarding metabolic content (compare position score plot in A)
126 J. Trygg et al.
genotypes, WT and max3. The PLS-DA model score plot is shown in Fig. 5A.
The score plot reveals the relationship among the samples. It is clear from the
figure that the model plane displays a clear separation of the two genotypes.
To validate the model results, predictions were made for the genotype max4,
using the calculated PLS-DA model based on the other sample-set (WT and
max3). The results, shown in the obtained PLS-DA score plot (Fig. 5B) pre-
dicted that the max4 is closer to max3 than WT. This is consistent with the facts
that max3 is very similar to the max4 genotype, where the MAX3 and MAX4
proteins use the same substrate (Schwarz et al. 2005). Interpretation of the first
weight vector (w1) from the PLS-DA model, as described by Trygg and Wold
(2002), together with the 99% confidence intervals calculated using jack-knifing
(Martens and Martens 2000), highlighted 64 significant variables (metabolites)
differing between WT and max4. The importance of these metabolites is a part
of biological validation of the data set. The statistical validation was done
by prediction of the max3 mutants into the WT/max4 model. Both type of
validation is of importance for validating the multivariate data set.
4 Summary and Future Prospectives
Multivariate projection methods, e. g. PCA and PLS, represent a useful and ver-
satile technology to modelling, monitoring and prediction of complex prob-
lems and data structures encountered within metabolomics and other ‘omics’
disciplines. The common denominator is that high complexity data tables are
generated and that these data tables can be analysed and interpreted by means
of chemometric methods. The principal component analysis (PCA) method
summarizes the variation in a data table X into a model plane (the scores T).
A scatter plot of these scores gives an overview of the samples (observations)
and how they relate to each other, e. g. if there are groupings or trends or
deviating samples and so on. In order to interpret the patterns found in a score
plot one examines the corresponding loading plot (P). The loadings P reveal
how each variable contributes to the separation among samples in the model
plane and also gives insights into the relative importance of each variable.
However, one fundamental property is that the data does contain relevant
information regarding our biological question. In other words, how to max-
imise the information content in the data? The traditional way to Change One
Factor at a Time, i. e. the COST approach, is not recommended. Design of Ex-
periments (DOE) is the methodology of how to conduct and plan experiments
in order to maximize information in the data in the fewest number of runs.
A proper experimental design will reveal the influence of each of the different
factors separately and also any interactions between them. DOE is the only
feasible approach to separate cause and effect from each other. Therefore is
DOE in combination with chemometrical analysis a powerful way of planning,
conducting and evaluating metabolomics experiments.
One common discussion point in the analysis of “omics” data is how to

correlate several types of data, usually with different data structures. Systems
biology seeks to integrate information from multiple parts of a biological
system in a holistic attempt to understand the whole system. There are still
many obstacles and hurdles to overcome in order to succeed. One of these
relates to how the actual integration of the different types of data will be done.
Hence, the advancement of systems biology depends heavily on the ability
to integrate multiple profiling techniques (e. g. transcriptomics, proteomics,
GC/MS, LC-NMR). The current multivariate statistical methods (e. g. the PLS
method) lacks the proper model structure to describe these types of data
structures, because they focus only on the correlation pattern among multiple
data tables (e. g. X = microarrays vs Y = metabolomics data) and not on the
non-correlated variation among these data tables which, in a biological sense,
can be of equal interest. It has also been demonstrated that, because of this,
the interpretation of these models are negatively affected (Trygg and Wold
2002), e. g. positive correlation patterns are interpreted as negligible or even
flipped and become negative. This is a fundamental problem as we certainly
cannot expect that all variation in transcript and metabolite levels co-vary.
Fortunately, recent advances in chemometrics provide the ability to compare
multiple data sets with each other. Novel extensions of the PLS method, called
O-PLS (Trygg and Wold 2002) and O2-PLS (Trygg 2002) contain the model
structure to support both these features. In addition, the O2-PLS method is bi-
directional which means that the flow of information can go in both ways, from
X (e. g. microarray) to Y (e. g. metabolomics) and vice versa. Hence, the O2-
PLS methodology will be important in selecting what genes or metabolites are
important to do further experimentation upon, e. g. understanding biomarker
patterns and selecting genes for knockout studies. The O2-PLS methodology
can also be extended to more than two data tables, hence it nicely fits into the
framework of a combined profiling approach.
Acknowledgements. The Swedish Research Council, Wallenberg Consortium North (WCN), the
Kempe foundation, EU strategic funding, Knut and Alice Wallenberg Foundation (JT) and Strate-
gic Research Funding (SSF) are acknowledged for financial support. Professor Ottoline Leyser,
York, UK, for allowing us to show data from the max-mutant project, and Dr. Miyako Kusano,
RIKEN Plant Science Centre, Yokohama, Japan for the initial analysis of metabolites in the
max-mutants.
References
Allen J, Davey HM, Broadhurst D, Heald JK, Rowland JJ, Oliver SG, Kell DB (2003) High-
throughput classification of yeast mutants for functional genomics using metabolic foot-
printing. Nature Biotechnol 21:692–696
Andreev VP, Rejtar T, Chen HS, Moskovets EV, Ivanov AR, Karger BL (2003) A universal denoising
and peak picking algorithm for LC-MS based on matched filtration in the chromatographic
time domain. Anal Chem 75:6314–6326
128 J. Trygg et al.
Dunn WB, Bailey NJC, Johnson HE (2005) Measuring the metabolome: current analytical tech-
nologies. Analyst 130:606–625
Eriksson L, Johansson E, Kettaneh-Wold N, Wold S (2001) Multi and megavariate data analysis.
Umetrics (www.umetrics.com), ISBN 91–973730-1-X
Gullberg J, Jonsson P, Nordström A, Sjöström, M, Moritz T (2004) Optimisation of preparation
of plant samples for metabolic profiling by GC-MS. Anal Biochem 331:283–295
Halket JM, Przyborowska A, Stein SE, Mallard WG, Down S, Chalmers RA (1999) Deconvolution
gas chromatography mass spectrometry of urinary organic acids - potential for pattern recog-
nition and automated identification of metabolic disorders. Rapid Commun Mass Spectrom
13:279–284
Idborg-Björkman H, Edlund, PO, Kvalheim OM, Schuppe-Koistinen I, Jacobsson SP (2003)
Screening of biomarkers in rat urine using LC/electrospray ionization-MS and two-way
data analysis. Anal Chem 75:4784–4792
Jackson JE (1991) A users guide to principal components. Wiley, New York
Jonsson P, Gullberg J, Nordström A, Kowalczyk M, Sjöström M, Moritz T (2004) A strategy for
extracting information from large series of non-processed complex GC/MS data. Anal Chem
76:1738–1745
Jonsson P, Johansson AI, Gullberg J, Trygg J, A J, Grung B, Marklund S, Sjöström M, Antti H,
Moritz T (2005a) Highthroughput data analysis for detecting and identifying differences
between samples in GC/MS-based metabolomic analyses. Anal Chem 77:5635–5642
Jonsson P, Bruce SJ, Moritz T, Trygg J, Sjostrom M, Plumb R, Granger J, Maibaum E, Nicholson JK,
Holmes E, Antti H (2005b) Extraction, interpretation and validation of information for
comparing samples in metabolic LC/MS data sets. Analyst 130:701–707
Lundstedt T, Seifert E, Abramo L, Thelin B, Nyström A, Pettersen J, Bergman R (1998) Experi-
mental design and optimization. Chem Intel Lab Systems 42:3–40
Martens H, Martens M (2000) Modified Jack-knife estimation of parameter uncertainty in bilinear
modelling by partial least squares regression (PLSR). Food Qual Pref 11:5–16
McSteen P, Leyser O (2005) Shoot branching. Annu Rev Plant Biol 56:353–374
Tunali U, Forbes MG, Willmitzer L et al (2005) GC-MS libraries for the rapid identification
of metabolites in complex biological samples. FEBS Lett 579:1332–1337
Schwartz S, Qin XQ, Loewen MC (2005) The biochemical characterization of two Carotenoid
cleavage enzymes from Arabidopsis indicates that a carotenoid-derived compound inhibits
lateral branching. J Biol Chem 279:46940–46945
Trygg J (2002) O2-PLS for qualitative and quantitative analysis in multivariate calibration.
J Chemometr 16:283–293
Trygg J, Wold S (2002) Orthogonal projections to latent structures (O-PLS). J Chemometrics
16:119–128
Wold S, Ruhe A, Wold H, Dunn WJ III (1984) The collinearity problem in linear regression.
The partial least squares (PLS) approach to generalized inverses. SIAM J Sci Statist Comput
5:735–743
II.3 Map Editor for the Atomic Reconstruction
of Metabolism (ARM)
M. Arita1,2 , Y. Fujiwara1 , and Y. Nakanishi3
1 Introduction
In the systems study of biological networks, computational analysis is expected

to contribute in three phases by (1) model selection, the formal definition of
each pathway’s role in the manifestation of the biological aspect under analysis,
(2) model refinement, the estimation of model parameters to refine constructed
mathematical models, and (3) simulation and feedback, the computer simula-
tion for the feedback of the predicted results for a better understanding of the
target mechanism(s) (Hood 2003; Arita et al. 2005). Of these, the first phase is
of prime importance because it determines the target of the analysis and its
abstraction level. In other words, in the model selection phase, an appropriate
model is searched and selected among all hypothetical candidates. The process
of model selection is often performed intuitively by researchers. For example,
glycolysis, the best-known pathway module in energy metabolism, contains
a sequence of ten biochemical reactions from glucose to pyruvate (Berg et al.
2002). Since the pathway is linear, it can be summarized as if it were a single
reaction:
Glucose + 2NAD+ + 2ADP + 2Pi → 2Pyruvate + 2ATP + 2NADH + 2H+
The behavior of the pathway may be mathematically described either as

a set of ten reactions, or as a single, abstract reaction. The abstract model is
preferred in explaining net ATP generation, whereas the ten-reaction model
is used for metabolic simulations. In choosing a model, we must be aware of
the trade-off between model accuracy and its description length. In general,
any model inevitably loses its fit to its corresponding natural mechanism as its
description becomes simpler. In glycolysis, the nine intermediate molecules
in the pathway can be eliminated to obtain the net ATP model in return for
sacrificing biochemical details (i. e., the intermediates) of the pathway.
However, when glycolysis is placed in a global metabolic network, the de-
scription of the intermediate molecules is no less important than that of the
gateway molecules such as glucose and pyruvate. The question arises as to
1 Department of Computational Biology, Graduate School of Frontier Sciences, The University of
Tokyo, 5–1–5 Kashiwanoha, Kashiwa, 277–8561 Japan, e-mail: [email protected]

2 Institutefor Advanced Biosciences, Keio University, and Precursory Research for Embryonic
Science and Technology, Japan Science and Technology Agency
3 Intec Web and Genome Informatics Corporation

130 M. Arita, Y. Fujiwara, and Y. Nakanishi
which criteria should be evaluated and chosen for an appropriate abstraction

of the given network. In biology, the focus has been directed at modularity in
terms of function and structure.
The introduction of modularity, i. e. the encapsulation of network details
by specifying the input/output, yields many advantages. First, it simplifies the
description of the given network and facilitates its understanding. In metabolic
networks, biologists have used intuitive, functional concepts such as ‘amino-
acid biosynthesis’ or ‘nitrogen assimilation’ without formal, logical definitions.
Second, the introduction of modularity simplifies the static verification of the
network, a required step that precedes quantitative analyses such as simulation.
The stoichiometric balance is one such property that can be verified statically.
In glycolysis, the structural and functional modularity is clear, mainly be-
cause the pathway is not branched; the pathway structure is linear and all
carbon atoms in glucose are mapped to pyruvate and the function is the de-
composition (lysis) of glucose. In general, modularity is less straightforward
in branching metabolic pathways. Since molecular moieties are split into or
merged with multiple molecules, it is not easy to trace carbon and other
atomic elements, let alone delineate modularity. In fact, the determination
of molecules to be regarded as intermediates is context-dependent: focusing
on different atoms changes the pathways to be traced and therefore the re-
sulting modular decomposition. For multiply branching pathways there is no
universally effective, definitive decomposition.
Can the modularity of metabolic networks be detected computationally?
Many automated methods exploit the stoichiometry of biochemical reactions
(Mavrovouniotis 1992); this is one solution for the formal decomposition into
metabolic modules. Currently, however, the most widely accepted method for
finding modularity is through network topology; verification is by visualiza-
tion with functional annotations (Ravasz et al. 2002; Ma et al. 2004). This is
why pathway modules remain intuitive. Although electronic circuits can be
verified using Boolean logic, no such formal system has been developed for bi-
ological systems. Indeed, many software programs for genomic and proteomic
networks address only visualization, and their formal analysis remains to be
solved.
The basic concept of our metabolic map editor, introduced in this chapter,
combines a formal description of metabolism (structural conversions and
their stoichiometric conditions) with intuitive network visualization. Many
visualization tools and databases provide a static view of a given metabolic
network (Mendes 2002; Kanehisa et al. 2004; Keseler et al. 2005). Because
a metabolic network is well investigated and has a traditionally accepted layout
for metabolites and enzymatic reactions, fully automated layout algorithms
that ignore such standard layouts do little to further our understanding of
its properties. Rather, an interactive drawing tool that can edit and modify
standard metabolic networks is needed. Only with interactive software can
biologists derive species-specific pathway images and visualize their intuitive
ideas.
Map Editor for the Atomic Reconstruction of Metabolism (ARM) 131
2 Definition of Metabolic Information
In this section, we introduce the concept of handling metabolic information at

the atomic scale, and show how atomic representation can advance the formal
understanding of metabolism.
2.1 Definition of Metabolic Pathways
Functions that should be computationally supported by a metabolic map editor

include (1) searching and adding alternative or new metabolic pathways, (2)
superimposing genomic, proteomic, and metabolomic data onto the network,
and (3) rearranging the network topology to accommodate the metabolism
of a particular species of interest. To fulfill these criteria, a formal definition
of metabolic pathways is required. Previous computational studies tended to
present a metabolic network as a graph where nodes and edges correspond to
metabolites and their biochemical reactions, and the pathway as a sequence of
graph edges (Ravasz et al. 2002; Ma et al. 2004). However, from a biochemical
perspective, edge sequences (or graph paths) defined in this manner do not
necessarily correspond to metabolic pathways. Since molecular structures are
transformed in the course of each reaction, adjacent graph edges may not share
the common structural moiety that corresponds to the metabolic (or atomic)
flux (Fig. 1).
To resolve the conflict between graph paths and biochemical pathways,
an additional constraint must be introduced (Arita 2003). In this chapter,
a metabolic pathway (pathway for short) from metabolite X to metabolite Y is
defined as a sequence of reactions through which at least one atom (carbon,
nitrogen, or sulfur) in X reaches Y. A metabolite Y is called reachable from X
if there is a pathway from X to Y. This is a rather strict constraint because the
Fig. 1. Example of a biochemically inappropriate pathway from citrate to malonate. Analysis

of molecular structures is required to enable the computer to detect that the citrate moiety is
transferred to citryl CoA, rather than to acetate
conserved moiety throughout a pathway may not consist of carbon, nitrogen,

or sulfur atoms; it may consist of oxygen or hydrogen atoms or even electrons.
The map editor deals with only three types of elements because they can be
computationally traced without ambiguity. The tracing of oxygen or hydrogen
atoms is virtually impossible because the water molecule is involved in many
reactions. The same is true for phosphates and metal ions that exist as free
inorganics in a cell.
2.2 Representation of Pathways and Networks
In the atomic representation of a metabolism, each reaction is decomposed into

a set of sub-structural correspondences called atomic mappings (Arita 2003).
Each atomic mapping represents the transfer of a certain structural moiety
in the course of biochemical reactions, and may be shared among multiple
reactions. For example, atomic mapping between ATP and ADP, and between
glutamate and α-keto-glutarate is prevalent in phosphate- and amino-transfer
reactions, respectively (Fig. 2).
Conventionally, a metabolic pathway is thought of as a sequence of catalytic
reactions classified by EC numbers. However, in the metabolic reconstruc-
tion from a whole genome sequence, a metabolic pathway is better viewed
Fig. 2. Three atomic mappings in the reaction glucose + ATP = glucose 6-phospahte + ADP. The
mapping between ATP and ADP (shown in solid lines) is common to all phosphorylation reactions
with ATP and ADP
as a sequence of atomic mappings rather than as EC reactions. There are

at least two reasons for this. First, the proposed atomic view can provide
more candidate pathways in the reconstruction. Reconstruction based only
on EC numbers often results in a set of incomplete pathways with multi-
ple gaps. Such superficial gaps, however, may be filled with other reactions
that share atomic mapping corresponding to the gap. Second, the proposed
atomic view provides flexibility in choosing coenzymes. In theoretical analy-
ses, the stoichiometric balance of reactions has been discussed as if coenzymes
were fixed for all reactions (as in the traditional metabolic map) (Papin et al.
2004). In practice, however, their balance should be determined considering
the net reaction balance in the network. For example, some NAD-dependent
enzymes can also catalyze reactions using NADP, and the overall balance of
their use depends on the total networking condition, not on individual reac-
tions.
Thus, at least for the computational reconstruction of pathways, a metabolic
network is better viewed as a set of atomic mappings rather than a set of
EC-numbered reactions. In the biosynthesis of isoleucine and valine, for ex-
ample, the same set of enzymes catalyzes 2-oxobutanoate and pyruvate to
form isoleucine and valine, respectively. The only difference between the two
pathways is a single alkyl group independent of the catalytic sequences in the
biosynthesis. The decomposition into atomic mappings can explicitly describe
such structural sharing between pathways.
3 Metabolic Map Editor
3.1 Overview
The design principle of the map editor is that users can flexibly integrate a se-
quence of atomic mappings (not reactions) into existing metabolic pathways
to form metabolic maps (Fig. 3). First, users are expected to search metabolic
pathways using the associated database that stores enzymatic reactions, their
atomic mappings and molecular structures. The searched pathways (sequences
of atomic mappings) are transferred to the main window where their layout
can be freely edited as in a conventional graphical drawing editor. The advan-
tage of our editor over conventional editors such as Microsoft PowerPoint is
that users can import metabolic objects (e. g. compound structures and re-
actions) from the background database: although on-screen it appears as if
only graphical objects for compound structures and reactions are imported,
more information is processed in the background. For example, importing
one enzymatic reaction on the screen implicitly invokes the integration of its
associated atomic mappings into the already drawn metabolic map so that the
route of any atom in the new reaction can be traced seamlessly on the resulting
metabolic map.
Fig. 3. Screen-shot of the Map Editor. Pathways are searched by typing molecular names in the
input fields (Number 1). The search results are listed (Number 2). Users can drag pathways into
the main canvas
3.2 Metabolic Database
The background database for atomic information stores three types of metabol-
ic data: molecular structures, reaction formulas, and their atomic mappings.
All data are freely available in text format from the website http://www.
metabolome.jp/download.html.
Molecular structures are registered in the MOL-file format (MDL Informa-
tion Systems; its description is downloadable from http://www.mdli.com/). The
MOL-file format is the de facto standard to describe molecular structures; an
example is shown in Fig. 4. Each MOL-file describes one molecular structure
as a list of atoms with their XYZ coordinates and their chemical bondings. The
chirality of carbon is specified using one integer value for each corresponding
carbon atom. Information on the display of chirality (in thick and shaded lines)
is specified using other integer values. The metabolic editor does not use the
XYZ coordinates written in the format; rather, it applies the original drawing
algorithm to assign XYZ positions (Arita 2005).
As in other metabolic databases, enzymatic reactions are described using
compound names. Reaction formulas were obtained from the Enzyme Nomen-
clature of the International Union of Biochemistry and Molecular Biology
(http://www.chem.qmw.ac.uk/iubmb/enzyme/). In each reaction, the order of
molecules on the left- and right-hand side was manually rearranged so that
the atomic mappings can be computationally detected by comparing molecu-
Fig. 4. MOL-file format for l-alanine. In the ARM database, carbon atoms in the atom-block
are ordered according to the IUPAC positions for molecular structures. Only carbon atoms are
correctly ordered
Fig. 5. Schematic view of reaction EC 6.3.5.2. The reaction formula is written as “ATP + XMP + l-
glutamine + H2 O =>AMP + GMP + l-glutamic acid + pyrophosphate” in the database so that the
molecular structures roughly correspond one-to-one (top to bottom)
lar structures sequentially from left to right (Fig. 5). For details on structure
comparison to compute atomic mappings refer to Arita (2003).
Using molecular structures, atomic mappings were pre-computed for all
registered reactions and, after manual verification, correct mapping results
were stored in the database. Details, including the accuracy of the mapping
computation, were described previously (Arita 2003). Although the mapping
was computed for all atomic elements except hydrogen, the results were reg-
istered only for carbon, nitrogen, and sulfur atoms due to ambiguities in the
mapping of the rest of the elements.
3.3 Drawing Maps from Pathways
The map editor is equipped with a search engine for metabolic pathways.
Given a source and target metabolites, the engine computes logically possible
pathways between these metabolites from the shortest- to pathways of any
length. Although pathway length is measured by the number of reaction steps,
an arbitrary value can be assigned in the algorithm used. In other words, the
engine can compute any pathway throughout which at least one carbon (or
nitrogen, sulfur) atom is conserved. An arbitrary combination of pathways can
be visualized by dragging a searched pathway into the main canvas window
(Fig. 3). When a pathway is dragged into the canvas, it is merged with the already
drawn network (Fig. 6). Although its initial layout is automatically assigned,
a user can freely rearrange the orientation or location of any metabolic object
by using the mouse.
Fig. 6. The network generated by merging two pathways from xylulose 5-phosphate (X5P) to
erythrose 4-phosphate. Every time a pathway is dropped, only the difference from the existing
network is drawn. Carbon 3 in X5P is traced in this example (shown with blank arrows)
The unique function of the map editor is its ability to trace a particular
atom on the map. Since each metabolic object on the map is linked with its
atomic information in the database, the logical tracing of each atomic position
is possible by transitive calculation of the atomic mappings in the network.
A user needs only to mouse-click a particular atom on the network to see its
traces (Fig. 6).
4 Applications
4.1 Carbon Flow in Cyclic Pathways
Metabolic pathways contain two types of cycles in terms of tracing atoms: cycles
where carbon atoms are exchanged in each round (e. g. the tricarboxylic acid
(TCA) cycle), and cycles where all carbon atoms are conserved (e. g. the urea
cycle). If reversible, a single reaction catalyzing two identical molecules (e. g.
2 pyruvate = 2-acetolactate +CO2 ) can form the former type of cycle by itself.
Likewise, the latter type of cycle can be formed by any reversible reaction. Cyclic
pathways may be biochemically meaningful, but in practice, their existence is
problematic in searching metabolic pathways. Since pathways are searched
and output according to the number of reaction steps in our system, short
cycles drastically increase the number of spurious pathways with local loops.
To eliminate such futile pathways, our pathway-search algorithm eliminates all
pathways that visit the same molecule multiple times. However, this constraint
is too strict for searching all possibly existing pathways. For example, users
studying the TCA cycle may want to analyze carbon traces that go round the
cycle multiple times.
To support the atomic analysis of cyclic pathways, a metabolic map editor
is indispensable. First, users search the pathways of interest and paste them
into the main window using the edit function (i. e., model selection). Then,
a particular atom can be interactively traced within the selected set of reactions.
Since the target model is highly constrained, it is feasible to compute pathways
visiting the same compound multiple times.
4.2 Carbon Flow in Energy Metabolism
Because of the shared metabolites between the glycolytic and pentose phos-
phate pathways, the atomic traces of a particular carbon atom often become
hard to follow. This is the case for the correspondence between the C-1 of
glucose and the C-1 of pentose 5-phosphate. By clicking the corresponding
atomic position in the map editor, the atomic trace can be grasped at a glance.
Although the entry point of the pentose phosphate pathway decarboxylates
the C-1 position of glucose, this position is identical to the C-1 of fructose 6-
phosphate through glycolysis. Thus, in the pentose phosphate pathway, the C-1
of fructose 6-phosphate corresponds to the C-1 of sedoheptulose 7-phosphate,

of xylulose 5-phosphate, and of ribose- and ribulose 5-phosphate. In fact, the
C-1 of glucose corresponds to both the C-1 and C-6 positions of fructose 6-
phosphate, and therefore to all C-1 and C-5 positions of pentose 5-phosphate.
These positions are invariable throughout the glycolytic and pentose phosphate
pathways.
4.3 Visualization of Lipid Metabolism
Recently, metabolome analysis has been facilitated due to the rapid techni-
cal progress made in mass spectrometry (MS). To detect lipid molecules, for
example, an effective strategy is to couple MS with liquid chromatography-
electrospray ionization (Houjou et al. 2004). More than 1000 glycerophospho-
lipid species can be quantitated in a single assay in less than 2 h (R. Taguchi,
personal communication). However, the efficient analysis of such large-scale
data sets poses a vexing problem. Network visualization remains the first step
for gaining an overview of the data; however, the traditional metabolic map
is not suitable for visualization because it contains abstract notations. For
example, the ‘phosphatidyl group’ contains two fatty acids of variable lengths
(usually 12∼24 carbon atoms) and degrees of unsaturation (usually 0∼6 double
bonds, depending on the length). Since experimentally confirmed fatty acids
in a phosphatidyl group are comprised of more than 30 species, the number
of actual phosphatidyl species may be as many as its square, i. e., ≈ 1000. To
visualize the distribution of the spectrum of molecular species, our map editor
supports an interactive instantiation of abstract moieties. For each abstract
notation using ‘R-group’, a user can assign a list of molecules as its possi-
ble instantiation. The map editor can also display an integer value for each
molecule (such as the concentration, mass, logP, etc.). Given the list of possible
instantiations for each R-group and their corresponding concentrations (i. e.,
metabolomic data), the editor can display the percentage fraction of candidate
molecules. When multiple R-groups exist, the amount to be displayed will be
the integration of all possible assignments. In a phosphatidyl group, various
fatty acids can be linked at R1 and R2 positions of glycerol phosphate. With
a mouse click, the candidate list for the R1 (or R2) position is displayed together
with the relevant percentage fractions. The percentage for a docosahexanoic
acid (DHA) in the R1 group, for example, is calculated as the sum of all phos-
phatidyl molecules that have DHA at R1. When DHA is chosen for R1, the
percentage list for R2 consists of fully instantiated molecular species that have
DHA at R1 and another fatty acid at R2.
Due to the abstract notations for molecules, lipid metabolism is a particu-
larly unspecific part in the traditional metabolic map. The computer-assisted
metabolic map is indispensable to visualize the metabolomic data of such
pathways.
5 Conclusions
The map editor is not only a tool for visualizing metabolic pathways, but
is a necessary component for the systematic and modular understanding of
species- and context-dependent metabolic networks. Since the software system
is linked with atomic-level information in the background database, users can
trace any atomic position on any metabolic network they draw. It is a desir-
able realization of a pathway database. Most web-based pathway databases do
not support the users’ own arrangement of networks, although in computer
science, the definition of a database system is ‘ a collection of information or-
ganized in such a way that a computer program can quickly select desired data
in a desired arrangement’. Our map editor compensates for this drawback, and
represents a step forward to a more flexible analysis of large-scale biological
information.
Acknowledgements. The ongoing analysis of lipid metabolism is a joint effort with Prof. Ryo
Taguchi at The University of Tokyo. The authors thank Ursula Petralia for editing the manuscript.
This work was supported by The Ministry of Education, Culture, Sports, Science and Technology
(MEXT), Grant-in-Aid for Scientific Research on Priority Areas.
References
Arita M (2003) In silico atomic tracing by substrate-product relationships in Escherichia coli
intermediary metabolism. Genome Res 13(11):2455–2466
Arita M (2005) Introduction to the ARM database: database on chemical transformations in
metabolism for tracing pathways. In: Tomita M, Nishioka T (eds) Metabolomics: the frontier
of systems biology. Springer, Berlin Heidelberg New York, pp 193–211
Arita M, Robert M, Tomita M (2005) All systems go: launching cell simulation fueled by integrated
experimental biology data. Curr Opin Biotechnol 16(3):344–349
Berg JM, Tymoczka JL, Stryer L (2002) Biochemistry, 5th edn. Freeman, New York
Hood L (2003) Systems biology: integrating technology, biology, and computation. Mech Ageing
Dev 124:9–16
Houjou T, Yamatani K, Nakanishi H, Imagawa M, Shimizu T, Taguchi R (2004) Rapid and selective
identification of molecular species in phosphatidylcholine and sphingomyelin by conditional
neutral loss scanning and MS3. Rapid Commun Mass Spectrom 18(24):3123–3130
Kanehisa M, Goto S, Kawashima S, Okuno Y, Hattori M (2004) The KEGG resource for deciphering
the genome. Nucleic Acids Res 32 (Database Issue):D277–D280
Keseler IM, Collado-Vides J, Gama-Castro S, Ingraham J, Paley S, Paulsen IT, Peralta-Gil M,
Karp PD (2005) EcoCyc: a comprehensive database resource for Escherichia coli. Nucleic
Acids Res 33 (Database Issue):D334–337
Ma HW, Zhao XM, Yuan YJ, Zeng AP (2004) Decomposition of metabolic network into func-
tional modules based on the global connectivity structure of reaction graph. Bioinformatics
20(12):1870–1876
Mavrovouniotis ML (1992) Computer-aided synthesis of biochemical pathways. Biotechnol Bio-
eng 36:1119–1132
Mendes P (2002) Emerging bioinformatics for the metabolome. Brief Bioinform 3(2):134–145
Papin JA, Reed JL, Palsson BO (2004) Hierarchical thinking in network biology: the unbiased
modularization of biochemical networks. Trends Biochem Sci 29(12):641–647
Ravasz E, Somera AL, Mongru DA, Oltvai ZN, Barabasi AL (2002) Hierarchical organization of
modularity in metabolic networks. Science 297(5586):1551–1555
Subject Index
ACBP 221 calibration 7

acetyl-CoA 212 – quantitative 7
acyl-ACP 220 candidate gene 190
acyl-CoA 212 capillary HPLC 50
acyltransferase 216, 246 carotenoid 229–231, 233–235, 237, 240
ajmalicine 283 – astaxanthin 231, 234, 237
alignment 14, 15 – β-carotene 230, 234, 236, 237
AMDIS 108 – canthaxanthin 234, 237
analyte 10 – echinenone 234
– alternate 11 – 4-ketozeaxanthin 237
– composite 11 – lutein 230
– definition 11 – lycopene 234
– major 11 – 3 -OH-echinenone 237
– minor 11 – phytoene 235
– preferred 11 – spectrum 236
– specific 11 – zeaxanthin 230, 237
anthocyanin 206–208 catharanthine 280, 283
Arabidopsis 125, 141 Catharanthus roseus 264, 280, 282,
Arabidopsis 158, 313, 328, 331 285, 287
Arabidopsis thaliana 66, 74, 82 – acid 268, 269
AraCyc 109, 141, 159 – ajmalicine 264
ArMet 107 – catharanthine 264
Atmospheric Pressure Chemical – chlorogenic 269
Ionisation (APCI) 38 – loganic 269
atomic mapping 132 – secologanin 268
ATTED-II 160 – vindoline 267
CE/MS 312
baccatin III 294, 296, 298 chemometrics 83, 117, 118
β-oxidation 212 – orthogonal signal correction
bioinformatics 200 (OSC) 85
biomarker 312 chromatographic method 235
biomolecular network 187–189 chromatography 261
BioPathAt 160 – GC 261
biosynthesis 144, 292 – HPLC 261
biotechnology 190 – TLC 261
BL-SOM 202, 204, 205 chromatography theory 24
B-Net 109 CID-MS 66, 74, 77
Brassica napus 217 classification analysis 117, 118
breeding 332, 336, 337 cold acclimation 313, 315
BROME 110, 111 crop improvement 336
342 Subject Index
cyanidin 42 flax 314, 318

cytochrome P450 246, 248, 251 – linoleic 314
– linoleic acid 318
2D NMR 270 – linolenic 314
– COSY 272 – linolenic acid 318
– HMBC 273 – linseed 315, 317
– HMQC 273 – oil 314
– HSQC 273 – solin 315, 317
– J-resolved NMR 270 Fourier Transform Ion Cyclotron
– TOCSY 272 Resonance 312
3D virtual reality visualization fractional factorial design 119
environment 150 free fatty acid 214
10-deacetylbaccatin III 298 FT-ICR MS 314, 315, 317, 320, 324
deconvolution 14, 121 FTICR-MS 312
– algorithm 14 FT-MS 200, 202, 206
– error 14 functional genomics 200, 208
– software 14
degradation/utilization/assimilation 144 gain of function analysis 190
desaturase 216 galactolipid 214
Design of Experiments (DOE) 118 gas chromatograph 213
2D-HPLC 55, 58 Gas Chromatography Mass
diacylglycerol 220 Spectrometry 3
diagnostic 187 GC/MS 311
direct flow injection (DFI) 38, 41, 46 – cost effectiveness 22
discriminant analysis 117, 118 – derivatization 22
discriminant partial least squares – profiling 22
(PLS-DA) 112 – separation efficiency 22
DOME 107, 108 GC-FID 219
dynamic (flux) profiling 193, 194 GCMS 216
GC-MS technology 3
Electro Spray Ionization (ESI) 38 – acquisition 5
electrospray ionisation 215 – derivatisation 5, 6
elicitation 281–283, 285 – acylation 7
elongase 216 – alkoxyamination 6
ELSD 219 – alkylation 7
evidence code 144 – silylation 7
exploratory analysis 117 – detection 5
expressed sequence tag [EST] 200, 245, – extraction 5
253, 254 – GC×GC-TOF-MS 15
extraction – ionisation 5
– solid phase 7 – separation 5
– vapour phase 7 – two dimensional 5
extraction procedure 234 gene expression microarray 150
gene expression profiling 149
FAME 213 GeneMaths 38, 39
fatty acid 211 GeneSpring 160
fatty acid amide hydrolase 247 genetic algorithm 112
FCModeler 150 genetic engineering 336
flavonoid 244, 319 genetic modification 230
Subject Index 343
genetic programming 112 – selectivity 24

genome segment introgression 190 – separation efficiency 24
genomics 332 – universal technique 23
– Arabidopsis 335, 336 – UPLC 27
– functional genomics 335
– overexpression 336 ICAT 54
– T-DNA 336 industrial application 328, 332, 335, 338
germination 213 integrative genomic 187, 190
GiGA 160 introgression line 230, 237, 240
glucosinolate 201, 204 ion exchange 59
glutamate dehydrogenase 320 ionization suppression 54
glycolysis 129, 138 isoflavone synthase 248
glycosyltransferase 246, 248, 251 isoflavonoid 247
– in vitro enzyme promiscuity 251 isopentenyl diphosphate 232
– in vivo substrate specificity 251 isoprenoid 232–234
Golm metabolome database (GMD) 17 – plastoquinone, tocopherols
graph clustering 179 and gibberellin 232
– graph clustering algorithm 166, 177
growth regimes 233 kaempferol 42
KaPPA-View 155
herb 338 KEGG 111
herbicide research 335 KEGG/PATHWAY 158
hierarchical clustering (HCA) 111, 316 Kennedy pathway 213
hierarchical ontology 144 ketocarotenoid 237
high pigment mutation 44 k-means clustering 111
high throughput metabolite profiling 332 known unknowns 328
– experimental design 332
– validation 332 LC 66
HILIC 59 – reversed phase 66
homogenisation procedure 233 LC/MS 311
HPLC 219, 235, 236 LC-ESI-MS 54
– analytical 23 LCMS 216
– capacity factor 24 LC-MS 65–67, 75, 77
– capillary 23 – accurate mass 71, 76
– chromatography theory 24 – deconvolution 70, 72
– column efficiency 24 – electrospray ionization 69
– column length 26 – extraction 67
– column permeability 27 – flow rate 67
– improving resolution 26 – hybrid mass spectrometer 70
– increasing resolution 26 – in-source fragment 77
– micro 23 – matrix effect 69, 70, 74–76
– mobile phase viscosity 27 – separation 68
– monolithic column 27 LIGAND 109
– nano 23 lignification 247
– peak capacity 24, 25 linear discriminant analysis 112
– preparative 23 Linum usitatissimum L. 314
– pressure 26 lipid 211
– reproducibility 23 lipid metabolism 138
– resolution 24 lipid synthesis 211
344 Subject Index
lipidomic 223 – selectivity 22

loading plot 122 – sensitivity 22
– PCA 84 – software 334
lock mass spray 42, 45 – spectral matching 22
– TOFMS 30
macroarray 200 matrix effect 7
manual curation 152 – discriminatory 7
MAPMAN 160 – inhibitory 7
MapMan 150 – stabilising 7
Markerlynx 38 Medicago sativa 248
mass analyzer Medicago truncatula 248
– mass accuracy 29 – cell suspension culture 250, 251
– mass resolution 29 medicinal herb 337
– scan speed 29 metabolic control analysis 186
mass detection 7 metabolic pathway 131
mass fragments 12 metabolic pathway database 141
– definition 12 metabolic profiling
– quantifying 12 – non-targeted 311, 324
mass isotopomer 10 – targeted 312
– chemically synthesised 10 metabolic regulation 187
– deuterated 8 metabolite 10, 155
– in vivo labelled 10 – definition 11
mass spectral tag (MST) 12 – developmentally modulated 314
– chimeric 14 – temperature modulated 314
– definition 12 metabolite pool size 12
– erroneous 14 metabolite profiling 3, 149, 278, 279, 286
– identification 15 – challenge 3
– inventory 15 – definition 3
– software 15 – limitation 6
mass spectrometer 215 – perspective (breakthrough) 4
mass spectrometry 236, 328 – renaissance 3
– capillary electrophoresis 329 – scope (compounds) 6
– compound identification 22 metabolite-organism relationship 166
– extraction method 334 metabolite-species relationship 166
– FTMS 30 metabolome 12, 21, 65, 311, 327
– gas chromatography (GC) 329 – complexity 22, 28
– GC-MS 332, 334 – plant 22
– ion mobility MS 28 – stable isotope labelled 12
– ion-trap 29 – technology 22
– Laboratory Information Management – visualization 22
System (LIMS) 334 metabolomics 21, 229, 311
– LC-ion mobility MS 28 – biological variance 25
– LC-MS 332, 334 – dynamic range 25
– liquid chromatography (LC) 329 – goal 25
– mass accuracy 29 – limitation 25
– mass resolution 29 – qualitative 25
– Orbitrap 30 – quantitative 25
– quadrupole 29 metadata 106, 107
– scan speed 29 MetAlign 37
Subject Index 345
methyl jasmonate 292 – multi-dimensional 93

MetNet 160 – NMR 266
MetNetDB 150 – spectral subtraction 97
MIAMET 107 – TOCSY 85
Micro Channel Plate (MCP) 34 – two-dimensional (2D) NMR 85
micro HPLC 49, 52, 53 nutrition 337, 338
microarray 207 – nutrigenomics 337
model selection 129
modularity 130 oil 211
MOL-file format 134 oil yield 223
MSRI library 17 O-methyltransferase 246, 248
multidimensional chromatography 28 Omics Viewer 149
– chip system 29 O-PLS 127
– GC-GC 28 O2-PLS 127
– LC-GC 28 outlier 122
– LC-LC 28
– MUDPIT 28
paclitaxel 291, 292
– multiplexed 29
partial least squares (PLS) 121
– on-line/off-line 29
PathMAPA 160
– parallel system 29
– peak capacity 28 pathway
– resolution 28 – pathway database 155
– selectivity 28 – pathway map 155
– unified chromatography 28 Pathway Processor 160
multidimensional HPLC 49, 59 Pathway Tools software package 148
multidimensional scaling 112 PCA 202, 204, 206
multivariate analysis 121 – correlation matrix 85
multivariate statistical method 111 – covariance matrix 85
– score 84
Nicotiana tabacum 320 peak capacity 55
NMR 81, 93, 266, 312, 328 PEDRo 106, 107
– 13 C NMR 269 peroxisome 221
– 13 CNMR 94 phosphatidylcholine 215
– 1 H 266 phospholipid 214
– 1 H inverse probe 97 photodiode array (PDA) 42
– 14 N 95 plant disease resistance 216
– 15 N 95 plant secondary metabolite 277, 278,
– chemical shift 96 280, 287
– 2D-J-resolved spectroscopy 81 plant stress response 192
– 1D-NMR 95 PLS Discriminant Analysis
– DOSY 86 (PLS-DA) 124
– 2D-spectra 95 polyketide synthase 246
– hetero-nuclear NMR 95 polyunsaturated fatty acid 222
– 1 H-NMR 94 power-law 172, 176, 177, 179
– HSQC 94 – power-law distribution 166, 172,
– in vivo 95 177, 179
– in vivo measurement 98 – power-law structure 176
– LC-NMR-MS 87 prediction model 117
– magic-angle spinning 94 principal 316
346 Subject Index
principal component analysis (PCA) seed development 217

39, 41, 84, 111, 121, 122, 273, 282, 284 self-organizing map (SOM) 111
– loading plot 274 sensitivity of NMR 99
– principal component 274 – cryogenetically cooled probe 99
– score 274 – low dielectric solvent 99
projection method 121, 123 – magnetic-field 99
proteomics 21 – preamplifier 99
– radio frequency detector 99
quadrupole 35 – S/N 99
quadrupole time of flight (QTOF) 33, 35 separation impedance 52
quantification 12 serine protease 246
– recovery 12 SFC 55
– relative 12 simulation 129
– response 12 size-exclusion 59
quercetin 42 solid-state NMR 99
– biomembrane 99
reductionism 186 – cell-wall component 99
regioisomer 219 – insoluble metabolite 99
regression analysis 117, 118 – starch 99
resolution 53 species-metabolite relation 172, 179
response species-metabolite relationship 172, 179
– average 13 sphingolipid 215
– ion current 12 stable isotope labelling 10, 86
– normalised 13 – 13 carbon (13 C) 10
– peak area 12 – deuterium 10
– peak height 12 – in vivo 10
– ratio 13 standardisation 13
– relative standard deviation 13 – chromatographic time 13
– response 13 – chromatography 13
response surface modeling (RSM) 120 – GC-MS technology 13
retention time index 13 – intensity 13
reversed-phase 59 – ion current 13
– mass to charge 13
score plot 122 starch biosynthesis pathway
search 166, 167, 169 – adg-1 83
– compounds in mass spectra 169 – pgm-1 83
– metabolite or organism 166 subcellular compartment 144
– molecular formula 166, 167, 169 super-pathway 144
– molecular weight 166, 169 supervised 111
– taxonomic tree and hierarchy 169 SVG 156
secondary metabolism 152 – Scalable Vector Graphics 156
secondary metabolite 65, 66, 73, 74, 327 system biology 335
– benzenoid 73 systems biology 106
– flavonoid 73, 74
– glucosinolate 74 tabersonine 283
– indole derivative 74 TAG remodelling 217
– phenylpropanoid 73 tandem MS 215
– polyketide 73 taxadiene 295
– terpene 73 – fragmentation of hydroxylated 295
Subject Index 347
– polyacetate 295 tomato 43

– polyol 295 transcript 155
taxane 291 transcriptomics 21, 199, 201, 207, 208
– diterpenoid 291 triacylglycerol 211
taxoid 291, 299, 306, 307 tricarboxylic acid (TCA) cycle 137
– biosynthesis 294 trichomes 253
– extraction 306 – mint 255
– GC/MS 305 – tobacco 255
– GC-MS analysis 307 – tomato 255
– HPLC analysis 307 triterpene cyclase 251
– mass spectral fragmentation 294 triterpene saponin 250, 251
taxol 291, 292, 298, 302, 305, 308
– biosynthesis 291, 305 UHPLC 55
Taxus 291, 305, 307 unsupervised 111
– brevifolia 304 UPLC
– callus culture 306 – frictional heating 27
– cell culture 304–306 – particle size (dp ) 27
– cell suspension culture 291 – peak capacity 27
– cell suspension cultures 292 – resolution enhancement 27
– cuspidata 298 – speed 27
– metabolic engineering 308
– metabolome 307 vinblastine 280, 283
– secondary metabolism 307 vincristine 280
– x media 298, 302, 305, 306 vindoline 280, 283
terpene cyclase 244
terpenoid indole alkaloid 280 wheat
theoretical plate 53 – red-seeded 319
time-of-flight (TOF) 33, 34 – Triticum aestivum L. 319
TLC 237 – white-seeded 319
tobacco 320, 321
– gdhA 320, 321, 323 yew 291
Volumes already published
Volume 1: Trees I (1986)

Volume 2: Crops I (1986)
Volume 3: Potato (1987)
Volume 4: Medicinal and Aromatic Plants I (1988)
Volume 5: Trees II (1989)
Volume 6: Crops II (1988)
Volume 7: Medicinal and Aromatic Plants II (1989)
Volume 8: Plant Protoplasts and Genetic Engineering I (1989)
Volume 9: Plant Protoplasts and Genetic Engineering II (1989)
Volume 10: Legumes and Oilseed Crops I (1990)
Volume 11: Somaclonal Variation in Crop Improvement I (1990)
Volume 12: Haploids in Crop Improvement I (1990)
Volume 13: Wheat (1990)
Volume 14: Rice (1991)
Volume 15: Medicinal and Aromatic Plants III (1991)
Volume 16: Trees III (1991)
Volume 17: High-Tech and Micropropagation I (1991)
Volume 18: High-Tech and Micropropagation II (1992)
Volume 19: High-Tech and Micropropagation III (1992)
Volume 20: High-Tech and Micropropagation IV (1992)
Volume 21: Medicinal and Aromatic Plants IV (1993)
Volume 22: Plant Protoplasts and Genetic Engineering III (1993)
Volume 23: Plant Protoplasts and Genetic Engineering IV (1993)
Volume 24: Medicinal and Aromatic Plants V (1993)
Volume 25: Maize (1994)
Volume 26: Medicinal and Aromatic Plants VI (1994)
Volume 27: Somatic Hybridization in Crop Improvement I (1994)
Volume 28: Medicinal and Aromatic Plants VII (1994)
Volume 29: Plant Protoplasts and Genetic Engineering V (1994)
Volume 30: Somatic Embryogenesis and Synthetic Seed I (1995)
Volume 31: Somatic Embryogenesis and Synthetic Seed II (1995)
Volume 32: Cryopreservation of Plant Germplasm I (1995)
Volume 33: Medicinal and Aromatic Plants VIII (1995)
Volume 34: Plant Protoplasts and Genetic Engineering VI (1995)
Volume 35: Trees IV (1996)
Volume 36: Somaclonal Variation in Crop Improvement II (1996)
Volume 37: Medicinal and Aromatic Plants IX (1996)
Volume 38: Plant Protoplasts and Genetic Engineering VII (1996)
Volume 39: High-Tech and Microprogation V (1997)
Volume 40: High-Tech and Microprogation VI (1997)
Volume 41: Medicinal and Aromatic Plants X (1998)
Volume 42: Cotton (1998)
Volume 43: Medicinal and Aromatic Plants XI (1999)
Volume 44: Transgenic Trees (1999)
Volume 45: Transgenic Medicinal Plants (1999)
Volume 46: Transgenic Crops 1I (1999)
Volume 47: Transgenic Crops II (2001)
Volume 48: Transgenic Crops III (2001)
Volumes 1-48 were edited by Y.P.S. Bajaj†
Volume 49: Somatic Hybridization in Crop Improvement II (2001)
T. Nagata and Y.P.S. Bajaj (Eds.)
Volume 50: Cryopreservation of Plant Germplasm II (2002)
L.E. Towill and Y.P.S. Bajaj (Eds.)
Volume 51: Medicinal and Aromatic Plants XII (2002)
T. Nagata and Y. Ebizuka (Eds.)
Volume 52: Brassicas and Legumes: From Genome Structure to Breeding (2003)
T. Nagata and S. Tabata (Eds.)
Volume 53: Tobacco BY-2 Cells (2004)
T. Nagata, S. Hasezawa, and D. Inzé (Eds.)
Volume 54: Brassica (2004)
E.C. Pua and C.J. Douglas (Eds.)
Volume 55: Molecular Marker Systems in Plant Breeding and Crop Improvement (2005)
H. Lörz and G. Wenzel (Eds.)
Volume 56: Haploids in Crop Improvement II (2005)
C.E. Palmer, W.A. Keller, and K.J. Kasha (Eds.)
Volume 57: Plant Metabolomics (2006)
K. Saito, R.A. Dixon, and L. Willmitzer (Eds.)
Volumes in preparation
Tobacco BY-2 Cells: A New Treatise

T. Nagata, K. Matsuoka, and D. Inzé (Eds.)
Transgenic Crops IV
E.C. Pua and M.R. Davey (Eds.)
Transgenic Crops V
Transgenic Crops VI

57 Plant Metabolomics (158pages) Biotechnology in Agriculture and Forestry

Uploaded by

Document Informationclick to expand document information

Document Informationclick to expand document information

Copyright:

Available Formats

57 Plant Metabolomics (158pages) Biotechnology in Agriculture and Forestry

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

57 Plant Metabolomics (158pages) Biotechnology in Agriculture and Forestry

Uploaded by

Copyright:

Available Formats

Biotechnology in Agriculture and Forestry

With 96 Figures, 29 in Color, and 10 Tables

Library of Congress Control Number: 2005936763

Metabolomics is a rapidly-emerging sector of post-genome research. The

January 2006 Kazuki Saito,

Section I Analytical Technology

I.1 Gas Chromatography Mass Spectrometry . . . . . . . . . . . . . . . . . . . . 3

I.2 Current Status and Forward Looking Thoughts

I.3 Plant Metabolomics Strategies Based upon Quadrupole Time

I.4 Capillary HPLC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

I.5 Capillary HPLC Coupled to Electrospray Ionization

I.6 NMR Spectroscopy in Plant Metabolomics . . . . . . . . . . . . . . . . . . . 81

I.7 Hetero-nuclear NMR-based Metabolomics . . . . . . . . . . . . . . . . . . . 93

II.1 Bioinformatics Approaches to Integrate Metabolomics

II.2 Chemometrics in Metabolomics – An Introduction . . . . . . . . . . . . 117

II.4 AraCyc: Overview of an Arabidopsis Metabolism Database

II.5 KaPPA-View: A Tool for Integrating Transcriptomic

3 Plant Metabolic Pathway Maps . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158

II.6 KNApSAcK: A Comprehensive Species-Metabolite

Section III Applications

III.1 Systems Biology: A Renaissance of the Top-down Approach

III.2 Systems-based Analysis of Plant Metabolism by Integration

III.3 Targeted Proﬁling of Fatty Acids and Related Metabolites . . . . . . . . 211

III.4 Metabolic Proﬁling and Quantiﬁcation of Carotenoids

III.5 Metabolomics and Gene Identiﬁcation

III.6 Metabolomic Analysis of Catharanthus roseus

III.7 Metabolomics of Plant Secondary Compounds:

III.8 The Taxus Metabolome and the Elucidation

III.9 The Use of Non-targeted Metabolomics in Plant Science . . . . . . . . . 311

III.10 Plant Metabolite Proﬁling for Industrial Applications . . . . . . . . . . . 327

Subject Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341

C.H. Ric de Vos

Biotechnology in Agriculture and Forestry, Vol. 57

(Weckwerth et al. 2004b), (iii) proﬁling of trace compounds, or signalling

constituents (Urbanczyk-Wochniak et al. 2003; Steinhauser et al. 2004; Kopka

2 GC-MS Proﬁling Technology in a Nutshell

Metabolite proﬁling with GC-MS involves six general steps:

unknown metabolites. As we are far from knowing all possible metabolites

2.1 Chemical Derivatisation and Chromatography

The principles of fast metabolic sample inactivation and nondestructive ex-

2. Silylation reagents classify into those which introduce either a trimethyl-

2.2 Mass Detection and Quantitative Calibration Techniques

cause discriminations, especially in view of the complex and rather crude

diverse, representative metabolites (e. g. Roessner-Tunali et al. 2003; Gullberg

3 Short Excursion into Nomenclature and Deﬁnitions

Concise and unambiguous description of GC-MS metabolite proﬁling results

3.1 Metabolite and Analyte

A metabolite may be described as a compound which is internalised, chem-

3.2 Mass Spectral Tag (MST) and Mass Fragment

GC-MS metabolite proﬁles resolve hundreds of analytes, which represent

Because in non-biased analysis of GC-MS proﬁles identiﬁed and unidenti-

3.3 Response and Relative Quantiﬁcation of Metabolite Pools

GC-MS metabolite proﬁling studies monitor relative changes in metabolite

by addition of a constant amount of a representative internal standard com-

4 Present Challenges of GC-MS Proﬁling

4.1 Standardisation of GC-MS Systems

GC-MS proﬁles, with the exception of GC×GC-TOF-MS data, are in essence

4.2 Deconvolution and Alignment of Mass Spectral Tags