Metabolic Profiling: Methods and Protocols

ME T H O D S IN MO L E C U L A R BI O L O G Y Series Editor John M. Walker School of Life Sciences University of Her...

Author: Thomas O. Metz

92 downloads 2106 Views 7MB Size Report

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

Report copyright / DMCA form

DOWNLOAD PDF

ME T H O D S

IN

MO L E C U L A R BI O L O G Y

Series Editor John M. Walker School of Life Sciences University of Hertfordshire Hatfield, Hertfordshire, AL10 9AB, UK

For other titles published in this series, go to www.springer.com/series/7651

TM

Metabolic Profiling Methods and Protocols

Edited by

Thomas O. Metz Biological Sciences Division, Pacific Northwest National Laboratory, Richland, WA, USA

Editor Thomas O. Metz Pacific Northwest National Laboratory Biological Sciences Division P.O. Box 999, MS K8-98 Richland, WA 99352 USA [email protected]

ISSN 1064-3745 ISBN 978-1-61737-984-0 DOI 10.1007/978-1-61737-985-7 Springer New York Dordrecht Heidelberg London © Springer Science+Business Media, LLC 2011 All rights reserved. This work may not be translated or copied in whole or in part without the written permission of the publisher (Humana Press, c/o Springer Science+Business Media, LLC, 233 Spring Street, New York, NY 10013, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection with any form of information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology now known or hereafter developed is forbidden. The use in this publication of trade names, trademarks, service marks, and similar terms, even if they are not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject to proprietary rights. Printed on acid-free paper Humana Press is part of Springer Science+Business Media (www.springer.com)

Preface After accepting the task to edit a volume of Methods in Molecular Biology devoted to metabolic profiling, I began to contemplate the definition of the term. Fiehn referred to “metabolic profiling” as the identification and quantification of a select number of metabolites in an entire metabolic pathway or intersecting pathways (1). Closely related disciplines were targeted metabolite analysis, metabolic fingerprinting, and metabolomics, the latter of which was defined as the quantitative measurement of perturbations in the metabolite complement of a biological system (2). These four terms are often used interchangeably; indeed, in reviewing the literature over the past 40 years, it is evident that these various disciplines of metabolite analysis are related via an evolution of methods and technology. For example, while the field of metabolomics is now 10 years old, the protocols and instrumentation that form the foundation for the myriad approaches of this discipline are based on those originally established for the diagnosis of inborn errors of metabolism and drug metabolite analysis. Thus, in compiling this volume, I have made an attempt to incorporate protocols that are illustrative of the evolution of metabolic profiling from single molecule analysis to global metabolome profiling. The constraints of this volume necessitate that its contents will be perspective based, rather than comprehensive. However, it is my hope that the methods contained herein will be a resource for both established and new investigators in the field of metabolic profiling. Thomas O. Metz

References 1. Fiehn, O. (2002) Metabolomics – the link between genotypes and phenotypes. Plant Mol Biol 48, 155–171.

metabolic responses of living systems to pathophysiological stimuli via multivariate statistical analysis of biological NMR spectroscopic data. Xenobiotica 29, 1181–1189.

2. Nicholson, J. K., Lindon, J. C., Holmes, E. (1999) ‘Metabonomics’: understanding the

v

Contents Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

v

Contributors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

ix

1.

Origins of Metabolic Profiling . . . . . . . . . . . . . . . . . . . . . . . . . . . Arthur B. Robinson and Noah E. Robinson

1

2.

Amino Acid Profiling for the Diagnosis of Inborn Errors of Metabolism . . . . . Monique Piraud, Séverine Ruet, Sylvie Boyer, Cécile Acquaviva, Pascale Clerc-Renaud, David Cheillan, and Christine Vianey-Saban

25

3.

Acylcarnitines: Analysis in Plasma and Whole Blood Using Tandem Mass Spectrometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . David S. Millington and Robert D. Stevens

55

Analysis of Organic Acids and Acylglycines for the Diagnosis of Related Inborn Errors of Metabolism by GC- and HPLC-MS . . . . . . . . . . . . . . Giancarlo la Marca and Cristiano Rizzo

73

4.

5.

HPLC Analysis for the Clinical–Biochemical Diagnosis of Inborn Errors of Metabolism of Purines and Pyrimidines . . . . . . . . . . . . . . . . . . . . Giuseppe Lazzarino, Angela Maria Amorini, Valentina Di Pietro, and Barbara Tavazzi

99

6.

Bile Acid Analysis in Various Biological Samples Using Ultra Performance Liquid Chromatography/Electrospray Ionization-Mass Spectrometry (UPLC/ESI-MS) . . . . . . . . . . . . . . . . . . . . . . . . . . 119 Masahito Hagio, Megumi Matsumoto, and Satoshi Ishizuka

7.

Analysis of Glycolytic Intermediates with Ion Chromatography- and Gas Chromatography-Mass Spectrometry . . . . . . . . . . . . . . . . . . . . . . . 131 Jan C. van Dam, Cor Ras, and Angela ten Pierick

8.

Analysis of the Citric Acid Cycle Intermediates Using Gas Chromatography-Mass Spectrometry . . . . . . . . . . . . . . . . . . . . . . . 147 Rajan S. Kombu, Henri Brunengraber, and Michelle A. Puchowicz

9.

Quantification of Pentose Phosphate Pathway (PPP) Metabolites by Liquid Chromatography-Mass Spectrometry (LC-MS) . . . . . . . . . . . . 159 Amber Jannasch, Miroslav Sedlak, and Jiri Adamec

10.

High-Performance Liquid Chromatography-Mass Spectrometry (HPLC-MS)-Based Drug Metabolite Profiling . . . . . . . . . . . . . . . . . . 173 Ian D. Wilson

vii

viii

Contents

11.

Gas Chromatography-Mass Spectrometry (GC-MS)-Based Metabolomics . . . . 191 Antonia Garcia and Coral Barbas

12.

The Use of Two-Dimensional Gas Chromatography–Time-of-Flight Mass Spectrometry (GC×GC–TOF-MS) for Metabolomic Analysis of Polar Metabolites . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205 Kimberly Ralston-Hooper, Amber Jannasch, Jiri Adamec, and Maria Sepúlveda

13.

LC-MS-Based Metabolomics . . . . . . . . . . . . . . . . . . . . . . . . . . . 213 Sunil Bajad and Vladimir Shulaev

14.

Capillary Electrophoresis–Electrospray Ionization-Mass Spectrometry (CE–ESI-MS)-Based Metabolomics . . . . . . . . . . . . . . . . . . . . . . . . 229 Philip Britz-McKibbin

15.

Liquid Chromatography-Mass Spectrometry (LC-MS)-Based Lipidomics for Studies of Body Fluids and Tissues . . . . . . . . . . . . . . . . . . . . . . 247 Heli Nygren, Tuulikki Seppänen-Laakso, Sandra Castillo, Tuulia Hyötyläinen, and Matej Orešiˇc

16.

Electrospray Ionization Tandem Mass Spectrometry (ESI-MS/MS)Based Shotgun Lipidomics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259 Giorgis Isaac

17.

Processing and Analysis of GC/LC-MS-Based Metabolomics Data . . . . . . . . 277 Elizabeth Want and Perrine Masson

18.

Nuclear Magnetic Resonance (NMR)-Based Drug Metabolite Profiling . . . . . 299 Eva M. Lenz

19.

Nuclear Magnetic Resonance (NMR)-Based Metabolomics . . . . . . . . . . . . 321 Hector C. Keun and Toby J. Athersuch

20.

Slow Magic Angle Sample Spinning: A Non- or Minimally Invasive Method for High-Resolution 1 H Nuclear Magnetic Resonance (NMR) Metabolic Profiling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335 Jian Zhi Hu

21.

Processing and Modeling of Nuclear Magnetic Resonance (NMR) Metabolic Profiles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365 Timothy M.D. Ebbels, John C. Lindon, and Muireann Coen

Subject Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389

Contributors CÉCILE ACQUAVIVA • Laboratoire des Maladies Héréditaires du Métabolisme et Dépistage Néonatal, Hospices Civils de Lyon, Centre de Biologie Est, Bron, France JIRI ADAMEC • Department of Biochemistry, University of Nebraska-Lincoln, Lincoln, NE, USA ANGELA MARIA AMORINI • Institute of Biochemistry and Clinical Biochemistry, Catholic University of Rome, Rome, Italy TOBY J. ATHERSUCH • Biomolecular Medicine, Department of Surgery and Cancer, Faculty of Medicine, Imperial College London, South Kensington, London, UK SUNIL BAJAD • Sutro Biopharma Inc., South San Francisco, CA, USA CORAL BARBAS • Faculty of Pharmacy, San Pablo-CEU, Campus Monteprincipe, Madrid, Spain SYLVIE BOYER • Laboratoire des Maladies Héréditaires du Métabolisme et Dépistage Néonatal, Hospices Civils de Lyon, Centre de Biologie Est, Bron, France PHILIP BRITZ-MCKIBBIN • Department of Chemistry and Chemical Biology, McMaster University, Hamilton, ON, Canada HENRI BRUNENGRABER • Department of Nutrition, Mouse Metabolic Phenotyping Center, Case Western Reserve University, Cleveland, OH, USA SANDRA CASTILLO • VTT Technical Research Centre of Finland, Espoo, Finland DAVID CHEILLAN • Laboratoire des Maladies Héréditaires du Métabolisme et Dépistage Néonatal, Hospices Civils de Lyon, Centre de Biologie Est, Bron, France PASCALE CLERC-RENAUD • Laboratoire des Maladies Héréditaires du Métabolisme et Dépistage Néonatal, Hospices Civils de Lyon, Centre de Biologie Est, Bron, France MUIREANN COEN • Biomolecular Medicine, Department of Surgery and Cancer, Faculty of Medicine, Imperial College, London, UK VALENTINA DI PIETRO • Institute of Biochemistry and Clinical Biochemistry, Catholic University of Rome, Rome, Italy TIMOTHY M.D. EBBELS • Biomolecular Medicine, Department of Surgery and Cancer, Faculty of Medicine, Imperial College, London, UK ANTONIA GARCIA • Faculty of Pharmacy, San Pablo-CEU, Campus Monteprincipe, Madrid, Spain MASAHITO HAGIO • Division of Applied Bioscience, Research Faculty of Agriculture, Hokkaido University, Sapporo, Japan JIAN ZHI HU • Pacific Northwest National Laboratory, Richland, WA, USA TUULIA HYÖTYLÄINEN • VTT Technical Research Centre of Finland, Espoo, Finland GIORGIS ISAAC • Bio Separation and Mass Spectrometry, Pacific Northwest National Laboratory, Richland, WA, USA; Water corporation, Mulford, MA SATOSHI ISHIZUKA • Division of Applied Bioscience, Research Faculty of Agriculture, Hokkaido University, Sapporo, Japan AMBER JANNASCH • Bindley Bioscience Center, Purdue University, West Lafayette, IN, USA HECTOR C. KEUN • Biomolecular Medicine, Department of Surgery and Cancer, Faculty of Medicine, Imperial College London, South Kensington, London, UK

ix

x

Contributors

RAJAN S. KOMBU • Department of Nutrition, Mouse Metabolic Phenotyping Center, Case Western Reserve University, Cleveland, OH, USA GIUSEPPE LAZZARINO • Division of Biochemistry and Molecular Biology, Department of Chemical Sciences, University of Catania, Catania, Italy EVA M. LENZ • AstraZeneca Pharmaceuticals, Mereside, Macclesfield, UK JOHN C. LINDON • Biomolecular Medicine, Department of Surgery and Cancer, Faculty of Medicine, Imperial College, London, UK GIANCARLO LA MARCA • Mass Spectrometry, Clinical Chemistry and Pharmacology Laboratory, Department of Pharmacology, University of Florence, Meyer Children’s Hospital, Florence, Italy PERRINE MASSON • Biomolecular Medicine, Department of Surgery and Cancer, Faculty of Medicine, Imperial College, London, UK MEGUMI MATSUMOTO • Meiji Dairies Research Chair, Creative Research Institution Sousei (CRIS), Hokkaido University, Sapporo, Japan D AVID S. MILLINGTON • DUMC Biochemical Genetics Laboratory, Department of Pediatrics, Duke University Medical Center, Durham, NC, USA HELI NYGREN • VTT Technical Research Centre of Finland, Espoo, Finland MATEJ OREŠI Cˇ • VTT Technical Research Centre of Finland, Espoo, Finland MONIQUE P IRAUD • Laboratoire des Maladies Héréditaires du Métabolisme et Dépistage Néonatal, Hospices Civils de Lyon, Centre de Biologie Est, Bron, France MICHELLE A. PUCHOWICZ • Department of Nutrition, Mouse Metabolic Phenotyping Center, Case Western Reserve University, Cleveland, OH, USA K IMBERLY RALSTON-HOOPER • Ecosystem Research Division, National Research Council Post-Doctoral Fellow, United States Environmental Protection Agency, Athens, GA, USA COR RAS • Department of Biotechnology, Delft University of Technology, Delft, The Netherlands CRISTIANO RIZZO • Metabolic Unit and Laboratories, Bambino Gesù Children’s Hospital, Rome, Italy ARTHUR B. ROBINSON • Oregon Institute of Science and Medicine, Oregon, OR, USA NOAH E. ROBINSON • Oregon Institute of Science and Medicine, Oregon, OR, USA SÉVERINE RUET • Laboratoire des Maladies Héréditaires du Métabolisme et Dépistage Néonatal, Hospices Civils de Lyon, Centre de Biologie Est, Bron, France MIROSLAV SEDLAK • Laboratory of Renewable Resources Engineering, Purdue University, West Lafayette, IN, USA; Department of Agricultural and Biological Engineering, Purdue University, West Lafayette, IN, USA TUULIKKI SEPPÄNEN-L AAKSO • VTT Technical Research Centre of Finland, Espoo, Finland M ARIA SEPÚLVEDA • Department of Natural Resources, Purdue University, West Lafayette, IN, USA VLADIMIR SHULAEV • Virginia Bioinformatics Institute, Virginia Polytechnic Institute and State University, Blacksburg, VA, USA ROBERT D. STEVENS • Sarah W. Stedman Nutrition and Metabolism Center, Duke University Medical Center, Durham, NC, USA BARBARA TAVAZZI • Institute of Biochemistry and Clinical Biochemistry, Catholic University of Rome, Rome, Italy

Contributors

xi

ANGELA TEN PIERICK • Department of Biotechnology, Delft University of Technology, Delft, The Netherlands JAN C. VAN DAM • Department of Biotechnology, Delft University of Technology, Delft, The Netherlands CHRISTINE VIANEY-SABAN • Laboratoire des Maladies Héréditaires du Métabolisme et Dépistage Néonatal, Hospices Civils de Lyon, Centre de Biologie Est, Bron, France ELIZABETH WANT • Biomolecular Medicine, Department of Surgery and Cancer, Faculty of Medicine, Imperial College, London, UK IAN D. WILSON • AstraZeneca, Macclesfield, UK

Chapter 1 Origins of Metabolic Profiling Arthur B. Robinson and Noah E. Robinson Abstract Quantitative metabolic profiling originated as a 10-year project carried out between 1968 and 1978 in California. It was hypothesized and then demonstrated that quantitative analysis of a large number of metabolites – selected by analytical convenience and evaluated by computerized pattern recognition – could serve as a useful method for the quantitative measurement of human health. Using chromatographic and mass spectrometric methods to measure between 50 and 200 metabolites in more than 15,000 human specimens, statistically significant and diagnostically useful profiles for several human diseases and for other systematic variables including age, diet, fasting, sex, and other variables were demonstrated. It was also shown that genetically distinct metabolic profiles for each individual are present in both newborn infants and adults. In the course of this work, the many practical and conceptual problems involved in sampling, analysis, evaluation of results, and medical use of quantitative metabolic profiling were considered and, for the most part, solved. This article is an account of that research project. Key words: Metabolic profiling, metabolomics, urine, breath, chromatography, mass spectrometry, aging, diagnostic medicine, preventive medicine.

1. Introduction Since the dawn of the age of modern chemistry, biochemistry has been of great interest. When molecular structure became established as an exact discipline, the minds of scientists naturally turned toward those molecules of which they themselves are made. Extensive cataloging and structure determination of these substances followed. As the role of proteins in catalyzing the chemical reactions of metabolism was revealed, progress was made in understanding the metabolites – the smaller molecules required for life that protein

T.O. Metz (ed.), Metabolic Profiling, Methods in Molecular Biology 708, DOI 10.1007/978-1-61737-985-7_1, © Springer Science+Business Media, LLC 2011

1

2

Robinson and Robinson

catalysts select from the many atomic combinations available and produce to make life possible. Detailed understanding of metabolism was not, however, possible until the discovery of carbon 14 (1) and the development of tracer methodology (2), which now includes both radioactive and stable isotopes. When it became possible to label the atoms of metabolites and trace their paths through living systems, a thorough understanding of metabolism was achievable. This understanding and the rapid advance of protein chemistry then led to explanations for some of the simplest metabolic diseases – genetic errors that cause well-defined inborn errors of metabolism. As analytical technology advanced, the list of known genetic illnesses expanded to include a large number of such diseases which, while individually rare, together cause much suffering. This work was further accelerated by findings that, in some cases such as phenylketonuria, understanding of the disease could lead to effective therapy. Simultaneously, improvements in analytical chemistry led to a search for single metabolites that are diagnostic of more prevalent diseases – including those with non-genetic components. An extensive armament of single-substance measurements entered the inventory of clinical laboratories – tests for both inborn errors and other illnesses. Businesses arose to measure these substances, primarily in blood and urine, which have now grown in the United States alone into a $100 billion industry. This work usually involved the correlation of one substance with a condition of interest in human health. Scientists searched for metabolites and proteins, the quantities of which contained sufficient information about health and disease to warrant their measurement. A few such measurements became standard in health screening of ordinary patients, while a much larger number were made available in clinical laboratories, available upon request by physicians for specific patients. While the many substances measureable in human samples were increasingly evident as analytical methods improved, no practical efforts were made to test the possibility that the simultaneous quantitative analysis of large numbers of metabolites followed by computerized pattern recognition could yield health information of significant value. Forty years ago, however, there arose in California an experimental project with the potential to cause a paradigm shift toward the use of simultaneous measurement of large numbers of metabolites for the quantitative measurement of human health. This effort was ahead of its time and, therefore, faced daunting challenges in the construction of analytical and computational capabilities. This work was known in the 1970s as “quantitative metabolic profiling.” It is now a growing part of “metabolomics.” While

Origins of Metabolic Profiling

3

metabolomics still contains substantial single-substance components, extraordinary advances in analytical and computational technology are rapidly moving this field toward metabolic profiling – a continuation of that 1970s’ effort with greatly superior modern analytical equipment and computers. The California work was funded by private donors, NIH grants, and the personal savings of some of the scientists themselves. This effort proved the enormous analytical power of metabolic profiling as applied to human tissues and developed new analytical and computational tools. It had its origin in a collaboration beginning in 1968 between Linus Pauling and Art Robinson at the University of California at San Diego (UCSD). Later, it continued at Stanford University and the Institute of Orthomolecular Medicine (later renamed the Linus Pauling Institute of Science and Medicine), which Pauling and Robinson cofounded in Menlo Park, California, in 1973.

2. Orthomolecular Psychiatry

Pauling hypothesized (3) that the distribution functions of optimum human nutritional requirements are very wide, leading to nutritional deficiencies and illness, especially mental illness, in many people. He invented the term “orthomolecular psychiatry” – meaning right molecule in the right amount for mental health – to designate the treatment of mental illness by means of megavitamin therapy. Later this was designated “orthomolecular medicine” to include treatment of other illnesses in a similar way. Having worked together at Caltech in 1962–1963 on the chemical basis of general anesthesia (4), both Pauling and Robinson were faculty members at UCSD when Pauling made this proposal. At the time, Pauling was developing a theory of the structure of the atomic nucleus, and Robinson and his students were studying the deamidation of asparginyl and glutaminyl residues in peptides and proteins. In addition to this ongoing work, in 1968 the two men began a collaboration to test Pauling’s ideas about orthomolecular psychiatry, with Robinson directing the experimental work and Pauling extending the theoretical aspects, which led eventually to his widely known hypotheses concerning the role of vitamin C in health and disease. Pauling initially proposed an experimental program using vitamin-loading tests, in which large doses of vitamins were given to experimental subjects – those having mental illnesses and control subjects – and the urinary excretion of the vitamins measured. It was postulated that those individuals with greater needs for the substances would retain more, excreting lesser amounts in

4

Robinson and Robinson

their urine. Robinson assembled a small research group and set up a laboratory for this purpose, while continuing to direct his own laboratory – the size of which was increased by UCSD to accommodate the new work. The initial experiments emphasized loading tests with ascorbic acid, niacin, and pyridoxine, and some interesting results were obtained. It soon became evident, however, that this approach was of less value than hoped. The experiments gave very limited information, and the necessary analytical procedures of that day were laborious, time consuming, and expensive, which diminished their practical value.

3. Origin of the Profiling Hypothesis

4. Scientists Who Tested the Hypothesis

In the course of this work, Robinson utilized a method for measurement of pyridoxine in chemically derivatized urine by means of packed-column gas chromatography, which involved resolution of the pyridoxine peak from the large number of metabolic products that are present in urine. During these experiments, Robinson began to think that the information they needed might be more readily available in the many metabolic constituents evident in the chromatograms rather than in the pyridoxine peak itself. He reasoned as follows. The fundamental need was for a method to measure health vs. the amounts of ingested nutrients, as is illustrated in Fig. 1.1. This required, however, a means of measuring metabolic health quantitatively. He hypothesized that the needed values might be obtained by measuring the amounts of a large sampling of urinary metabolites and statistically correlating the patterns in these profiles with various states of human health and disease. Robinson, therefore, initiated an experimental program, with Pauling’s support, to test the hypothesis that quantitative metabolic profiles contained sufficient information for this purpose. As this work progressed, he assembled a skilled group of co-workers for this project.

These included Roy Teranishi, Dick Mon, and Robert Flath – highly skilled experts in gas chromatography; Martin Turner and Carl Boehme – engineers who built and maintained the PDP-11 vintage computer hardware used for lab automation and data collection; Laurelee Robinson – who wrote the computer software

Origins of Metabolic Profiling

5

Fig. 1.1. Diagrammatic representation of the nutritional problem that led to the development of metabolic profiling. The goal was to measure health as a function of nutrition quantitatively for (a) a single individual and (b) groups of individuals. It was hypothesized – and then demonstrated – that quantitative analyses of a large set of easily measurable metabolites contained sufficient information to determine the numbers needed for the vertical axis in (a) and that the same set could be used for a wide variety of human conditions. Reprinted with permission from the Proceedings of the 8th Annual Conference of the National Society for Autistic Children (15).

used for lab automation, data calculation, and profile evaluation; Fred Westall – Salk Institute chemist who provided samples for multiple sclerosis, muscular dystrophy, and Huntington’s disease work and helped organize the sample bank project; Bill Aberth – physicist who built the molecular ion mass spectrometers; Robert Melville – National Institutes of Health Administrator who arranged for and supervised the NIH support; glassblower Paul

6

Robinson and Robinson

Yeager; chemists Henri Dirren, Kent Matsumoto, and Lowell Brenneman; and biochemist Milton Winitz – who invented the synthetic diet Vivonex 100 and supplied it to the project in large quantities without charge, making possible the diet control studies. These people were aided by several technicians, including Sue Oxley, Maida Bergeson, Janet Tussey, Betsy Dore, Mark Weiss, and Walter Reynolds. Additionally helped by their own students and former students, Pauling’s – John Cheronis and Ian Keaveny – and Robinson’s – Fred Westall, Paul Cary, David Partridge, and Alan Sheets – and by other colleagues from UCSD, Stanford, Kaiser Permanente, and other institutions, a remarkable laboratory was built. Human samples were obtained from many institutions and physicians, including especially Drs. John Mann and George Ellison. Key to the laboratory’s success – from its completely automated, custom-made analytical equipment to its advanced computerized fund-raising systems (donations were obtained from more than 50,000 private individuals) – were four PDP-11 computer systems, utilizing machine language and Fortran programs that Art Robinson’s chemist and systems programmer wife Laurelee wrote in the 8-year period between 1971 and 1978. This group of people gradually built the finest physiological fluid analysis laboratory of its time – fully automated and designed for analysis of large numbers of samples. Ultimately, this laboratory measured, during a period of 8 years, more than 15,000 metabolic profiles – each profile including quantitative measurement of between 50 and 200 substances, primarily in human urine.

5. Accomplishments of the 10-Year Project

Most of the analytical work was by automated gas and ion exchange chromatography, with molecular ion mass spectrometry added during the later years and fragmentation mass spectrometry used for chemical characterizations of chromatographically resolved constituents. The early gas chromatographic work utilized 6-ft-long packed columns. These were replaced by 1000ft-long open-bore stainless steel columns, which were used for analysis of urine vapor and breath. The ion exchange system utilized the ninhydrin-positive physiological fluid analytical procedures developed by Dionex, with added PDP-11 automation. The identification work was carried out with a Finnegan gas chromatograph–fragmentation analysis mass spectrometer.

Origins of Metabolic Profiling

7

Between its beginning in 1969 and the end of the work in 1978, this research effort definitively showed the advantages of metabolic profiling – demonstrating unique statistically significant metabolic profiles for aging, multiple sclerosis, Duchenne dystrophy, Huntington’s disease, breast cancer, fasting, sex, diurnal variation, and chemical birth control use. The numbers of correlating metabolites found in these profiles ranged between 10 and 60, depending upon analytical technique and sample type (5–24). Reference 5 provides a review of much of this work, and References 25 and 26 provide more recent perspective. They also demonstrated, in experiments on 1,000 newborn infants, that the ninhydrin-positive metabolites in human urine are not unimodally distributed at birth. About half of these compounds are bimodally and trimodally distributed, apparently reflecting distinct genetic variations. In addition, they showed that the urinary profiles of adult humans are individually unique so that people can be identified by their urinary profiles just as with fingerprints. In experiments on themselves, they discovered the simplification of urinary profiles that occurs with strict chemical diet control but additionally showed that useful metabolic profiles can be obtained in most instances without diet control. Also demonstrated were metabolic profiles for physiological age in fruit flies (22), mice (21), and men (25), with the finding that about one-third of the metabolites in human urine are age correlated. This discovery opened the way toward quantitative evaluation of dietary and other measures that may decrease the degenerative diseases of aging. Some human conditions show no profiles. For example, their experiments on Stanford students failed to find a urinary metabolic pattern correlating with grade point average (5) – a useful comparison to the patterns that were found in other groups. The work of these scientists and engineers included the following: 1. Building the first fully automated metabolic profiling laboratory. Using PDP-11 computer technology, all of the chromatography and the later molecular ion mass spectrometry equipment that they built were entirely computer controlled and all of the data produced were computer collected and analyzed with very little manual intervention. 2. Building the first chromatographic breath and urine vapor metabolic analyzer (9). This machine contained four 1000ft-long stainless steel open-bore capillary columns. Each column was capable of resolution and quantitative analysis of about 200 volatile substances in a urine or a breath sample with a 6-h cycle time. In routine operation, it could analyze 16 samples per day.

8

Robinson and Robinson

3. The conceptual development and computerized implementation of mathematical tools that produced baselinecorrected integration of the chromatograms; corrected for chromatographic drift and automatically matched all metabolites within sets of chromatograms; normalized the experimental values to correct for variable physiological dilution and other systematic variables; computed nonparametric probabilities of significance for all metabolites in an experiment; corrected these probabilities for random correlation; tabulated and plotted cumulative probability distribution functions to determine the numbers of significant correlating substances; and, by means of a conceptually unique method (13, 25), computed the “diagnostic power” of any discovered metabolic pattern. 4. Origination and implementation of the concept of metabolic profiling by means of molecular ion mass spectrometric separation and quantitative analysis, without prior sample preparation or subdivision. This was done before the invention of the electrospray or laser ionization methods commonly in use today. In 1971, Stanford Research Institute experimental physicists William Aberth and Cap Spindt suggested a new way of producing ions for mass spectrometry without fragmentation, which made use of an array of hollow volcano-shaped structures with a grid registered above it, and Robinson suggested that this be used as a profiling device. With the help of Robert Melville at NIH, Aberth, Robinson, and Pauling received funding to build this device. Aberth ultimately built two such mass spectrometers – one at SRI with this NIH funding and one, later, at the Institute of Orthomolecular Medicine, with the second machine fully automated by Walter Reynolds, Carl Boehme, and Laurelee Robinson. These spectrometers had an initial resolution of 1 mass unit, with a design potential of 0.1 mass units in the mass range from about 50 to 1000. 5. Use of metabolic profiling in urine, breath, and cultured human cells. Similar profiling techniques can, of course, be applied to blood, saliva, and other sources of metabolites. The cultured cell work was undertaken to test the hypothesis that cell cultures from single individuals, monitored by metabolic profiles, might serve as individualized experimental systems in which therapeutic procedures for single individuals such as dietary requirements could be tested.

Origins of Metabolic Profiling

6. Goals and Requirements for Quantitative Metabolic Profiling

9

While the advance of technology during the past 30 years now makes a wonderful amount of fundamental biochemical knowledge available through metabolic profiling, these early workers were motivated almost entirely by a practical, empirical goal – in Robinson’s words, “improvement of the quality, length, and quantity of human life,” or, in Pauling’s, “diminution of the amount of human suffering.” Figure 1.2 summarizes these goals. In order to achieve these objectives, practical empirical metabolic profiling requires several things: 1. Suitable reproducible, quantitative, and low-cost analytical methods. 2. Control and experimental sample sets for test and calibration that are sufficiently free of uncontrolled systematic variables to allow reliable characterization of the differences under consideration. 3. Computerized mathematical methods that allow objective evaluation of the experimental results, without, through unnecessary complexity, separating the experimenters and their scientific intuitions from their data and results. During the 10-year duration of this project, these three problems were successfully addressed. The emphasis of the single-substance-orientated clinical chemistry industry of the 1970s was then, as it is today, primarily upon diagnosis of overt disease by technologically obsolete methods. Physicians are offered the amounts of single substances in human samples and comparisons with so-called normal values, typically two-standard-deviation ranges for the general population. Measurements of a couple of dozen such substances are included in ordinary analyses, and single substances beyond the normal range are noted and considered in patient evaluation. A large suite of additional single-substance measurements is available in industrial laboratories, which the physician can order to extend or confirm his diagnosis. This paradigm is expensive, so the number of substances measured is low and the application is limited to patients already exhibiting disease symptoms. Moreover, it entirely misses the metabolic patterns available from groups of substances that have values within the normal ranges – patterns that require computer analysis to discover. Quantitative metabolic profiling of analytically convenient metabolites allows a single analytical procedure, measuring a single large set of metabolites, to diagnose essentially all disease

10

Robinson and Robinson

Fig. 1.2. The primary health goal of “metabolic profiling” or, now, “metabolomics.” The gradually sloping curve of human survival, beginning to decrease at early ages, represents great amounts of suffering and lost years of human life, as illustrated here by the lifespan distribution of men in the United States in 1974 (a). The first goal of metabolomics should be the “squaring” of this curve (b) so that most people live a long, disease-free life during their intrinsic lifespan of about 90 years. A second goal should be extension of the healthful human lifespan (c). Reprinted with permission from (23).

Origins of Metabolic Profiling

11

conditions with one inexpensive procedure. Moreover, by including computerized pattern recognition, metabolic profiling extracts far more complete and valuable medical information than does the traditional method. The low cost and much greater information content of metabolic profiling permits its use in preventive medicine, allowing the individual to combat the probability of disease rather than overt disease itself. It also provides a convenient and inexpensive means of quantitative measurement of illness so that therapeutic procedures can be evaluated in real time – a capability almost entirely absent from current therapeutic medicine. Moreover, the finding that physiological age can be quantitatively measured by metabolic profiling opens the way toward the conduct of objective research for the evaluation of the effects of various adjustable nutritional and other parameters on aging and, when refined in the future, will allow single individuals to monitor their own rate of aging and probabilities of disease as a function of time and their own habits. The low cost and therefore increased availability of health evaluation that metabolic profiling makes possible can save the lives of many people that are now lost because current methods – imbedded in an expensive, inconvenient health system and providing inferior information – fail to diagnose their illnesses in time. “Health” is a concept that varies with individual objectives. Optimum health means different things to an athlete, to an artist, to a mathematician, or to a soldier. Each seeks to optimize different aspects of his abilities. The quantitative measurement of health that quantitative health profiling should eventually make possible will allow each person to optimize those abilities that he considers most valuable. It was these possibilities and other similar objectives that motivated Art Robinson and his colleagues in the 1970s to develop the techniques of quantitative metabolic profiling.

7. Analytical Methods Requirement 1, the choice of analytical methods is simplified by the fact that the many substances in a living metabolism are interlinked in synthesis and function, with each substance providing information about some of the others. When a large subset of these substances are quantitatively analyzed, the metabolites measured can be chosen by analytical convenience and economy rather than maximum information content per metabolite. Instead of seeking, as had been historically customary, one or two

12

Robinson and Robinson

high-information-content substances that are diagnostic of a disease or useful for some other practical health purpose, metabolic profiling combines the information in a large number of substances with lesser information content per substance. The profiler asks, “How many metabolic substances can be measured for a practical cost?” instead of “How can a single diagnostic substance be found?” This markedly reduces overall cost, an especially important goal for preventive medicine. A single cost-selected multi-substance profile is much less expensive than separate analyses of hundreds of singly selected substances. Thus, the same set of substances – between 50 and 200 during the original work in the 1970s and potentially thousands with current technology – can be used for many purposes, such as diagnosing disease, monitoring disease therapy, measuring optimum health as a function of diet, and other health goals. Quantitative measurement of a large number of metabolic substances, appropriate normalization, and computerized pattern recognition combine to make many new things possible. For example, quantitative measurement of individual human physiological age allows the graphing of physiological age vs. time. With this capability, the effect on physiological age of diet, exercise, and other adjustable parameters can be objectively monitored – both for groups and for single individuals. In the case of disease, metabolic profiling makes possible the quantitative measurement of the probability of a specific illness by comparing the profile of an individual with that of diseased individuals and those who will later become ill. During disease therapy, an individual can be measured as he moves along this probability axis, allowing the objective evaluation of therapy. Moreover, assessment of disease probabilities rather than overt symptoms opens the option of fighting the probability of disease rather than disease itself. While these and other similar health goals have become a part of our custom and culture, they have remained largely in the realm of qualitative discussion and guesswork, without the benefit of objective science and technology. The reason for this has been the lack of a means for measuring health quantitatively. Metabolic profiling provides this means. In the 1970s, chromatography was a slow but inexpensive method, while mass spectrometry was fast but expensive. Chromatography linked to mass spectrometry was, therefore, slow and expensive. Robinson imposed a condition upon the analytical methods chosen by that research group that no profiling tool would be used that could not, in industrial application, be offered to the general public at a cost of $5–$10 or less. This was necessary to the health goals of the project and is still a useful limitation.

Origins of Metabolic Profiling

13

8. Sampling Requirement 2, sampling has always been and still remains the weak link in the development of metabolic profiling. Even the simplest experiment – comparison of profiles of individuals known to be sick with a specific illness to those of well individuals – suffers from the danger of systematic error. There is always a possibility that the two groups also differ in ways not fundamental to the illness such as lifestyle, diet, therapeutic drugs, and other factors – differences that may exhibit profiles that are mistaken for those of interest. Moreover, comparison of profiles of individuals to those of groups of other individuals markedly decreases the information content of the profiles because it introduces biological variation into the observations. Longitudinal measurements wherein an individual serves as his own control and differences in his own profile with time are measured with reference to known population profiles contain far more information. Metabolic profiling – as a result of the enormous amount of information that it makes available – provides, for the first time, an objective means of fighting the probability of disease rather than disease itself. By reducing the data to a linear probability axis extending between sick and well, the probability of illness can be measured and the individual’s position on the axis determined – and moved empirically toward the well end of the axis by means of preventive measures before disease symptoms are evident. This cannot be done, however, unless the techniques can be calibrated by means of longitudinal samples obtained from people before they became ill. Realizing this need for sample sets from people that include serial samples from the same individuals before illness was evident, Robinson, Westall, and Pauling, in collaboration with colleagues at Kaiser Permanente, proposed in 1976 that a sample bank be created in which blood and urine samples were collected at 6-month intervals from 50,000 people in ordinary health and stored at –80◦ C. After 5 years, with current disease incidences, this bank would have provided statistically significant prospective sample sets for all diseases with incidences equal to or greater than multiple sclerosis. If collection continued beyond 5 years, suitable sample sets from more rare conditions would be available. Kaiser Permanente agreed to pay the costs of this project if their contribution were matched by $5 million from the National Institutes of Health. After 2 years of lobbying by Robinson, NIH decided in 1978 to fund an initial project with 10,000 subjects, with the possibility of later expansion to 50,000. Unfortunately, just after

14

Robinson and Robinson

this approval, work on profiling at the Pauling Institute collapsed, so this sample bank was not built. The sampling problem is today still the greatest impediment to progress in metabolomics. This sample bank should be built – with urine, breath, saliva, blood, and culturable human tissue collected from at least 100,000 ordinary people at regular intervals and stored at –80◦ C. After a few years, with such a bank in hand, a great increase in progress in the application of metabolic profiling for the improvement of human health would become possible.

9. Computation Requirement 3, computation was more difficult with the PDP11s of the 1970s, which were primitive computers by today’s standards. They were slow, expensive, and required continual repair. They were, however, a great advance over the mainframe computer systems that then existed at most research centers. The PDP-11s permitted dedicated computer systems for specific purposes, and they were not subject to the frequent outages and other limitations that rendered institutional mainframe systems unsuitable for metabolic profiling work. For real-time automation and data collection, the PDP-11s required programming in machine language in order to meet time requirements. For data analysis, Fortran programming was sufficient because time was not a constraint. Most of the multivariate computational methods in use today were available in the 1970s. These, however, proved less than ideal for the profiling work. First, although designed to give attractive displays such as pseudo two-dimensional plots of metabolically similar groupings, the outputs of these systems did not lend themselves to practical, medically useful decision making. Second, especially during research and development, computer procedures that move beyond the intuition of the experimentalists pose great risks of error. The danger of false positives in profile detection through the use of too many adjustable parameters, improper, uncorrected statistical analysis, introduction of systematic errors, and other factors must always be avoided – avoidance that becomes more difficult if “trust” is placed in computer calculations that are beyond checking by simple manual methods or mental calculations. Since there is a large amount of data to be collected and much repetitive calculation required, computers are essential for metabolic work, but they must not be allowed to become experimental variables in the research itself. For this reason, Laurelee and Art Robinson developed specialized

Origins of Metabolic Profiling

15

calculation procedures for this work (Fig. 1.3). These included the following: 1. Fully automatic baseline fitting, integration, and peak matching software to extract the necessary data from the chromatograms. While today, online mass spectrometry can do this, in the 1970s, automatic peak matching in highresolution chromatograms was an especially difficult problem, which they successfully solved. 2. After normalization to weighted averages of the measured values – a procedure they introduced to eliminate systematic errors that affect all of the values such as urine volume – probabilities of correlation with the phenomenon under study were calculated for each metabolite with nonparametric statistics. Since the distribution functions of most metabolites are not Gaussian and many are multi-modally distributed, parametric statistics is unsuitable. So, a method based on modified Wilcoxon statistics was employed. Cumulative distribution functions of these probabilities were then constructed to correct for random correlations and to test for the existence of metabolic profiles in the experiments, as illustrated in Fig. 1.4. 3. After measuring a set of n substances that correlate with a condition of interest, the values from this set were compressed to one dimension because most actions that would be based upon the analysis – for example, medical therapy – are one dimensional. So, simple mathematical means for making this compression were devised as diagrammatically represented in Fig. 1.5. 4. Since experiments to discover metabolic profiles do not inherently contain information about how these profiles will be used, a general method for evaluating profile strength is also needed. In medical practice, the detection of a deadly disease that will be treated by a safe therapy requires that false positives be risked at the expense of avoiding false negatives. Conversely, a mild disease with a relatively dangerous therapy requires the opposite bias. These and other similar interpretive biases are not, however, inherently a part of a metabolic profile. Therefore, a pattern strength methodology was invented, which averages all possible uses, as illustrated in Fig. 1.6. With the correlation indices of individual subjects for the n correlating metabolites compressed on a linear axis extending between the averaged profiles for the two groups being compared, that axis is divided at all possible points and the results graphed as shown. If no diagnostic power is present in the profiles, the values will lie along the diagonal line. If the diagnostic power of the metabolic profile is perfect, the values comprise

16

Robinson and Robinson

Fig. 1.3. The highest resolution metabolic profiler utilized in the 1970s (a). In 6 h, this device performed completely automated quantitative measurement of about 200 volatile compounds (b) in four different urine or breath samples. The equipment was completely automated with PDP-11 computers. All data analysis was also automated in another PDP-11 system, pictured here with Laurelee Robinson, who wrote the software for the 10-year project. Laurelee died in 1988 at the age of 43 – a death that could have been prevented had the metabolic profiling that she helped to develop been in routine medical use. Personal photographs – Art Robinson.

Origins of Metabolic Profiling

17

Fig. 1.4. Cumulative distribution functions of nonparametric probabilities of correlation. In this example of 72 ninhydrin-positive urinary metabolites as a function of sex in 71 male and 77 female Stanford students (a), the diagonal line corrects for random correlation. About 30 of the 72 substances correlate with sex, as is shown by deviation from the diagonal line of random correlation. A similar experiment with 57 male Stanford students (b) shows that, since no profile is present for grade point average, the measured values fall along the diagonal line. Alternatively, a tabular presentation of number of substances below each probability vs. the number expected below that probability can be used. Reprinted with permission from Clinical Chemistry (5). In ordinary singlesubstance clinical chemistry, it is more difficult to rigorously evaluate the significance of a reported correlation because there is no suitable way to correct for the unknown number of unreported experiments that have failed to detect a correlation. Metabolic profiling, in which each experiment evaluates a statistically significant number of potential correlations, has the advantage that correction for random correlation can be made from the experimental data itself.

18

Robinson and Robinson

Fig. 1.5. Diagrammatic representation of three important uses of metabolic profiling. The n-dimensional profiles of substances that correlate with physiological age, probability of illness, and severity of illness are computationally compressed onto one-dimensional axes suitable for empirical evaluation of interventions for preventive and diagnostic medicine and for therapeutic monitoring. This one dimensional compression is also used in the calculation of profile “diagnostic power” as illustrated in Fig. 1.6. Reprinted with permission from Mechanisms of Ageing and Development (23).

Fig. 1.6. Diagnostic power of the sex-correlated ninhydrin-positive compounds illustrated in Fig. 1.4a. The n correlates are computationally compressed to one dimension. The one-dimensional axis extending between the male profile and the female profile as diagrammatically illustrated in Fig. 1.5 has one value for each of the 148 students. In the computation of these values, the profile of the student being calculated is always omitted from the calculation of the profiles to which he is being compared. Reprinted with permission from Clinical Chemistry (5). The resulting linear axis is divided at all of the 149 possible places between and beyond the 148 values and the errors plotted as shown. If there were no diagnostic power, the values would lie along the diagonal line. If the diagnosis is perfect, the graph is reduced to a point in the origin. The fraction of the area between the diagonal and perfect correlation that has been successfully eliminated is the “diagnostic power,” in this case 0.93. This quantitative experimental endpoint is useful in evaluating the strength of a profile, in comparing the relative values of alternative profiling techniques, and in optimizing experimental and computational parameters, such as the coefficients used to normalize the data.

Origins of Metabolic Profiling

19

a point at the origin. The “diagnostic power” is defined as the percentage of the area between the diagonal line and the origin that has been successfully eliminated by the technique. If the data are from more than one replicated experiment or from groups of subjects that are fundamentally impossible to completely separate into unique groups, the probability and diagnostic power distributions are non-linear, so appropriately curved expected distributions are used (13, 24) rather than the linear ones illustrated in Figs. 1.4 and 1.6. The invention of this method (13) permitted further optimization of the metabolic profiling experiments and calculation methods because it provided a quantitative score for the entire experiment and calculation. For example, data normalization coefficients could be optimized by iterative calculations to maximize diagnostic power. This “diagnostic power” method has proved to be generally useful in many unexpected ways. For example, it was recently used to optimize the parameters involved in predicting the deamidation rates in asparginyl residues for proteins of known threedimensional structure (27–30).

10. Summary of the Project In 1978, after 10 years of effort, the initial hypothesis about metabolic profiling had been verified. In every case in which the profiling laboratory completed an experiment intended to detect a profile distinctive of human disease and in many other experiments on variables of interest, a distinctive, diagnostically useful profile had been found. A set of 1000 humans had been profiled at birth with the goal of correlating their genetically unique urinary profiles with their health in later life. Analytical methods using chromatography and direct injection molecular ion mass spectrometry had been refined and automated, and suitable computational tools developed. Moreover, funding had been obtained for the regular sampling of 10,000 humans and permanent storage of the samples at –80◦ C so that profiling technologies could be tested and calibrated with longitudinal controls. Also, the various metabolic profiles discovered over the 10 years of work had been correlated with a single database.

11. Termination of the Project During the 10 years, all of the conceptual and computational questions that arose had been answered. So, the discipline of

20

Robinson and Robinson

metabolic profiling was firmly established. One metabolic profiling laboratory was well funded, equipped, and staffed with welltrained and experienced scientists and engineers and was ready to move forward. Robinson, 36 years old, was then research professor and president of the Pauling Institute and Pauling, 77 years old, was research professor and chairman of the board. Unfortunately, a disagreement arose between these two men regarding a series of experiments on nutrition and the growth rate of squamous cell carcinoma in hairless mice that had been carried out under Robinson’s direction. Involving experiments with about 2,000 mice, these experiments showed that sub-optimum nutrition decreased the cancer growth rate, while better nutrition increased it. The growth rate of squamous cell carcinoma in these mice was varied over a range of 20-fold by means of diet alone, an important finding that may have importance in the management of human cancer (30). General dietary restriction or provision of single nutritional components such as protein in amounts above or below those providing optimum overall health decreased the cancer growth rate, a finding that included vitamin C. With a supplement of the human equivalent of about 3 g/d (a dose Pauling was recommending to the public), vitamin C increased the cancer growth rate by twofold as compared with no supplement – a result that, unfortunately, Pauling believed harmful to his personal campaign to promote vitamin C. Exacerbated by several individuals who sought to gain from this disagreement, this controversy caused the Institute’s research to collapse, and all further work on metabolic profiling ended. Pauling and his attorneys who gained control of the Institute succeeded in prohibiting Robinson and his colleagues access to their equipment and research data, and the very high cost of rebuilding the necessary technology elsewhere prevented them from continuing the work. Accounts of a large part of their work in the 1970s have, however, been published (refs. 5–24 are a complete listing) and should prove useful to those in the field of metabolomics who are now rapidly advancing by means of the wonderful new analytical technology that is available today.

12. Some Thoughts About the Future

In the 1970s, the goal of metabolic profiling was to obtain as much quantitative health information as conveniently and economically as possible. At that time, urine analysis in central laboratories was the best path to this goal. Recognizing that other tissues and fluids were potentially complimentary or superior

Origins of Metabolic Profiling

21

sources, small forays into breath, saliva, blood, and cultured tissue were also undertaken. With regard to the substances analyzed, volatile compounds were easiest with the technology of that day, but non-volatile compounds appeared to contain more information per substance. While nucleic acids, proteins, and metabolites are all now of interest as a result of technological advance, only metabolites were within realistic analytical reach in the 1970s. Even today, it is likely that metabolites are still superior for these purposes. Nucleic acids and proteins are the blueprints for the biological structure and the machines that comprise and operate it. Quantitative health determination based on these molecules is, however, still in its infancy. Metabolites provide immediate real-time access to biochemical information reflected in the chemical output of basic processes in the living system. This profiling is becoming especially valuable, since the computer revolution now makes possible the transfer of diagnostic technology away from centralized laboratories and into each individual’s personal environment (25, 26). This transfer will markedly lower cost and increase convenience and, thereby, make possible a continuous flow of very useful information to each individual. We are not far from a time when each person’s personal computer will constantly acquire data about his health and submit that data for interpretation to the free-enterprise entrepreneurs of his choice – anywhere in the world. Which technologies will be ascendant in the future? All relevant health measurement technologies as applied to a wide variety of samples will obviously compete among research laboratories and in the marketplace. Only this competitive environment can sort out which approaches will provide the greatest health benefits for the greatest numbers of people. Many human factors as well as scientific factors will be important, such as the crucial issue of human compliance. Miraculous machines may arise for the analysis of drops of urine or blood, for example, but how many people would routinely utilize them? While we surely do not know the answers to these questions, breath metabolite analysis seems especially promising. On the basis of information content and ease of analysis, current techniques of breath analysis are much inferior to, for example, urine or blood analysis, but the human compliance problem is fundamentally much easier. Miniaturized mass spectrometers or other devices, for example, purchased as computer peripherals or perhaps routinely included on ordinary personal computer motherboards could continuously sample the breath of a single computer user or the air in a room frequented by a few people, without distraction of the people being monitored. During its birth in the 1970s, metabolic profiling required custom-built, slow, and expensive analytical machines coupled to

22

Robinson and Robinson

primitive, costly, and temperamental computer systems. Still, even then, it could have been incorporated as a very useful tool in centralized clinical laboratories. It is tragic that this was not done. Now, the advance of technology has removed these limitations. Analytical technology is greatly improved and is becoming less expensive and more reliable, while powerful interlinked computers have brought the computational requirements of metabolic profiling within reach of virtually every individual in the developed world. The applications of this new technology to increase the quality, quantity, and length of human life during the coming decades should be spectacular indeed. References 1. Kamen, M. D., Ruben, S. (1940) Production and properties of carbon 14. Phys Rev 58, 194. 2. Kamen, M. D. (1957) Isotopic Tracers in Biology: An Introduction to Tracer Methodology, Academic Press, New York, NY. 3. Pauling, L. (1968) Orthomolecular psychiatry. Science 160, 265–271. 4. Robinson, A. B., Manly, K. F., Anthony, M. P., Catchpool, J. F., Pauling, L. C. (1965) Anesthesia of artemia larvae: method for quantitative study. Science 149, 1255–1258. 5. Robinson, A. B., Pauling, L. C. (1974) Techniques of orthomolecular diagnosis. Clin Chem 20, 961–965. 6. Teranishi, R., Mon, T. R., Robinson, A. B., Cary, P., Pauling, L. C. (1972) Gas chromatography of volatiles from breath and urine. Anal Chem 44, 18–20. 7. Pauling, L. C., Robinson, A. B., Teranishi, R., Cary, P. (1971) Quantitative analysis of urine vapor and breath by gas– liquid partition chromatography. Proc Natl Acad Sci USA 68, 2374–2376. 8. Robinson, A. B., Pauling, L. C. (1973) Quantitative chromatographic analysis in orthomolecular medicine, in W H Freeman & Co (Hawkins, D., ed.), Orthomolecular Psychiatry, pp 35–53. 9. Robinson, A. B., Partridge, D., Turner, M., Teranishi, R., Pauling, L. C. (1973) An apparatus for the quantitative analysis of volatile compounds in urine. J Chromatogr 85, 19–29. 10. Matsumoto, K. E., Partridge, D. H., Robinson, A. B., Pauling, L. C., Flath, R. A., Mon, T. R., Teranishi, R. (1973) The identification of volatile compounds in human urine. J Chromatogr 85, 31–34. 11. Pauling, L. C., Robinson, A. B. (1973) Techniques of orthomolecular medicine. In First

12.

13.

14.

15.

16.

17.

18.

19.

20.

Conference on the Analysis of Multicomponent Mixtures and Their Application to HealthRelated Problems 1, 1–7. Robinson, A. B., Cary, P., Dore, B., Keaveny, I., Brenneman, L., Turner, M., Pauling, L. (1973) Orthomolecular diagnosis of mental retardation and diurnal variation in normal subjects by low-resolution gas–liquid chromatography of urine. Int Res Commun Syst 70, 3. Robinson, A. B., Westall, F. C. (1974) The use of urinary amine measurement for orthomolecular diagnosis of multiple sclerosis. J Orth Psychol 3, 1–10. Robinson, A. B., Westall, F. C., Ellison, G. W. (1974) Multiple sclerosis: urinary amine measurement for orthomolecular diagnosis. Life Sci 14, 1747–1753. Robinson, A. B. (1974) Orthomolecular medicine – diagnosis and therapy. In Proceedings of the 8th Annual Conference National Society for Autistic Children, 1–8. Robinson, A. B. (1975) Looking for optimum health: A guided tour through the Linus Pauling Institute. Prevention 89–96. Robinson, A. B., Weiss, M., Reynolds, W. E., Robinson, L. R. (1975) Use of mass spectrometry for Orthomolecular diagnosis. In Proceedings Twenty-Third Annual Conference on Mass Spectrometry and Allied Topics, 182–184. Dirren, H., Robinson, A. B., Pauling, L. C. (1975) Sex-related patterns in the profiles of human urinary amino acids. Clin Chem 21, 1970–1975. Rosenberg, R. N., Robinson, A. B., Partridge, D. (1975) Urine vapor pattern for olivopontocerebellar degeneration. Clin Biochem 8, 365–368. Robinson, A. B., Dirren, H., And Sheets, A., Miquel, J., Lundgren, P. R. (1976) Quanti-

Origins of Metabolic Profiling

21.

22. 23.

24.

25.

tative aging pattern in mouse urine vapor as measured by gas–liquid chromatography. Exp Gerontol 11, 11–16. Robinson, A. B., Willoughby, R., Robinson, L. R. (1976) Age dependent amines, amides, and amino acid residues in Drosophila melanogaster. Exp Gerontol 11, 113–120. Robinson, A. B., Pauling, L. C., Aberth, W. (1977) A controversy: diagnosis of infectious hepatitis. Clin Chem 23, 908–910. Robinson, A. B. (1979) Molecular clocks, molecular profiles, and optimum diets: three approaches to the problem of aging. Mech Ageing Dev 9, 225–236. Robinson, A. B., Robinson, L. R. (1991) Quantitative measurement of human physiological age by profiling of body fluids and pattern recognition. Mech Ageing Dev 59, 47–67. Robinson, A. B. (2007) Revolutionizing 21st century medicine with consumer-based

26. 27.

28.

29. 30.

23

diagnostics and the internet. J Am Phys Surg 12(1), 14–21. Robinson, A. B. (2007) Human health in the Telecosm. Gilder/Forbes Telecosm 10. Available at www.oism.org/health Robinson, N. E., Robinson, A. B. (2001) Prediction of protein deamidation rates from primary structure and three-dimensional structure. Proc Natl Acad Sci USA 98, 4367–4372. Robinson, N. E., Robinson, A. B. (2001) Deamidation of human proteins. Proc Natl Acad Sci USA 98, 12409–12413. Robinson, N. E. (2002) Protein deamidation. Proc Natl Acad Sci USA 99, 5283–5288. Robinson, A. B., Hunsberger, A., Westall., F. C. (1994) Suppression of squamous cell carcinoma in hairless mice by dietary nutrient variation. Mech Ageing Dev 76, 210–214.

Chapter 2 Amino Acid Profiling for the Diagnosis of Inborn Errors of Metabolism Monique Piraud, Séverine Ruet, Sylvie Boyer, Cécile Acquaviva, Pascale Clerc-Renaud, David Cheillan, and Christine Vianey-Saban Abstract The diagnosis of inherited metabolic disorders of amino acid (AA) metabolism is based on the qualitative and/or the quantitative analysis of AAs, mainly in blood and urine. For years, the most widespread technique in use was ion-exchange chromatography followed by post-column derivatization with ninhydrin, a method which is the basis of numerous automated AA analyzers with a throughput of about eight samples/day. The emergence of tandem mass spectrometry (MS/MS) coupled to liquid chromatography (LC) has made possible the measurement of many metabolites for the diagnosis of inborn errors of metabolism. The LC-MS/MS method described here allows the clinical diagnosis of AA disorders by analysis of underivatized AAs and derivative molecules in various biological samples prepared by methanol precipitation. AAs are separated by ion-pairing reversed-phase LC, using perfluorocarboxylic acid as an ion-pairing agent. Each AA is detected in MS/MS-positive ionization mode by its specific transition. The method allows the analysis of about 40 biological samples/day. Key words: Amino acids, inborn errors of metabolism, tandem mass spectrometry, liquid chromatography.

1. Introduction The diagnosis of inherited metabolic disorders of amino acid (AA) metabolism is based on the qualitative and/or the quantitative investigation of about 80 AAs or derivative molecules in biological fluids, mainly blood and urine, but also cerebrospinal fluid (CSF), leukocytes, and amniotic fluid (AF). For years, the most T.O. Metz (ed.), Metabolic Profiling, Methods in Molecular Biology 708, DOI 10.1007/978-1-61737-985-7_2, © Springer Science+Business Media, LLC 2011

25

26

Piraud et al.

widespread technique for this purpose was ion-exchange chromatography followed by post-column derivatization with ninhydrin, a method which is the basis of numerous automated AA analyzers with a throughput of about eight samples/day (1). Tandem mass spectrometry (MS/MS) coupled to liquid chromatography (LC) is still a developing technology capable of replacing classical methods in clinical biology (2) for the diagnosis of inborn errors of metabolism (3–5). Indeed, LC-MS/MS has been applied to newborn screening and other numerous metabolic conditions (6). An LC-MS/MS method has been developed allowing for the analysis of underivatized AAs in various biological samples for clinical diagnosis (7–9). The samples (plasma, urine, CSF, leukocytes, AF) are prepared by methanol precipitation. AAs are separated by ion-pairing reversed-phase liquid chromatography, using perfluorocarboxylic acid as an ion-pairing agent. Each AA is detected in MS/MS-positive ionization mode by its specific transition. The method allows a much higher throughput (about 40 biological samples/day) compared to traditional methods, and the technique can be adapted to other kinds of samples.

2. Materials Reagents used for daily analysis of AAs are underlined. All of them are made ready for analysis. 2.1. Common Reagents

1. Acetonitrile. 2. HPLC-grade methanol. 3. Sterile deionized water.

2.2. LC

1. Solvent A: 0.5 mM Tridecafluoroheptanoic acid (TDFHA). Store at room temperature. For preparation of the stock solution, see Note 1. 2. Solvent B: Acetonitrile. Store at room temperature. 3. Octadecyl-bonded silica gel LC column: Modulo-cart QS Uptisphere, 120 Å, 3 μm BP2, 50 mm × 2 mm (Interchrom; Interchim, Montluçon, France) (see Note 2). 4. Pre-columns of the same stationary phase (Interchim): 10 mm × 2 mm. 5. On-line filters (Interchim): 2 μm. 6. Direct connectors for 10-mm pre-column (Interchim). 7. Two Series 200 micropumps (Perkin-Elmer, Norwalk, CT, USA) (see Note 3).

Amino Acid Profiling for the Diagnosis of Inborn Errors of Metabolism

27

8. Series 200 autosampler (Perkin–Elmer) (see Note 3). 9. The temperature of the LC column is controlled with a Croco-Cil (Cluzeau Info Labo, Sainte-Foy-la-Grande, France). 2.3. MS/MS

1. API 2000 triple-quadrupole mass spectrometer (AB Sciex, Toronto, Canada) equipped with a TurboIonSpray source. 2. Balston generator (Parker Hannifin International Ltd, Milan, Italy) to produce nitrogen from compressed (8 bars) air for desolvation and collision. 3. Analyst 1.3.1 software (AB Sciex).

2.4. Amino Acids and AA Qualitative Standards

Beside the individual stock solution for each molecule, a solution containing 26 common AAs is prepared for daily checking of the chromatographic separation. 1. Investigated AAs and other derivative molecules and their manufacturers are listed in Table 2.1, with abbreviations used. 2. Stock solutions of each molecule (5 mM in water (see Note 4)). Store in aliquots at –20◦ C for a maximum of 1 year. 3. L-Cysteine-L-homocysteine disulfide (Cys-Hcy) is not commercially available (see Note 5). 4. Anhydrides of L-argininosuccinic acid (ASA anhydrides) are not commercially available (see Note 6). 5. 50 μM stock solution (26 AAs): Combine 1 mL of each 5 mM AA solution (see list in Table 2.2) and complete to 100 mL with water. Store in 1 mL aliquots at –20◦ C for a maximum of 1 year. 6. 25 μM solution (26 AAs): Dilute 1 mL of the 50 μM stock solution with 1 mL of 1.25 mM TDFHA in a 2-mL injection vial. This solution should be prepared every other week. 7. S-2-Aminoethylcysteine (S2AE, see Note 7) 1 mM stock solution in methanol: prepare in a glass flask with grinded cap by dissolving 20.1 mg of S2AE in 100 mL methanol. Store in a glass flask with grinded cap tightly closed at 4◦ C (maximum 6 months, see Note 8). Do not let the flask open longer than necessary when preparing the deproteinization reagent (see Note 8).

28

Piraud et al.

Table 2.1 List of amino acids and derivative molecules of biological interest for the diagnosis of inherited disorders of amino acids metabolism Name

Abbreviation MM

Manufacturer Biological interest

Sample

Ethanolamine, HCl

EA

61

Calbiochem

Hypoxia

P,U

Glycine

Gly

75

Merck

Nonketotic hyperglycinemia, organic acidurias

P,U,CSF

L -Alanine

Ala

89

Calbiochem

Hyperlactacidemia

P,U

Sarcosine

Sar

89

Sigma

MAD (ETF or ETF-QO deficiency)

P,U

β-Alanine

β-Ala

89

Merck

Hyper-β-alaninemia

U

L -α-Amino-n-

Abu

103

Calbiochem

Protein intake

P

N,NDimethylglycine

(Me)2 -Gly

103

Sigma

MAD (ETF or ETF-QO deficiency)

P,U

β-Aminoisobutyric acid

β-AIB

103

Sigma

Neoplasia, excessive breakdown

γ-Aminobutyric acid L -Serine

GABA

103

Calbiochem

Ser

105

Calbiochem

CNS disorders, neurotrans- P,U,CSF mitter disorders Vitamin B6 deficiency, B6 P,U, CSF synthesis deficiency, serine deficiency disorders

1 -Pyrroline5-carboxylic acid

PC

113

Sigma

Hyperprolinemia (type II)

L -Proline

Pro

115

Calbiochem

Hyperprolinemia (types I, II) P,U

butyric acid

tissue U

U

L -Valine

Val

117

Merck

Maple syrup urine disease

L -Homoserine

Hse

119

Sigma

Interference (neuroblastoma) P, U

L -Threonine

Thr

119

Calbiochem

Hepatic failure

P,U

Taurine

Tau

125

Sigma

Sulfite oxidase deficiency

P,U

Pip

129

Sigma

Peroxisomal disorders

P,U

Pyro-L-glutamic acid

pGlu

129

Fluka

Pyroglutamic aciduria

P,U

ε-Aminocaproic acid

eCap

131

Calbiochem

Interference (therapeutic agent)

δ-Aminolevulinic acid

dALA

131

Sigma

Tyrosinemia type I, porphyrias

U

4-Hydroxy- Lproline

Hyp

131

Calbiochem

Hydroxyprolinemia

P,U

L -Leucine

Leu

131

Calbiochem

Maple syrup urine disease

P,U

L -allo-Isoleucine

aIle

131

Sigma

Maple syrup urine disease

P,U

L -Pipecolic

acid

P,U

Amino Acid Profiling for the Diagnosis of Inborn Errors of Metabolism

Table 2.1 (continued) Name

Abbreviation MM

Manufacturer Biological interest

Sample

L -Isoleucine

Ile

131

Calbiochem

Maple syrup urine disease

P,U

Orn

132

Merck

Urea cycle disorders, HHH, P,U hyperornithinemia

L -Asparagine

Asn

132

Fluka

Asparaginase treatment (leukemia)

P,U

L -Aspartic

Asp

133

Merck

Asparaginase treatment (leukemia)

P,U

L -Homocysteine

Hcy

135

Sigma

Homocystinuria, Cbl C/D P,U deficiency

L -Glutamine

Gln

146

Merck

Hyperammonemia, urea cycle disorders

L -Lysine,

Lys

146

Calbiochem

Hyperlysinemia, cystinuria– P,U lysinuria

Glu

147

Calbiochem

Urea cycle disorders

P,U

L -Methionine

Met

149

Calbiochem

Homocystinuria, hypermethioninemia

P,U

L -Histidine,

HCl

His

155

Calbiochem

Histidinemia

P,U

L -α-Aminoadipic

Aad

161

Sigma

α-Aminoadipic aciduria

P,U

5-Hydroxy-L-lysine, HCl

Hyl

162

Sigma

Hydroxylysinuria

U

N-Acetyl-L-cysteine

NAcCys

163

Merck

Interference (therapeutic agent)

P,U

L -Phenylalanine

Phe

165

Merck

Phenylketonuria, hyperphenylalaninemia

P,U

N3 -Methyl-Lhistidine

His-1Me

169

Calbiochem

Renal insufficiency

P,U

N1 -Methyl-Lhistidine

His-3Me

169

Calbiochem

Renal insufficiency, denutri- P,U tion, starvation

Glycyl-L-proline

Gly-Pro

171

Sigma

Iminodipeptiduria

L -Ornithine,

HCl

acid

HCl

L -Glutamic

acid

P,U

acid

U

L -Arginine

Arg

174

Merck

Urea cycle disorders

P,U

Formimino-Lglutamic acid

FIGLU

174

Sigma

FIGLU aciduria

U

N-Acetyl-Lornithine

NAcOrn

174

Sigma

Interference (hyperornithinemia)

L -Citrulline

Cit

175

Merck

Urea cycle disorders

P,U

L -Tyrosine◦

Tyr

181

Calbiochem

Tyrosinemia type I or II

P,U

L -Homocitrulline

Hci

189

ICN

HHH

P,U

L -3-(3,4-

Dopa

197

Sigma

Tyrosinemia, neurotrans- U mitter disorders

Dihydroxyphenyl)alanine◦

29

30

Piraud et al.

Table 2.1 (continued) Name

Abbreviation MM

Manufacturer Biological interest

Sample

S-Sulfo-L-cysteine

S-Cys

201

Sigma

Sulfite oxidase deficiency

P,U

N G ,N G Dimethylarginine, HCl

asym(Me)2 Arg

202

Sigma

Physiological

U

L -Tryptophan◦

Trp

204

Calbiochem

Hartnup disease

P,U

L -Kynurenine

Kyn

208

Sigma

Vitamin B6 deficiency

U

5-Hydroxy-Ltryptophan

Hyt

220

Sigma

Therapeutic agent, neuro- U transmitter disorders

Hcy(Ala)

222

Sigma

Cystathioninuria, neuroblastoma

P,U

3-Hydroxy-Lkynurenine

Hyk

224

Sigma

Vitamin B6 deficiency

U

L -Carnosine

Car

226

Calbiochem

Carnosinemia, physiological U (meat)

L -Anserine

Ans

240

Sigma

Carnosinemia, physiological U (meat)

L -Cystine

(Cys)2

240

Merck

Cystinuria, cystinuria– P,U lysinuria, homocystinuria, sulfite oxidase deficiency

L -Cysteine- L-

Cys-Hcy

254

∗

Cystinosis Homocystinuria, cystinuria

L P,U

(Hcy)2

268

Sigma

Homocystinuria

P,U

ASA anh

272

∗

Urea cycle disorders (AS aciduria)

P,U

L -Saccharopine

Sac

276

Sigma

Saccharopinuria

P,U

L -Argininosuccinic

ASA

290

Sigma

Urea cycle disorders (AS aciduria)

P,U

Glutathione (reduced form)

GSH

307

Sigma

γGT deficiency, GSH synthase deficiency

P,U

β-Aspartylglucosamine

GlcNAcAsn

335

Sigma

Aspartylglucosaminuria

U

L -Cystathionine◦

homocysteine disulfide L -Homocystine L -Argininosuccinic

anhydrides

acid

Amino Acid Profiling for the Diagnosis of Inborn Errors of Metabolism

31

Table 2.1 (continued) Name

Abbreviation MM

Manufacturer Biological interest

Sample

S-Adenosyl-L homocysteine

Ad-Hcy

384

Sigma

Homocystinuria, hyperme- P,U thioninemia

S-Adenosyl-L methionine

Ad-Met

398

Sigma

Homocystinuria, hyperme- P,U thioninemia

Glutathione (oxidized form)

GSSG

612

Sigma

γGT deficiency, GSH syn- P,U thase deficiency

∗ : for preparation, see text. ◦ : molecules for which 2 drops of 37% HCl are added per 10 mL of 5 mM stock solution

for dissolution. MM, molecular mass; MAD, multiple acyl-CoA dehydrogenase deficiency; ETF, electron transfer flavoprotein; ETF-QO, ETF-ubiquinone oxidoreductase; CNS, central nervous system; Cbl, cobalamine; HHH, hyperammonemia, hyperornithinemia, homocitrullinuria; P, plasma; U, urine; CSF, cerebrospinal fluid; L, leukocytes. Reproduced with permission from ref. (7).

2.5. Stable Isotope-Labeled AAs (Internal Standards, AAs∗ )

1. Stable isotope-labeled standards for MS/MS are listed in Table 2.3, along with their abbreviations (see Note 9). 2. Stock solutions of each molecule (25 mM for Gly∗ , 15 mM for Gln∗ , 10 mM for Ala∗ and Pro∗ ; 5 mM for others) in water (see Note 4). Store in aliquots at –20◦ C for a maximum of 1 year. 3. Working solutions: 17 AAs∗ at 200 μM (except Pro∗ and Ala∗ at 400 μM, Gln∗ at 600 μM, and Gly∗ at 1 mM). In a 50-mL flask, add 2 mL of each AA∗ stock solution. Complete to 50 mL with water, mix, and store in aliquots at – 20◦ C for a maximum of 1 year.

2.6. AA Quantification Standards

Two quantitative standards are used daily for quantification: one is prepared from a commercially available AA standard supplemented in order to obtain a standard with all the most commonly measured physiological AAs for the diagnosis of inborn errors of metabolism (100 and 50 μM AAs standards) and one is homemade and contains all the molecules to be quantified that are not present in the other standard (supplemental AAs standard). 1. “Amino acid standards, physiological acidics, neutrals and basics for calibrating amino acid analyzers” (Sigma-Aldrich) containing several AAs at 500 μM. 2. AA additional stock solution: Gln 6.25 mM, Gly 5 mM, Asn 1.25 mM. Store in single-use aliquots at –20◦ C for a maximum of 6 months.

β-Ala

Abu

β-Alanine

L -α-Amino-n-

(Me)2 Gly

β-AIB

GABA

N,NDimethylglycine

β-Aminoisobutyric acid (1)

γ-Aminobutyric acid

butyric acid

Ala

Sar

L -Alanine

Gly

Glycine

Sarcosine

EA

Ethanolamine

Abbreviation

20 20

104 > 87

20

20

20

20

20

20

10

20

DP

104 > 86

104 > 86

104 > 58

104 > 58

90 > 72

90 > 44

90 > 44

76 > 30

62 > 44

Monitored transition

14

14

14

18

18

12

20

20

16

16

CE

x

9.5 9.5

Lys∗ Lys∗

3.7

Pro∗

1.3

8.0

Val∗

Gly∗

1.6

Ala∗ x

1.8

x

Ala∗ 2.4

8.8

Expected RT

Gly∗

26 AAs

Leu∗

AA∗

100

100

100

100

100

100

500

100

200

200

Conc. in Conc. in 100 Supplemental µM AA stan- AA standard dard (in µM) (in µM)

3

3

3

1

1–2

2–3

1

1

1

3

Monitoring period

Table 2.2 List of unlabeled AAs and derivative molecules of biological interest for the diagnosis of inherited disorders of amino acid metabolism. Corresponding specific transitions used for the identification of AAs in MS/MS positive ionization mode are given with their specific declustering potential (DP, in V) and collision energy (CE, in eV). Concentrations (in µM) of analytes in the 100 µM AA standard solution and in the supplemental standard solution are indicated, as well as AAs present in the 25 µM solution (26 AAs). Transitions are monitored either in period 1 (0–3.5 min of the chromatogram), in period 2 (3.5–9 min), and/or in period 3 (9–20 min). #: transition 132 > 86 is common to Hyp, Leu, Ile, aIle and dALA; transition 175 > 130 is common to Arg and FIGLU; RT: retention time; In italics: molecules for which quantitation could not be validated. Reproduced with permission from ref. (9)

32 Piraud et al.

Pip

pGlu

eCap

dALA

Hyp

Leu

aIle

Ile

Pyroglutamic acid

ε-Aminocaproic acid

δ-Aminolevulinic acid

4-Hydroxy-L-proline

L -Leucine

L -allo-isoleucine

L -Isoleucine

acid

L -Pipecolic

Hse

L -Homoserine

Thr

Val

L -Valine

Tau

Pro

L -Proline

L -Threonine

PC

1-Pyrroline-5carboxylic acid

Taurine

Ser

Abbreviation

L -Serine

Table 2.2 (continued)

20

16

132 > 86# (2) 25

26

36

30

26

25

30

30

16

16

24

20

20

16

32

32

32

16

20

20

14

CE

25

132 > 69

132 > 69

132 > 43

132 > 68

20

132 > 114 132 > 114

30

30

30

40

132 > 79

130 > 84

130 > 84

126 > 108

20

20

120 > 74 120 > 74

20

25

20

25

10

DP

120 > 44

118 > 72

116 > 70

114 > 68

106 > 60

Monitored transition

1.1

Glu∗

1.3

x x x x

Asp∗ Leu∗ Leu∗ Leu∗

9.8

9.2

10.4

9.6

Leu∗

11.2

5.2

Val∗

1.9 0.77

x

1.9

Ala∗

Ala∗

x

Val∗ 6.0

2.3

x

Pro∗

1.6 2.0

x

Ser∗

Expected RT

Val∗

26 AAs

AA∗

100

100

100

100

100

100

100

100

200

200

200

200

Conc. in Conc. in 100 Supplemental µM AA stan- AA standard dard (in µM) (in µM)

2–3

2–3

3

1

3

3

1

2

1

1

1

1

2

1

1

1

Monitoring period

Amino Acid Profiling for the Diagnosis of Inborn Errors of Metabolism 33

Asn

Asp

L -Asparagine

L -Aspartic

Gln

Lys

Glu

L -Glutamine

L -Lysine

L -Glutamic

His

Aad

L -Histidine

L -α-Aminoadipic

Hyl

Phe

His-3Me

His-1Me

Gly-Pro

Arg

FIGLU

5-Hydroxy-L-lysine

L -Phenylalanine

N1 -Methylhistidine

N3 -Methylhistidine

Glycyl-L-proline

L -Arginine

Formimino-Lglutamic acid

acid

Met

L -Methionine

acid

Hcy

L -Homocysteine

acid

Orn

Abbreviation

L -Ornithine

Table 2.2 (continued)

22

175 > 130# 20

32

16

20

20

20

18

24

18

14

32

20

15

30

30

30

5

20

20

10

24

36

24

24

16

18

18

22

CE

25

175 > 84

175 > 70

173 > 116

170 > 126

170 > 124

166 > 120

163 > 128

162 > 98

156 > 110

150 > 104

15

20

147 > 67

148 > 84

20

20

10

25

25

10

DP

147 > 84

147 > 84

136 > 90

134 > 74

133 > 74

133 > 70

Monitored transition

x x

Met∗ Lys∗

x

Lys∗

Glu∗

3.5

9.7 12.7

Arg∗

11.9

Leu∗ x

x

Lys∗

12.0

12.4 11.2

x

Phe∗

3.9

12.0

6.5

2.0

12.6

1.74

4.3

1.3

1.5

12.1

Expected RT

Lys∗

Pro∗

x

Glu∗

x

x

Asp∗

Lys∗

x

Asp∗

x

x

Orn∗

Gln∗

26 AAs

AA∗

100

100

100

100

100

100

100

100

100

500

100

100

100

200

200

Conc. in Conc. in 100 Supplemental µM AA stan- AA standard dard (in µM) (in µM)

2

2

3

2–3

3

3

3

3

1–2

3

2

1

3

3

1

2

1

1

3

Monitoring period

34 Piraud et al.

Tyr

Hci

Dopa

L -Tyrosine

L -Homocitrulline

L -3-(3,4-

asym(Me)2 Arg 203 > 70

Trp

Kyn

Hyt

Hcy(Ala)

Hyk

Car

Ans

NG ,NG Dimethylarginine

L -Tryptophan

L -Kynurenine

5-Hydroxy-Ltryptophan

L -Cystathionine

3-Hydroxy-Lkynurenine

L -Carnosine

L -Anserine

241 > 109

227 > 110

225 > 208

223 > 88

221 > 204

209 > 192

205 > 188

S-Cys

202 > 120

198 > 152

190 > 173

182 > 165

176 > 159

Monitored transition

S-Sulfo-l-cysteine

Dihydroxyphenyl)alanine

Cit

Abbreviation

L -Citrulline

Table 2.2 (continued)

30

30

20

15

20

20

10

20

15

10

20

25

10

DP

35

30

14

40

10

14

18

42

18

20

16

12

14

CE

12.4 11.9

Lys∗ Lys∗

3.5 10.0 14.0 13.8

Ala∗ Lys∗ Lys∗ Lys∗

10.4

0.6 12.8

Lys∗

7.0

Ala∗

4.5

8.5

2.6

Expected RT

Met∗

x

x

26 AAs

Val∗

Tyr∗

Ala∗ , Pro∗

AA∗

100

100

100

100

100

200

200

200

200

200

200

200

Conc. in Conc. in 100 Supplemental µM AA stan- AA standard dard (in µM) (in µM)

3

3

3

1–2

3

3

3

3

1

2

2

2–3

1–2

Monitoring period

Amino Acid Profiling for the Diagnosis of Inborn Errors of Metabolism 35

acid

GlcNAc-Asn

Ad-Hcy

Ad-Met

GSSG

S2AE

β-Aspartylglucosamine

S-Adenosyl-lhomocysteine

S-Adenosyl-Lmethionine

Glutathione (oxidized form)

S-2-Aminoethylcysteine

165 > 120

613 > 355

399 > 250

385 > 88

336 > 126

308 > 179

291 > 70

277 > 84

273 > 70

269 > 136

241 > 74 255 > 134

Monitored transition

8

20

15

30

15

25

30

25

20

10

25 100

DP

18

30

20

64

24

18

48

36

55

12

40 18

CE

1.6 12.7 15.3 8.7 11.1

Lys∗ Val∗ Lys∗

4.1

Lys∗

11.5

Val∗ Asp∗

5.3

Lys∗

12.4

Val∗

11.8

2.1 5.0

Expected RT

Lys∗

x

26 AAs

(Hcy)2 ∗

(Cys)2 ∗ (Cys)2 ∗

AA∗

100

100

200

200

200

200

Conc. in Conc. in 100 Supplemental µM AA stan- AA standard dard (in µM) (in µM)

(1) BAIB is not measurable when GABA is detectable. (2) Common transition to Leu, Ile, aIle, dALA, and Hyp. It cannot be used for Ile quantification when dALA is present.

GSH

ASA

Glutathione (reduced form)

L -Argininosuccinic

L -Saccharopine

Sac

ASA anh

L -Argininosuccinic

anhydrides

(Hcy)2

(Cys)2 Cys-Hcy

Abbreviation

L -Homocystine

homocysteine disulfide

L -Cysteine- L-

L -Cystine

Table 2.2 (continued)

2

3

3

1

2

3

2

3

3

1 2

Monitoring period

36 Piraud et al.

(5,5-D2 ), 2HCl

(3,3,3’,3’-D4 )

DL -Cystine

151 > 87 151 > 88 152 > 88 153 > 107 171 > 125 177 > 70 184 > 167 245 > 74 277 > 140

Gln∗ Met∗ Phe∗ Arg∗ Tyr∗ (Cys)2 ∗ (Hcy)2 ∗

10

25

25

20

30

10

20

20

15

25

10

25

20

10

20

12

40

12

32

20

14

24

24

24

18

22

36

16

16

20

14

20

16

CE

CIL

CIL

CIL

CIL

CIL

Euriso-top

CIL

Euriso-top

CIL

Euriso-top

CIL

Euriso-top

CIL

CIL

CIL

CIL

CIL

Manufacturer

5

5

5

5

5

5

15

5

5

5

5

5

5

5

10

5

10

25

Stock solution (mM)

Corresponding specific transitions used for their identification in MS/MS positive ionization mode are given with their specific declustering potential (DP, in V) and collision energy (CE, in eV). CIL, Cambridge Isotope Laboratories. Reproduced with permission from ref. (8).

(3,3,3’,3’,4,4,4’,4’-D8 )

(ring 3,5-D2 )

L -Tyrosine

DL -Homocystine

(guanido-15 N2 ), HCl

L -Arginine

(ring-D5 )

(methyl-D3 )

L -Phenylalanine

L -Methionine

(2,3,3,4,4-D5 )

(4,4,5,5-D4 ), 2HCl

L -Glutamine

DL -Lysine

acid (2,4,4-D3 )

acid (2,3,3-D3 )

DL -Glutamic

L -Aspartic

L -Ornithine

Lys∗

135 > 46

(5,5,5-D3 )

Glu∗

135 > 89

Leu∗

L -Leucine

DL -Valine-D8

(2,3,3,4,4,5,5-D7 )

137 > 75

126 > 80

Val∗

DL -Proline

135 > 72

123 > 77

Pro∗

Asp∗

30

109 > 63

Orn∗

25

94 > 48

Ser∗

(2,3,3- D3 )

(2,3,3,3-D4 )

DL -Alanine

DL -Serine

Ala∗

10

78 > 32

Gly∗

Glycine (2,2-D2 )

DP

Monitored transition

Abbreviation

Table 2.3 List of the 17 stable isotope-labeled standards (AA∗ ) Amino Acid Profiling for the Diagnosis of Inborn Errors of Metabolism 37

38

Piraud et al.

3. AA diluted solution: Gln 625 μM, Gly 500 μM, Asn 125 μM. Add 1 mL AA additional stock solution to 9 mL of water. Use this solution within the day of preparation. 4. 100 μM AA standard (see list in Table 2.2): all AAs at 100 μM (except Gly and Gln at 500 μM). Add 300 μL of the Sigma standard (“amino acid standards, physiological acidics, neutrals, and basics for calibrating amino acid analyzers”) to 1200 μL of AA diluted solution. Store in aliquots at –20◦ C for a maximum of 1 month. When thawed, store at 4◦ C for a maximum of 1 week. 5. 50 μM AA standard: Dilute the 100 μM AA solution. Prepare weekly and use for a maximum of 1 week with storage at 4◦ C. 6. Supplemental AA standard (see list in Table 2.2): several molecules at 200 μM, which are not present in the 100 μM standard. Combine 1 mL of each 5 mM stock solution and complete to 25 mL. Store in 1 mL aliquots at –20◦ C for a maximum of 2 years. When thawed, store at 4◦ C for a maximum of 1 week. 7. 0 μM standard is a water blank. 2.7. Materials for Preparation of Standards and Samples

1. 2-mL injection vials with screw-top caps (Interchim). 2. 9-mm silicone/PTFE slit screw caps (Interchim). 3. Microtest tube rack (Brand, Wertheim, Germany). 4. Microtubes PP 1.5 mL with attached PP safety cap (Sarstedt France, Marnay, France). 5. Deproteinization reagent: 20 μM S2AE in methanol (see Note 7). Dilute the S2AE stock solution 1:49 with methanol in a 50-mL glass flask with grinded cap. Store tightly closed at 4◦ C for a maximum of 1 month. This reagent can be aliquoted to small glass flasks with grinded caps. Open only when used and close immediately after use (see Note 8). 6. 1.25 mM TDFHA working solution (for preparation of the stock solution, see Note 1) 7. Plasma: Sample blood with heparin, then centrifuge 10 min at 5000×g at 4◦ C. Transfer plasma (200 μL minimum) in a tube. Deproteinize immediately or store at –80◦ C. 8. Urine: Determine creatinine concentration for each urine sample with any classical creatinine determination method for this purpose (e.g., Jaffé reaction) and freeze one aliquot at –20◦ C for AA determination (1 mL minimum). 9. Cerebrospinal fluid (CSF) and amniotic fluid (AF): Freeze the sample (300 μL minimum) at –20◦ C immediately.

Amino Acid Profiling for the Diagnosis of Inborn Errors of Metabolism

39

10. Leukocytes: Isolate the leukocytes pellet from 15–20 mL of blood as soon as possible (maximum 48 h) after sampling with citric acid–citrate–dextrose as anticlotting agent, according to usual procedures currently in use in the laboratory for leukocytes isolation. For preparation of the leukocyte mixture and analysis, see Note 10. 2.8. Quality Control Samples

1. External QAP (Quality Assurance Program): ERNDIM amino acid scheme (www.erndimqa.nl). 2. External QAP: ERNDIM cystine in white blood cell scheme (www.erndimqa.nl). 3. Internal QAP: SKML control material amino acids (www.erndimqa.nl).

3. Methods Samples for AA analysis are first deproteinized with methanol containing a known quantity of S2AE for control of the correct dilution/precipitation of the sample. Ion-pairing reversed-phase liquid chromatography is used for separation of AAs, and the use of volatile mobile phases (ACN, TDFHA, water) allows MS/MS detection, particularly with the TurboIonSpray source adapted to evaporation of water containing mobile phases. This chromatographic system must be used in stable conditions as described (with respect to rinsing periods) in order to obtain repeatability of the amino acid retention times and of the quantification results. As necessary for MS/MS quantitative analysis, 17 stable isotope-labeled AAs are used as IS. Quantification of the 17 corresponding AAs has been validated in these conditions. Validation of quantification has been obtained for many other AAs but not all. Several AAs for which quantitative validation has not been obtained are described anyway as qualitative markers useful for the diagnosis of some inherited disorders of AA metabolism. 3.1. Sample Preparation

1. Push 1.5-mL microtubes firmly onto the microtube rack. 2. AA standards (0, 50, and 100 μM, AA Suppl. STD 200 μM), plasma, CSF, AF, and quality control samples are used undiluted. For leukocytes, see Note 10. 3. Do not dilute urine samples if the creatinine concentration is below 4 mM. If the creatinine concentration exceeds 4 mM, then dilute urine with water as follows: a. 4–8 mM creatinine, dilute 1:1. b. 8–12 mM creatinine, dilute 1:2.

40

Piraud et al.

c. 12–16 mM creatinine, dilute 1:3. d. 16–20 mM creatinine, dilute 1:4. e. ≥20 mM creatinine, dilute 1:9. 4. Mix 150 μL of each sample (standards, plasma, CSF, AF, controls or adequately diluted urine) with 600 μL of deproteinization reagent, close the cap, and mix thoroughly for 2 min. 5. Store the samples at room temperature for 5 min. 6. Centrifuge the samples at 17,500×g for 5 min at +4◦ C, then transfer the supernatants to new 1.5-mL microtubes, and store at –80◦ C until used. 7. Prior to analysis, thaw and homogenize the deproteinized samples and standards. 8. Combine 200 μL of each supernatant with 40 μL of AA∗ working solution and 160 μL of 1.25 mM TDFHA in an injection vial and mix. 3.2. Liquid Chromatography

1. Assemble the column and the pre-column using the connectors and add a pre-filter. 2. Maintain the column at 26◦ C. 3. Before the first use, rinse the column with ACN at 200 μL/min for 2 h. 4. Separations are carried out at a flow rate of 200 μL/min. 5. Inject 5 μL of sample or standard. 6. Gradient elution is as follows: from 0 to 15% B in 1 min; maintain 15% B for 5 min; from 15 to 25% B in 3 min; maintain 25% B for 6 min; from 25 to 0% B in 1 min; maintain 0% B for 15 min to re-equilibrate the column before a new injection. 7. The sample throughput of the method is 2 samples/h. The LC effluent is directed to the mass spectrometer for the first 20 min of the separation, after which it can be directed to waste.

3.3. Tandem Mass Spectrometry

1. Set the ionization source to positive ionization mode. 2. Set the TurboIonSpray source to 450◦ C. 3. Set the curtain gas (CUR) to 20. 4. Set the collision gas to 2. 5. Set the ion source nebulization gas (GS1) to 25 (arbitrary units are given by the manufacturer). 6. Set the auxiliary gas (GS2) to 40. 7. Set Q1 and Q3 to unit resolution. 8. Set the ion spray voltage (IS) to 5000 V.

Amino Acid Profiling for the Diagnosis of Inborn Errors of Metabolism

41

9. Set the focusing potential (FP) to 200 V. 10. Set the entrance potential (EP) to –10 V. 11. Specific transitions optimized for each unlabeled (Table 2.2) or labeled (Table 2.3) compound are listed (see Note 11) with their specific declustering potential (DP, in V) and collision energy (CE, in eV) parameters (see Note 12). 12. For each sample, acquisition of data is performed during the first 20 min of the gradient analysis, according to the chosen transitions in positive ionization mode and specific parameters (Tables 2.2 and 2.3). During the re-equilibration time, the flow can be diverted to waste. 13. The predicted retention time for each molecule is indicated (see Table 2.2). 14. For qualitative analysis (25 μM solution (26 AAs)), the acquisition method contains all the involved transitions in a single period with a dwell time of 25 ms each. 15. For quantification using the Analyst 1.3.1 software, the acquisition time is divided into three periods (period 1 is from 0 to 3.5 min; period 2 is from 3.5 to 9 min; period 3 is from 9 to 20 min), during which only the relevant molecules are managed with their own transition (for the corresponding period, see Table 2.2). For each transition, the dwell time is 25 ms, except for Gly and Gly∗ (100 ms). 16. Some molecules are monitored over two periods in case there is a slight shift in RT (see Table 2.2). 3.4. LC-MS/MS Analysis

1. After rinsing the LC column, inject samples in the following order (see Note 13): a. One 0.5 mM TDFHA sample (not usable). b. One 25 μM solution (26 AAs) (for checking the quality of the analysis). c. 12–14 standards or samples (for quantification). d. Rinse the column on line 1 h with ACN at 200 μL/min. 2. The same sequence (from a to d) can be repeated two more times. Modifying or avoiding the rinsing periods may alter the chromatogram and cause shift in retention times. 3. The 0 and 100 μM AA standards are analyzed in each sequence. 4. The 50 μM and supplemental AA standards are analyzed at least two times every 24 h. 5. After each use (maximum 24 h, three sequences, 36–42 samples), rinse the column (out-line, reverse flow) with methanol/water (85:15) for 16 h at 100 μL/min.

42

Piraud et al.

Fig. 2.1. Overall chromatogram obtained with the 26 AA sample, with an MS/MS method integrating the transitions of each molecule during all the analysis time (20 min). Column: Uptisphere 50 mm × 2 mm ID 3 μm. Elution gradient: solvent A: 0.5 mM TDFHA in water and solvent B: acetonitrile; from 0 to 15% B in 1 min, 1–6 min 15% B maintained, 6–9 min linear gradient to 25% B, 9–15 min 25% B maintained, 15–16 min gradient back to 0% B, then 16–31 min 0% B to equilibrate the column before a new injection. Injection volume, 5 μL; concentration of molecules in mixtures, 25 μM each.

6. Check the quality of the chromatography based on the 25 μM solution (26 AAs) (see Fig. 2.1). 7. Identify each AA based on its transition and its retention time. 8. The overall aspect of the chromatogram and the elution order of the AA must remain the same. The observed retention time for each AA must correspond to the expected value (Table 2.2 and Fig. 2.2; see Note 14). 9. Visually inspect the separation of the critical pairs of AAs (see Fig. 2.2), i.e., mainly Glu and Gln (see Fig. 2.2c), Asp and Asn (see Fig. 2.2d), aIle, Ile, and Leu (see Fig. 2.2g; see Note 15). 10. Verify the sensitivity of the system daily by comparing peak intensities of AAs with those of the previous days.

Amino Acid Profiling for the Diagnosis of Inborn Errors of Metabolism

(a)

Fig. 2.2. (continued)

(b)

Fig. 2.2. (continued)

43

44

Piraud et al.

(c)

Fig. 2.2. (continued)

(d)

Fig. 2.2. (continued)

Amino Acid Profiling for the Diagnosis of Inborn Errors of Metabolism

(e)

Fig. 2.2. (continued)

(f)

Fig. 2.2. (continued)

45

46

Piraud et al.

(g)

Fig. 2.2. Extracted ion current (XIC) chromatograms of the amino acid pairs that have similar MS/MS characteristics. Column: Uptisphere 50 mm × 2 mm ID, 3 μm. Elution gradient: solvent A: 0.5 mM TDFHA in water and solvent B: acetonitrile; from 0 to 15% B in 1 min, 1–6 min 15% B maintained, 6–9 min linear gradient to 25% B, 9–15 min 25% B maintained, 15–16 min gradient back to 0% B, then 16–31 min 0% B to equilibrate the column before a new injection. Injection volume, 5 μL; concentration of molecules in mixtures, 25 μM each. (a) Ala, Sar (90 > 44); N-acetyl-L-ornithine (NAcOrn), Arg (175 > 70). (b) Abu, (Me)2 Gly (104 > 58). (c) Gln (147 > 84), Glu, Lys (148 > 84). (d) Asn (133 > 74), Asp (134 > 74). (e) Pip, pGlu (130 > 84). (f) Cys (122 > 76), Hcy (136 > 90), (Cys)2 (241 > 74), (Hcy)2 (269 > 136), Cys-Hcy (255 > 134). (g) Isobar molecules giving a parent ion at m/z 132: Leu, Ile, aIle, dALA, Hyp, and εCap. Reproduced with permission from ref. (8).

3.5. Quantification

1. Perform quantification of AA after acquisition of all the samples in the three sequences using Analyst 1.3.1 software. 2. Use the corresponding AAs∗ as internal standards (see Table 2.2, Note 16) (peak area ratio), and the 0 and 100 μM AA standards (or 200 μM supplemental AA standard) as external standards according to a linear calibration curve. 3. The 50 μM AA standard is used as a control sample. 4. The retention times of AAs and AAs∗ are noted, and the AA retention time/AA∗ retention time ratio (ratio of retention times) is calculated. 5. Confirm the quality of the peak integrations manually (see Note 17). 6. Quantitative validation has been obtained for most AAs measured by this method (see Note 17). However, quantita-

Amino Acid Profiling for the Diagnosis of Inborn Errors of Metabolism

47

tive validation could not be obtained for some of them (Table 2.2). Therefore, their analysis is only qualitative (see Note 18). 7. In several cases, qualitative identification of the disease marker is sufficient for a diagnosis (e.g., ASA, pGlu, FIGLU, S-Cys, Cys-Hcy). Using the stable isotope-labeled IS analogues of the compounds when available will improve the results. 8. Direct quantification of Cys-Hcy and ASA is not possible because of the lack of quantitative standards (see Notes 19 and 20). 9. Validation of the series is achieved if quantitative results match with the calculated consensus values of the internal control SKML amino acids.

4. Notes 1. Preparation of the 13.73 mM Tridecafluoroheptanoic acid (TDFHA) stock solution: Prepare by dissolving 5 g of TDFHA in 1 L water and store at room temperature (maximum 6 months). 2. Other ultrapure octadecyl-bonded silica gel columns of the same characteristics could be used with no or slight modifications to the gradient shape, provided that the separation of critical pairs of AAs is obtained (see 7, Section 3.4, Fig. 2.2 and Notes 14 and 15). 3. Any other HPLC pumps/autosampler can be used for solvent delivery and automated sample introduction, if adapted to the given flow/sample introduction volume. 4. When necessary, stock solutions are acidified with 2 drops of 37% HCl per 10 mL, in order to enhance the solubility of the compound (see Table 2.1). 5. A qualitative standard solution containing Cys-Hcy can be obtained by heating a solution containing 0.5 M Cys and 0.5 M Hcy at 50◦ C for 2 h. 6. A 0.5 mM ASA anhydrides solution can be obtained by heating a 0.5 M argininosuccinic acid solution at 100◦ C for 60 min. 7. S2AE is used as an internal standard for the control of the correct dilution/precipitation of samples.

48

Piraud et al.

8. Methanol is an efficient reagent for protein precipitation. Its use limits ion suppression that can occur with other precipitating reagents. This solution is used as a standard for deproteinization. Methanol is highly volatile. If evaporation occurs, the concentration may be altered. If the solution stays open longer than necessary, then it will be best to discard the flask and use a new one. 9. The use of stable isotope-labeled ISs of AAs to be quantified corrects the ion suppression phenomenon occurring when several molecules co-elute, assuming that the signal modification present in each sample is the same for the analyte and its corresponding IS. Due to their cost, the number of AAs∗ has been limited to 17, and they have been chosen according to the necessity of a precise quantification of the AAs for the diagnosis of metabolic disorders. For the remaining AAs, one of the 17 nonhomologous AAs∗ has been used as a surrogate IS after careful validation of their quantification in these conditions. In some cases, the quantification could not be validated, but the method can be used for qualitative analysis of the marker. If necessary, other AAs∗ can be easily added to the method (e.g., Cit∗ ), in order to improve quantification results and/or to extend the linear dynamic range of the measurement. 10. Preparation and analysis of leukocyte samples for cystine measurements are slightly modified when compared to those for other samples, in order to enhance the sensitivity of the method for this purpose. The method is modified as follows: a. A leukocyte mixture is prepared by mixing the leukocyte pellet with 250 μL water and sonicating according to usual procedures currently in use in the laboratory for this purpose. b. Protein concentration is determined. c. The 10, 5, and 2.5 μM AA standards for cystine measurement in leukocytes are prepared by diluting adequately the 100 μM AA standard (see Step 4 of Section 2.6) in water. d. Leukocytes extracts, standards, and controls (ERNDIM external QAP for cystine in white blood cells and SKML internal QAP) are immediately deproteinized in 1.5-mL microtubes, as described in Step 4 of Section 3.1, with the exception that 200 μL of leukocyte extracts, standard (0, 2.5, 5, and 10 μM), or controls are mixed with 600 μL of deproteinization reagent.

Amino Acid Profiling for the Diagnosis of Inborn Errors of Metabolism

49

e. After deproteinization, the supernatant is kept frozen at –80◦ C. f. Samples for injection are prepared by mixing 200 μL of each supernatant with 50 μL of AA∗ working solution in an injection vial. g. 5 μL is injected. h. The injection sequence is similar to that described in Step 1 of Section 3.4, with the exception that adequate standards and control samples are used. As leukocyte samples are usually very few, these samples can be included in another sample series, provided the leukocyte standards are also included in the same series. i. Quantification (see Section 3.5) involves only the corresponding transitions for Cys and Cys∗ and can be done with a specific quantification method. 11. Transitions are pairs of [M+H]+ precursor ion > fragment ion obtained before (precursor) and after (fragment) fragmentation of the AA and specific of it, e.g., 132 > 86 corresponds to the precursor ion (m/z = 132) and fragment ion (m/z = 86) produced by leucine. For each AA, the most sensitive/specific transition has been chosen. In some cases, the monitoring of two transitions for one AA is necessary to ensure the confident identification/quantification of the AA to be measured. Every molecule that can give the same fragmentation as another is a potential interference when these two molecules are present in the same mixture. This interference phenomenon is mainly due either to isobaric molecules, arising from in-source, collision-induced fragmentation, or to natural isotopic contributions (i.e., mainly 1.1% 13 C and 4.5% 34 S). Interference occurring between molecules analyzed in this method is reported in (7), but most of them have been separated from the AA of interest in the described chromatographic system. 12. Specific transitions have been determined with the API 2000 triple-quadrupole mass spectrometer. However, these transitions can be used in any triple-quadrupole mass spectrometer. DP and CE reported here are specific to the API 2000. The use of another triple-quadrupole mass spectrometer will necessitate the preliminary study of the specific parameters of the new apparatus for each AA. Various parameters (e.g., dwell time) for the data processing should be adapted to the software. Contact the manufacturer for any questions concerning this technology transfer. 13. This method uses ion-pairing, reversed-phase liquid chromatography, a technique which must be run in strictly standardized conditions. Repeatability of the retention time is obtained only when the system is equilibrated with a few

50

Piraud et al.

gradients. Frequent rinsing is necessary in order to limit the fouling of the column (by samples and ion-pairing agent) and thus the peak drift. In a series, the shape of the gradient must not be modified nor the duration of the L re-equilibration time. The first sample injected after rinsing the column cannot be used for analysis. The second sample (25 μM solution (26 AAs) is used to check the quality of the chromatography in the sequence. No more than three sequences must be analyzed in a series. When the column has to be maintained on-line for more than 24 h, it is possible to rinse the column on-line (10 h with 100 μL/min ACN) and use it again for a 24-h period (three sequences, 36–42 samples). 14. Retention times can be slightly different if the conditions of the chromatography are slightly different than those described (HPLC pumps, column, etc.). Calculating the ratios of the retention time of the analyte to that of the IS is helpful. For example, in instances of slight shifts of the retention times after several runs or on some samples, this ratio has been found to be very stable and allows correct identification of the peaks (see Note 15). 15. Gln/Glu (see Fig. 2.2c) and Asn/Asp (see Fig. 2.2d) present interference due to the 13 C isotope contribution and must present as separated peaks. Asp and Glu may be overestimated if they are not baseline separated from Asn and Gln, respectively (mainly Glu, due to the important concentration of Gln in biological fluids). aIle, Ile and Leu share a common 132 > 86 transition and must give three clear peaks with baseline separation (see Figs. 2.1 and 2.2 g). dALA coelutes with Ile. Thus, the common 132 > 86 transition cannot be used for the measurement of Ile when dALA is detectable. GABA and β-AIB cannot be separated by this method. β-AIB is measurable only if GABA is undetectable in the sample. Hse and Thr cannot be separated by this method. Thr is measurable only if Hse is undetectable in the sample. However, in our experience, no Hse has been detected among more than 10,000 analyzed biological samples. 16. When the homologous labeled AA is not available, another AA∗ , as close as possible to the target AA in terms of retention time and structure, is used (see Table 2.2). 17. For every series, the following parameters must be systematically checked: a. Each peak of each AA must be correctly integrated by the software. If not, manually correct the integration. b. The AA retention time/AA∗ retention time ratio must be 100% for each AA for which the corresponding AA∗ is

Amino Acid Profiling for the Diagnosis of Inborn Errors of Metabolism

51

present. For other molecules, the AA retention time/IS retention time ratio is the same in the standards and in the samples. c. The surface of the S2AE peak should be stable among all the standards and samples. If not, then repeat the preparation for injection and analysis. d. The identification of Leu, Ile, and aIle peaks must be correct: Leu/Leu∗ retention time ratio is 100%; Ile/Leu∗ retention time ratio is about 95%, and aIle/Leu∗ retention time ratio is about 90%. All of these can be quantified using the 132 > 86 common transition (135 > 89 for Leu∗ ) except for Ile when dALA is present. The 132 > 43 specific transition can be used for Leu quantification (135 > 46 for Leu∗ ). The 132 > 69 specific transition can be used for Ile and aIle quantification (135 > 46 for Leu∗ ). e. Hyp can be quantified either with the common 132 > 86 or with the 132 > 68 specific transition. f. The Asn/Asp∗ retention time ratio must be about 123%. g. The Gln/Gln∗ retention time ratio must be 100%. h. The Gln/Glu retention time ratio must be about 88%. Because of the prominent concentration of Gln in biological fluids (mainly plasma), Glu may be present as a double peak, with one peak at 100% of Glu∗ (corresponding to Glu) and one peak at 88% of Glu∗ (corresponding to Gln interference). Check that only the Glu part (as well as the Glu∗ part) is integrated for Glu quantification. i. The Sar/Ala∗ retention time ratio must be about 68%. j. The 76 > 30 transition used for Gly is the most sensitive of those characteristic for this AA, but it is not as sensitive relative to transitions for other AA. However, Gly concentrations are elevated in biological fluids, allowing for its accurate quantification. In CSF, the concentration of Gly is reduced, but elevated concentrations found in nonketotic hyperglycinemia can be detected. k. The His-3Me/Lys∗ retention time ratio is about 93%. An interfering peak is present at 88%, which may not be considered as His-3Me. l. The basic AA Arg, His, Lys, Orn elute late in the chromatographic system; a small interfering peak is present in the blank (0 μM AA standard) which must be taken into account in blanks for their measurement.

52

Piraud et al.

18. Quantitative validation, including the determination of the limit of quantification (LOQ), linear dynamic range, intra-day and inter-day precision, and accuracy (correlation and/or recovery experiments) have been obtained for all the AAs for which the corresponding AAs∗ are available (8). It has also been obtained for many AAs for which the corresponding AAs∗ are not used (9) (see Table 2.2). LOQs given in (7) are those determined with the API 2000. LOQs are highly instrument dependent, and a linearity limit has been found up to 3 mM for AAs for which the corresponding AAs∗ are used in the method. For other AAs for which AAs∗ were not used, the linearity limit was found up to 1 mM (9). For some AAs, the quantitative validation could not be obtained (9) (Table 2.2). 19. Cys-Hcy can be qualitatively evaluated by the presence/absence of the peak. A semi-quantitative estimation is possible using (Cys)2 as external standard and (Cys)2 ∗ as IS for the management of patients affected with homocystinuria. 20. The dual form of ASA (acid and anhydrides) is the cause of poor quantitative results for ASA and ASA anhydride by this method. Complete transformation of ASA to ASA anhydride is obtained by heating standards and samples at 100◦ C for 60 min. Total ASA anhydride can thus be measured and allow for the management of patients affected with argininosuccinic aciduria.

Acknowledgments The authors would like to thank Luc Anselmini for reading over this manuscript. The authors thank the journal Rapid Communication in Mass Spectrometry and Wiley Publishing for authorization to reproduce tables and figures from refs.(7–9). References 1. Moore, S., Spackman, D. H., Stein, W. H. (1958) Automatic recording apparatus for use in the chromatography of amino acids. Anal Chem 30, 1190–1206. 2. Vogeser, M., Seger, C. (2008) A decade of HPLC-MS/MS in the routine clinical laboratory – goals for further developments. Clin Biochem 9, 649–662.

3. Dooley, K. C. (2003) Tandem mass spectrometry in the clinical chemistry laboratory. Clin Biochem 36, 471–481. 4. Rashed, M. S. (2001) Clinical applications of tandem mass spectrometry: ten years of diagnosis and screening for inherited metabolic diseases. J Chromatogr B Biomed Sci Appl 758, 27–48.

Amino Acid Profiling for the Diagnosis of Inborn Errors of Metabolism 5. Gelb, M. H., Turecek, F., Scott, C. R., Chamoles, N. A. (2006) Direct multiplex assay of enzymes in dried blood spots by tandem mass spectrometry for the newborn screening of lysosomal storage disorders. J Inherit Metab Dis 29, 397–404. 6. Dietzen, D. J., Rinaldo, P., Whitley, R. J., Rhead, W. J., Hannon, W. H., Garg, U. C., et al. (2009) National academy of clinical biochemistry laboratory medicine practice guidelines: follow-up testing for metabolic disease identified by expanded newborn screening using tandem mass spectrometry executive summary. Clin Chem 55, 1615–1626. 7. Piraud, M., Vianey-Saban, C., Petritis, K., Elfakir, C., Steghens, J. P., Morla, A., Bouchu, D. (2003) ESI-MS/MS analysis of underivatised amino acids: a new tool for the diagnosis of inherited disorders of amino acid metabolism. Fragmentation study of 79

53

molecules of biological interest in positive and negative ionisation mode. Rapid Commun Mass Spectrom 17, 1297–1311. 8. Piraud, M., Vianey-Saban, C., Petritis, K., Elfakir, C., Steghens, J. P., Bouchu, D.. (2005) Ion-pairing reversed-phase liquid chromatography/electrospray ionization mass spectrometric analysis of 76 underivatized amino acids of biological interest: a new tool for the diagnosis of inherited disorders of amino acid metabolism. Rapid Commun Mass Spectrom 19, 1587–1602. 9. Piraud, M., Vianey-Saban, C., Bourdin, C., Acquaviva-Bourdain, C., Boyer, S., Elfakir, C., Bouchu, D. (2005) A new reversed-phase liquid chromatographic/tandem mass spectrometric method for analysis of underivatised amino acids: evaluation for the diagnosis and the management of inherited disorders of amino acid metabolism. Rapid Commun Mass Spectrom 19, 3287–3297.

Chapter 3 Acylcarnitines: Analysis in Plasma and Whole Blood Using Tandem Mass Spectrometry David S. Millington and Robert D. Stevens Abstract The acylcarnitine profile is a diagnostic test for inherited disorders of fatty acid and branched-chain amino acid catabolism. Patients with this type of metabolic disorder accumulate disease-specific acylcarnitines that correlate with the acyl coenzyme A compounds in the affected mitochondrial metabolic pathways. For example, propionylcarnitine accumulates in patients with both propionic and methylmalonic acidemias. The test identifies and quantifies the species of acylcarnitines in the whole blood or blood plasma of patients at risk for or suspected of having such a disorder. The acylcarnitines are analyzed using electrospray ionization–tandem mass spectrometry. The instrument is used in the precursor ion scan mode to record the molecular species giving rise to fragment ions at m/z 99, derived specifically from the methylated acylcarnitines within the specimen. Quantification is based on the principle of stable isotope dilution, whereby concentrations are derived from the response ratio of each acylcarnitine species to that of a deuterium-labeled acylcarnitine standard. Interpretation of the acylcarnitine profile requires recognition of abnormal concentrations of specific analytes or patterns of analytes and knowledge of their metabolic origin. Key words: Acylcarnitines, tandem mass spectrometry, inherited metabolic disease, metabolomics, fatty acid oxidation.

1. Introduction Acylcarnitines are the catabolic end products of fatty acids and several branched-chain amino acids that are utilized to generate cellular energy. They are derived from their corresponding acyl coenzyme A (acyl-CoA) analogs through exchange of acyl groups between coenzyme A and L-carnitine by the action of a series of carnitine acyl-transferases. These transferases have overlapping T.O. Metz (ed.), Metabolic Profiling, Methods in Molecular Biology 708, DOI 10.1007/978-1-61737-985-7_3, © Springer Science+Business Media, LLC 2011

55

56

Millington and Stevens

chain length specificity for the various acyl groups, ranging from 2 carbons (acetyl) to over 18 carbons (stearoyl) in length. Unlike their corresponding acyl-CoA analogs, acylcarnitines can cross mitochondrial and cell membranes, and are readily detectable in plasma. The plasma acylcarnitine profile is thus a fair reflection of the intramitochondrial acyl-CoA status at the time of analysis (1). Under normal metabolic circumstances, the acylcarnitine pattern in fasting plasma or serum is relatively stable and consists primarily of acetylcarnitine, plus relatively minor amounts of species derived from the branched-chain amino acids (C3–C5 acylcarnitines) and still lower concentrations of fatty acid intermediates (2) (Fig. 3.1a). Whole blood also exhibits significant concentrations of long-chain acylcarnitines in proportion to their corresponding dietary precursors (2) (C16, C18:1, C18:2, etc.; Fig. 3.1b). The presence of a block in the catabolic pathway of either the fatty acids or branched-chain amino acids results in the accumulation of one or more acyl-CoA intermediates. This in turn elevates the corresponding acylcarnitine concentrations. Analysis of plasma acylcarnitines can detect more than 25 metabolic disorders of fatty acid and branched-chain catabolism (3, 4), and it has become a frontline diagnostic test for these

100

C2

(a)

%

*

* *

*

*

* C14:1 C16

C18:1

0 m/z 200 220 240 260 280 300 320 340 360 380 400 420 440 460 480 500

100

(b)

C2

%

* *

*

*

*

C16

*

C18:1

C14

0 m/z 200 220 240 260 280 300 320 340 360 380 400 420 440 460 480 500

Fig. 3.1. (a) Mass spectrum of plasma acylcarnitines derivatized as their methyl esters, generated from a precursor ion scan of m/z 99 (sum of approximately 50 individual scans) after baseline subtraction, peak smoothing, and centroiding. Internal standard signals are indicated by (∗ ). Note the unit mass resolution at all points in the mass scale. (b) Mass spectrum of blood acylcarnitines derivatized as their methyl esters, generated from a precursor ion scan of m/z 99 (sum of approximately 50 individual scans).

Acylcarnitines: Analysis in Plasma and Whole Blood Using Tandem Mass Spectrometry

57

types of disorders. The value of this test has gained international recognition as the basis for the so-called expanded newborn screening panel (5). All babies born in the United States, most European, and several other countries are now routinely screened by analysis of acylcarnitines and amino acids in dried blood spots on filter paper. Those detected as at risk for an inherited metabolic disorder are referred for confirmatory metabolic testing that includes analysis of plasma acylcarnitines (6, 7). Thus the acylcarnitine profile is widely used as a follow-up test to an abnormal newborn screen, as well as for the evaluation and monitoring of patients suspected of having a metabolic disorder (4). Results from the test are usually correlated with urine organic acid analysis performed at the same time. In vitro tests based on the investigation of defects in pathways using metabolic probes with cultured cell lines, particularly fibroblasts, also employ acylcarnitine analysis by tandem mass spectrometry (8, 9). More recently, acylcarnitine analysis has become integrated into targeted metabolomics platforms that have aided the discovery of animal models of human disease and the discovery of new mechanisms of insulin resistance (10, 11). Here we describe a popular method for the analysis of acylcarnitines as their methyl esters using tandem mass spectrometry. Although the preferred specimen type is plasma (or serum), satisfactory (though less accurate) results can also be achieved from dried blood or plasma spots on filter paper. The method is straightforward, robust, highly sensitive to changes in acylcarnitine concentration outside the control range and has a rapid turnaround time. It relies on the molecular specificity of the tandem mass spectrometer and the reproducibility of the analytical procedure. Six isotope-labeled internal standards are used to enable pseudo-quantitative analysis of the acylcarnitines. It should be noted that free carnitine cannot be reliably quantified by this method because the derivatization step partially hydrolyzes acylcarnitines to free carnitine. A separate procedure is required for accurate analysis of free and total carnitine by tandem mass spectrometry (12).

2. Materials 2.1. Standards and Internal Standards

1. Acetyl-L -carnitine and palmitoyl-L-carnitine hydrochloride salts (Sigma Chemical Co, St. Louis, MO) and octanoyl- Lcarnitine (Larodan, Malmo, Sweden). These standards are stored in airtight screw-cap vials within a sealed container with self-indicating desiccant at a temperature of –70◦ C or below. Storage time is 5 years.

58

Millington and Stevens

2. D3-L -Acetylcarnitine, D3-L-propionylcarnitine, D3-L D3-L butyrylcarnitine, D9-L-isovalerylcarnitine, octanoylcarnitine, and D3-L-palmitoylcarnitine hydrochloride salts (Cambridge Isotope Laboratories, Andover, MA). These standards are stored in airtight screw-cap vials within a sealed container with selfindicating desiccant at a temperature of –70◦ C (see Note 1). Storage time is 5 years. 3. Stock solutions of each standard and internal standard (except for C16) are prepared in deionized water by rapidly weighing approximately 25–50 mg of each and making up to a concentration of 0.1 M by addition of the exact volume of deionized water (DI-H2 O) required (see Note 1). The palmitoylcarnitine (C16) standard and internal standard solutions are prepared in MeOH:DI-H2 O (75:25 v/v). Stock solutions are stored at –20◦ C for up to 1 year. 4. Quality control stock solutions for the plasma assay are prepared by first mixing the standard solutions in correct proportion and diluting the mixture in MeOH:DI-H2 O (1:1, v/v) to provide a stock solution (5 mL) with final concentrations of 3 mM acetylcarnitine, 1 mM octanoylcarnitine, and 2 mM of palmitoylcarnitine. This quality control stock solution is stored at –20◦ C for up to 1 year. 5. Working quality control solution for the plasma assay is made first by a 1:5 dilution of the stock solution with MeOH:DI-H2 O (1:1, v/v) to make the working mixture (0.1 mL). Then 0.06 mL of the working mixture is equilibrated with 12 mL bovine adult serum (BAS) in a-15 mL centrifuge tube by agitating for 30 min on a sample rocker at ambient temperature. 6. 100 μL aliquots of the working quality control solution, containing final concentrations of 3 μM acetylcarnitine, 1 μM octanoylcarnitine, and 2 μM palmitoylcarnitine, are stored at –20◦ C. QC samples are analyzed at least in duplicate within each batch of samples. A “patient QC” is also selected to be analyzed within each sample run. This is a sample selected at random from a previously analyzed batch of patient samples. 7. Quality controls for the blood spot assay are prepared from a fresh specimen of whole blood (10 mL, heparinized). This whole blood specimen can be spiked with octanoyl-L carnitine at a concentration of 1 μM for comparison with the plasma QC standard. Then 50 μL aliquots are pipetted onto the pre-defined circles on newborn screening specimen cards. The spots are allowed to air-dry for at least 6 h and are then stored in paper envelopes within a sealed

Acylcarnitines: Analysis in Plasma and Whole Blood Using Tandem Mass Spectrometry

59

plastic bag in the presence of self-indicating desiccant at –70◦ C. 8. Stock internal standard mixture is prepared by first mixing the internal standard solutions and diluting the mixture in MeOH:DI-H2 O (1:1, v/v) to provide final concentrations of 10 mM D3-acetylcarnitine, 4 mM D3palmitoylcarnitine, and 2 mM of each of the other internal standards. The stock internal standard solution is stored at –20◦ C for up to 1 year. 9. Working internal standard solution for the plasma assay is prepared from stock by serial dilution in MeOH:DIH2 O (1:1, v/v) to provide final concentrations of 0.1 mM D3-acetylcarnitine, 0.04 mM D3-palmitoylcarnitine, and 0.02 mM of each of the other internal standards (see Note 2). The working internal standard solution is stored for up to 1 month at 0–4◦ C. 10. Working internal standard solution for the dried blood spot assay is prepared from stock by serial dilution in MeOH:H2 O (1:1, v/v) to provide final concentrations of 12 μM D3-acetylcarnitine, 4.8 μM D3-palmitoylcarnitine, and 2.4 μM of each of the other internal standards. The working internal standard solution is stored for up to 1 month at 0–4◦ C. 11. A mass spectrometry tuning solution is prepared from the internal standard stock solution as follows. 30 μL of stock (containing 1 mM D3-acetyl-, 0.2 mM D3-octanoyl-, and 0.4 mM D3-palmitoyl-carnitines) is pipetted into a 5-mL glass vial. Evaporate to dryness under nitrogen. Add 100 μL of 3 M HCl in MeOH, cap and vortex mix, then incubate for 15 min at 50◦ C. Uncap the vial, evaporate to dryness, and reconstitute in 2 mL of final matrix (MeOH:H2 O, 85:15, v/v). Store at 0–4◦ C for up to 2 weeks. 2.2. Solvents and Reagents

1. Anhydrous methanolic hydrogen chloride (Supelco, Bellefonte, PA): 3 M. It is advisable to aliquot this reagent as rapidly as possible upon receipt into small screw-cap glass vials intended for single use (1–2 mL), wrap with parafilm, and store with desiccant in a sealed plastic bag at –20◦ C (see Note 3). 2. Anhydrous butanolic hydrogen chloride (Supelco): 3 M. It is advisable to aliquot this reagent as rapidly as possible upon receipt into small screw-cap glass vials intended for single use (1–2 mL), wrap with parafilm, and store with desiccant in a sealed plastic bag at –20◦ C (see Note 3). 3. HPLC-grade methanol.

60

Millington and Stevens

4. Deionized water (resistivity >17 M/cm). 5. Mobile phase: MeOH:DI-H2 O (15:85, v/v). This should be filtered prior to use through a 0.2-μm nylon membrane (Grace, Deerfield, IL). 2.3. Equipment and Other Materials

1. Tandem quadrupole mass spectrometer (Quattro Micro, Waters Associates, Milford, MA) equipped with ESI source, solvent delivery, and autosampler systems (Acquity, Waters). Equivalent systems from other vendors are acceptable. 2. Plate dryer, 96-well (Biotage, CA), preferably equipped R -coated stainless steel head. with Teflon 3. Harvard pump model HA11 (Instech Labs, Inc., Plymouth Meeting, MA). 4. A set of high-quality variable pipettes (0.5–10; 5–50, 20–200; 100–1000 μL; e.g., from Finnpipette or Eppendorf) are required, plus appropriate disposable pipette tips. 5. Microtiter plates (96-well; Evergreen, Los Angeles, CA). 6. A good quality steel hole puncher, 3/16 in. diameter. 7. Aluminum foil (7.5-cm-wide roll; Fisher Scientific, Pittsburgh, PA) and adhesive film (microadhesive film; USA Scientific, Ocala, FL), used to cover 96-well plates. 8. Plastic microcentrifuge and centrifuge tubes of 1.5, 2, and 15 mL capacity. 9. Laboratory benchtop centrifuge (e.g., Fisher model 235C from Fisher Scientific, Pittsburgh, PA or equivalent). 10. Vortex mixer. 11. Orbital shaker. 12. Incubator/oven.

3. Methods The general objective of this method is to generate a report showing the concentrations of up to 30 acylcarnitine species relative to the upper limits, and for a few analytes the lower limits, of a control range for each of them. It is not a rigorous quantitative analysis. The general principles are as follows. Isotope-labeled internal standards are added to the specimen as early as possible during the sample preparation procedure to establish a fixed ratio of their concentrations to those of the target analytes. When the sample is a dried blood spot (DBS), the internal standards are added to an extract from the DBS. Thus, the accuracy of the

Acylcarnitines: Analysis in Plasma and Whole Blood Using Tandem Mass Spectrometry

61

method is inherently lower for DBSs than for plasma, which is compounded by the fact that the volume of blood in a DBS is estimated from the diameter of the punch and by the variable hematocrit. Similarly, dried plasma spots on filter paper, although facilitating specimen collection and transfer from remote or challenging locations, do not provide the most accurate results. For these reasons, liquid plasma (or serum) is regarded as the best and preferred specimen for this test. DBSs collected by the heelstick method are widely used to screen newborns using a method similar to that described here (6). The errors inherent in this approach are minimized in the diagnostic laboratory by using a larger sample quantity and a more rigorous analytical technique. The amounts of the added internal standards are designed to provide high sensitivity at the upper limits of the normal analyte concentration ranges, thus enhancing the discrimination between normal individuals and those at risk for a metabolic disorder. After precipitation of proteins, the acylcarnitines are converted to methyl esters to enhance sensitivity, which is especially important for dicarboxylic species (see Note 4). Their analysis is performed by flow injection of the derivatized sample directly into the electrospray ion source of a tandem quadrupole mass spectrometer operated at unit mass resolution, using a precursor ion scan of m/z 99. The precursor ions are recorded as a single accumulated spectrum from m/z 200 to 500 (Fig. 3.1a). After baseline subtraction, peak smoothing, and centroiding, the signal ratios for the analytes and their respective internal standards (see Table 3.1) are generated, converted to an approximate concentration by multiplying by the final concentration of internal standard added to the aliquot of sample, and inserted into a final report (see examples in Tables 3.2 and 3.3). In some cases, in order to distinguish between isomers (Section 3.4), samples may be reanalyzed as their butyl esters, in which case the precursor ion scan of m/z 85 is used and the precursor ions are recorded from m/z 250 to 550 (6).

Table 3.1 Pseudo-molecular ion masses of the target analytes and their respective internal standards (in bold characters) as methyl esters and butyl esters Species

Methyl ester

Butyl ester

C2

218

260

C2–IS

221

263

C3

232

274

C3–IS

235

277

C4

246

288

62

Millington and Stevens

Table 3.1 (continued) Species

Methyl ester

Butyl ester

C4–IS

249

291

C5:1

258

300

C5

260

302

C4–OH

262

304

C5–IS

269

311

C6

274

316

C5–OH

276

318

C3–DC

276

360

C4–DC

290

374

C8:1

300

342

C8

302

344

C5–DC

304

388

C8–IS

305

347

C6–DC

318

402

C10:2

326

368

C10:1

328

370

C10

330

372

C8–DC

346

430

C12:1

356

398

C12

358

400

C14:2

382

424

C14:1

384

426

C14

386

428

C14:1–OH

400

442

C14–OH

402

444

C16

414

456

C16–IS

417

459

C16–OH

430

472

C18:2

438

480

C18:1

440

482

C18

442

484

C18:2–OH

454

496

C18:1–OH

456

498

C16–DC

458

542

C18:1–DC

484

568

Acylcarnitines: Analysis in Plasma and Whole Blood Using Tandem Mass Spectrometry

63

Table 3.2 Example of a final report from the acylcarnitine profile analysis of a plasma specimen, showing comparison of patient results (bold characters) with control values and interpretation Nanoliter range (nmol/mL)

Result (nmol/mL)

Status

21.48

NL

2.41

NL

<0.41

0.28

NL

(Tiglyl/Me-crotonyl)

<0.03

(BQL)

NL

(Isovaleryl/2Me-butyryl)

<0.35

0.22

NL

∗ <0.24

0.11

NL

<0.03

0.12

Elevated

Species

(Acyl group)

C2

(Acetyl)

C3

(Propionyl)

<2.57

C4

(Butyryl/isobutyryl)

C5:1 C5 OH–C4

(3-OH-butyryl)

C6

(Hexanoyl)

2.0–28.3

OH–C5

(3-OH-isovaleryl/malonyl)

∗ <0.19

0.13

NL

BZL

(Benzoyl)

∗ <0.01

(BQL)

NL

(Methylmalonyl/succinyl)

∗ <0.62

0.29

NL

C4–DC C8:1

(Octenoyl)

<0.24

0.23

NL

C8

(Octanoyl)

<0.11

0.10

NL

C5–DC

(Glutaryl)

∗ <0.02

(BQL)

NL

C6–DC

(Adipoyl/Me-glutaryl)

∗ <0.05

0.10

Elevated

C10:2

(Decadienoyl)

∗∗ <0.02

0.03

Elevated

C10:1

(Cis-4-decenoyl)

<0.14

0.08

NL

C10

(Decanoyl)

<0.20

0.13

NL

∗ <0.05

0.09

Elevated

C8–DC

(Suberyl)

C12:1

(Dodecenoyl)

<0.08

0.03

NL

C12

(Dodecanoyl)

<0.14

0.10

NL

C14:2

(Tetradecadienoyl)

<0.07

0.06

NL

C14:1

(Tetradecenoyl)

<0.14

0.05

NL

C14

(Tetradecanoyl)

<0.30

0.25

NL

0.04

Elevated

(BQL)

OH–C14:1

(3-OH-C14:1)

∗ <0.03

OH–C14

(3-OH-C14)

∗ <0.02

C16

(Palmitoyl)

OH–C16

(3-OH-palmitoyl)

C18:2

(Linoleoyl)

0.3–2.4 ∗ <0.02

<1.33 0.4–2.7

NL

1.83

NL

(BQL)

NL

1.48

Elevated

C18:1

(Oleoyl)

1.30

NL

OH–C18:2

(3-OH-linoleoyl)

∗ <0.03

(BQL)

NL

OH–C18:1

(3-OH-oleoyl)

∗ <0.02

(BQL)

NL

C16–DC

(C16-dicarboxylic)

∗ <0.01

(BQL)

NL

C18:1–DC

(C18:1-dicarboxylic)

∗ <0.01

(BQL)

NL

Comments: several mildly elevated values – pattern is nonspecific. BQL, Below quantitation limit; NL, normal. ∗ These values are ratios to internal standard; not nmol/mL. ∗∗ New analyte added to report.

64

Millington and Stevens

Table 3.3 Example of a final report from the acylcarnitine profile analysis of a blood spot specimen, showing comparison of patient results (bold characters) with control values and interpretation Nanoliter range (nmol/mL)

Result (nmol/mL)

Status

1.07

Low

0.06

NL

<0.47

0.09

NL

(Tiglyl/Me-crotonyl)

<0.07

(BQL)

NL

(Isovaleryl/2Me-butyryl)

<0.44

0.05

NL

∗ <0.18

(BQL)

NL

<0.13

(BQL)

NL

(3-OH-isovaleryl/malonyl)

∗ <0.04

(BQL)

NL

(BQL)

NL

(BQL)

NL

<0.55

0.06

NL

Species

(Acyl group)

C2

(Acetyl)

C3

(Propionyl)

<0.90

C4

(Butyryl/isobutyryl)

C5:1 C5 OH–C4

(3-OH-butyryl)

C6

(Hexanoyl)

OH–C5

2.1–15.7

BZL

(Benzoyl)

∗ <0.03

C4–DC

(Methylmalonyl/succinyl)

∗ <0.07

C8:1

(Octenoyl)

C8

(Octanoyl)

<0.25

(BQL)

NL

C5–DC

(Glutaryl)

∗ <0.06

(BQL)

NL

C6–DC

(Adipoyl/Me-glutaryl)

∗ <0.15

C10:2

(Decadienoyl)

C10:1 C10 C8–DC

0.03

NL

∗∗ <0.07

(BQL)

NL

(Cis-4-decenoyl)

<0.37

(BQL)

NL

(Decanoyl)

<0.50

(BQL)

NL

0.03

NL

(Suberyl)

∗ <0.15

C12:1

(Dodecenoyl)

<0.17

(BQL)

NL

C12

(Dodecanoyl)

<0.21

0.04

NL

C14:2

(Tetradecadienoyl)

<0.14

0.20

Elevated

C14:1

(Tetradecenoyl)

<0.23

0.66

Elevated

C14

(Tetradecanoyl)

<0.09

0.22

Elevated

0.03

NL

(BQL)

NL

0.21

NL

OH–C14:1

(3-OH-C14:1)

∗ <0.03

OH–C14

(3-OH-C14)

∗ <0.03

C16

(Palmitoyl)

OH–C16

(3-OH-palmitoyl)

C18:2

(Linoleoyl)

C18:1

(Oleoyl)

OH–C18:2

(3-OH-linoleoyl)

<0.23 ∗ <0.03

(BQL)

NL

<0.21

0.06

NL

<0.33

0.12

NL

(BQL)

NL

∗ <0.03

Acylcarnitines: Analysis in Plasma and Whole Blood Using Tandem Mass Spectrometry

65

Table 3.3 (continued) Result (nmol/mL)

Status

∗ <0.03

(BQL)

NL

(C16-dicarboxylic)

∗ <0.03

(BQL)

NL

(C18:1-dicarboxylic)

∗ <0.03

(BQL)

NL

Species

(Acyl group)

OH–C18:1

(3-OH-oleoyl)

C16–DC C18:1–DC

Nanoliter range (nmol/mL)

Comments: modest elevations of C14:1, C14:2, C14 species. Elevated C14:1/C12:1 ratio Low acetyl signal Pattern is consistent with VLCADD and possible carnitine insufficiency BQL, Below quantitation limit; NL, normal ∗ These values are ratios to internal standard; not nmol/mL. ∗∗ New analyte added to report.

3.1. Sample Preparation for Plasma/Serum

1. Allow all specimens, reagents, quality controls, and the working internal standard mixture to equilibrate to ambient laboratory temperature for at least 30 min. This is especially true for the methanolic HCl, for which it is imperative that exposure to moisture is limited. 2. Vortex each specimen tube, quality control, and internal standard vial for at least 10 s to ensure homogeneity. 3. Pipette 100 μL of each sample into a numbered plastic microcentrifuge tube (2.0 mL capacity with snap-top cap) containing 5 μL of the plasma working internal standard solution, then gently vortex mix each tube for 5 s. Treat the pre-aliquoted quality controls in the same manner. 4. Add methanol (800 μL) to each tube, cap, and vortex mix vigorously for 30 s, then centrifuge for 5 min in order to pellet the protein precipitate. 5. Remove and uncap each tube in turn and transfer 200 μL of supernatant liquid to a pre-assigned well of a microtiter plate. 6. Cap the sample tubes and store at 0–4◦ C for possible repeat or reflex tests. Steps 7–11 below must be performed under a fume hood. 7. Transfer the microtiter plate to a 96-well plate dryer/incubator and evaporate the solvent under nitrogen flow at 50◦ C for 20 min. Then remove the plate and allow it to cool for 2 min. 8. Open a vial of reagent (3 M HCl-MeOH) and transfer 100 μL to each well as rapidly as possible, then place an

66

Millington and Stevens

adhesive film over the plate, and carefully seal. Discard unused reagent. 9. Place the plate in an incubator oven pre-heated to 50◦ C for 15 min. Use a timer, and do not exceed the specified time. 10. Allow the plate to cool for 2 min, then carefully remove the cover, and without delay evaporate the samples to dryness on the 96-well plate dryer (20 min at 50◦ C). If for any reason the samples cannot be analyzed within 1 h, cover and seal the plate with adhesive film and store at 0–4◦ C for up to 48 h. 11. Add final matrix (200 μL) to each sample well, then cover, and seal the plate with aluminum foil. Transfer the plate to the autosampler of the tandem mass spectrometer system for analysis. Typically, 20 μL of the sample is injected directly via a loop injector. 3.2. Sample Preparation for Dried Blood Spots (DBSs)

1. For each DBS sample and quality control, use a hole puncher to punch out two discs of 3/16 in. diameter from a single blood spot (equivalent to 15.2 μL whole blood) into a numbered microcentrifuge tube (1.5 mL capacity with snap-top cap). Punch a circle from a clean area of the filter paper to clean the hole puncher between sampling each specimen. 2. To the inside wall of each sample tube, pipette 6 μL of the DBS internal standard working solution, then add methanol (400 μL) to each tube, and vortex mix for 5 s. 3. Agitate the sample tubes on an orbital shaker for a timed 30 min at ambient temperature. Do not exceed the extraction time. 4. Centrifuge the sample tubes for 5 min and then transfer 200 μL aliquots of the supernatants to pre-assigned wells of a microtiter plate. 5. Finally, follow Steps 7–11 in Section 3.1 with the exception that 60 μL of reagent is required for derivatization instead of 100 μL. Note that plasma and DBS extracts can be prepared on the same microtiter plate.

3.3. Sample Preparation for Dried Plasma Spots (DPSs)

1. For each DPS sample, use a hole puncher to punch out four circles of 3/16 in. diameter from the plasma spots (equivalent to 15.2 μL whole blood) into a numbered microcentrifuge tube (1.5 mL capacity with snaptop cap). Punch a circle from a clean area of the filter paper to clean the hole puncher between sampling each specimen. 2. Follow Steps 2–5 in Section 3.2.

Acylcarnitines: Analysis in Plasma and Whole Blood Using Tandem Mass Spectrometry

67

3.4. Sample Preparation with Modification to Form Butyl Esters

This procedure is used at the laboratory director’s discretion to distinguish between isomeric hydroxylated and dicarboxylic acylcarnitine species with two fewer carbon atoms when such species are elevated. It is used reflexively to retest all specimens (except from previously diagnosed patients) with elevated C5–OH/ C3–DC species (precursor m/z 276) and on occasion to distinguish between long-chain dicarboxylic species (C12–DC to C18:1–DC) and isomeric long-chain hydroxylated species (C14–OH to C20–OH). In such cases, the same procedures are followed in Sections 3.1, 3.2, and 3.3, depending on the specimen type, on stored residual extracts when available. The only modifications required in the sample preparation are to substitute butanolic HCl (3 M) for methanolic HCl as the reagent and to incubate at 65◦ C for 15 min instead of 50◦ C.

3.5. Analysis of Prepared Samples by Tandem Mass Spectrometry (MS/MS)

There is no specific recommendation as to the equipment manufacturer for the major item equipment required for the assay. The important requirements are that the mass analyzer is a triple-quadrupole equipped with ESI source and a minimum mass range of 0–1000 amu and there is an integrated solvent delivery system (at least a single HPLC pump) and an autosampler that can accommodate 96-well microtiter plates. The software package must enable a full precursor ion scan and be able to generate a summed spectrum. It must also generate ion ratios and either export them to a spreadsheet or incorporate software to perform the necessary calculations. There is no LC column; however, an inline filter is recommended to remove particulate matter. 1. Tune the mass spectrometer for optimal performance for acylcarnitine analysis using the tuning solution (Step 11 of Section 2.1). Ideally, this solution should be delivered continuously to the ion source via a syringe infusion pump. 2. Set up a run sequence in the order QC#1, blank, patient QC, patient specimens, QC#2 and analyze the first three samples prior to running any other sample. Generate reports from these analyses and review the acceptance criteria. These criteria must at a minimum address mass resolution and accuracy, absolute sensitivity, carryover, quantitative accuracy (for at least three analytes), and signal-to-noise ratio (see Note 5). The run is terminated if QC criteria are not met, and the problem must be corrected. 3. Generate and print the raw mass spectra (precursors of m/z 99) from each patient sample, acquiring data for about 1 min during the elution of the injected sample (Fig. 3.1a, b) and review each for overall quality. Generate the final reports (Tables 3.2 and 3.3) and add interpretative comments as appropriate (see Note 6).

68

Millington and Stevens

4. In order to establish initial analyte control ranges, it is necessary to analyze at least 50–100 specimens from age-matched patients considered at risk for an underlying metabolic disorder, but in whom no underlying disorder is clearly manifest in the acylcarnitine or the urine organic acid profile, if available. All the absolute analyte “levels” (i.e., the ratios of the analyte signals to their respective internal standards) are tabulated into a single spreadsheet and valid statistical methods applied to establish the upper limits for each analyte (also referred to as the “cutoffs”). Typically, the distribution of levels for each analyte is skewed and should be normalized by taking the logarithms of the ion ratios. Lognormal distributions, after removing outliers, allow for an accurate calculation of the upper limit (mean + 2SD). In plasma, it is helpful to establish a lower control limit (mean – 2SD) for acetylcarnitine (Table 3.2). Values below the low cutoff indicate either low flux through the mitochondrial oxidative pathways or carnitine insufficiency. In the DBS, acetyl-, C16and C18:1-acylcarnitines are provided with both upper and lower limits (mean ± 2SD) (Table 3.3). 5. In addition to the analytes, certain biochemically rational analyte ratios have been found helpful in assigning risk for a metabolic disorder. It is recommended that control ranges for C3/C2, C8/C10, and C14:1/C12:1 be compiled alongside the analyte ranges because their elevation, in combination with elevation of the primary analytes C3, C8, C14:1, is an added risk factor for metabolic disease (2, 4, Millington DS, unpublished). Mild isolated elevations of C3, C8, and C14:1 are rather commonly observed and are often unrelated to metabolic disease. 6. The control ranges generally stabilize after about 250 results are compiled and remain stable unless there is a significant change in either the sample population or the assay conditions (see Note 7). It is advisable to monitor these ranges by continuously compiling results and recalculating control ranges at 6-month intervals. 3.6. Interpretation

Providing interpretation and recommendations for the referring physician is an integral part of the analysis. This requires considerable experience and knowledge of the biochemistry of inherited metabolic disorders of fatty acid and amino acid catabolism, and continual review of the literature. In the authors’ laboratory, salient comments are initially made on the acylcarnitine report by a trained analyst, noting any significant findings. Then a final interpretation is provided by a professional (PhD or MD, PhD) who specializes and is certified in biochemical and/or medical genetics. Comments on the final report may

Acylcarnitines: Analysis in Plasma and Whole Blood Using Tandem Mass Spectrometry

69

include recommendations for further testing or treatment, taking into consideration other test results, if applicable, plus any clinical and dietary information provided with the specimen. The acylcarnitine profile as performed by this method is not a complete analysis because it does not resolve isomeric or isobaric species, such as isobutyryl and butyryl (C4), isovaleryl and 2-methylbutyryl (C5), tiglyl and 3-methylcrotonyl (C5:1), methylmalonyl and succinyl (C4DC), and 30H-isovaleryl and 2-methyl-3OH-isobutyryl (C5OH). Although methods exist for separation of these isomers using chromatography in combination with tandem mass spectrometry (13), the added complexity, time, and validation required are not rewarded by the added information they provide because the acylcarnitine analysis is almost always part of a more extensive biochemical workup that includes, at a minimum, urine organic acid analysis. All known potential diagnostic conflicts arising from isomerism in the acylcarnitine profile are resolved by concomitant urine organic acid analysis. Although a comprehensive guide to interpretation is outside the scope of this report, the following general guidelines, plus comments elsewhere (2, 4), may be helpful. 1. The cutoffs in the diagnostic laboratory (mean + 2SD) are more aggressive than those used in newborn screening (typically mean + 4SD). Therefore, it is typical to observe abnormalities in more than 10% of all profiles. Most of these are mild and are typically the result of either random statistical variation or dietary artifacts. The interpreter’s challenge is to assess the biochemical significance of these abnormalities and make recommendations, often without the benefit of clinical information. 2. Review the report and note any elevated signals from the primary analytes. The primary analytes are those having potential diagnostic significance and include C3, C4, C5, C5:1, C5–OH/C3–DC, C5–DC, C8, C10:2, C14:1, C16, and C16–OH. Isolated elevations of analytes other than the primaries are not diagnostically significant. 3. Isolated elevations of primary analytes greater than the upper limit of the control range may be of potential diagnostic significance. The possible diagnoses should be mentioned, as well as recommendations for further diagnostic evaluation. 4. In addition to the primary analytes, certain secondary analytes and biochemically rational analyte ratios, such as the ratio of C3/C2 and C8/C10, are noted, since they add an element of risk for an underlying metabolic disorder. These ratios are computed within the same spreadsheet and compared with control ratios but are not normally reported.

70

Millington and Stevens

5. Literature review on a continuous basis is necessary because new discoveries continue to add to the scope of metabolic disorders detectable by the acylcarnitine profile. Examples of recent discoveries include the detection of peroxisomal disorders (14).

4. Notes 1. Standards and internal standards are supplied as solid hydrochloride salts. They are inherently chemically unstable and spontaneously decompose to carnitine hydrochloride and the corresponding fatty acid, both in the dry state and in solution. The rate of decomposition is increased by temperature and the presence of moisture. Hence precautions to avoid ingress of moisture during weighings, including equilibration to room temperature, rapid transfer to the weighing balance, immediate recapping, and freezing of the stock, and dry storage conditions are essential. 2. When the stock internal standards or solutions thereof expire or degrade and have to be replaced, their absolute concentrations may not be the same as in the original mixture. New internal standard mixtures must be rigorously checked alongside the previous lot using a standard mixture of acylcarnitine standards and quality controls before they are placed in service. To avoid re-establishing the control ranges from scratch, it is acceptable to initially adjust the response factors based on these comparisons. 3. The derivatizing reagents are extremely hygroscopic and very corrosive. They must be purchased from a reputable supplier, kept under anhydrous conditions to avoid premature loss of potency, and used under a fume hood. The stock should be stored in a refrigerator designed for storage of flammables. 4. Alternative methods for acylcarnitine analysis include butylation, which in the authors’ experience is less sensitive than methylation and offers no significant advantage except to differentiate between isomers when necessary, as described. Analysis of free acylcarnitines without derivatization is possible but not recommended. The esterification of acylcarnitine carboxyl groups enhances sensitivity both by producing preformed cations (analogous to surfactants) and by avoiding multiple cationization (from sodium, potassium, protium, ammonium ions in solution) that dilutes the signal of the targeted protonated form. This is particularly important for

Acylcarnitines: Analysis in Plasma and Whole Blood Using Tandem Mass Spectrometry

71

the dicarboxylic species such as glutarylcarnitine, a marker for glutaryl-CoA dehydrogenase deficiency that may be only slightly elevated above background in affected patients (15). 5. Use of multiple reaction monitoring (MRM) for acylcarnitine analysis is unacceptable. The quality of each recorded precursor ion mass spectrum (after processing the raw data) must be individually reviewed and assessed. Although there are few known interfering substances in this assay, occasionally spurious signals occur in the profile, possibly from unusual intravenous fluids, medications, or dietary additives (2, 4), and their impact on the assay can be assessed only in the full scan mode. The isotope peaks at adjacent masses must be completely resolved, and the signal-to-noise ratio calculated for every analysis. Degradation of the signal-tonoise ratio can lead to spurious elevations of analytes and misinterpretation (see Note 7). 6. The ion ratios used to generate the pseudo-concentration values are those of the target analytes and their corresponding internal standards (usually the nearest m/z value). Thus several analytes may “share” the same internal standard (e.g., C6 to C12, C5–DC to C8–DC, and C5–OH all use D3–C8 as the internal standard; all analytes higher in m/z value than the C12 acylcarnitine species share the D3–C16 internal standard). 7. The method is semi-quantitative and relies heavily upon the reproducibility of the analytical procedure. In the authors’ experience, the addition of more internal standards than the six recommended here does not significantly add to the analytical precision. Because several analytes of diagnostic significance (including C3–DC, C5–DC, and C16–OH) are undetectable in normal individuals, and there are no commercially available standards or internal standards available to establish quantitative parameters, the control limits for such analytes are merely a reflection of baseline electronic “noise” of the instrument. Hence the need to avoid spurious signal elevations by rigorous adherence to quality control procedures.

References 1. Kler, R. S., Jackson, S., Bartlett, K., Bindoff, L. A., Eaton, S., Pourfarzam, M., Frerman, F. E., Goodman, S. I., Watmough, N. J., Turnbull, D. M. (1991) Quantitation of acyl-coA and acylcarnitine esters accumulated during abnormal mitochondrial fatty acid oxidation. J Biol Chem 266, 22932–22939.

2. Millington, D. S. (2003) Tandem mass spectrometry in clinical diagnosis, in (Blau, N., Duran, M., Blaskovics, M. E., Gibson, M., eds.), Physician’s Guide to the Laboratory Diagnosis of Metabolic Diseases, 2nd edn, pp. E57–75. Springer, Berlin.

72

Millington and Stevens

3. Millington, D. S., Chace, D. H. (1992) Carnitine and acylcarnitines in metabolic disease diagnosis and management, in (Desiderio, D. M., ed.), Mass Spectrometry: Clinical and Biomedical Applications, Vol. 1, pp. 199–219. Plenum Press, New York, NY. 4. Rinaldo, P., Cowan, T. M., Matern, D. (2008) Acylcarnitine profile analysis. Genet Med 10, 151–156. 5. Sweetman, L., Millington, D. S., Therrell, B. L., Hannon, W. H., Popovich, B., Watson, M. S., Mann, M. Y., Loyd-Puryear, M. A. (2006) Naming and counting disorders (conditions) included in newborn screening panels. Pediatrics 117, S308–S314. 6. Chace, D. H., Kalas, T. A., Naylor, E. W. (2003) Use of tandem mass spectrometry for multianalyte screening of dried blood spot specimens from newborns. Clin Chem 49, 1797–1817. 7. Frazier, D. M., Millington, D. S., McCandless, S. E., Koeberl, D. D., Weavil, S. D., Chaing, S. H., Muenzer, J. (2006) The tandem mass spectrometry newborn screening experience in North Carolina: 1997– 2005. J Inherit Metab Dis 29, 76–85. 8. Ventura, F. V., Costa, C. G., Struys, E. A., Ruiter, J., Allers, P., Ijlst, L., Tavares de Almeida, I., Duran, M., Jakobs, C., Wanders, R. J. (1999) Quantitative acylcarnitine profiling in fibroblasts using [U-13C] palmitic acid: an improved tool for the diagnosis of fatty acid oxidation defects. Clin Chim Acta 281, 1–17. 9. Roe, C. R., Roe, D. S. (2000) Detection of gene defects in branched-chain amino acid metabolism by tandem mass spectrometry of carnitine esters produced by cultured fibroblasts. Methods Enzymol 324, 424–431. 10. Kao, H.-J., Cheng, C.-F., Chen, Y.-H., Hung, S.-I., Huang, C.-C., Millington, D.,

11.

12.

13.

14.

15.

Kikuchi, T., Wu, J.-Y., Chen, Y.-T. (2006) ENU mutagenesis identifies mice with cardiac fibrosis and hepatic steatosis caused by a mutation in the mitochondrial trifunctional protein beta-subunit. Hum Mol Genet 15, 3569–3577. Newgard, C. B., An, J., Bain, J. R., Muehlbauer, M. J., Stevens, R. D., Lien, L. F., Haqq, A. M., Shah, S. H., Arlotto, M., Slentz, C. A., Rochon, J., Gallup, D., Ilkayeva, O., Wenner, B. R., Yancey, W. S. Jr., Eisenson, H., Musante, G., Surwit, R., Millington, D. S., Butler, M. D., Svetkey, L. P. (2009) A branched-chain amino acidrelated metabolic signature that differentiates obese and lean humans and contributes to insulin resistance. Cell Metab 9, 311–326. Stevens, R. D., Hillman, S., Worthy, L. S., Sanders, D., Millington, D. S. (2000) Assay for free and total carnitine in human plasma using tandem mass spectrometry. Clin Chem 46, 727–729. Minkler, P. E., Stoll, M. S., Ingalls, S. T., Yang, S., Kerner, J., Hoppel, C. L. (2008) Quantification of carnitine and acylcarnitines in biological matrices by HPLC electrospray ionization–mass spectrometry. Clin Chem 54, 1451–1462. Rizzo, C., Boenzi, S., Wanders, R. J., Duran, M., Caruso, U., Dionisi-Vici, C. (2003) Characteristic acylcarnitine profiles in inherited defects of peroxisome biogenesis: a novel tool for screening diagnosis using tandem mass spectrometry. Pediatr Res 53, 1013–1018. Gallaher, R. C., Cowan, T. M., Goodman, S. I., Enns, G. M. (2005) Glutaryl-CoA dehydrogenase deficiency and newborn screening: retrospective analysis of a low excretor provides further evidence that some cases may be missed. Mol Genet Metab 86, 417–420.

Chapter 4 Analysis of Organic Acids and Acylglycines for the Diagnosis of Related Inborn Errors of Metabolism by GC- and HPLC-MS Giancarlo la Marca and Cristiano Rizzo Abstract The analysis of organic acids in urine is commonly included in routine procedures for detecting many inborn errors of metabolism. Many analytical methods allow for both qualitative and quantitative determination of organic acids, mainly in urine but also in plasma, serum, whole blood, amniotic fluid, and cerebrospinal fluid. Liquid–liquid extraction and solid-phase extraction using anion exchange or silica columns are commonly employed approaches for sample treatment. Before analysis can be carried out using gas chromatography-mass spectrometry, organic acids must be converted into more thermally stable, volatile, and chemically inert forms, mainly trimethylsilyl ethers, esters, or methyl esters. Key words: Organic acids, inborn errors of metabolism, gas chromatography-mass spectrometry, organic acidurias.

1. Introduction Organic acids are small molecules continuously produced, metabolized, and excreted by healthy individuals, as well as by those in abnormal states. Their origins include the metabolic pathways of amino acids, carbohydrates, lipids, and steroids, as well as drug metabolism, bacterial decomposition of food in the intestine, artifacts due to storage, sample preparation, and analysis of the sample (see Note 1). A list of the most characteristic diagnostic organic acids and their associated disorders is reported in Table 4.1. To date, by the analytical methodologies available, more than 250 different compounds (mainly organic acids) and glycine conjugates can be identified in the urine of a normal T.O. Metz (ed.), Metabolic Profiling, Methods in Molecular Biology 708, DOI 10.1007/978-1-61737-985-7_4, © Springer Science+Business Media, LLC 2011

73

74

la Marca and Rizzo

Table 4.1 Characteristic organic acids profile in metabolic disorders Disorders

Metabolites

Amino acid disorders Maple syrup urine disease

2-Keto-isocaproic acid 2-Keto-isovaleric acid 2-Keto-3-methylvaleric acid 2-Hydroxy-isocaproic acid 2-Hydroxy-isovaleric acid 2-Hydroxy-3-methylvaleric acid

Tyrosinemia type I

4-Hydroxy-phenyllactic acid Succinylacetone 4-Hydroxy-phenylpyruvic acid

Tyrosinemia type II

4 Hydroxy-phenyllactic acid 4-Hydroxy-phenylpyruvic acid

Phenylketonuria

Phenyllactic acid Phenylpyruvic acid Mandelic acid di-TMS

Organic acidurias Isovaleric acidemia

3-Hydroxy-isovaleric acid Isovalerylglycine Isovalerylglutamic acid

3-Methylcrotonyl-CoA carboxylase deficiency

3-Hydroxy-isovaleric acid 3-Methylcrotonylglycine

Multiple carboxylase deficiency

3-Hydroxy-isovaleric acid Lactic acid Methylcitric acid 3-Hydroxy-propionic acid 3-Methylcrotonylglycine

3-Hydroxy-3-methylglutaric aciduria

3-Hydroxy-3-methylglutaric acid 3-Methylglutaconic acid 3-Methylglutaric acid 3-Hydroxy-isovaleric acid 3-Methylcrotonylglycine

3-Methylglutaconic aciduria

3-Methylglutaconic acid 3-Methylglutaric acid

3-Ketothiolase deficiency

2-Methyl-3-hydroxy-butyric acid 2-Methyl-acetoacetic acid 3-Hydroxy-butyric acid Tiglylglycine

2-Methylbutyryl-CoA dehydrogenase deficiency

2-Methylbutyrylglycine

2-Methyl-3-hydroxy-butyryl-CoA dehydrogenase deficiency

2-Methyl-3-hydroxy-butyric acid Tiglylglycine

Propionic acidemia

3-Hydroxy-propionic acid Methylcitric acid Propionylglycine Tiglylglycine

Analysis of Organic Acids and Acylglycines

Table 4.1 (continued) Disorders

Metabolites

Methylmalonic aciduria

Methylmalonic acid 3-Hydroxy-propionic acid Methylcitric acid

Glutaric aciduria type I

Glutaric acid 3-Hydroxy-glutaric acid Glutaconic acid

Beta-oxidation fatty acid defects Glutaric aciduria type II

Glutaric acid Ethylmalonic acid Adipic acid Suberic acid 2-Hydroxy-glutaric acid Isovalerylglycine Isobutyrylglycine 2-Methylbutyrylglycine

Short chain acyl-CoA dehydrogenase deficiency

Ethylmalonic acid Butyrylglycine Methylsuccinic acid Adipic acid Suberic acid Sebacic acid

Medium chain acyl-CoA dehydrogenase deficiency

5-Hydroxy-hexanoic acid 7-Hydroxy-octanoic acid Adipic acid Suberic acid Sebacic acid Octanedioic acid Decanedioic acid Hexanoylglycine Phenylpropionylglicine Suberylglycine

Very long-chain acyl-CoA dehydrogenase deficiency

Suberic acid Sebacic acid

Long-chain 3-hydroxy-acyl-CoA dehydrogenase deficiency

Adipic acid

Mitochondrial trifunctional protein deficiency

Suberic acid Sebacic acid 2-Hydroxy-adipic acid 3-Hydroxy-adipic acid 3-Hydroxy-octenedioic acid 3-Hydroxy-suberic acid 3-Hydroxy-decanedioic acid 3-Hydroxy-sebacic acid 3-Hydroxy-dodecenedioic acid 3-Hydroxy-dodecanedioic acid 3-Hydroxy-tetradecenedioic acid 3-Hydroxy-tetradecanedioic acid

Short-chain 3-hydroxy-acyl-CoA dehydrogenase deficiency

3,4-Dihydroxy-butyric acid 3-Hydroxy-glutaric acid

75

76

la Marca and Rizzo

Table 4.1 (continued) Disorders

Metabolites

Translocase deficiency

Adipic acid Suberic acid Sebacic acid Dodecanedioic aid Tetradecanedioic acid

Urea cycle defects HHH syndrome, OTCD, argininemia

Orotic acid

Citrullinemia, argininosuccinic aciduria

Uracil

Carbamoyl phosphate synthetase deficiency

3-Methylglutaconic acid

Carbohydrate metabolism Fructose 1,6-bisphosphatase deficiency

Lactic acid Glycerol Glycerol 3-phosphate

Glycogenosis type I

Lactic acid 3-Methylglutaconic acid 3-Metylglutaric acid

Galactosemia

4-Hydroxy-phenyl-lactic acid

Hereditary fructose intolerance

4-Hydroxy-phenyl-lactic acid

Pyruvate dehydrogenase (subunit E3) deficiency

Lactic acid 2-Ketoglutaric acid Succinic acid Malic acid 2-Keto-isocaproic acid 2-Keto-isovaleric acid 2-Keto-3-methylvaleric acid 2-Hydroxy-isocaproic acid 2-Hydroxy-isovaleric acid 2-Hydroxy-3-methylvaleric acid

Pyruvate carboxylase deficiency

Lactic acid Pyruvic acid Succinic acid Malic acid Fumaric acid

Miscellaneous Lactic acidosis

Lactic acid Pyruvic acid 2-Hydroxy-butyric acid 2-Hydroxy-isobutyric acid 4-Hydroxy-phenyl-lactic acid

Hyperoxaluria type I

Glycolic acid Oxalic acid

Hyperoxaluria type II

Oxalic acid Glyceric acid

D -glyceric

D -glyceric

aciduria

acid

Analysis of Organic Acids and Acylglycines

77

Table 4.1 (continued) Disorders

Metabolites

Glycerol kinase deficiency

Glycerol

3-Hydroxy-isobutyric aciduria

3-Hydroxy-isobutyric acid

Fumaric aciduria Malonic aciduria

Fumaric acid Malonic acid Methylmalonic acid Succinic acid

Mevalonic aciduria

Mevalonic acid

Alkaptonuria

Homogentisic acid

Canavan disease

N-acetylaspartic acid

Acylase I deficiency

N-acetyl-alanine N-acetylaspartic acid N-acetylglutamic acid N-acetyl-isoleucine N-acetyl-methionine N-acetyl-serine N-acetyl-valine

Pyroglutamic aciduria

Pyroglutamic acid

Xanthurenic aciduria

Xanthurenic acid

Peroxisomal biogenesis disorders

3,6-Epoxyoctanedioc acid 3,6-Epoxydecanedioc acid 3,6-Epoxydodecanedioc acid 3,6-Epoxytetradecanedioc acid 2-Hydroxy-sebacic acid 4-Hydroxy-phenyllactic acid Adipic acid Suberic acid Sebacic acid

Formiminoglutamic aciduria

Hydantoin-3-hydroxy-propionic acid

2-Hydroxy-glutaric aciduria

2-Hydroxy-glutaric acid

Aromatic L-amino acid decarboxylase deficiency

Vanyllactic acid

Dihydropyrimidine dehydrogenase deficiency

Thymine Uracil

Molybdenum cofactor deficiency

Xanthine

individual (1), although a great variability due to age and diet occurs (2). More than 70 inborn errors of metabolism are known to yield a characteristic urinary organic acids pattern, which is essential for diagnosis and follow-up (3). Organic acids can be detected in many biological matrices (e.g., blood, plasma, serum, amniotic fluid, and cerebrospinal fluid), but urine is the best fluid. This is mainly due to the fact that organic acids are concentrated in the urine by the kidneys, and this matrix contains very small quantities of proteins, reducing potential drawbacks during sample analysis.

78

la Marca and Rizzo

There are three main steps for the determination of urinary organic acids: (1) oximation; (2) extraction procedures: liquid– liquid extraction or ion-exchange chromatography or solid-phase extraction; (3) derivatization (usually silylation) followed by gas chromatography with mass detection in scan and/or single ion monitoring modes data acquisition. In 2002, Pitt and colleagues (4) reported a new tandem mass spectrometry (MS/MS)-based method to detect some organic acids and acylglycines in urine with a simple dilution of sample by mobile phase followed by direct injection into the mass spectrometer. In 2006, Rebollido and colleagues proposed a modified Pitt method as a neonatal screening tool for the detection of organic acids, acylglycines, acylcarnitines, and amino acids from urine spotted on filter paper (5). Herein, we describe in detail how to perform the analysis of organic acids and acylglycines in urine for the biochemical diagnosis of organic acidemias.

2. Materials

2.1. Reagent Preparation for Picrate Determination of Creatinine

All reagents (Siemens Healthcare Diagnostics Inc., Newark, NJ) are dissolved in water and ready to use and are of reagent grade unless stated otherwise. Their storage temperature is 2–8◦ C. These reagents are delivered and mixed automatically by the Dimension Chemistry System during the determination of creatinine. 1. Lithium picrate 25 mM. 2. Sodium hydroxide 100 mM. 3. Potassium ferricyanide 0.13 mM.

2.2. Solvents, Standards, and Reagents (See Note 1)

All solvents are high-purity chromatography grade. 1. Ethyl acetate. 2. Pyridine. 3 Hydroxyl chloride acid, 30%. 4. Methanol. 5. N-butanol. 6. Formic acid. 7. Acetic acid. 8. Diethyl ether. 9. Ba(OH)2 . 10. H2 SO4 .

Analysis of Organic Acids and Acylglycines

79

11. Hydroxylamine hydroxychloride (25 g/L), dissolved in pyridine. 12. Stock solutions (10 mM) of each organic acid (SigmaAldrich, St. Louis, MO) are prepared in water or methanol. a. Glycolic acid. b. Lactic acid. c. 3-Hydroxy-propionic acid. d. 3-Hydroxy-butyric acid. e. 4-Hydroxy-butyric acid. f. 2-Hydroxy-butyric acid. g. Glyceric acid. h. Fumaric acid. i. Maleic acid. j. 3-Hydroxy-isovaleric acid. k. 2-Hydroxy-isovaleric acid. l. Succinic acid. m. Methylmalonic acid. n. Glutaric acid. o. Ethylmalonic acid. p. Methylsuccinic acid. q. Phosphoethanolamine. r. 3-Methylglutaconic acid. s. 2-Ketoglutaric acid. t. 3-Methylglutaric acid. u. Adipic acid. v. 2-Hydroxy-glutaric acid. w. Mevalonic acid. x. 3-Hydroxy-glutaric acid. y. 4-Hydroxy-phenylacetic acid. z. Succinylacetone. aa. 3-Hydroxy-3-methylglutaric acid. bb. Homogentisic acid. cc. Suberic acid. dd. N-acetylaspartic acid. ee. Homovanillic acid. ff. 4-Hydroxy-phenylacetic acid. gg. Sulfocysteine. hh. Sialic acid.

80

la Marca and Rizzo

ii. Uracil. jj. Thymine. kk. Xanthine. ll. Orotic acid. mm. Uric acid. 13. Acylglycines were synthesized and purchased from Dr. H. Ten Brink (Academic Medical Hospital, the Netherlands). Stock solutions (10 mM) of acylglycines are prepared in water or methanol. a. Propionylglycine. b. Butyrylglycine. c. Isobutyrylglycine. d. 3-Methylcrotonylglycine. e. Tiglylglycine. f. Methylbutyrylglycine. g. Isovalerylglycine. h. Hexanoylglycine. i. 3-Phenylpropionylglycine. j. Suberylglycine. 14. Stock solutions (10 mM) of isotopically labeled and unlabeled internal standards are prepared in water or methanol. The storage temperature of the internal standards is –20◦ C. Under these conditions, the stability of the standard solutions is at least 6 months. a. Tricarballylic acid (TCA) (Sigma-Aldrich). b. 2-Oxocaproic acid (OA) (Sigma-Aldrich). c. 2-Phenylbutyric acid (PBA) (Sigma-Aldrich). d. Dimethylmalonic acid (DMMA) (Sigma-Aldrich). e. Tropic acid (TPA) (Sigma-Aldrich). f. Pentadecanoic acid (PDA) (Sigma-Aldrich). g. 2 H3 -methylmalonic acid (Cambridge Isotopes Laboratories, Andover, MA). h. 2 H3 -3-hydroxy-isovaleric acid (C.D.N. Pointe-Claire, Quebec, Canada). i. j.

Isotopes,

13 C -adipic acid (Cambridge Isotopes Laboratories). 6 2 H -3-hydroxy-3-methylglutaric acid (C.D.N. Iso3

topes). k. 2 H3 -acetylaspartic acid (C.D.N. Isotopes). l. 1,3 15 N2 -uric acid (Cambridge Isotopes Laboratories). m.

15 N -uracil 2

(C.D.N. Isotopes).

Analysis of Organic Acids and Acylglycines

81

n. 2 H4 -thymine (C.D.N. Isotopes). o. p.

15 N

2 -orotic acid (C.D.N. Isotopes). 2 H -propionylglycine (Dr. H. Ten 3

Brink, Academic

Medical Hospital, the Netherlands). q. 2 H3 -hexanoylglycine (Dr. H. Ten Brink, Academic Medical Hospital, the Netherlands). 15. Silylating reagents. a. N,O-bis-(trimethylsilyl)trifluoroacetamide (Supelco, Bellefonte, PA).

(BSTFA)

b. N-methyl-N-trifluoroacetamide (MSTFA) (Supelco). c. N,O-bis-(trimethylsilyl)acetamide (BSA) (Supelco). d. N-methyl-N-(tert-butyldimethylsilyl)trifluoroacetimide (MTBSTFA) (Supelco). 2.3. Glassware

1. Extraction tubes (see Note 1): 8 mL screw-capped pyrex glass tubes (14 × 100 mm) with Teflon-lined caps (Barcorword Scientific, Ltd., Stone Staff, UK). 2. GC-MS 2 mL glass vials (9 mm) and caps with siliconized rubber septa (Chromacol, Ltd. Welwyn Garden City, UK) (see Note 1).

2.4. Columns

2.5. Internal Standards for Semiquantitative and Quantitative Determination of Organic Acids and Acylglycines in Urine by GC-MS

1. Gas chromatography separation column: Agilent J&W ULTRA2 (methyl siloxane 5% phenyl), 25 m × 0.20 mm × 0.33 μm fused silica capillary column or an HP-5 ms (methyl siloxane 5% phenyl), 25 m × 0.25 mm × 0.25 μm fused silica capillary column. A stock calibration solution containing 100 mg/L of each internal standard in ethanol/water (1:1, v/v) is made and stored at 4◦ C or less (see Note 2).The final makeup volume of the working calibration solution is 50 μL; therefore, the concentration of the working calibration solution is 1 μg/mol creatinine when the added internal standard volume is 10 μL and 2 μg/mol creatinine when the added internal standard volume is 20 μL. 1. Dimethylmalonic acid (DMMA). Working calibration solution contains 10 μL of DMMA (100 mg/L) in ethanol/water (1:1, v/v). Store at 4◦ C or less. 2. Tropic acid (TA). Working calibration solution contains 20 μL of TA (100 mg/L) in aqueous solution. Store at 4◦ C or less. 3. Pentadecanoic acid (PDA). Working calibration solution contains 20 μL of PDA (100 mg/L) in ethanol. Store at 4◦ C or less.

82

la Marca and Rizzo

4. Tricarballylic acid (propane-1,2,3-tricarboxylic acid). Working calibration solution contains 10 μL of tricarballylic acid (100 mg/L) in aqueous solution. Store at 4◦ C or less. 5. 2-Phenylbutyric acid. Working calibration solution contains 10 μL of a 2-phenylbutyric acid (100 mg/L) in aqueous solution. Store at 4◦ C or less. 6. 2-Oxocaproic acid (2-oxohexanoic acid). Working calibration solution contains 10 μL of a 2-oxocaproic acid (100 mg/L) in aqueous solution. Store at 4◦ C or less. 7. 2 H3 -methylmalonic acid. Working calibration solution contains 10 μL of 2 H3 -methylmalonic acid (100 mg/L) in ethanol. Store at 4◦ C or less. 8. 2 H3 -3-hydroxy-isovaleric acid. Working calibration solution contains 10 μL of 2 H3 -3-hydroxy-isovaleric acid (100 mg/L) in ethanol. Store at 4◦ C or less. 9.

13 C -adipic acid. Working calibration 6 μL of 13 C6 -adipic acid (100 mg/L) 4◦ C or less.

solution contains 10 in ethanol. Store at

10. 2 H3 -3-hydroxy-3-methylglutaric acid. Working calibration solution contains 10 μL of 2 H3 -3-hydroxy-3methylglutaric acid (100 mg/L) in ethanol. Store at 4◦ C or less. 11. 2 H3 -acetylaspartic acid. Working calibration solution contains 10 μL of 2 H3 -acetylaspartic acid (100 mg/L) in ethanol. Store at 4◦ C or less. 12. 2 H3 -propionylglycine. Working calibration solution contains 20 μL of 2 H3 -propionylglycine (100 mg/L) in ethanol. Store at 4◦ C or less. 13. 2 H3 -hexanoylglycine. Working calibration solution contains 20 μL of 2 H3 -hexanoylglycine (100 mg/L) in ethanol. Store at 4◦ C or less. 2.6. Internal Standards and Calibrators for Semiquantitative and Quantitative Determination of Organic Acids and Acylglycines in Urine by HPLC-MS/MS

A stock calibration solution containing 100 mg/L of each internal standard in ethanol is made and stored at 4◦ C or less (see Note 2). The final makeup volume of the working calibration solution is 200 μL; therefore, the concentration of the working calibration solution is 5 μg/mol creatinine when the added internal standard volume is 10 μL and 10 μg/mol creatinine when the added internal standard volume is 20 μL. 1. 2 H3 -methylmalonic acid. Working calibration solution contains 10 μL of 2 H3 -methylmalonic acid (100 mg/L) in ethanol. Store at 4◦ C or less.

Analysis of Organic Acids and Acylglycines

83

2. 2 H3 -3-hydroxy-isovaleric acid. Working calibration solution contains 10 μL of 2 H3 -3-hydroxy-isovaleric acid (100 mg/L) in ethanol. Store at 4◦ C or less. 3.

13 C

6 -adipic acid. Working calibration solution contains 10 μL of 13 C6 -adipic acid (100 mg/L) in ethanol. Store at 4◦ C or less.

4. 2 H3 -3-hydroxy-3-methylglutaric acid. Working calibration solution contains 10 μL of 2 H3 -3-hydroxy-3methylglutaric acid (100 mg/L) in ethanol. Store at 4◦ C or less. 5. 2 H3 -acetylaspartic acid. Working calibration solution contains 10 μL of 2 H3 -acetylaspartic acid (100 mg/L) in ethanol. Store at 4◦ C or less. 6. 2 H3 -propionylglycine. Working calibration solution contains 20 μL of 2 H3 -propionylglycine (100 mg/L) in ethanol. Store at 4◦ C or less. 7. 2 H3 -hexanoylglycine. Working calibration solution contains 20 μL of 2 H3 -hexanoylglycine (100 mg/L) in ethanol. Store at 4◦ C or less. 8. 1,3-15 N2 -uric acid. Working calibration solution contains 10 μL of 1,3 15 N2 -uric acid (100 mg/L) in ethanol. Store at 4◦ C or less. 9.

2 -uracil. Working calibration solution contains 10 μL of 15 N2 -uracil (100 mg/L) in water. Store at 4◦ C or less. 15 N

10. 2 H4 thymine. Working calibration solution contains 10 μL of 2 H4 thymine (100 mg/L) in water. Store at 4◦ C or less. 11.

15 N

2 -orotic acid. Working calibration solution contains 10 μL of 15 N2 -orotic acid (100 mg/L) in water. Store at 4◦ C or less.

12. Calibration solutions are prepared in an aqueous matrix to simulate the composition of urine samples. The aqueous matrix is composed of the following: a. 30 mM urea (Sigma-Aldrich). b. 11.4 mM sodium chloride (Sigma-Aldrich). c. 7.8 mM potassium chloride (Sigma-Aldrich). d. 1.5 mM potassium dihydrogen orthophosphate (SigmaAldrich). e. 7.6 mg/L bovine serum albumin (Sigma-Aldrich). 13. For monitoring recoveries and for quality control purposes, a pooled urine sample is prepared and diluted to a creatinine concentration of 1.0 mM. Two enriched urine samples are prepared by diluting the same pooled urine sample to a creatinine concentration of 1.0 mM and adding abnormal amounts of metabolites.

84

la Marca and Rizzo

2.7. Instruments

1. Dimension RxL Max Integrated Chemistry System (Siemens Healthcare Diagnostics Inc., Newark, NJ). 2. Gas chromatograph 6890 N with a single-quadrupole mass spectrometer (5975) as detector (Agilent Technologies, Palo Alto, CA). 3. 1100 Series Cap Pump HPLC (Agilent Technologies, Waldbronn, Germany) coupled to an Agilent Micro ALS Autosampler. 4. API 4000 bench-top triple-quadrupole mass spectrometer equipped with the TurboIonSpray source (AB–Sciex, Toronto, Canada).

3. Methods 3.1. Sample Collection

1. Urine samples should be collected over 24 h when possible. However, considering that prolonged urine collections are quite difficult, especially for newborns or very young babies, a random sample, preferably the first morning voiding, is acceptable. Thus, it is possible to obtain results similar to those obtained in a 1-day collection (6), although the amount of many organic acids, excreted per milligram of creatinine, varies considerably according to age. Newborns and small babies have not yet developed complete renal tubular function, so they normally excrete a larger amount of aliphatic acids. 2. Any urine samples must be accompanied by a form containing information about clinical history, therapy, and diet details over the last 48 h. This is a critical point for a correct interpretation of data (see Note 1). 3. Urine should be collected in clean glass containers to avoid contamination by plastic compounds (e.g., phthalates) (7), and the containers should be preferably sterile. 4. Collect 24-h urine samples by freezing each single sampling. Other protein-free physiological fluids should be collected in the same way (see Note 1). Considering that most patients are acutely ill, this procedure is almost impractical.

3.2. Storage

1. Analyze urine samples immediately or freeze at –70◦ C until analysis. 2. If a –70◦ C freezer is not available, freeze the urine samples at –20◦ C. Indeed, if containers are not sterile, prolonged exposure at room temperature leads to bacterial growth, production of artifacts, and the formation of new organic acids (8) or loss of volatile acids (see Note 1). Urine samples should

Analysis of Organic Acids and Acylglycines

85

be transported in a deeply frozen state, preferably packed in solid carbon dioxide at –70◦ C. 3. Some authors suggested adding 2 mL of chloroform or merthiolate [o-(carboxyphenyl)thioethylmercury] sodium salt to urine sample as a preservative (see Note 1). Merthiolate is toxic by inhalation, ingestion, and contact with skin (EC hazard symbol T+), and it has cumulative effects. It is also very toxic to aquatic organisms and may cause longterm adverse effects in aquatic environments. Therefore, the addition of choloroform or merthiolate as a preservative is very seldom used. 3.3. Creatinine Assay

Creatinine (2-imino-1-methyl-4-imidazolidinone) concentration is the reference to which each organic acid concentration is typically normalized for quantitative purposes. However, this molecule is often declared unsuitable as a reference standard (8), due to its increasing excretion with age (9) and wide variation between individuals, especially small babies and newborns (10). Moreover many differences between males and females have been reported (9). Thus, the accurate determination of creatinine prior to urinary organic acid analysis is extremely important. The alkaline picrate-based method (a modified Jaffe method) has been reported to be less susceptible than conventional methods to interference from non-creatinine Jaffe-positive compounds (11). Sampling, reagent delivery, mixing, and processing of the analysis are automatically carried out by the Dimension Chemistry System. 1. Add about 100 μL of urine to a cuvette. Creatinine reacts with picrate (lithium picrate 25 mM) in presence of a strong base, like sodium hydroxide (NaOH 100 mM), forming a red chromophore (λ 510 nm). The rate of increasing absorption is directly proportional to the creatinine concentration in the sample. 2. Read the value of creatinine. It is measured by using a bichromatic (absorption wavelengths: λ 510 and λ 600 nm) rate technique (Dimension Chemistry System or similar) (see Note 3).

3.4. Oximation

1. Dilute urine sample corresponding to 0.5 or 1 mmol of creatinine with distilled water to 1,500 mL. 2. Adjust the pH to 14 with 5 M NaOH. 3. Add 0.5 mL of aqueous hydroxylamine solution. 4. Heat the urine and hydroxylamine mixture at 60◦ C for 30 min, then cool the solution to room temperature, and acidify to pH 1 with 6 M HCl. 5. After acidification, proceed with extraction as described below.

86

la Marca and Rizzo

3.5. Extraction Procedure

In our experience, samples capped well with silicon caps are stable about 1 week at 4◦ C, prior to the derivatization step (12). The solvent extraction procedure described below is widely used even if polar acids are not efficiently extracted. Moreover, many non-acid compounds are extracted, such as urea and alcohols, and these compounds can interfere with the chromatographic separation. Great attention also must be paid to the contaminants potentially present in the organic extraction solvent (see Note 1). a. Place a volume of urine corresponding to 0.5 or 1 mmol of creatinine in a 14 × 100-mm screw-cap culture tube. b. Dilute the sample with deionized water to 1.5 mL and adjust the pH to 1 by dropwise addition of 6 M HC1 or 2.5 M H2 SO4 . c. Add internal standards: 10 μL of DMMA (100 mg/L) in ethanol/water (1:1, v/v); 20 μL of TA (100 mg/L) in water; and 20 μL of PDA (100 mg/L) in ethanol. d. Extract the acidified sample three or four times with 2-mL aliquots of ethyl acetate (or diethyl ether), shaking the tube vigorously. The extraction efficiency has been measured by Tanaka and colleagues (13) who reported that hydroxy acids and low molecular weight dicarboxylic acids were recovered with much higher yields with ethyl acetate than with diethyl ether. e. After shaking, centrifuge the mixture at 1,620×g for 10 min. f. Combine the organic layers in a second tube (see Note 4) and then dry the combined extract over about 1–1.5 g of anhydrous Na2 SO4 for 45–60 min. g. Wash the Na2 SO4 twice with 1-mL portions of ethyl acetate. Optionally, the sample can be filtered with filter paper into a third tube prior to washing the Na2 SO4 . h. Dry the extract under a nitrogen stream at 60◦ C. When the volume is about 1–1.5 mL, transfer the solution to a 2-mL GC-MS glass vial and dry the remaining solvent.

3.6. Derivatization Procedures

After extraction procedures, in order to make the acids more volatile and stable for gas chromatography, a derivatization step is performed. Most laboratories use trimethylsilylation. 1. Add 50 μL of bis-trimethylsilyl-trifluoroacetamide (BSTFA) to the evaporated sample in a 2-mL GC-MS glass vial, vortex-mix, and incubate at 60◦ C for 30 min. It is also possible to use N-methyl-N-trifluoroacetamide (MSTFA), N,O-bis[trimethylsilyl]acetamide (BSA), or N-methyl-N[tert-butyldimethyl-silyl]trifluoroacetimide (MTBSTFA) as alternative derivatization reagents. All of these agents can be used at full strength or diluted with a suitable solvent such as pyridine.

Analysis of Organic Acids and Acylglycines

3.7. Gas Chromatography/Mass Spectrometry Analysis

87

1. The vialed samples should be kept at room temperature until analysis and analyzed no later than 24 h after preparation. 2. Generally, the GC analyses are performed on a gas chromatograph coupled to a single-quadrupole mass spectrometer as analyzer. 3. The capillary column is typically a nonpolar column such as an HP-5 ms or J&W ULTRA2 or similar. 4. The carrier gas is helium at 1 mL/min. 5. The column oven initial temperature is 70◦ C; the temperature gradient is 10◦ C/min to 280◦ C, followed by 280◦ C for 5 min. 6. The injection volume is 1 μL (see Note 5) with a split ratio of 10:1, and the front inlet temperature is 280◦ C. 7. The detector temperature is typically about 280–300◦ C. 8. The mass spectra are recorded both in selected ion monitoring (SIM) and in scan mode over a mass range of 40– 550 m/z. A list containing the most characteristic ions and molecular weights of the main organic acids is reported in Table 4.2. Table 4.3 contains the most diagnostic acylglycines molecular ions and characteristic fragments. 9. The retention times are obtained using chemical organic acid standards. Both standards and urine samples are analyzed under the same chromatographic conditions. 10. Mass calibration (see Note 6) of the instrument by using a mixture containing the target compounds at different concentrations is a critical step. The authors calibrated the mass spectrometer using a mixture containing about 100 different organic acids using a 5-point curve containing 10, 25, 50, 100, and 250 μM of each. 11. The calibration must be performed monthly or more frequently (see Note 6). 12. For quantitation purposes, a mixture containing 50 μM of standards is analyzed weekly, and if the quantitation report differs from the expected values by more than 25%, a new calibration curve is made. 13. The GC-MS system must have a regular maintenance schedule to avoid artifacts and generation of new compounds in the organic acids profile (see Note 7). 14. Many labs choose to perform a semiquantitative determination of urinary organic acids using non-labeled internal standards. These standards are non-physiological compounds that elute in proximity to the organic acids to be quantified. In our own experience, dimethylmalonic

88

la Marca and Rizzo

Table 4.2 The most characteristic ions and molecular weights of the main organic acids Characteristic urinary organic acid

MW

M-15

Ion 1

Ion 2

2-Keto-isovaleric acid (oxime) di-TMS

275

260

232

158

2-Keto-isocaproic acid (oxime) di-TMS

289

274

200

172

2-Keto-3-methylvaleric acid (oxime) di-TMS

289

274

200

172

2-Ketoglutaric acid (oxime) di-TMS

377

362

260

2-Hydroxy-3-methylvaleric acid

276

261

159

2-Hydroxy-adipic acid

378

363

2-Hydroxy-butyric acid di-TMS

248

233

131

2-Hydroxy-glutaric acid

364

349

247

2-Hydroxy-isobutyric acid di-TMS

248

233

95

2-Hydroxy-isocaproic acid di-TMS

276

261

159

103

2-Hydroxy-isovaleric acid di-TMS

262

247

219

145

203

2-Hydroxy-sebacic acid tri-TMS

419

391

317

2-Methyl-3-hydroxy-butyric acid di-TMS

262

247

117

2-Methyl-acetoacetic acid di-TMS

260

245

3-Hydroxy-3-methylglutaric acid

378

363

273

3-Hydroxy-adipic acid

378

363

247

3-Hydroxy-butyric acid

248

233

191

117

3-Hydroxy-isobutyric acid

248

233

177

103

3-Hydroxy-decanedioic acid

432

417

233

3-Hydroxy-dodecanedioic acid

462

447

233

3-Hydroxy-dodecanedioic acid

460

445

233

3-Hydroxy-glutaric acid

364

349

259

185

3-Hydroxy-isovaleric acid

262

247

205

131

3-Hydroxy-octanedioic acid

404

389

233

217

3-Hydroxy-propionic acid di-TMS

234

219

204

177

3-Hydroxy-sebacic acid tri-TMS

434

419

233

217

3-Hydroxy-suberic acid tri-TMS

406

391

233

217

3-Hydroxy-tetradecanedioic acid tri-TMS

490

475

233

217

3-Hydroxy-tetradecanedioic acid tri-TMS

488

473

233

217

247

3-Methylglutaconic acid di-TMS

288

273

198

183

3-Methylglutaric acid di-TMS

290

275

204

69

3,4-Dihydroxy-butyric acid tri-TMS

336

233

189

3,6-Epoxyoctanedioc acid di-TMS

332

317

201

174

3,6-Epoxydecanedioc acid di-TMS

360

345

201

174

3,6-Epoxydodecanedioc acid di-TMS

388

372

201

174

3,6-Epoxytetradecanedioc acid di-TMS

416

401

201

174

4-Hydroxy-butyric acid di-TMS

248

233

204

117

Analysis of Organic Acids and Acylglycines

89

Table 4.2 (continued) Characteristic urinary organic acid

MW

M-15

Ion 1

Ion 2

4-Hydroxy-phenyllactic acid tri-TMS

398

383

308

179

4-Hydroxy-phenylpyruvic acid tri-TMS

396

381

325

4-Hydroxy-phenylpyruvic acid (oxime) tri-TMS

412

396

277

190

4,5-Dihydroxy-hexanoic tri-TMS (erythro and threo)

364

349

247

129

4,5-Dihydroxy-hexanoic lactone (erythro and threo)

202

187

158

143

5-Hydroxy-hexanoic acid di-TMS

276

261

204

171

7-Hydroxy-octanoic acid di-TMS

304

289

260

Adipic acid di-TMS

290

275

172

111 119

Decanedioic acid di-TMS

344

329

164

Dodecanedioic aid

374

359

243

Ethylmalonic acid di-TMS

276

261

217

Fumaric acid di-TMS

260

245

147

Glutaconic acid di-TMS

274

259

217

Glutaric acid di-TMS

276

261

158

Glycerol tri-TMS

308

293

218

205

Glycerol 3-phosphate tetra-TMS

460

445

357

299

Glyceric acid tri-TMS

322

307

292

205

Hydantoin-3-hydroxy-propionic acid tri-TMS

388

373

257

230

Homogentisic acid tri-TMS

384

341

252

Isovaleryl-glutamic acid di-TMS

375

360

258

156

Lactic acid di-TMS

234

219

190

117

Malic acid tri-TMS

350

319

245

233

Malonic acid di-TMS

248

233

133

Mandelic acid di-TMS

296

281

253

179

Methylcitric acid tetra-TMS

494

479

389

361

Methylmalonic acid di-TMS

262

247

218

157

Methylsuccinic acid di-TMS

276

261

217

Mevalonic acid tri-TMS

364

349

247

233

Mevalonic lactone mono-TMS

202

187

145

115

Mevalonic lactone di-TMS

274

259

187

175

N-acetyl-alanine di-TMS

275

260

232

158

N-acetylaspartic acid tri-TMS

391

376

274

230

N-acetylaspartic acid di-TMS

319

304

202

158

N-acetylglutamic di-TMS

333

318

216

158

N-acetyl-isoleucine tri-TMS

245

230

128

86

N-acetyl-methionine di-TMS

335

320

274

261

N-acetyl-serine di-TMS

291

276

261

186

90

la Marca and Rizzo

Table 4.2 (continued) Characteristic urinary organic acid

MW

M-15

Ion 1

Ion 2

N-acetyl-valine di-TMS

303

288

260

186 137

Octanedioic acid di-TMS

316

301

136

Orotic acid tri-TMS

372

357

254

Oxalic acid di-TMS

234

219

190

Phenyllactic acid di-TMS

310

295

220

Phenylpyruvic acid (oxime)

323

308

189

Pyroglutamic acid di-TMS

273

258

156

Pyruvic acid (oxime) di-TMS

247

232

204

Sebacic acid di-TMS

346

331

345

Suberic acid di-TMS

318

303

169

Succinic acid di-TMS

262

247

172

Succinylacetone (oxime)

227

212

182

Tetradecanedioic acid di-TMS

402

387

271

Uracil di-TMS

256

241

99

Vanillactic acid tri-TMS

428

413

338

Xanthurenic acid tri-TMS

420

406

318

193

138

209

acid, tropic acid, pentadecanoic acid, tricarballylic acid, 2-oxocaproic acid, and 2-phenylbutyrric acid can be used as internal standards. Their use is advantageous only in terms of cost per analysis (see Note 8). 15. Quantitation is performed according to either Internal Standard Method or the Absolute Calibration Curve Method, generally using the peak height or peak area. 16. Internal Standard Method a. Prepare several standard solutions containing a constant amount of the specified labeled or unlabeled internal standard and known graded amounts of the compound to be quantified. b. With each of the chromatograms obtained by injecting a constant volume of each standard solution, calculate the ratio of the peak height or peak area of the compound to be quantified to that of the internal standard. c. Prepare a calibration curve by plotting these ratios on the ordinate and the ratios of each amount of the compound to be quantified to the amount of the internal standard, or simply the amount of the compound to be quantified, on the abscissa. The calibration curve is usually a straight line through the origin.

Analysis of Organic Acids and Acylglycines

91

Table 4.3 The most diagnostic acylglycines molecular ions and characteristic fragments Acylglycine mono-TMS derivatives

[M]+

[M-15]+

[M-117]+ [M-146]+ [M-59]+

[M-45]+ [M-44]+

2-Methylbutyrylglycine

231

216

114

85

172

186

187

3-Methylcrotonylglycine

229

214

112

83

170

184

185

Butyrylglycine

217

202

100

71

158

172

173

Hexanoylglycine

245

230

128

99

186

200

201

Isobutyrylglycine

217

202

100

71

158

172

173

Isovalerylglycine

231

216

114

85

172

186

187

Phenylpropionylglycine

270

255

86

57

144

158

159

Propionylglycine

203

188

86

57

144

158

159

Tiglylglycine

229

214

112

83

170

184

185

Suberylglycine

300

285

183

154

241

255

256

Aclglycine di-TMS derivatives

[M]+

[M-15]+

[M-117]+ [M-131]+ [M-218]+

2-Methylbutyrylglycine

303

288

186

172

85

3-Methylcrotonylglycine

301

286

184

170

83

Butyrylglycine

289

274

172

158

71

Hexanoylglycine

317

302

200

186

99

Isobutyrylglycine

289

274

172

158

71

Isovalerylglycine

303

288

186

172

85

Phenylpropionylglycine

343

328

226

212

125

Propionylglycine

275

260

158

144

57

Tiglylglycine

301

286

184

170

83

Suberylglycine

375

360

256

242

155

[M-15]+ =[M-CH3 ]+ [M-117]+ =[M-COOSi(CH3 )3 ]+ [M-146]+ =[RCO]+ R=alkyl radical [M-59]+ =[M-CH2 COOH]+ [M-45]+ =[M-COOH]+ [M-44]+ =[M-COO]+ [M-131]+ =[M-CH2 COOSi(CH3 )3 ]+ [M-218]+ =[RCO]+ R=alkyl radical

d. Prepare the urine sample containing the same amount of the internal standard as directed in the individual monograph. e. Record a chromatogram under the same conditions as for the preparation of the calibration curve. f. Calculate the ratio of the peak height or peak area of the compound to be quantified to that of the internal standard.

92

la Marca and Rizzo

g. Perform the quantitation using the calibration curve. 17. Absolute Calibration Curve Method a. Prepare standard solutions containing a graded amount of the compound to be quantified and inject a constant volume of each standard solution. b. With the chromatograms obtained, prepare a calibration curve by plotting the peak heights or peak areas of the compound to be quantified on the ordinate and the amounts of the compound to be quantified on the abscissa. The calibration curve is usually a straight line through the origin. c. Prepare the urine sample as directed in the individual monograph. d. Record a chromatogram under the same conditions used for the preparation of the calibration curve. e. Measure the peak height or peak area of the compound to be quantified. f. Perform the quantitation using the calibration curve. 3.8. Liquid Chromatography/Mass Spectrometry Analysis

1. Loop injections (20 μL) are made via a HPLC autosampler into a mobile phase of acetonitrile–water (1:1 by volume). 2. The mobile phase and sample are infused at a flow rate of 30 μL/min by an HPLC pumping system into the electrospray ion source of a triple-quadrupole mass spectrometer. 3. The ion source operated with a capillary voltage of 5.0 kV (positive-ion mode) or 4.5 kV (negative-ion mode) and a temperature of 250◦ C. Nitrogen is used as collision gas at a pressure of 10 mTorr. 4. Fragmentation transitions, declustering potentials, and collision energies should be optimized during manual loop injections of pure compounds. 5. Negative-ion multiple reaction monitoring (MRM) for 49 metabolites and 11 labeled internal standards is performed under the conditions shown in Table 4.4. 6. Data are acquired for 1.3 min after injection, and the total cycle time between injections is 2.1 min. 7. Data are processed with MRM signals averaged between 0.6 and 1.3 min and baseline subtraction of signals from 0.15 to 0.45 min. 8. Raw data are exported to an Excel spreadsheet file (Microsoft) for calculations.

Analysis of Organic Acids and Acylglycines

93

Table 4.4 Precursor (Q1), product ion (Q3) masses, and optimized voltages for the 60 compounds studied in negative-ion mode Metabolites

Q1/Q3 transition

DP (V)

CE (V) Internal standard

1 Glycolic acid

75 > 47

–30

–20

14

2 Lactic acid

89 > 43

–40

–20

14

3 3-Hydroxy-propionic acid

89 > 59

–30

–15

14

4 3-Hydroxy-butyric acid

103 > 59

–30

–15

14

5 4-Hydroxy-butyric acid

103 > 57

–15

–20

14

6 2-Hydroxy-butyric acid

103 > 57

–17

–25

14

7 Glyceric acid

105 > 75

–25

–15

14

8 Fumaric acid

115 > 71

–18

–12

14

9 Maleic acid

115 > 71

–19

–13

14

10 3-Hydroxy-isovaleric acid

117 > 59

–21

–14

15

11 2-Hydroxy-isovaleric acid

117 > 71

–31

–16

14

12 Succinic acid

117 > 73

–12

13

14 14

13 Methylmalonic acid

117 > 73

–10

–13

14 2 H3 -Methylmalonic acid 15 2 H3 -3-Hydroxy-isovaleric acid

120 > 76

–10

–13

123 > 59

–21

–15

16 Glutaric acid

131 > 87

–10

–13

28

17 Ethylmalonic acid

131 > 87

–10

–12

28

18 Methylsuccinic acid

131 > 87

–11

–13

28

19 Phosphoethanolamine

140 > 79

–20

–15

28

20 3-Methylglutaconic acid

143 > 99

–5

–9

28

21 2-Ketoglutaric acid

145 > 101

–20

–13

28

22 3-Methylglutaric acid

145 > 101

–22

–16

28

23 Adipic acid

145 > 83

–10

–18

28

24 2-Hydroxy-glutaric acid

147 > 129

–22

–13

28

25 Mevalonic acid

147 > 59

–25

–21

28

26 3-Hydroxy-glutaric acid

147 > 85

–11

–13

31

27 4-Hydroxy-phenylacetic acid

151 > 107

–8

–20

28

28 13 C6 -Adipic acid

151 > 88

–10

–18

29 Succinylacetone

157 > 99

–10

–12

46

30 3-Hydroxy-3-methylglutaric acid

161 > 99

–13

–14

31

31 2 H3 -3-Hydroxy-3-methylglutaric acid

164 > 102

–13

–14

32 Homogentisic acid

167 > 123

–34

–15

28

33 Suberic acid

173 > 111

–31

–18

28

34 N-acetylaspartic acid

174 > 88

–17

–19

35

35 2 H3 -acetylaspartic acid

177 > 89

–17

–19

36 Homovanillic acid

181 > 137

–19

–13

28

37 4-Hydroxy-phenyllactic acid

181 > 163

–30

–16

28

94

la Marca and Rizzo

Table 4.4 (continued) Metabolites 38 Sulfocysteine

Q1/Q3 transition

DP (V)

CE (V) Internal standard

200 > 136

–30

–14

28

39 Sialic acid

308 > 87

–33

–23

28

40 Uracil

111 > 42

–20

–20

41

41 15 N2 -uracil

113 > 43

–17

–20

42 Thymine

125 > 42

–22

–31

43 Thymine D4

129 > 42

–20

–31

44 Xanthine

151 > 108

–27

–25

48

45 Orotic acid

155 > 111

–13

–13

46

46 15 N2 -orotic acid

157 > 113

–13

–15

47 Uric acid

167 > 124

–54

–25

48 1,3 15 N2 -uric acid

169 > 125

–54

–25

49 Propionylglycine

130 > 74

–30

–13

50 2 H3 -propionylglycine

133 > 74

–30

–13

51 Butyrylglycine

144 > 74

–45

–15

43

48 50 50

52 Isobutyrylglycine

144 > 74

–45

–15

50

53 3-Methylcrotony lglycine

156 > 74

–45

–15

50 50

54 Tiglylglycine

156 > 75

–45

–15

55 Methylbutyry lglycine

158 > 74

–45

–15

56 Isovalerylglycine

158 > 74

–50

–15

58

57 Hexanoylglycine

172 > 74

–45

–15

58

58 2 H3 -hexanoylglycine 59 3-Phenylpropiony lglycine

175 > 74

–45

–15

206 > 74

–45

–15

58

60 Suberylglycine

230 > 74

–45

–15

58

9. Ratios of the MRM signals of the metabolites relative to the internal standards are used to construct calibration curves and calculate concentrations in the urine samples. The internal standards used for each metabolite are listed in Table 4.4. 10. Six calibrators with concentrations 0 (i.e., blank), 10, 25, 50, 100, and 250 μM are run with every batch of samples. The unenriched and enriched urine samples are also included with each batch. 11. Imprecision and analytical recovery from the enriched urine samples are determined from consecutive batches run over a period of 5 weeks. The presence of salts in the urine can lead to significant suppression of negativeion signals, which in turn affects the linearity, inter-batch variability, and recovery for metabolites quantified without their corresponding labeled internal standards. This ion

Analysis of Organic Acids and Acylglycines

95

Table 4.5 Isobaric interferences in negative ionization mode Metabolites

Q1/Q3 transition

4-Hydroxy-butyric acid

103 > 57

2-Hydroxy-butyric acid

103 > 57

Succinic acid

117 > 73

Methylmalonic acid

117 > 73

Glutaric acid

131 > 87

Ethylmalonic acid

131 > 87

Methylsuccinic acid

131 > 87

Butyrylglycine

144 > 74

Isobutyrylglycine

144 > 74

2-Ketoglutaric acid

145 > 101

3-Methylglutaric acid

145 > 101

Tiglylglycine

156 > 74

3-Methylcrotonylglycine

156 > 74

Isovalerylglycine

158 > 74

Methylbutyrylglycine

158 > 74

suppression effect is minimized using several labeled internal standards. Apart from routine ion source cleaning performed at approximately weekly intervals, no other precautions are taken to minimize ion source contamination. 12. A careful selection of appropriate MRM transitions makes it possible to measure several isomeric metabolites. In some cases, no suitable transitions to solve isomeric compounds are found (Table 4.5) (5). This condition can be solved in the future by coupling the tandem mass spectrometer with HPLC separation. 13. The primary application of the LC tandem mass spectrometer urine screening method is as a first test for suspected inborn errors of metabolism. While the tandem mass spectrometry method can be used to test all urine samples, GCMS-based organic acid analysis is typically used to confirm abnormal LC tandem mass spectrometer results or to distinguish isomers.

4. Notes 1. The 24-h collected urine should be considered a first choice for the analysis of organic acids. The majority of patients with suspicion of IEM are in neonatal age; thus, it is important to

96

la Marca and Rizzo

perform a rapid test, preferably on the first morning voiding. When possible, urine samples must be stored in a freezer at –20◦ C if a –70◦ C system is not available. In fact, poor preservation of samples can lead to (1) bacterial conversion of some organic acids to artifacts such as the conversion of all keto acids to their respective hydroxy acids, (2) increased levels of succinic acid from bacterial degradation of glutamine, and (3) an abnormal concentration of pyroglutamic acid due to conversion from glutamine. The concentration of such volatile compounds as small molecule organic acids (e.g. short-chain organic acids) can be shortened if the sample is exposed to high temperature (especially in summer). The administration of drugs should be suspended in patients undergoing organic acids analysis to avoid artifacts due to drugs or drug metabolism. The following are some examples of the interference of drugs on the analysis of organic acids. Valproic acid administration causes an abnormal excretion of 3-hydroxy-isovaleric acid, tiglylglycine, dicarboxylic acids (even saturated, e.g., suberic, adipic, and sebacid acids), 2methylbutyrylglycine, 7-hydroxy-octanoic acid, 5-hydroxyhexanoic acid, and hexanoylglycine. An increase of pivalic acid can be due to the administration of the antibiotics pivampicillin and pivmecillinam. Fluvoxamine maleate, a selective serotonin reuptake inhibitor (SSRI) used to treat obsessive–compulsive disorder and depression, can cause an increment of maleic acid. A therapy with the antihyperuricemic drug allopurinol can increase orotate urinary excretion. Artifactual formation of new compounds during sample preparation can also occur. For example, during extraction with ethyl acetate, decarboxylation of keto acids can occur (14). Additionally, it is essential to pay attention to solvent contaminants, plasticizers (phthalate, adipate, and sebacate esters), glassware-cleaning agents, lubricants, resins, and bleeding from the stationary phase of the chromatographic columns. 2. IS work solution stability is only at 4◦ C or below. For example, the peak intensity of PDA-TMS is considerably larger when analyzed 1 week after preparation and storage at room temperature than when immediately analyzed. This is due to solvent evaporation at room temperature (12). 3. Bilirubine, a potential interferent, is oxidized by using potassium ferricyanide K3 Fe(CN)6 at 0.13 mmol/L. 4. Do not mix organic and aqueous phases during liquid/liquid extraction of urine with solvents. 5. Clean the injector syringe everyday to avoid encrustation, purging, and bleeding.

Analysis of Organic Acids and Acylglycines

97

6. Incorrect mass spectrometer calibration can result in abnormal qualitative and quantitative results. 7. Clean the mass spectrometer source and change the liner every 500 samples. Contamination is usually identified by an abnormal background. In a GC-MS instrument, there are sources of contamination within the gas chromatograph (bleeding septa, dirty injection port liners, air leaks, etc.) and the mass spectrometer (ion source leaks, fluid/oil, manifold dirt). Unfortunately, not all contamination can be removed through running clean carrier gas overnight. When this is the case, the instrument may need extensive cleaning. Cleaning the mass spectrometer ion source to remove contamination is critical to restore the electrostatic properties of the ion source lens. The authors suggest that the ion source be cleaned every 500 runs. Another common source of mass spectrometer contamination is from inadequate venting and maintenance of the diffusion pump. Preventive maintenance ensures that diffusion pump oil/fluids are topped to the correct levels. It is important to maintain proper fluid levels to avoid pump failure and ensure optimum performance of the vacuum system. Air leaks can occur if a seal becomes damaged or is not correctly fastened. This is another common problem for any instruments that use a vacuum and can be identified by a higher than normal vacuum manifold pressure, low relative ion abundance, and poor sensitivity. 8. The use of labeled internal standards and isotope dilution allows one to obtain a correct and absolute quantitation of organic acids, but it can be very expensive. References 1. Björkman, L., McLean, C., Steen, G. (1976) Organic acids in urine from human newborns. Clin Chem 22, 49–52. 2. Chalmers, R. A., Healy, M. J., Lawson, A. M., Watts, R. W. (1976) Urinary organic acids in man. II. Effects of individual variation and diet on the urinary excretion of acidic metabolites. Clin Chem 22, 1288–1291. 3. Kumps, A., Duez, P., Mardens, Y. (2002) Metabolic, nutritional, iatrogenic, and artifactual sources of urinary organic acids: a comprehensive table. Clin Chem 48, 708–717. 4. Pitt, J. J., Eggington, M., Kahler, S. G. (2002) Comprehensive screening of urine samples for inborn errors of metabolism by electrospray tandem mass spectrometry. Clin Chem 48, 1970–1980.

5. Rebollido, M., Cocho, J. A., Castiñeiras, D. E., Bóveda, M. D., Fraga, J. M. (2006) Aplicación de la espectrometría de masas en tándem al análisis de aminoácidos, acilcarnitinas, acilglicinas y ácidos orgánicos en muestras de orina en papel. Quím Clín 25, 64–74. 6. Chalmers, R. A., Healy, M. J., Lawson, A. M., Hart, J. T., Watts, R. W. (1976) Urinary organic acids in man. III. Quantitative ranges and patterns of excretion in a normal population. Clin Chem 22, 1292–1298. 7. Perry, T. L., Hansen, S. (1974) Artifacts and pitfalls in interpretation of gas chromatograms in Application of Gas Chromatography-Mass Spectrometry to the Investigation of Human Disease. In Proceedings of a workshop, Montreal, May 1973. McGill University-Montreal Children’s

98

8. 9. 10.

11. 12.

la Marca and Rizzo Hospital Research Institute, Montreal, Quebec, Canada, 89–102. Paterson, N. (1967) Relative constancy of 24-hour urine volume and 24-hour creatinine output. Clin Chim Acta 18, 57–58. Zorab, P. A. (1969) Normal creatinine and hydroxyproline excretion in young persons. Lancet 29, 1164–1165. Applegarth, D. A., Hardwick, D. F., Ross, P. M. (1968) Creatinine excretion in children and the usefulness of creatinine equivalents in amino acid chromatography. Clin Chim Acta 22, 131–134. Larsen, K. (1972) Creatinine assay by a reaction-kinetic principle. Clin Chim Acta 41, 209–217. Tanaka, K., West-Dull, A., Hine., D. G., Lynn, T. B., Lowe, T. (1980) Gas-

chromatographic method of analysis for urinary organic acids. II. Description of the procedure, and its application to diagnosis of patients with organic acidurias. Clin Chem 26, 847–853. 13. Tanaka, K., Hine, D. G., West-Dull, A., Nad Lynn, T. B. (1980) Gas-chromatographic method of analysis for urinary organic acids. I. Retention indices of 155 metabolically important compounds. Clin Chem 26, 1839–1846. 14. Thompson, R. M., Belanger, B. G., Wappner, R. S., Brandt, I. K. (1975) An artifact in the gas chromatographic analysis of urinary organic acids from phenylketonuric children: decarboxylation of the phenylpyruvic acid during extraction. Clin Chim Acta 61, 367–374.

Chapter 5 HPLC Analysis for the Clinical–Biochemical Diagnosis of Inborn Errors of Metabolism of Purines and Pyrimidines Giuseppe Lazzarino, Angela Maria Amorini, Valentina Di Pietro, and Barbara Tavazzi Abstract The determination of purines and pyrimidines in biofluids is useful for the clinical–biochemical characterization of acute and chronic pathological states that induce transient or permanent alterations of metabolism. In particular, the diagnosis of several inborn errors of metabolism (IEMs) is accomplished by the analysis of circulating and excreted purines and pyrimidines. It is certainly advantageous to simultaneously determine the full purine and pyrimidine profile, as well as to quantify other compounds of relevance (e.g., organic acids, amino acids, sugars) in various metabolic hereditary diseases, in order to screen for a large number of IEMs using a reliable and sensitive analytical method characterized by mild to moderate costs. Toward this end, we have developed an ion-pairing HPLC method with diode array detection for the synchronous separation of several purines and pyrimidines. This method also allows the quantification of additional compounds such as N-acetylated amino acids and dicarboxylic acids, the concentrations of which are profoundly altered in different IEMs. The application of the method in the analysis of biological samples from patients with suspected purine and pyrimidine disorders is presented to illustrate its applicability for the clinical–biochemical diagnosis of IEM. Key words: Purines, pyrimidines, HPLC, body fluids, clinical–biochemical diagnosis, inborn errors of metabolism (IEMs), head injury, neurodegeneration, ischemia and reperfusion, multiple sclerosis.

1. Introduction In the clinical biochemistry setting, the terms “purines” and “pyrimidines” identify several low molecular weight compounds deriving from nucleic acid and nucleotide catabolism. Most of the T.O. Metz (ed.), Metabolic Profiling, Methods in Molecular Biology 708, DOI 10.1007/978-1-61737-985-7_5, © Springer Science+Business Media, LLC 2011

99

100

Lazzarino et al.

circulating and excreted purines and pyrimidines are uncharged, with various degrees of water solubility, and have excellent absorption coefficients in the UV region (230–300 nm). Depending on the compound considered, purines and pyrimidines have broad ranges of concentrations (from 0 to the millimolar levels) in bodily fluids of normal subjects; these levels are influenced by various factors such as age, sex, diet. Very often, those absent or present physiologically in trace amounts are subject to tremendous concentration fluctuations due to several pathologic states for which the analysis of purines and pyrimidines in urine and plasma provides a strong indication of an altered metabolic state. In fact, the determination of purines and pyrimidines could be of particular clinical relevance in a number of acute and chronic diseases, such as myocardial ischemia (1, 2), traumatic brain injury (3, 4), and neurodegenerative disorders (multiple sclerosis, amyotrophic lateral sclerosis, Alzheimer’s disease, and Parkinson’s disease) (5–7). Under these conditions, the determination of certain compounds of these two classes is useful to monitor changes in the metabolic status of the patient, thus giving information on the progression of the disease and the effectiveness of possible pharmacological treatments. Purine and pyrimidine concentrations in bodily fluids are also dramatically altered in various inborn errors of metabolism (IEMs). In fact, a large number of IEMs, characterized by various degrees of disability and neurological disturbances, include several enzymatic defects involving pathways of purine and pyrimidine catabolism, recovery, and de novo biosynthesis. For many of these IEMs, an early diagnosis may greatly reduce consequences for the patients but, since clinical signs are not always discriminatory, the final diagnosis is not simple and is frequently delayed. It is therefore evident that purine and pyrimidine profiling in biological fluids plays a critical role for both clinicians and IEM-affected patients (8–10). Based on the physical–chemical characteristics of those purines and pyrimidines with clinical–biochemical significance, various methods have been proposed during the last decade based on the use of high-performance liquid chromatography (HPLC) coupled with either UV-VIS (11–18) or mass spectrometric (19– 23) detectors, capillary electrophoresis (24, 25), thin-layer chromatography (TLC) (26), and gas chromatography–mass spectrometry (GC–MS) (27). We have recently described a highly sensitive, reproducible, easy to use, and low-cost ion-pairing HPLC method coupled with a diode array detector (DAD) for the determination of these compounds in biological fluids (28), suitable for the clinical– biochemical diagnosis of IEMs of purines and pyrimidines (29). This method also allows for the separation of compounds other than purines and pyrimidines and is therefore useful for the diagnosis of additional IEMs of clinical relevance (30). The method is

HPLC Analysis for the Clinical–Biochemical Diagnosis

101

characterized by minimal manipulation of biological fluids (urine, plasma/serum, cerebrospinal fluid, amniotic fluid) and allows the separation of purines (hypoxanthine, xanthine, uric acid, inosine, guanosine, adenosine, adenine, S-adenosylhomocysteine, S-adenosylmethionine, adenylosuccinate, succinylaminoimidazole carboxamide riboside, aminoimidazole carboxamide riboside) and pyrimidines (cytosine, cytidine, thymine, thyimidine, uracil, beta-pseudouridine, uridine, orotic acid), simultaneously with N-acetylated amino acids (N-acetylaspartate, N-acetylglutamate, N-acetylaspartylglutamate), dicarboxylic acids (propionic acid, malonic acid, methylmalonic acid), creatinine, reduced and oxidized glutathione, and ascorbic acid. Herein, we describe in detail how to carry out the analysis of the aforementioned compounds in biological fluids (serum and urine) of healthy subjects and of two selected patients each suffering from an IEM of purine metabolism, i.e., hypoxanthine phosphoribosyltransferase (HPRT) deficiency and adenylosuccinate lyase (ADSL) deficiency. A comprehensive list of the genetic disorders based on metabolic defects of the aforementioned compounds, for which the application of this HPLC method might be helpful in clinical diagnoses is reported in Table 5.1.

2. Materials 2.1. Chemicals and Standards

1. Tetrabutylammonium hydroxide is supplied as a 55% water solution (Nova Chimica, Milan, Italy), stored at room temperature, and light protected (see Note 1). 2. HPLC-grade methanol. 3. HPLC-grade water. 4. Monobasic potassium phosphate (KH2 PO4 ). A 1 M stock solution is prepared by dissolving the proper amount of the compound in HPLC-grade water, with stirring and moderate heating; this solution is stable for up to 1 month at 4◦ C. 5. Ultrapure standards for HPLC (Sigma, St. Louis, MO, USA). Stock solutions (1 mM final concentration) of the following compounds are prepared by dissolving the proper amount of each substance with HPLC-grade water. All solutions, with the exceptions of ascorbic acid and GSH, can be stored in aliquots and maintained at –80◦ C with no significant loss (decrease in concentration <5%) for up to 4 weeks. Ascorbic acid and GSH solutions must be prepared and used within the same day.

Analysis of succinyladenosine, adenylosuccinate, and adenosine (serum and urine) Analysis of the purines hypoxanthine, xanthine, uric acid, inosine, guanosine and of the pyrimidines uracil, uridine, thymine, thymidine, orotic acid (serum and urine) Analysis of inosine, guanosine, hypoxanthine, xanthine (serum and urine) Analysis of S-adenosyl-homocysteine (serum and urine) Analysis of thymine and uracil (serum and urine)

Adenylosuccinate lyase deficiency

Dysfunction of phosphoribosyl-pyrophosphate synthetase

Purine nucleoside phosphorylase deficiency

S-Adenosyl-homocysteine hydrolase deficiency

Dihydropyrimidinase deficiency (dihydropyrimidinuria)

Analysis of the pyrimidines uracil, uridine, thymine, thymidine, and orotic acid (serum and urine) Analysis of 5-amine-imidazole-4-carboxamide ribonucleotide and ribonucleoside (AICAR) and of inosine (serum and urine)

Isocratic separation, 100% Buffer A (25 min)

Analysis of orotic acid (serum and urine)

Orotate phosphoribosyl transferase, orotate monophosphate decarboxylase, or uridine monophosphate synthetase deficiency (orotic acidemia/uria) Uridine-5 -monophosphate hydrolase 1 deficiency

5-Amine-imidazole-4-carboxamide ribonucleotide formyltransferase and IMP cyclohydrolase deficiency

Isocratic separation, 100% Buffer A (25 min)

Analysis of malonic and methylmalonic acids (serum and urine)

Malonyl-CoA decarboxylase deficiency (malonic acidemia/uria)

Isocratic separation 100% Buffer A (25 min)

Step gradient separation (40 min)

Isocratic separation, 100% Buffer A (25 min)

Isocratic separation, 100% Buffer A (25 min)

Step gradient separation (80 min)

Step gradient separation (50 min)

Isocratic separation, 100% Buffer A (25 min)

Isocratic separation, 100% Buffer A (25 min)

Analysis of adenine and 2,8-dihydroxyadenine (serum and urine)

Adenine phosphoribosyl transferase deficiency (2,8-dihydroxyadenine urolithiasis)

HPLC conditions

Biochemical marker

Enzymatic defect

Table 5.1 Main inborn errors of metabolism that can be diagnosed using the present HPLC method

102 Lazzarino et al.

Analysis of thymine, uracil, thymidine (serum and urine)

Analysis of adenosine (serum, urine) Analysis of N-acetylaspartate (serum, urine, cerebrospinal fluid) Analysis of thymine and uracil (serum and urine) Analysis of xanthine, hypoxanthine, uric acid (serum and urine)

Thymidine phosphorylase deficiency (hepatocerebral variant of the myopathy due to mitochondrial DNA depletion or MNGIE syndrome)

Adenosine deaminase deficiency (severe combined immunodeficiency, SCID)

N-acetylaspartoacylase deficiency (Canavan disease) Dihydropyrimidine dehydrogenase deficiency (familiar pyridinemia, or pyrimidinuria, or thyminuria, or uraciluria)

Xanthine dehydrogenase deficiency (xanthinuria)

Gout due to altered PRPS1 activity

Isocratic separation, 100% Buffer A (25 min)

Isocratic separation, 100% Buffer A (25 min)

Isocratic separation, 100% Buffer A (28 min)

Step gradient separation (40 min)

Isocratic separation, 100% Buffer A (25 min)

Isocratic separation, 100% Buffer A (25 min)

Isocratic separation, 100% Buffer A (25 min)

Isocratic separation, 100% Buffer A (25 min)

Analysis of uric acid and of correlated purines (hypoxanthine, xanthine, inosine, guanosine) (serum and urine) Analysis of uric acid and of correlated purines (hypoxanthine, xanthine, inosine, guanosine) (serum and urine) Analysis of uric acid and of correlated purines (hypoxanthine, xanthine, inosine, guanosine) (serum and urine)

deficiency

Isocratic separation, 100% Buffer A (25 min)

Analysis of uric acid and of correlated purines (hypoxanthine, xanthine, inosine, guanosine) (serum and urine)

(renal

Urate transporter hypouricemia)

Hypoxanthine phosphoribosyl transferase deficiency (including Lesch–Nyhan syndrome, Kelley–Seegmiller syndrome, generic hyperuricemia, hyperuricemia with neurologic symptoms) Uromodulin deficiency (juvenile familiar hyperuricemic nephropathy)

HPLC Analysis for the Clinical–Biochemical Diagnosis 103

104

Lazzarino et al.

5a. Creatinine. 5b. Cytosine. 5c. Adenine. 5d. Adenosine. 5e. Inosine. 5f. Cytidine. 5g. Uracil. 5h. Uridine. 5i. Orotic acid. 5j. Thymine. 5k. Thymidine. 5l. Adenylosuccinate. 5m. Aminoimidazole carboxamide riboside (AICAR). 5n. Succinylaminoimidazole carboxamide riboside (SAICAR). 5o. S-adenosylhomocysteine (SAH). 5p. S-adenosylmethionine (SAM). 5q. Ascorbic acid. 5r. Reduced glutathione (GSH). 5s. Oxidized (GSSG) glutathione. 6. Ultrapure standards for HPLC (Sigma). Stock solutions (10 mM final concentration) of the following compounds are prepared by dissolving/diluting the proper amount of each substance with HPLC-grade water. These solutions can be stored in aliquots and maintained at –80◦ C with no significant loss (decrease in concentration <5%) for up to 4 weeks. 6a. N-acetylaspartate (NAA). 6b. N-acetylglutamate (NAG). 6c. N-acetylaspartylglutamate (NAAG). 6d. Propionic acid. 6e. Malonic acid. 6f. Methylmalonic acid. 7. Ultrapure standards for HPLC (Sigma). Stock solutions (1 mM final concentration) of the following compounds are prepared by dissolving the proper amount of each substance in 100 μM NaOH (see Note 2). All solutions, with the exception of uric acid, can be stored in aliquots and maintained at –80◦ C with no significant loss (decrease in concentration <5%) for up to 4 weeks. The uric acid solution must be prepared and used within the same day. 7a. Hypoxanthine. 7b. Xanthine. 7c. Uric acid. 7d. Guanosine. 8. Ultrapure beta-pseudouridine for HPLC (ICN-MP Biochemicals, Irvine, CA, USA). Stock solution (1 mM final concentration) of beta-pseudouridine is prepared by dissolving the proper amount with HPLC-grade water and can be

HPLC Analysis for the Clinical–Biochemical Diagnosis

105

stored in aliquots and maintained at –80◦ C with no significant loss (decrease in concentration <5%) for up to 4 weeks. 9. Three and 10 kDa molecular weight cut-off filters R Centrifugal Devices; Pall Gelman Laboratory, (Nanosep Ann Arbor, MI, USA). 2.2. HPLC Equipment

1. SpectraSystem P4000 HPLC pump, UV6000LP diode array detector equipped with a 5 cm light path flow cell, and refrigerated AS3000 autosampler (ThermoScientific Italia, Milan, Italy). 2. Hypersil C-18, 250 × 4.6 mm, 5-μm particle size HPLC column, and guard column (ThermoScientific Italia).

2.3. HPLC Buffers

1. Buffer A: 12 mM tetrabutylammonium hydroxide, 10 mM KH2 PO4 , 0.125% methanol; stir the solution and adjust the pH to 7.00 by adding 1 M HCl (see Note 3). 2. Buffer B: 2.8 mM tetrabutylammonium hydroxide, 100 mM KH2 PO4 , 30% methanol; stir the solution and adjust the pH to 5.50 by adding 1 M HCl (see Notes 3 and 4). 3. Filter both buffers through a Millipore filtering system equipped with acetate filters (pore diameter of 0.22 μm) to ensure proper degassing and to improve shelf life (see Note 5).

2.4. Preparation of the Standard Mixture for HPLC Separation

1. Freshly prepared standard mixtures with known concentrations are prepared daily and assayed by ion-pairing HPLC. 2. The standard mixture has a final volume of 1 mL. The following standards are used at a final concentration of 5 μM (add 5 μL of the corresponding 1 mM solution): cytosine, cytidine, creatinine, uridine, adenine, hypoxanthine, xanthine, guanosine, thymine, thymidine, and ascorbic acid; the following standards are used at a final concentration of 10 μM (add 10 μL of the corresponding 1 mM solution): uracil, beta-pseudouridine, inosine, and orotic acid; the following standards are used at a final concentration of 20 μM (add 20 μL of the corresponding 1 mM solution): uric acid, adenylosuccinate, SAH, SAM, AICAR, and SAICAR; the following standards are used at a final concentration of 25 μM (add 25 μL of the corresponding 1 mM solution): GSH and GSSG; the following standards are used at a final concentration of 50 μM (add 50 μL of the corresponding 1 mM solution): NAA, NAG, and NAAG; the following standards are used at a final concentration of 100 μM (add 10 μL of the corresponding 10 mM solution): propionic acid, malonic acid, and methylmalonic acid. 3. Adjust the final volume of the standard mixture to 1 mL with the proper volume of HPLC-grade water (see Note 6).

106

Lazzarino et al.

3. Methods 3.1. Pre-treatment of the MembraneEquipped Tubes

1. The 3 and 10 kDa molecular weight cut-off filters are equipped with regenerated cellulose membranes and require an appropriate washing treatment prior to their use with biological samples. 2. To extensively rinse the membranes, add about 0.5 mL of HPLC-grade water to each membrane followed by centrifugation at 15,000×g for 5 min at 4◦ C. This step should be repeated twice.

3.2. Preparation of Serum/Plasma Samples

1. No significant differences in concentrations of purines, pyrimidines, and other compounds between serum and plasma have been evidenced; consequently, it is preferable that serum be utilized in order to speed up the deproteinization process (see Note 7). 2. Dilute 200 μL of serum with 400 μL of HPLC-grade water R tube fitted (see Note 8) and place the sample in a Nanosep with a 10 kDa molecular weight cut-off membrane. Centrifuge the sample at 15,000×g at 4◦ C until complete passage of the liquid through the membrane (time required about 40 min). 3. In the case of plasma, the process of dilution and centrifugation is identical, but the time required for completion of the centrifugation step is increased to 50–60 min. 4. Transfer the deproteinized and ultrafiltered serum/plasma R to a Nanosep tube equipped with a 3 kDa molecular weight cut-off membrane and centrifuge again at 15,000×g at 4◦ C until complete passage of the liquid through the membrane (time required about 20 min). 5. Finally, inject the deproteinized and doubly ultrafiltered serum/plasma directly onto the HPLC column.

3.3. Preparation of Urine Samples

1. Urine samples are prepared similarly to serum/plasma samples. Dilute 100 μL of urine with 2400 μL of HPLC-grade R tube fitted with water (see Note 8) and place in a Nanosep a 3 kDa molecular weight cut-off membrane. Centrifuge the sample at 15,000×g at 4◦ C until complete passage of the liquid through the membrane (time required about 15 min) (see Note 9). 2. Inject the deproteinized and ultrafiltered urine sample directly onto the HPLC column.

HPLC Analysis for the Clinical–Biochemical Diagnosis

3.4. HPLC Equipment and Chromatographic Run Conditions

107

1. The diode array detector should be set up for data acquisition between 200 and 300 nm. Equilibrate the chromatographic column with Buffer A for 20 min at 1.2 mL/min and 10◦ C prior to the injection of any standards or biological sample.

Fig. 5.1. HPLC separation of an ultrapure standard mixture. In Panel A, the assignment of peaks to cytosine (1), creatinine (2), uracil (3), beta-pseudouridine (4), cytidine (5), hypoxanthine (6), thymine (7), xanthine (8), ascorbic acid (9), uridine (10), adenine (11), SAM (14), inosine (15), uric acid (16) guanosine (17), thymidine (18) orotic acid (19), SAH (24), AICAR (25), adenosine (26), SAICAR (28), succinyladenosine (29), and adenylosuccinate (31) was performed at 260 nm. In Panel B, the assignment of peaks to propionic acid (12), GSH (13), malonic acid (20), methylmalonic acid (21), NAA (22), NAG (23), GSSG (27), and NAAG (30) was performed at 206 nm.

0.3

0.1

0.1

0.1

0.1

0.1

0.1

0.1

0.1

0.1

0.1

0.2

0.1

0.2

Inosine

Guanosine

Adenosine

Adenine

SAH

SAM

Adenylosuccinate

SAICAR

AICAR

Cytosine

Cytidine

Thymine

Thymidine

0.1

Xanthine

Uric acid

0.1

Hypoxanthine

1

0.5

1

0.5

0.5

0.5

0.5

0.5

0.5

0.5

0.5

0.5

0.5

1.5

0.5

0.5

2

1

2

1

1

1

1

1

1

1

1

1

1

3

1

1

10

5

10

5

5

5

5

5

5

5

5

5

5

15

5

5

100

50

100

50

50

50

50

50

50

50

50

50

50

150

50

50

1

1

0.999

1

0.989

0.995

0.998

0.999

0.998

0.999

1

1

0.993

0.999

0.999

0.999

Final concentration Final concentration Final concentration Final concentration Final concentration Correlation in the mixture (µM) in the mixture (µM) in the mixture (µM) in the mixture (µM) in the mixture (µM) coefficient (R2 )

Table 5.2 Calibration curves and linearity of the method described

108 Lazzarino et al.

0.1

0.1

0.3

0.7

0.7

0.36

4

4

4

0.1

0.3

0.1

0.1

β-Pseudouridine

Uridine

Orotic acid

NAA

NAG

NAAG

Propionic acid

Malonic acid

Methylmalonic acid

Creatinine

GSH

GSSG

Ascorbic acid

0.5

0.5

1.5

0.5

20

20

20

1.8

3.5

3.5

1.5

0.5

0.5

0.5

1

1

3

1

40

40

40

3.6

7

7

3

1

1

1

5

5

15

5

200

200

200

18

35

35

15

5

5

5

50

50

150

50

2000

2000

2000

180

350

350

150

50

50

50

0.993

0.999

0.9999

1

0.991

0.990

0.992

0.998

0.997

0.999

1

1

1

0.999

SAH = S-adenosylhomocysteine; SAM = S-adenosylmethionine; SAICAR = succinylaminoimidazole carboxamide riboside; AICAR = aminoimidazole carboxamide riboside; NAA = N-acetylaspartate; NAG = N-acetylglutamate; NAAG = N-acetylaspartylglutamate; GSH = reduced glutathione; GSSG = oxidized glutathione.

0.1

Uracil

HPLC Analysis for the Clinical–Biochemical Diagnosis 109

110

Lazzarino et al.

Fig. 5.2. HPLC separation of deproteinized and ultrafiltered serum (Panel A) and urine (Panel B) samples from a healthy control subject. Identified purine or pyrimidine peaks follow the numbering reported in the legend to Fig. 5.1: Panel A, uracil (3), beta-pseudouridine (4), hypoxanthine (6), xanthine (8), ascorbic acid (9), uridine (10), inosine (15), and uric acid (16); Panel B, uracil (3), beta-pseudouridine (4), hypoxanthine (6), xanthine (8), and uric acid (16).

2. After sample injection, a step gradient is formed as follows: 25 min of 0% Buffer B; ramp to 20% Buffer B over 8 min; ramp to 30% Buffer B over 10 min; ramp to 45% Buffer B over 12 min; ramp to 60% Buffer B over 11 min; ramp to 85% Buffer B over 9 min; ramp to 100% Buffer B over 5 min; hold for 10 min at 100% Buffer B. This last step is lengthened to 30 min in the case of a biological sample.

HPLC Analysis for the Clinical–Biochemical Diagnosis

111

Fig. 5.3. HPLC separation of deproteinized and ultrafiltered serum (Panel A) and urine (Panel B) samples from an IEM patient suffering from partial HPRT deficiency (Kelley–Seegmiller syndrome). Only those compounds whose concentrations were changed by disease are indicated, and their identities follow the numbering reported in the legend to Fig. 5.1: Panel A and Panel B, hypoxanthine (6), xanthine (8), and uric acid (16).

Re-equilibrate the column at 100% Buffer A for 15 min before the next injection (see Note 10). 3. Maintain the flow rate and the column temperature at 1.2 mL/min and 10◦ C, respectively, for the duration of the analysis.

112

Lazzarino et al.

Fig. 5.4. HPLC separation of deproteinized and ultrafiltered serum (Panel A) and urine (Panel B) samples from an IEM patient suffering from partial ADSL deficiency. Only those compounds whose concentrations were changed by disease are indicated, and their identities follow the numbering reported in the legend to Fig. 5.1: Panel A and Panel B, SAICAR (28) and succinyladenosine (29).

4. Data are acquired and analyzed by a PC using the R software package provided by the HPLC ChromQuest manufacturer. 5. The following compounds are monitored at 260 nm: cytosine, cytidine, uracil, uridine, beta-pseudouridine, adenine,

HPLC Analysis for the Clinical–Biochemical Diagnosis

113

hypoxanthine, xanthine, inosine, guanosine, adenosine, AICAR, SAICAR, SAH, adenylosuccinate, ascorbic acid, thymine, thymidine, uric acid, and orotic acid. 6. The following compounds are monitored at 206 nm: NAA, NAG, NAAG, propionic acid, malonic acid, methylmalonic acid, GSH, and GSSG. 7. Creatinine is monitored at 234 nm. Figure 5.1 shows chromatograms of a standard mixture. 8. Calibration curves for each compound are obtained by running standard mixtures in increasing concentrations. As shown in Table 5.2, a very high linearity is observed for each compound in the concentration ranges indicated. These curves are then used to calculate the concentrations of compounds in unknown samples, as indicated below. 9. Peak assignment in deproteinized and ultrafiltered samples of biological fluids is carried out by comparing peak retention times and absorption spectra with those of freshly prepared ultrapure standards. The quantification of the different compounds in chromatographic runs of biological samples is effected by comparing their peak areas with the corresponding peak areas from chromatographic runs of standard mixtures with known concentrations. Figure 5.2 shows chromatograms of deproteinized and ultrafiltered serum (Panel A) and urine (Panel B) from a healthy control subject. Figure 5.3 illustrates chromatographic runs of deproteinized and ultrafiltrated serum (Panel A) and urine (Panel B) of an IEM patient suffering from partial HPRT deficiency (Kelley–Seegmiller syndrome), while deproteinized and ultrafiltered serum (Panel A) and urine (Panel B) of an IEM patient suffering from partial ADSL deficiency are shown in Fig. 5.4. 10. After the last chromatographic run, first wash the column with HPLC-grade water for 60 min at 1.2 mL/min followed by extensive washing with methanol/water (60/40, v/v; no less than 60 min at 1.2 mL/min at room temperature) in order to increase the column life span (see Note 11).

4. Notes 1. Tetrabutylammonium hydroxide (TBA) is utilized as the pairing reagent in both HPLC buffers; it is necessary to manage it with care because it is harmful (irritant). Highpurity TBA for HPLC applications is supplied as a very

114

Lazzarino et al.

viscous 55% solution. It is therefore recommended to pipet TBA with care and to avoid cleaning the extra liquid left on the outer wall of the pipet tip (it could alter the final concentration of TBA in the buffer solution). It is also advisable to put a reasonable amount of HPLC-grade water in a beaker and to add TBA after having cleaned the outer wall of the pipet tip. Then pipet several times to carefully rinse the inside of the pipet tip. 2. The use of NaOH is necessary because of the low solubility of the keto-tautomeric forms of these compounds in water. Prepare the 100 μM NaOH solution by dissolving a proper amount of the compound in HPLC-grade water. 3. Let the solutions stand for 5–10 min and recheck the pH; the value must be stable with no stirring at 7.00 for Buffer A and at 5.50 for Buffer B. 4. When preparing Buffer B, add methanol and water first in a beaker. Mix the solution by stirring and then add the proper amount of KH2 PO4 . Since KH2 PO4 is added from a 1 M stock solution to have a final concentration of 100 mM, the phosphate would precipitate if it were directly added to undiluted methanol. 5. In the case of Buffer A, the solution is stable for up to 4 days at 4◦ C, while in the case of Buffer B, the solution is stable for up to 4 weeks at 4◦ C. Moreover, in order to avoid any bacterial growth, it is recommended that glassware be subjected to a cycle of sterilization by autoclaving on a weekly basis. 6. After correcting the final volume of the standard mixture, it is important to check its pH value, since several purines are dissolved in 100 μM NaOH (pH can be checked with pH paper); when significantly higher than neutral, correct the pH by adding 2–5 μL of 1 M HCl. 7. This procedure avoids the use of complex, time-consuming sample manipulations, which may significantly alter the real concentrations of the various compounds in the sample. Furthermore, the method allows the direct detection of all compounds with no sample derivatization prior to HPLC analysis. 8. The sample dilution indicated is suitable for biological samples from healthy controls. In the case of patients with IEM, this dilution may result in a chromatogram with one or more peaks out of scale. In these cases, a further dilution of the sample and a new chromatographic run will be necessary in order to perform the correct quantification of the compound(s) in excess.

HPLC Analysis for the Clinical–Biochemical Diagnosis

115

9. The filtering procedure adopted for the urine preparation can also be used, with proper differences in sample dilution, for cerebrospinal or amniotic fluid preparation. In the case of cerebrospinal fluid, 200 μL of sample is diluted with 200 μL of HPLC-grade water (1:1, v/v); in the case of amniotic fluid, 200 μL of sample is diluted with 400 μL of water (1:2, v/v). 11. Since the majority of purines and pyrimidines of interest for the clinical–biochemical diagnosis of IEM are eluted isocratically during the first 25 min of the run, it is possible to perform a significantly shorter analysis with which the following compounds are separated: cytosine, creatinine, cytidine, uracil, uridine, beta-pseudouridine, adenine, hypoxanthine, xanthine, inosine, guanosine, ascorbic acid, thymine, thymidine, uric acid and orotic acid, NAA, propionic acid, malonic acid, methylmalonic acid, and GSH. When using this isocratic separation, it is possible to perform a simple column wash with methanol/water (60:40, v/v) at 1.2 mL/min for 15 min after elution of the last compound of interest. In this case, due to the high concentration of methanol in the washing solution, the reequilibration procedure should be lengthened to 25 min before a new sample is injected. 12. It is worth noting that Buffer B has a high phosphate concentration, rendering direct passage to the washing solution of methanol/water (60:40, v/v) unfeasible. If this happened, an immediate precipitation of phosphate crystals into the HPLC system would occur, with dangerous consequences for pistons, column, and tubing. An extensive wash with HPLC-grade water is therefore needed in between Buffer B and the washing solution of methanol/water (60:40, v/v).

Acknowledgments We wish to thank Sig. Salvatore Meo for his technical help in preparing the figures for this chapter. This work has been supported by research funds of the University of Catania and the Catholic University of Rome (grant D1-2004-2008).

References 1. Giardina, B., Penco, M., Lazzarino, G., Romano, S., Tavazzi, B., Fedele, F., Di Pierro, D., Dagianti, A. (1993) Effective-

ness of thrombolysis is associated with a time-dependent increase of malondialdehyde in peripheral blood of patients with acute

116

2.

3.

4.

5.

6.

7. 8.

9.

10.

11.

Lazzarino et al. myocardial infarction. Am J Cardiol 71, 788–793. Lazzarino, G., Raatikainen, P., Nuutinen, M., Nissinen, J., Tavazzi, B., Di Pierro, D., Giardina, B., Peuhkurinen, K. (1994) Myocardial release of malondialdehyde and purine compounds during coronary bypass surgery. Circulation 90, 291–297. Cristofori, L., Tavazzi, B., Gambin, R., Vagnozzi, R., Signoretti, S., Amorini, A. M., Fazzina, G., Lazzarino, G. (2005) Biochemical analysis of the cerebrospinal fluid: evidence for catastrophic energy failure and oxidative damage preceding brain death in severe head injury: a case report. Clin Biochem 38, 97–100. Cristofori, L., Tavazzi, B., Gambin, R., Vagnozzi, R., Vivenza, C., Amorini, A. M., Di Pierro, D., Fazzina, G., Lazzarino, G. (2001) Early onset of lipid peroxidation after human traumatic brain injury: a fatal limitation for the free radical scavenger pharmacological therapy?. J Invest Med 49, 450–458. Amorini, A. M., Petzold, A., Tavazzi, B., Eikelenboom, J., Keir, G., Belli, A., Giovannoni, G., Di Pietro, V., Polman, C., D’Urso, S., Vagnozzi, R., Uitdehaag, B., Lazzarino, G. (2009) Increase of uric acid and purine compounds in biological fluids of multiple sclerosis patients. Clin Biochem 42, 1001– 1006. Keizman, D., Ish-Shalom, M., Berliner, S., Maimon, N., Vered, Y., Artamonov, I., Tsehori, J., Nefussy, B., Drory, V. E. (2009) Low uric acid levels in serum of patients with ALS: further evidence for oxidative stress?. J Neurol Sci 285, 95–99. Schlesinger, I., Schlesinger, N. (2008) Uric acid in Parkinson’s disease. Mov Disord 23, 1653–1657. Morris, G. S., Simmonds, H. A., Davies, P. M. (1986) Use of biological fluids for the rapid diagnosis of potentially lethal inherited disorders of human purine and pyrimidine metabolism. Biomed Chromatogr 3, 109–118. Vidotto, C., Fousert, D., Akkermann, M., Griesmacher, A., Muller, M. M. (2003) Purine and pyrimidine metabolites in children’s urine. Clin Chim Acta 335, 27–32. Duran, M., Dorland, L., Meuleman, EE., Allers, P, Berger, R (1997) Inherited defects of purine and pyrimidine metabolism: laboratory methods for diagnosis. J Inherit Metab Dis 20, 227–236. Bennet, M. J., Carpenter, K. H. (1984) Experience with a simple high performance liquid chromatography method for the anal-

12.

13.

14.

15.

16.

17.

18.

19.

20.

21.

ysis of purine and pyrimidine nucleosides and bases in biological fluids. Ann Clin Biochem 21, 131–136. Boulieu, R., Bory, C., Baltassat, P., Divry, P. (1984) Hypoxanthine and xanthine concentration determined by high performance liquid chromatography in biological fluids from patients with xanthinuria. Clin Chim Acta 142, 83–89. Di Pierro, D., Tavazzi, B., Perno, C. F., Bartolini, M., Balestra, E., Caliò, R., Giardina, B., Lazzarino, G. (1995) An ion-pairing high-performance liquid chromatographic method for the direct simultaneous determination of nucleotides, deoxynucleotides, nicotinic coenzymes, oxypurines, nucleosides, and bases in perchloric acid cell extracts. Anal Biochem 231, 407–412. Kazoka, H. (2002) Analysis of purines and pyrimidines by mixed partition-adsorption normal-phase high-performance liquid chromatography. J Chromatogr A 942, 1–10. Czauderna, M., Kowalczyk, J. (1997) Simultaneous measurement of allantoin, uric acid, xanthine and hypoxanthine in blood by highperformance liquid chromatography. J Chromatogr B 704, 89–98. Minniti, G., Caruso, U., Cerone, R., de Toni, E. (1998) Purines and pyrimidines determination in urine using high-performance liquid chromatography. Adv Exp Med Biol 431, 843–848. Katayama, M., Matsuda, Y., Shimokawa, K., Tanabe, S., Kaneko, S., Hara, I., Sato, H. (2001) Simultaneous determination of six adenyl purines in human plasma by highperformance liquid chromatography with fluorescence derivatization. J Chromatogr B Biomed Sci Appl 760, 159–163. Kojima, T., Nishina, T., Kitamura, M., Kamatani, N., Nishioka, K. (1987) Liquid chromatography with multichannel ultraviolet detection used for studying disorders of purine metabolism. Clin Chem 33, 2052–2056. Lemr, K., Adam, T., Frycak, P., Friedecky, D. (2000) Mass spectrometry for analysis of purine and pyrimidine compounds. Adv Exp Med Biol 486, 399–403. Frycak, P., Huskova, R., Adam, T., Lemr, K. (2002) Atmospheric pressure ionization mass spectrometry of purine and pyrimidine markers of inherited metabolic disorders. J Mass Spectrom 37, 1242–1248. la Marca, G., Casetta, B., Malvagia, S., Pasquini, E., Innocenti, M., Donati, M. A., Zammarchi, E. (2006) Implementing tandem mass spectrometry as a routine tool

HPLC Analysis for the Clinical–Biochemical Diagnosis

22.

23.

24.

25.

26.

for characterizing the complete purine and pyrimidine metabolic profile in urine samples. J Mass Spectrom 41, 1442–1452. Ito, T., van Kuilenburg, A. B., Bootsma, A. H., van Cruchten, A., Wada, Y., van Gennip, A. H. (2000) Rapid screening of high-risk patients for disorders of purine and pyrimidine metabolism using HPLC-electrospray tandem mass spectrometry of liquid urine or urine-soaked filter paper strips. Clin Chem 46, 445–452. Hartmann, S., Okun, J. G., Schmidt, C., Langhans, C. D., Garbade, S. F., Burgard, P., Haas, D., Sass, J. O., Nyhan, W. L., Hoffmann, G. F. (2006) Comprehensive detection of disorders of purine and pyrimidine metabolism by HPLC with electrospray ionization tandem mass spectrometry. Clin Chem 52, 1127–1137. Friedecky, D., Adam, T., Bartak, P. (2002) Capillary electrophoresis for detection of inherited disorders of purine and pyrimidine metabolism: a selective approach. Electrophoresis 23, 565–571. Wang, S. P., Liao, C. S. (2004) Comparison of ion-pair chromatography and capillary zone electrophoresis for the assay of organic acids as markers of abnormal metabolism. J Chromatogr A 1051, 213–219. van Gennip, A. H., Grift, J., Bree-Blom, E. J., van Ketting, D., Wadman, S. K. (1979) Urinary excretion of methylated purines in

27.

28.

29.

30.

117

man and in the rat after the administration of theophylline. J Chromatogr 163, 351–362. Kuhara, T. (2002) Diagnosis and monitoring of inborn errors of metabolism using urease-pretreatment of urine, isotope dilution, and gas chromatography–mass spectrometry. J Chromatogr B Biomed Sci Appl 781, 497–517. Tavazzi, B., Lazzarino, G., Leone, P., Amorini, A. M., Bellia, F., Janson, C. G., Di Pietro, V., Ceccarelli, L., Donzelli, S., Francis, J. S., Giardina, B. (2005) Simultaneous high performance liquid chromatographic separation of purines, pyrimidines, N-acetylated amino acids, and dicarboxylic acids for the chemical diagnosis of inborn errors of metabolism. Clin Biochem 38, 997–1008. Di Pietro, V., Perruzza, I., Amorini, A. M., Balducci, A., Ceccarelli, L., Lazzarino, G., Barsotti, P., Giardina, B., Tavazzi, B. (2007) Clinical, biochemical and molecular diagnosis of a compound homozygote for the 254 bp deletion-8 bp insertion of the APRT gene suffering from severe renal failure. Clin Biochem 40, 73–80. Di Pietro, V., Gambacurta, A., Amorini, A. M., Finocchiaro, A., D’Urso, S., Ceccarelli, L., Tavazzi, B., Giardina, B., Lazzarino, G. (2008) A new T677C mutation of the aspartoacylase gene encodes for a protein with no enzymatic activity. Clin Biochem 41, 611–615.

Chapter 6 Bile Acid Analysis in Various Biological Samples Using Ultra Performance Liquid Chromatography/Electrospray Ionization-Mass Spectrometry (UPLC/ESI-MS) Masahito Hagio, Megumi Matsumoto, and Satoshi Ishizuka Abstract In recent years, bile acids (BAs) have received much attention as signaling molecules as well as biosurfactants for lipophilic nutrients. To understand exact BA behavior, the precise distribution of BAs in vivo must be determined. However, to date, it has been difficult to know the precise roles of BA due to variations in BA molecules including conjugated forms. Thus, we reconsidered BA extraction methodology and introduced an ultra performance liquid chromatography/electrospray ionization-mass spectrometry (UPLC/ESI-MS) technique for BA analysis. Consequently, we established a rapid and reliable method, using UPLC/ESI-MS, for the analysis of BAs in various biological samples including liver, bile, blood, intestinal contents, and feces. This method enables us to determine the BA profile, including conjugation status, in a single 30 min run. This technique will be a useful tool for the investigation of the roles of BA metabolism in physiological and pathological conditions. Key words: Ultra performance liquid chromatography, electrospray ionization-mass spectrometry, bile acids.

1. Introduction Bile acids (BAs) are synthesized from cholesterol by various hepatic enzymes in the liver (1). They are indispensable compounds for the absorption of hydrophobic nutrients due to their amphipathic structure. BAs are absorbed mainly from the ileal epithelial cells and return to the liver through the portal vein. BAs flowing into the large intestine are converted to secondary BAs (SBAs). Normally, the amount of BA lost by fecal excretion is compensated for by neosynthesis in the liver, and the homeostasis T.O. Metz (ed.), Metabolic Profiling, Methods in Molecular Biology 708, DOI 10.1007/978-1-61737-985-7_6, © Springer Science+Business Media, LLC 2011

119

120

Hagio, Matsumoto, and Ishizuka

of the BA pool is closely maintained. The polarity of each BA depends on the number and direction of hydroxyl groups, and their hydrophobicity is closely related to the ability of the forming micelle and to cytotoxicity against epithelial cells in the gut. For example, the amount of solubilized cholesterol in aqueous solution depends on BA type. Solubility of cholesterol in sodium deoxycholate solution is greater than in sodium ursodeoxycholate solution (2). The hydroxyl group in the BA molecule is thought to affect the homeostasis of epithelial cells, particularly with regard to apoptosis or cell proliferation in the large intestine (3–6). However, it is difficult to understand the precise influence of BAs on epithelial cells, as a variety of BAs exist at the same time in various tissues and intestinal contents. Furthermore, changes in BA metabolism could induce some inborn errors, such as cerebrotendinous xanthomatosis (7), Zellweger’s cerebrohepatorenal syndrome (8), defective BA conjugation (9), deficiency of various enzymes in BA biosynthesis (10), and cholestatic diseases (11). To date, gas chromatography (GC) or GC-mass spectrometry (GC-MS) techniques have been widely used for the analysis of BA composition. However, GC analysis requires many complicated steps, such as methylation or trimethylsilylation, to obtain volatile and stabilized BA molecules. Moreover, the chemicals used in these steps are usually harmful. Further, this procedure requires the deconjugation of BAs, which prevents simultaneous detection of conjugated and unconjugated BAs. Recently, HPLC has been used for composition analysis of BAs (12–14). The UPLC technique has further improved the efficiency of BA analysis in terms of time and separation ability (15).

2. Materials 2.1. Chemicals

1. BA standards (Steraloids, Newport, RI, USA). 1a. Cholic acid (5β-cholanic acid-3α,7α,12α-triol, CA) 1b. α-muricholic acid (5β-cholanic acid-3α,6β,7α-triol, αMCA) 1c. β-muricholic acid (5β-cholanic acid-3α,6β,7β-triol, βMCA) 1d. ω-muricholic acid (5β-cholanic acid-3α,6α,7β-triol, ωMCA) 1e. Chenodeoxycholic acid (5β-cholanic acid-3α,7α-diol, CDCA) 1f. Deoxycholic acid (5β-cholanic acid-3α,12α-diol, DCA) 1g. Hyodeoxycholic acid (5β-cholanic acid-3α,6α-diol, HDCA)

Bile Acid Analysis in Various Biological Samples

121

1h. Ursodeoxycholic acid (5β-cholanic acid-3α,7β-diol, UDCA) 1i. Lithocholic acid (5β-cholanic acid-3α-ol, LCA) 1j. Taurocholic acid [5β-cholanic acid-3α,7α,12α-triol-N(2-sulphoethyl)-amide, TCA] 1k. Tauro-α-muricholic acid [5β-cholanic acid-3α,6β, 7α-triol-N-(2-sulphoethyl)-amide, TαMCA] 1l. Tauro-ω-muricholic acid [5β-cholanic acid-3α,6α, 7β-triol-N-(2-sulphoethyl)-amide, TωMCA] 1m. Taurochenodeoxycholic acid [5β-cholanic acid-3α, 7α-diol-N-(2-sulphoethyl)-amide, TCDCA] 1n. Taurodeoxycholic acid [5β-cholanic acid-3α,12α-diolN-(2-sulphoethyl)-amide, TDCA] 1o. Taurohyodeoxycholic acid [5β-cholanic 6α-diol-N-(2-sulphoethyl)-amide, THDCA]

acid-3α,

1p. Taurolithocholic acid [5β-cholanic acid-3α-ol-N-(2sulphoethyl)-amide, TLCA] 1q. Glycocholic acid [5β-cholanic acid-3α,7α,12α-triol-N(carboxymethyl)-amide, GCA] 1r. Glycochenodeoxycholic acid [5β-cholanic acid-3α, 7α-diol-N-(carboxymethyl)-amide, GCDCA] 1s. Glycodeoxycholic acid [5β-cholanic acid-3α,12α-diolN-(carboxymethyl)-amide, GDCA] 1t. Glycohyodeoxycholic acid [5β-cholanic acid-3α, 6α-diol-N-(carboxymethyl)-amide, GHDCA] 1u. Glycoursodeoxycholic acid [5β-cholanic acid-3α, 7β-diol-N-(carboxymethyl)-amide, GUDCA] 1v. Glycolithocholic acid [5β-cholanic (carboxymethyl)-amide, GLCA]

acid-3α-ol-N-

2. BA internal standard: 23-nordeoxycholic acid (23-nor-5βcholanic acid-3α,12α-diol, NDCA) (Steraloids). 3. Standard mixture solution including all above BAs. A calibration curve is constructed using the standard mixture solution over the concentration range 100 nM to 50 μM. 2.2. Reagents

1. Ion-exchanged and redistilled water. 2. High-grade ethanol. 3. HPLC-grade acetonitrile. 4. HPLC-grade methanol. 5. Solvent A: acetonitrile/water (20:80, v/v) containing 10 mM ammonium acetate (NH4 Ac). 6. Solvent B: acetonitrile/water (80:20, v/v) containing 10 mM NH4 Ac (see Note 1).

122

Hagio, Matsumoto, and Ishizuka

2.3. Instruments

1. 2-mL micro tubes. 2. Ultra S Homogenizer VP-15S (Taitec Corp., Saitama, Japan). 3. Water bath. 4. Oasis HLB cartridge 1 cc/10 mg (Waters, Milford, MA). 5. Acquity UPLC system (Waters). 6. BEH C18 column (1.7 μm, 100 mm × 2.0 mm i.d.; Waters). 7. Quattro Premier XE quadrupole tandem MS (Waters) equipped with an electrospray ionization (ESI) probe.

3. Methods Biological samples including blood plasma, bile, liver, intestinal contents, and feces are lyophilized before starting extraction so as to eliminate moisture as much as possible (see Note 2). Although chloroform/methanol is generally used in lipid extractions, it is not required for BA extraction due to the amphipathic nature of BAs. 3.1. Bile Acid Extraction from Biological Samples

1. Grind lyophilized samples thoroughly. 2. Add 1 mL of ethanol (high grade) to 100 mg of the ground samples. 3. Add 50 μL of 500 μM NDCA solution (25 nmol) in methanol as an internal standard to each sample and vortex thoroughly (see Note 3). 4. Sonicate the samples (constant, 40 cycles, control 2.5, twice for 10 s) (see Note 4). 5. Heat samples at 60◦ C for 30 min in a water bath, and then heat the samples at 100◦ C for 3 min in a water bath (see Note 5). 6. After cooling down to room temperature, vortex the samples thoroughly. 7. Centrifuge the samples at 1600×g for 10 min at 4◦ C. 8. Collect the supernatants into a new tube. 9. Add 1 mL of ethanol to the precipitates. 10. Vortex thoroughly. 11. Centrifuge the samples at 11,200×g for 1 min at 4◦ C. 12. Transfer the supernatants into the tubes from Step 8.

Bile Acid Analysis in Various Biological Samples

123

13. Repeat these extraction steps (Steps 9–12) once more. 14. Evaporate the pooled extracts. 15. Store at –40◦ C until analysis. 3.2. Sample Preparation for LC-MS

1. Add 1 mL of methanol (HPLC grade) to the evaporated extract fractions and mix well (see Note 6). 2. Purify the methanol solutions using a HLB cartridge according to the manufacturer’s instructions (see Note 7). 3. Evaporate recovered eluates. 4. Add 1 mL of methanol to evaporated eluents and suspend thoroughly (see Note 8). 5. Introduce dissolved extracts to LC.

3.3. LC Settings

1. The column oven and the auto sampler are maintained at 40 and 15◦ C, respectively (see Note 9). The solvent flow rate is 400 μL/min, and the sample injection volume is 5 μL. 2. The gradient program is as follows: 5% solvent B from 0 to 5 min; linear ramp to 15% solvent B over 10 min; linear ramp to 25% solvent B over 5 min; linear ramp to 75% solvent B over 2 min. Solvent B is kept at 75% for 1 min and is then decreased linearly to 5% over 1 min. The 5% of solvent B is then maintained at 5% for 3 min. 3. The column eluent is introduced to the MS.

3.4. MS Settings

1. The capillary voltage, source temperature, and desolvation temperature are −3200 V, 120◦ C, and 400◦ C, respectively. The cone voltage is 35 V for all BAs, both conjugated and unconjugated forms. The desolvation and cone gas flow rates are 800 and 50 L/h, respectively. 2. Each BA is detected by monitoring the appropriate deprotonated molecule [M-H]– under selected ion-recording (SIR) mode. For unconjugated BAs, the m/z values of the product ions from the mono-, di-, and tri-hydroxylated forms are 375.6, 391.6, and 407.6, respectively. For taurineconjugated BAs, the m/z values of the product ions from the mono-, di-, and tri-hydroxylated forms are selected as 482.7, 498.7, and 514.7, respectively. For glycine-conjugated BAs, the m/z values of the product ions from the mono-, di-, and tri-hydroxylated forms are selected as 432.6, 448.6, and 464.6, respectively. The m/z of the NDCA internal standard is 377.5. A summary of ions monitored during MS analysis of BAs is shown in Table 6.1. A chromatogram of all detected BAs is shown in Fig. 6.1.

124

Hagio, Matsumoto, and Ishizuka

Table 6.1 Detected m/z, applied ranges of analysis, correlation coefficient of calibration curves, and accuracy of successive analysis in standard mixture and feces for bile acids SD (%) [M-H]–

Range (µM)

Correlation coefficiency

STD

Feces

Unconjugated CA

407.6

0.1–50

0.9997

3.70

5.01

αMCA

407.6

0.1–50

1.0000

3.60

3.29

βMCA

407.6

0.1–50

0.9998

4.11

3.57

ωMCA

407.6

0.1–50

1.0000

2.12

2.44

HDCA

391.6

0.1–50

1.0000

7.12

1.04

UDCA

391.6

0.1–50

0.9995

4.46

2.39

CDCA

391.6

0.1–50

0.9991

3.94

9.58

DCA

391.6

0.1–50

0.9998

1.79

1.62

LCA

375.6

0.1–50

0.9976

3.48

3.17

Taurine conjugated CA

514.7

0.1–50

0.9995

3.30

–

αMCA

514.7

0.1–50

0.9998

6.33

–

ωMCA

514.7

0.1–50

0.9992

5.26

–

HDCA

498.7

0.1–50

0.9999

1.95

–

CDCA

498.7

0.1–50

0.9997

2.78

–

DCA

498.7

0.1–50

0.9997

5.29

–

LCA

482.7

0.1–50

0.9997

3.03

–

Glycine conjugated CA

464.6

0.1–50

1.0000

3.51

–

HDCA

448.6

0.1–50

0.9999

5.88

–

UDCA

448.6

0.1–50

0.9998

3.61

–

CDCA

448.6

0.1–50

0.9999

5.52

–

DCA

448.6

0.1–50

1.0000

2.64

–

LCA

432.6

0.1–50

0.9986

7.12

–

377.5

0.1–50

0.9991

3.31

3.88

Internal standard NDCA

In this analytical protocol using UPLC/ESI-MS, these values of m/z for each BA are applied in selective ion-recording mode. We repeatedly try to adjust the dilution ratio of sample extracts to match the detected concentrations of each BA within these ranges, because, especially in biological samples, the concentration of each BA is widespread. Correlation coefficient is calculated by analyzing 0.1, 1, 5, 10, 25, and 50 μM standard mixtures (STD). The values of SD are shown as a percentage of the standard deviation against the mean value of each BA concentration in five successive analyses of STD and fecal extracts from DA rats.

Bile Acid Analysis in Various Biological Samples 100

%

TIC 7.25e6 m/z = 100 –1000

0 2.00

4.00

6.00

8.00

10.00

12.00

14.00

16.00

18.00

20.00

100

22.00 6

24.00

26.00

SIR 4.27e5 m/z = 407.6, 377.5, 375.6

%

1 23

4

5

0 2.00

4.00

6.00

8.00

10.00

12.00

14.00

16.00

18.00 9

100

20.00 10

22.00

24.00

26.00

SIR 1.36e5 m/z = 391.6

8

%

7

0 2.00

4.00

6.00

8.00

10.00

12.00

14.00

16.00

18.00

20.00

100

%

Relative intensity (%)

125

15

13 14

11 12

22.00 17

24.00

26.00

SIR 1.56e5 m/z = 514.7, 498.7, 482.7

16

0 2.00

4.00

6.00

8.00

10.00

12.00

14.00

16.00

18.00

20.00

%

100

18 19

21

22.00 23

24.00

26.00

SIR 2.30e5 m/z = 448.6, 464.6, 432.6

22

20

0

Time 2.00

4.00

6.00

8.00

10.00

12.00

14.00

16.00

18.00

20.00

22.00

24.00

26.00

Time (min)

Fig. 6.1. Selected ion-recording (SIR) chromatograms for BA standard mixture solutions (5 μM each) obtained from UPLC/ESI-MS analysis. (1) ωMCA, (2) αMCA, (3) βMCA, (4) CA, (5) NDCA, (6) LCA, (7) UDCA, (8) HDCA, (9) CDCA, (10) DCA, (11) TωMCA, (12) TαMCA, (13) THDCA, (14) TCA, (15) TCDCA, (16) TDCA, (17) TLCA, (18) GUDCA, (19) GHDCA, (20) GCA, (21) GCDCA, (22) GDCA, (23) GLCA.

3. Concentrations of the individual BAs are calculated by comparing the area under the peak of each BA with that of the internal standard NDCA (see Note 10). 4. As an illustration, a chromatogram of the jejunal contents is shown in Fig. 6.2. The concentrations of each BA in the jejunal contents and feces are shown in Table 6.2.

4. Notes 1. In preparing solvents A and B, it is preferable to dissolve NH4 Ac in water in advance and then mix acetonitrile with the water containing NH4 Ac, as NH4 Ac cannot be dissolved in acetonitrile. 2. In cases of blood plasma or bile, 100 μL of plasma or 50 μL of bile samples is lyophilized and extracted without the heating step. Especially for blood plasma, this step is advantageous in eliminating unwanted substances, such as proteins.

126

Hagio, Matsumoto, and Ishizuka 100

%

TIC 1.33e8 m/z = 100 –1000

0 2.00

4.00

6.00

8.00

10.00

12.00

14.00

16.00

18.00

20.00

22.00

24.00

5

%

SIR 1.19e5 m/z = 407.6, 377.5, 375.6

3

0 2.00

4.00

6.00

8.00

10.00

12.00

14.00

16.00

18.00

20.00

22.00

24.00

26.00

8

%

100

SIR 1.02e4 m/z = 391.6

9 10

7

0 2.00

4.00

6.00

8.00

10.00

12.00 14

100

%

Relative intensity (%)

26.00

4

100

14.00

16.00

18.00

20.00

22.00

24.00

SIR 8.93e5 m/z = 514.7, 498.7, 482.7

13

12

15

16

26.00

17

0 2.00

4.00

6.00

8.00

%

100

10.00 20

12.00

14.00

16.00

18.00

20.00

22.00

24.00

SIR 3.43e5 m/z = 448.6, 464.6, 432.6

19 21

18

26.00

22

0

Time 2.00

4.00

6.00

8.00

10.00

12.00

14.00

16.00

18.00

20.00

22.00

24.00

26.00

Time (min)

Fig. 6.2. Chromatograms of BAs in the jejunal contents of rats using the selected ion-recording (SIR) mode. (3) βMCA, (4) CA, (5) NDCA, (7) UDCA, (8) HDCA, (9) CDCA, (10) DCA, (12) TαMCA, (13) THDCA, (14) TCA, (15) TCDCA, (16) TDCA, (17) TLCA, (18) GUDCA, (19) GHDCA, (20) GCA, (21) GCDCA, (22) GDCA.

3. It is not always necessary to add this amount of NDCA to samples, because dilution ratio of introducing samples to UPLC/ESI-MS is widespread (see Note 8). In our case, the amount of NDCA whose diluted concentration is from 10 to 25 μM is added to samples. 4. We only use sonication to make the suspensions. It is preferable to perform this step on ice. 5. As the caps on the microtubes can sometimes open during heating, we always place cap holders on the microtubes. 6. When methanol is added to the extracts, some debris might be found in the tube. If there is a large precipitant in the tube, it might be better to use the supernatant after centrifugation. 7. Although the sample for injection is normally dissolved in the same solvent as that used for cartridge equilibration, we usually dissolve the samples in methanol. Briefly, after conditioning cartridges with 1 mL of methanol, we use 1 mL of water containing 10 mM NH4 Ac as an equilibration step. Then, 1 mL of water containing 10 mM NH4 Ac is again added into the cartridge. Soon after this step, 100 μL of sample in methanol (from the solution described in Step 1 in Section 3.2) is added and gently mixed with

Bile Acid Analysis in Various Biological Samples

127

Table 6.2 Bile acid concentrations in the jejunal contents and feces Jejunum

Feces

1.52 ± 0.59

0.14 ± 0.03

Concentrations (μmol/g dry contents) Unconjugated CA αMCA

0.08 ± 0.04

0.03 ± 0.02

βMCA

0.22 ± 0.09

0.60 ± 0.17

ωMCA

0.03 ± 0.02

2.79 ± 0.80

HDCA

0.09 ± 0.04

1.96 ± 0.63

UDCA

0.03 ± 0.01

0.04 ± 0.01

CDCA

0.06 ± 0.03

<0.01

DCA

0.03 ± 0.01

0.58 ± 0.14

LCA

<0.01

0.29 ± 0.04

CA

6.63 ± 1.13

ND

αMCA

6.10 ± 1.72

ND

ωMCA

0.57 ± 0.09

ND

Taurine conjugated

HDCA

0.67 ± 0.21

<0.01

CDCA

0.35 ± 0.05

<0.01

DCA

0.31 ± 0.05

ND

LCA

ND

ND

0.47 ± 0.21

ND

Glycine conjugated CA HDCA

0.04 ± 0.03

ND

UDCA

<0.01

ND

CDCA

0.01 ± 0.01

ND

DCA

0.01 ± 0.01

<0.01

LCA

ND

ND

TBA

17.24 ± 2.41

6.45 ± 0.90

Conjugation

85.3 ± 6.1

<0.1

SBA/TBA

11.4 ± 1.6

87.2 ± 2.7

Percentages (%)

Male DA/Slc rats (6 weeks old, n = 8) (Japan SLC, Inc., Hamamatsu, Japan) had free access to a purified diet and water for 7 weeks. The diet composition was as follows (16): 52.95% dextrin, 20% casein, 10% sucrose, 7% soybean oil, 5% cellulose, 3.5% mineral mixture (AIN 93G), 1% vitamin mixture (AIN 93G), 0.3% L-cystine, and 0.25% choline chloride. TBA; total BA, SBA; secondary BA (BAs except for CA, CDCA, αMCA, and βMCA, regardless of conjugation status), ND; not detected. Values are expressed as means ± SEM.

128

Hagio, Matsumoto, and Ishizuka

the water containing 10 mM NH4 Ac on the top of the cartridge. By doing this, the sample can be applied almost wholly to water containing 10 mM NH4 Ac. 8. At this point, dilution of the extracts has to be determined independently for each sample. 9. The temperature in the auto sampler is kept at 15◦ C in a default condition in the LC. However, evaporation of methanol in samples is strictly limited because the vials are tightly covered with a cap. 10. The range of concentrations in the standard curve for each BA has to be determined individually. In five successive analyses of a standard mixture sample, the ratios of standard deviation (SD) against the mean values of BA concentrations (23 BAs) were all below 8%. In the case of a biological sample, the ratios of SD against the mean values of unconjugated BA concentrations (nine BAs) in the feces of rats were all below 10% (Table 6.1). References 1. Russell, D. W. (2003) The enzymes, regulation, and genetics of bile acid synthesis. Ann Rev Biochem 72, 137–174. 2. Ninomiya, R., Matsuoka, K., Moroi, Y. (2003) Micelle formation of sodium chenodeoxycholate and solubilization into the micelles: comparison with other unconjugated bile salts. Biochim Biophys Acta 1634, 116–125. 3. Haza, A. I., Glinghammar, B., Grandien, A., Rafter, J. (2000) Effect of colonic luminal components on induction of apoptosis in human colonic cell lines. Nutr Cancer 36, 79–89. 4. Araki, Y., Fujiyama, Y., Andoh, A., Nakamura, F., Shimada, M., Takaya, H., Bamba, T. (2001) Hydrophilic and hydrophobic bile acids exhibit different cytotoxicities through cytolysis, interleukin-8 synthesis and apoptosis in the intestinal epithelial cell lines. IEC-6 and caco-2 cells. Scand J Gastroenterol 36, 533–539. 5. Booth, L. A., Gilmore, I. T., Bilton, R. F. (1997) Secondary bile acid induced DNA damage in HT29 cells: are free radicals involved?. Free Radic Res 26, 135–144. 6. Journe, F., Durbecq, V., Chaboteaux, C., Rouas, G., Laurent, G., Nonclercq, D., Sotiriou, C., Body, J. J., Larsimont, D. (2008) Association between farnesoid X receptor expression and cell proliferation in estrogen receptor-positive luminal-like breast

7.

8.

9.

10.

11.

12.

13.

cancer from postmenopausal patients. Breast Cancer Res Treat 115, 523–535. Keren, Z., Falik-Zaccai, T. C. (2009) Cerebrotendinous xanthomatosis (CTX): a treatable lipid storage disease. Pediatr Endocrinol Rev 7, 6–11. Kase, B. F. (1989) Role of liver peroxisomes in bile acid formation: inborn error of C27steroid side chain cleavage in peroxisome deficiency (Zellweger syndrome). Scand J Clin Lab Invest 49, 1–10. Hofmann, A. F., Strandvik, B. (1998) Defective bile acid amidation: predicted features of a new inborn error of metabolism. Lancet 2, 311–313. Sundaram, S. S., Bove, K. E., Lovell, M. A., Sokol, R. J. (2008) Mechanisms of disease: inborn errors of bile acid synthesis. Nat Clin Pract Gastroenterol Hepatol 5, 456–468. Colombo, C., Okolicsanyi, L., Strazzabosco, M. (2000) Advances in familial and congenital cholestatic diseases. Clinical and diagnostic implications. Dig Liver Dis 32, 152–159. Shaw, R., Smith, J. A., Elliott, W. H. (1978) Bile acids LIII. Application of reverse-phase high-pressure liquid chromatography to the analysis of conjugated bile acids in bile samples. Anal Biochem 86, 450–456. Siow, Y., Schurr, A., Vitale, G. C. (1991) Diabetes-induced bile acid composition changes in rat bile determined by high performance liquid chromatography. Life Sci 49, 1301–1308.

Bile Acid Analysis in Various Biological Samples 14. Ando, M., Kaneko, T., Watanabe, R., Kikuchi, S., Goto, T., Iida, T., Hishinuma, T., Mano, N., Goto, J. (2006) High sensitive analysis of rat serum bile acids by liquid chromatography/electrospray ionization tandem mass spectrometry. J Pharm Biomed Anal 40, 1179–1186. 15. Hagio, M., Matsumoto, M., Fukushima, M., Hara, H., Ishizuka, S. (2009) Improved

129

analysis of bile acids in tissues and intestinal contents of rats using LC/ESI-MS. J Lipid Res 50, 173–180. 16. Reeves, P. G., Nielsen, F. H., Fahey, G. C.. Jr. (1993) AIN-93 purified diets for laboratory rodents: final report of the American institute of nutrition ad hoc writing committee on the reformulation of the AIN-76a rodent diet. J Nutr 123, 1939–1951.

Chapter 7 Analysis of Glycolytic Intermediates with Ion Chromatography- and Gas Chromatography-Mass Spectrometry Jan C. van Dam, Cor Ras, and Angela ten Pierick Abstract In this chapter, we describe a method for the quantitative analysis of glycolytic intermediates using ion chromatography-mass spectrometry (IC-MS) and gas chromatography (GC)-MS as complementary methods. With IC-MS-MS, pyruvate, glucose-6-phosphate, fructuse-6-phosphate, fructose-1,6-bisphosphate, phosphoenolpyruvate, and the sum of 2-phosphoglyceraldehyde + 3phosphoglyceraldehyde can be quantified. With GC-MS using selected ion monitoring, glyceraldehyde3-phosphate, dihydroxyacetonephosphate, 2-phosphoglyceraldehyde, and 3-phosphoglyceraldehyde can be analyzed. The derivatization for GC-MS is performed in two steps. In the first step, the keto and the aldehyde groups are oximated. In the next step, a silylation with N-methyl-Ntrimethylsilyltrifluoroacetamide (MSTFA) is performed, giving TMS-MOX derivatives of the glycolytic intermediates. The derivatives are separated on a GC column and detected with MS in SIM mode. Key words: Glycolytic intermediates, anion exchange chromatography, ion suppressor, IC-MS-MS, TMS-MOX derivatives, GC-MS, isotope dilution mass spectrometry.

1. Introduction The glycolytic intermediates play an important role in many biological processes. By mapping these processes much insight has been gained on the mechanisms that play an important role in life and in the evolution of life. Although we still do not understand life itself, much can be said about the processes involved in life. Glycolysis is a main metabolic route in central carbon metabolism, and, therefore, the need to analyze the glycolytic intermediates is evident. Indeed, profiling of the glycolytic intermediates is applied T.O. Metz (ed.), Metabolic Profiling, Methods in Molecular Biology 708, DOI 10.1007/978-1-61737-985-7_7, © Springer Science+Business Media, LLC 2011

131

132

van Dam, Ras, and ten Pierick

in several scientific fields, such as bio-engineering, metabolic disorders, inborn diseases, and metabolic modeling. All the glycolytic intermediates are phosphorylated compounds except pyruvate. Due to the ionic nature of these metabolites, ion chromatography (IC) is an obvious choice for the analysis. Another method which is gaining in popularity is GC-MS, in which the intermediates have to be derivatized in order to make them volatile. IC-MS/MS (multiple reaction monitoring, MRM) and GC-MS can be used as complementary techniques for the analysis of the glycolytic intermediates. In this chapter, we describe in detail the application of these two analytical methods for analysis of the glycolytic intermediates (1, 2). Most of the glycolytic intermediates can be analyzed with the IC method, with the exception of glyceraldehyde-3-phosphate (GAP) and dihydroxyacetonephosphate (DHAP). Apparently, these compounds decompose in the column due to high sodium hydroxide concentrations used in the gradient. However, GAP and DHAP can be analyzed with the GC-MS method. In the IC method, 2-phosphoglycerate (2PG) and 3-phosphoglycerate (3PG) coelute, but, due to the fact that the concentration of 3PG is approximately ten times higher than that of 2PG and that 2PG and 3PG are assumed to be in equilibrium, we measure the sum of both intermediates by using a 3PG standard. However, 2PG and 3PG can be separated and analyzed with the GC-MS method.

2. Materials 2.1. Chemicals

All glycolytic intermediates are purchased from Sigma-Aldrich (St. Louis, MO). Pyruvate is stored at 4◦ C and all other compounds at –20◦ C (see Note 1). 1. Glucose-6-phosphate (G6P; 98.5%). 2. Fructose-6-phosphate (F6P; 100%). 3. Fructose-1,6-bisphosphate (FBP; 99%). 4. Phosphoenolpyruvate (PEP; 100%). 5. 2-Phoshoglycerate (2PG; 99%). 6. 3-Phosphoglycerate (3PG; 99%). 7. Glyceraldehyde-3-phosphate (GAP; 260–320 mM). 8. Dihydroxyacetonephosphate (DHAP; >95%). 9. Pyruvate (PYR; 100%). 10. Erythritol (99%). 11. Methoxyamine-HCl (MOX; Pierce, Rockford, IL).

Analysis of Glycolytic Intermediates

12. N-methyl-N-trimethylsilyltrifluoroacetamide Pierce).

133

(MSTFA;

13. Pyridine. 14. Methanol HPLC grade. 15. HPLC solvent A: 95% water/5% methanol. 16. HPLC solvent B: 0.6 mM NaOH. 17. HPLC solvent C: 0.06 M NaOH. 18. HPLC solvent D: 0.3 M NaOH. 19. U-13 C glucose (99%, Cambridge Isotope Laboratories, Inc., Andover, MA). 2.2. Equipment

1. Milli-Q water purification system (Millipore, Billerica, MA): This system is used to produce deionized and filtered water for the preparation of solutions and eluents. 2. Heating module (Reacti-Therm III, Pierce) with aluminum heating block. Eighty-one samples can be derivatized simultaneously. 3. Centrifuge. 4. RapidVap (Labconco Corporation, Kansas City, MO).

2.3. IC-MS/MS Instrumentation

1. Waters 2795 HPLC (Waters, Milford, MA). 2. IonPac AS11 (250 mm × 4 mm) anion exchange column (Dionex, Sunnyvale, CA). 3. XTerra MS C18 (5 μm, 3.0 mm × 20 mm) guard column (Waters). 4. ASRS 300 4 mm ion suppressor (Dionex). This system removes post-column sodium ions from the eluent. 5. Two-way LabPro solvent selector valve (Rheodyne, Cotati, CA). 6. Ten-port, two-position LabPro valve (Rheodyne). 7. Quattro LC triple quadrupole mass spectrometer (Waters). 8. MassLynx 4.0 software.

2.4. GC-MS Instrumentation

1. Trace GC Ultra (Thermo Finnigan, Boston, MA). 2. Trace DSQ single quadrupole mass spectrometer (Thermo Finnigan). 3. Zebron ZB-50 column (30 m × 250 μm internal diameter, 0.25 μm film thickness; Phenomenex, Torrance, CA) 4. Straight glass liners with CarboFrit (Restek, Bellefonte, PA). 5. BTO septa (Restek). These septa are low-bleed and hightemperature resistant (400◦ C).

134

van Dam, Ras, and ten Pierick

6. Needle to puncture the PTFE septa, any injection needle will work. 7. GC vials, glass short thread vials, 32 × 11.6 mm, wide opening (Grace, Dearfield, IL). 8. Septa GC vial for freeze-drying: ultra pure PTFE thickness 0.2 mm (Grace, Dearfield, IL). 9. Septa GC vial for analysis: natural rubber red/orange/ PTFE transparent (Grace, Dearfield, IL).

3. Methods Although sampling (3) and sample handling are not part of the analysis method, it is very clear that these two subjects are very important for reliable analysis results. Storage conditions are therefore important because biological samples can easily degrade. Freezing and thawing of samples several times can also degrade samples. If you thaw a sample and only use a portion, thaw completely and mix before dividing the sample into several aliquots. Keep in mind that even a good analytical method cannot compensate for poor sampling or sample handling. The methods described here are mainly used in analyses of intracellular and extracellular metabolites from fermentations. The intracellular samples contain the entire cell matrix, with the exception of any material that is precipitated during the extraction procedure with boiling ethanol (75%). 3.1. Preparation of Standards for IC-MS/MS and GC-MS

The electrospray ionization efficiency of a metabolite is dependent on the concentrations of other coeluting metabolites and matrix components. As matrix effects – observed with electrospray ionization – are compound dependent, quantitative studies require an internal standard for each target metabolite. Otherwise, the ion suppression during electrospray could result in nonlinear responses. With isotope dilution mass spectrometry (IDMS), isotopically labeled analogues of each metabolite are used as internal standards in the sample solution and in the calibration mixtures. In most cases, these labeled materials are not easily available from commercial suppliers. However, it is possible to prepare labeled standards by growing cells on fully labeled media followed by cell extraction. For the measurement of the glycolytic intermediates, labeled isotopomers are prepared by growing microorganisms on uniformly 13 C-labeled glucose. Cells are then harvested and the metabolites extracted, giving a 13 C-labeled extract containing the whole labeled metabolome of the cell (4). The unlabeled intermediate and its isotopologue behave similarly during

Analysis of Glycolytic Intermediates

135

chromatography separation and coelute. A fixed amount of the labeled internal standard extract is added to both the calibration standards and the samples (see Note 2). After the elution of the intermediate and its isotopologue, the two compounds are detected and measured by MS. The peak area ratios are obtained by dividing the peak area of each intermediate by the peak area of its corresponding isotopologue. The peak area ratios are used for the construction of calibration curves and for quantification. Using the IDMS correction method, calibration curves are linear over a much wider range. In addition, the method also corrects for incomplete derivatization and for losses due to sample handling. In the case of GC-MS (not MS/MS), the signal of the internal standard (the 13 C analogue) should be corrected for the contribution of the naturally present labeled intermediates. 3.1.1. Preparation of 13 C Cell Extract

1. Run a yeast fermentation feeding the microorganism with U-13 C glucose and sample the fermentor. 2. Centrifuge the sample at 5000×g for 5 min and decant the liquid. Keep the cell pellet (in our fermentations, 1 mL broth contains ∼15 mg of yeast dry weight). 3. Add the cell pellet from 1 mL of yeast to 5 mL of boiling ethanol (75%). 4. Boil the sample for 3 min and then cool the sample on ice. 5. Evaporate the extract to dryness in a RapidVap. 6. Dissolve the dry extract in 500 μL water. 7. Centrifuge the extract for 5 min at 15,000×g. 8. Transfer the clear liquid to a separate tube. The is ready for use or storage.

3.1.2. Preparation of Standards

13 C

extract

1. Prepare aqueous stock solutions of 100 mM for each metabolite. 2. Prepare an aqueous mixed working solution containing the metabolites of interest by dilution of the standard stock solutions. This solution is the calibration mixture. Store aliquots of this mixture at –80◦ C (see Note 3). The concentrations of the metabolites in the calibration mixture will depend on the linear dynamic range of the IC-MS platform used, as well as the concentrations of the metabolites to be measured in the samples of interest. The concentrations of the intermediates in our calibration mixture are as follows: 150 μM PYR, 250 μM G6P, 50 μM F6P, 50 μM PEP, 100 μM 3PG, and 500 μM FBP. 3. By diluting the calibration mixture, ten standards are prepared.

136

van Dam, Ras, and ten Pierick

4. Standard 1 (0.30 μM PYR, 0.50 μM G6P, 0.10 μM F6P, 0.10 μM PEP, 0.20 μM 3PG, and 1.00 μM FBP): Pipette 2 μL of calibration mixture to a vial, add 998 μL of water, and mix the liquid. 5. Standard 2 (0.75 μM PYR, 1.25 μM G6P, 0.25 μM F6P, 0.25 μM PEP, 0.50 μM 3PG, and 2.50 μM FBP): Pipette 5 μL of calibration mixture to a vial, add 995 μL of water, and mix the liquid. 6. Standard 3 (1.50 μM PYR, 2.50 μM G6P, 0.50 μM F6P, 0.50 μM PEP, 1.00 μM 3PG, and 5.00 μM FBP): Pipette 10 μL of calibration mixture to a vial, add 990 μL of water, and mix the liquid. 7. Standard 4 (3.00 μM PYR, 5.00 μM G6P, 1.00 μM F6P, 1.00 μM PEP, 2.00 μM 3PG, and 10.00 μM FBP): Pipette 20 μL of calibration mixture to a vial, add 980 μL of water, and mix the liquid. 8. Standard 5 (7.50 μM PYR, 12.50 μM G6P, 2.50 μM F6P, 2.50 μM PEP, 5.00 μM 3PG, and 25.00 μM FBP): Pipette 50 μL of calibration mixture to a vial, add 950 μL of water, and mix the liquid. 9. Standard 6 (15.00 μM PYR, 25.00 μM G6P, 5.00 μM F6P, 5.00 μM PEP, 10.00 μM 3PG, and 50.00 μM FBP): Pipette 100 μL of calibration mixture to a vial and add 900 μL of water and mix the liquid. 10. Standard 7 (30.0 μM PYR, 50.0 μM G6P, 10.0 μM F6P, 10.0 μM PEP, 20.0 μM 3PG, and 100.0 μM FBP): Pipette 200 μL of calibration mixture to a vial, add 800 μL of water, and mix the liquid. 11. Standard 8 (45.0 μM PYR, 75.0 μM G6P, 15.0 μM F6P, 15.0 μM PEP, 30.0 μM 3PG, and 150.0 μM FBP): Pipette 300 μL of calibration mixture to a vial, add 700 μL of water, and mix the liquid. 12. Standard 9 (67.5 μM PYR, 75.0 μM G6P, 22.5 μM F6P, 22.5 μM PEP, 45.0 μM 3PG, and 225.0 μM FBP): Pipette 450 μL of calibration mixture to a vial, add 550 μL of water, and mix the liquid. 13. Standard 10 (90.0 μM PYR, 150.0 μM G6P, 30.0 μM F6P, 30.0 μM PEP, 60.0 μM 3PG, and 300.0 μM FBP): Pipette 600 μL of calibration mixture to a vial, add 400 μL of water, and mix the liquid. 14. Fill LC injection vials with 80 μL of standard or sample, add 20 μL of labeled internal standard, and mix with a vortex mixer.

Analysis of Glycolytic Intermediates

LC

AS11 column

MS

Ion suppressor MeOH 80%

137

Waste

Fig. 7.1. General set up of the LC-MS system.

pump

LC

2

solvent selector

LC

1 10

3

9 8

4 5 6 7

9 8

4

MS

solvent selector

1 10

3

waste AS11

(A)

2 guard

pump

guard

5 6 7

waste AS11

MS

(B)

Fig. 7.2. Valve switching positions A and B. In position A the sample is transferred to the analytical column (AS11) and in position B the guard column is cleaned and the chromatography performed.

3.2. IC-MS/MS

This section describes the anion exchange IC-MS/MS method for the glycolytic intermediates and is an adaptation of the method described by van Dam et al. (1). Figure 7.1 shows the general setup of the IC-MS/MS method and Fig. 7.2 gives detailed information for the column switching. The ten-port valve and the solvent selector are switched with event switches in the LC. The tubing used to connect the parts in the setup should be as short as possible to avoid extra dispersion of the intermediates in the liquid flow, which results in peak broadening and thus lower peaks and higher detection limits.

3.2.1. Ion Chromatography

1. Set the LC flow rate to 1 mL/min with the following initial conditions: A: 95%, B: 5% (see Note 4), C: 0%, and D: 0%. 2. Set the event switches as follows: event switch 1: On, event switch 2: Off, and event switch 3: On. Event switch 1 switches the ten-port valve and event switches 2 and 3 switch the solvent selector valve. 3. Due to the instability of the standards and samples, the temperature of the sample tray is set to 4◦ C. 4. Set the pump for 80% methanol – located between the AS11 column and the ion suppressor (Fig. 7.1) – to 0.7 mL/min (see Note 5).

138

van Dam, Ras, and ten Pierick

5. Set the pump for the solvent selector (Fig. 7.2) to 0.9 mL/min. 6. Start the three pumps. 7. Switch the ion suppressor on. The ion suppressor removes the sodium ions from the eluate via an electrical current (see Note 6) and replaces the sodium with hydrogen (see Note 7). After the ion suppressor (Fig. 7.1), the effluent is split by a T union, resulting in an approximate flow to the mass spectrometer of 100 μL/min. This flow reduction not only increases the observed sensitivities but also prevents quick contamination of the MS (see Note 8). 8. The chromatography is started with the column switching valve in position A (the guard column is switched in front of the AS11 column, see Fig. 7.2) and the solvent selector giving a 5% methanol solution. 9. Inject 10 μL of sample or standard in the liquid chromatograph. 10. At 1.5 min, switch the guard column to position B with event switches as follows: event switch 1: Off, event switch 2: On, and event switch 3: Off. Set the solvent selector to 80% methanol. 11. At 1.51 min, set the LC solvents to A: 95%, B: 0%, C: 5%, and D: 0%. 12. Start the sodium hydroxide gradient from 1.51 min to 17.00 min, with an exponential curve (curve 7) and to the solvent settings of A: 75%, B: 0%, C: 25%, and D: 0%. The exponential gradient improves the separation of the isomers of G6P, which are F6P, glucose-1phosphate, fructose-1-phosphate, mannose-1-phosphate, and mannose-6-phosphate. 13. From 17.00 to 22.00 min, use a linear gradient from A: 75%, B: 0%, C: 25%, and D: 0% to A: 0%, B: 0%, C: 100%, and D: 0%. This linear gradient separates PEP, 2PG+3PG, and FBP. 14. At 22.00 min, switch the solvent selector to 5% methanol with event switch 2: Off and event switch 3: On. The guard column is now reconditioned (see Note 9). 15. From 22.00 to 23.00 min, use a linear gradient from A: 0%, B: 0%, C: 100%, and D: 0% to A: 20%, B: 0%, C: 0%, and D: 80%. 16. From 23.00 to 26.00 min, maintain the conditions A: 20%, B: 0%, C: 0%, and D: 80%. This is the AS11 column cleaning step (see Note 10). 17. From 26.00 to 27.00 min, return to initial conditions using a linear gradient: A: 95%, B: 5%, C: 0%, and D: 0%.

Analysis of Glycolytic Intermediates

139

18. From 27.00 to 34.00 min, the AS11 column is reconditioned. 19. At 34.00 min, switch the guard column to position A with event switch 1: On. The LC is ready for a new injection. 3.2.2. Mass Spectrometry

1. Set the capillary voltage to –2.70 kV for negative ionization. 2. Set the source block temperature to 90◦ C. 3. Set the desolvation temperature to 250◦ C. 4. Set the nebulation gas flow to 69 L/h. 5. Set the desolvation gas flow to 620 L/h. 6. The data acquisition for the intermediates is grouped into functions. Each function contains several channels, and there is an inter-channel delay of 0.03 s between each channel. The functions used in this method are described below. 7. Function 1. Start time: 0.00 min. End time: 7.50 min. 8. PYR: MS1: 86.8 m/z, MS2: 42.8 m/z, dwell time: 0.50 s, cone (V): 18, and collision energy (eV): 12. 9.

13 C-PYR:

MS1: 89.8 m/z, MS2: 45.8 m/z, dwell time 0.50 s, cone (V): 18, and collision energy (eV): 12.

10. Function 2. Start time: 7.5 min. End time: 19.00 min. 11. G6P and F6P: MS1: 259.0 m/z, MS2: 96.8 m/z, dwell time 0.50 s, cone (V): 20, and collision energy (eV): 18. 12.

13 C-G6P

and 13 C-F6P: MS1: 265.0 m/z, MS2: 96.8 m/z, dwell time 0.50 s, cone (V): 20, and collision energy (eV): 18.

13. Function 3. Start time: 19.00 min. End time: 25.00 min. 14. PEP: MS1: 167.0 m/z, MS2: 78.7 m/z, dwell time 0.25 s, cone (V): 18, and collision energy (eV): 20. 15.

13 C-PEP:

MS1: 167.0 m/z, MS2: 78.7 m/z, dwell time 0.25 s, cone (V): 18, and collision energy (eV): 20.

16. 2PG + 3PG: MS1: 185.0 m/z, MS2: 78.7 m/z, dwell time 0.25 s, cone (V): 20, and collision energy (eV): 20. 17.

13 C-2PG

+ 13 C-3PG: MS1: 188.0 m/z, MS2: 78.7 m/z, dwell time 0.25 s, cone (V): 20, and collision energy (eV): 20.

18. Function 4. Start time: 0.00 min. End time: 25.00 min. 19. H34 SO4 : MS1: 98.8 m/z, MS2: 81.8 m/z, dwell time 0.02 s, cone (V): 40, and collision energy (eV): 20 (see Note 11). From the MRM settings, the most abundant fragment of the parent ion is used. It is clear that the most abundant fragments are mostly produced by the phosphate groups. Although this does

140

van Dam, Ras, and ten Pierick

not give selectivity, it works fine. Mass 97 m/z is the phosphate group and mass 79 m/z is the phosphate group after losing water. These fragments do not contain carbon atoms, so there will be no 13 C-labeled fragments in this case. In contrast, the pyruvate fragment is labeled giving rise to different masses for the 12 C and the 13 C intermediate fragments (see Note 12). Figure 7.3 shows representative extracted ion chromatograms (EICs) of the glycolytic intermediates using the described method. The two bottom chromatograms demonstrate the coelution of FBP and its internal reference standard (the 13 C isotopomer); all the other reference standards are measured but are not shown. The EIC of HSO4 − is added as an example of a coeluting salt, since many biological samples contain high concentrations of non-volatile salts or buffers, which may cause ionization suppression or contaminate the ion source and the mass analyzer. Therefore, non-volatile additives and buffers should be replaced by those that are volatile and MS-compatible. Aside from

%

100

4.37

PYR

0 4.00

6.00

8.00

10.00

12.00

100

14.00 16.00 15.04

%

HSO4

18.00

20.00

22.00

18.00

20.00

22.00

18.00

20.00

22.00 21.28

–

0 4.00

6.00

8.00

10.00

12.00

100

14.00

%

G6P

16.00 15.35

F6P

16.09

0 4.00

6.00

8.00

10.00

12.00

14.00

16.00

100

%

PEP

0 4.00

6.00

8.00

10.00

12.00

14.00

16.00

18.00

20.00

22.00

20.58

100

%

2PG+3PG

0 4.00

6.00

8.00

10.00

12.00

14.00

16.00

18.00

100

20.00

22.00 22.52

%

FBP

0 4.00

6.00

8.00

10.00

12.00

14.00

16.00

18.00

100

20.00

%

13

22.00 22.46

C-FBP

0 4.00

6.00

8.00

10.00

12.00

14.00

16.00

18.00

20.00

22.00

Time (min.)

Fig. 7.3. IC-MS-MS chromatogram showing extracted ion chromatograms of PYR, sulfate, G6P, and F6P (second peak; the third peak is mannose-6-phoshate), PEP, 2PG+3PG, FBP, and 13 C-labeled FBP.

Analysis of Glycolytic Intermediates

141

causing ionization suppression, high concentrations of sulfate or other salts can influence the ion chromatography due to overloading of the column. Based on the data shown in Fig. 7.3, it is clear that G6P will be easily influenced by a sulfate overloaded column. The peak shape in that case is no longer symmetrical and a retention time shift is also observed. Due to the high sulfate concentrations, ionization suppression also occurs for G6P (see Note 13). 3.3. GC-MS

Prior to derivatization, samples and standards are freeze-dried in order to remove water, because water reacts with the derivatization reagents. Two separate derivatizations – oximation and silylation – are performed prior to GC-MS analysis to increase the volatility of metabolites. In the oximation step, methoxyamineHCl reacts with the carbonyl groups of ketones and aldehydes as follows: −C=O + CH3 ONH2 → C=N−O−CH3 This reaction protects the functional groups of ketones and aldehydes and reduces the number of sugar isomers. In the silylation step, MSTFA (X−Si(CH3 )3 ) reacts with hydroxyl (–OH) and thiol (–SH) groups (in acids, amino acids, and alcohols) to form trimethylsilyl (TMS) groups on the oxygens or sulfurs: −OH + X−Si(CH3 )3 → −O−Si(CH3 )3 + HX The final derivatives are indicated as TMS-MOX derivatives. During the derivatization process, two oximated isomers are generated: syn and anti. The heights of the two isomer peaks always appear in the same ratio in the GC-MS chromatograms independent of the concentration (2). Safety precautions should be considered throughout the derivatization procedure. Use a fume hood, safety glasses, and gloves. During sample handling, open the GC vials only when they have reached room temperature and add reagents only when they are at room temperature, since moisture will react with reagents and can influence your derivatization results.

3.3.1. Derivatization

1. Fill a glass GC vial with 100 μL of sample or standard. 2. Add 20 μL of 13 C-labeled internal standard to the vial. 3. Add 20 μL of 40 μM erythritol to each GC vial for quality control of GC-MS performance (see Note 14). 4. Close the GC vials with a screw cap containing a Teflon seal. 5. Freeze the samples at –80◦ C.

142

van Dam, Ras, and ten Pierick

6. Remove the samples from the freezer after 15 min and puncture the Teflon seal twice with a needle (e.g., injection needle). 7. Freeze-dry the samples overnight. 8. Replace the punctured Teflon screw cap with a screw cap of Teflon containing natural rubber (see Note 15). Samples and standards can easily be stored when they are dry in a –80◦ C freezer. 9. Resuspend the dried samples in 50 μL of 240 mM methoxyamine-HCl in pyridine (see Note 16). 10. Place the GC vials in an aluminum heating block and incubate for 50 min at 65◦ C. Due to heating, the contents of the GC vials are mixed. 11. Remove the vials from the heating block. Cool the samples to room temperature and add 80 μL of MSTFA reagent to each standard or sample. Fresh solutions of MSTFA should be used for the silylation. 12. Incubate the GC vials again for 50 min at 65◦ C in the heating block. 13. Cool the samples to room temperature and transfer the contents of the GC vials to glass vials that can be centrifuged. 14. Centrifuge the glass vials for 1 min at 10,000×g. 15. Transfer the samples to glass inserts and place the inserts back in the previously used GC vials and close the vial with a rubber/Teflon septum. 16. For duplicate analyses, fill two inserts and put them in separate GC vials. Samples are stable for about 5 days. The contents of punctured vials will degrade soon. Ideally, the samples should be analyzed immediately after preparation. 3.3.2. GC-MS Settings

1. Operate the programmed temperature vaporization (PTV) in PTV SL mode. 2. Set the purge flow rate to 200 mL/min. 3. Set the helium carrier gas flow rate to 1 mL/min. 4. Set the temperature of the PTV to 70◦ C with a splitless time of 0.7 min. 5. Prior to injection, rinse the needle of the injector three times with isooctane. 6. After the injection of 1 μL sample or standard, raise the temperature of the PTV to 220◦ C at 10◦ C/s and hold for 5 min, then raise the temperature to 300◦ C at 14.5◦ C/s to clean the injector.

Analysis of Glycolytic Intermediates

143

7. Program the temperature gradient as follows: 70◦ C for 1 min, ramp to 76◦ C at 1◦ C/min, ramp to 150◦ C at 10◦ C/min, and finally ramp to 320◦ C at 30◦ C/min. 8. Rinse the needle of the injector five times with toluene. 9. Set the temperature of the transfer line from the GC to the MS and the ion source at 250 and 280◦ C, respectively. 10. Operate the MS in electron impact mode (EI) with 70 eV. 11. Set the mass resolution to 1 mass unit throughout the mass range of 1–1050 amu. 12. For quantitative measurements, operate the MS in selected ion monitoring (SIM) mode. 13. MS masses (m/z) for SIM analysis: GAP: 328, 13 C-GAP: 331, DHAP: 400, 13 C-DHAP: 403, 2PG: 459, 13 C-2PG: 462, 3PG: 357, 13 C-3PG: 359, G6P: 357, 13 C-G6P: 359, F6P: 217, and 13 C-F6P: 220. Due to the natural labeling of each fragment, there is a contribution of this labeling to the 13 C fragment. The 13 C fragment area therefore must be corrected for this contribution in the IDMS method. The higher the mass difference between 12 C and 13 C intermediate fragments, the lower the influence of the natural labeling on the 13 C internal standard signal. The fragments we have selected for the analysis are the optimal fragments for our instrument. Some instruments are more sensitive at higher masses. In that case, it is better to use higher masses, since the background noise is lower at higher masses and lower detection limits can be obtained (see Note 17).

4. Notes 1. The actual purities of the intermediates can be found on the manufacturers’ websites. Reference the lot numbers on the bottles to find the purities. 2. Adding the internal reference standard immediately after the sampling corrects for sample handling and degradation of intermediates during extraction and storage. 3. Store the stock solution in a –80◦ C freezer; –20◦ C is not low enough for long-term storage. The solution is stable for more than 1 year. 4. The methanol in solvent A prevents the growth of microorganisms inside the HPLC and in the tubings. Solvent B increases the pH of the injected sample above 9, so that the acids to be analyzed will be ionized. The XTerra guard column traps mainly proteins, peptides, and lipids at pH > 9. The XTerra guard column will not be damaged at this pH.

144

van Dam, Ras, and ten Pierick

Solvent B has to be replaced regularly (every 3–4 weeks) to reduce the appearance of strange peaks in front of G6P, which influence its peak shape. 5. Methanol improves the ionization in the MS. We have also evaluated the effect of adding methanol after the ion suppressor, but the best results were obtained when methanol was added before the ion suppressor. 6. The electrical current through the ion suppressor has to be optimized for the optimal signal. However, the ion suppressor lifetime is reduced at higher electrical currents. 7. The sodium concentration at the end of the chromatographic separation is 240 mM. If high sodium concentrations are introduced to the MS, one would lose almost all signals due to ionization suppression. The ion suppressor electrically removes the sodium via a membrane, and the hydroxide is neutralized by the hydrogen producing water. 8. Generally speaking, a mass spectrometer remains cleaner if you split the eluent before it enters the MS, because the load of salts and compounds is less. The MS signal arising from electrospray ionization is concentration dependent but not amount dependent, so eluent splitting does not influence the analysis result (if the flow to the MS is >10 μL/min). 9. There is an original guard column, AG11, from Dionex, but this guard column does not protect the analytical column longer than 8–10 samples when analyzing intracellular samples. After 80 sample injections, the analytical column is usually dirty and needs to be cleaned with a special procedure: fill the column with 1 M NaOH and store for 24 h at 40◦ C; wash the column with water and repeat the procedure with 1 M HCl. By using the XTerra guard column, hundreds of samples can be injected without clear performance loss of the AS11 column. 10. The concentration of sodium hydroxide should not be excessively high for a long time, because this will reduce the lifetime of the ion suppressor. On the other hand, a high concentration of sodium hydroxide is good for cleaning the analytical column. Otherwise FBP shows too much memory effect on the analytical column. The duration of this cleaning step does not significantly reduce the lifetime of the ion suppressor. 11. This molecular ion of sulfate contains the natural labeled sulfur isotope: 34 S. The reason for using this isotope is the overload of the MS detector for the 32 S sulfate. 34 S is 4% of 32 S. Note that the total dwell time in a function is ∼ 1 s. So there is one data point per second for each MRM transition.

Analysis of Glycolytic Intermediates

145

12. In analytical methods with complex samples, one prefers more selectivity, but, due to separation of isomeric intermediates, one can only expect non-isomeric intermediates to coelute. We have evaluated this and have not found any evidence for interferences. 13. All of our samples contain sulfate. The source of sulfate is the medium applied in a fermentation process. The replacement of sulfate with chloride was not successful since it produced memory effects in the LC column and suppressed ionization of all intermediates. 14. Erythritol is added as a standard to correct for the injection volume in the GC. Erythritol does not correct for discrimination in the injector. In contrast, IDMS does correct for discrimination in the injector. Note that if erythritol is a compound in your sample matrix – as it is in Penicillium chrysogenum – you cannot use it as internal standard. 15. For GC sample bottles, do not use silicon rubber instead of natural rubber caps, because silicon rubber reacts easily with the derivatization reagents. This can lead to sticky septa, a dirty needle, or even a bent needle. Storage of these dried samples has only been evaluated at –80◦ C, which did not show degradation. 16. Store the derivatization agents in a dry place. We store them in a desiccator at 4◦ C. 17. General remarks. The liner in the GC system has to be replaced (cleaned) regularly. The frequency is strongly dependent on the sample type. Some laboratories replace the liner after each injection due to the analysis of very dirty samples. In our laboratory, we mainly analyze intracellular samples – dirty samples – and we replace the liner after 30 sample injections. The MS (DSQ) which we use for the analysis is known to have fouling problems with TMS-MOX derivatives, which is observed during tuning. The design of the ion source is the main source for the fouling. Newer instruments have a better design. Check the performance of the column regularly (the frequency depends on the type of samples). Testing with a standard is in most cases sufficient. Due to dirty samples, the first part of the column becomes dirty, which is mostly seen as tailing of peaks or missing peaks in the chromatogram. The first part of the column can be cutoff and the performance can be checked. If this is not helpful, then replace the column. One of the major problems in the derivatization is moisture. If samples or solutions are coming out of the fridge or freezer, moisture can condense on the outside. Wait until they are at room temperature before opening.

146

van Dam, Ras, and ten Pierick

References 1. van Dam, J. C., Eman, M. R., Frank, J., Lange, H. C., van Dedem, G. W. K., Heijnen, S. J. (2002) Analysis of glycolytic intermediates in Saccharomyces cerevisiae using anion exchange chromatography and electrospray ionization with tandem mass spectrometric detection. Anal Chim Acta 460, 209–218. 2. Cipollina, C., ten Pierick, A., Canelas, A. B., Seifar, R. M., van Maris, A. J. A., van Dam, J. C., nd Heijnen, J. J. (2009) A comprehensive method for the quantification of the nonoxidative pentose phosphate pathway intermediates in Saccharomyces cerevisiae by GCIDMS. J Chromatogr B 877, 3231–3236.

3. Canelas, A. B., Ras, C., ten Pierick, A., Seifar, R., van Dam, J. C., van Gulik, W. M., Heijnen, J. J. (2009) Quantitative evaluation of intracellular metabolite extraction techniques for microbial metabolomics. Anal Chem 81, 7379–7389 4. Mashego, M. R., Wu, L., van Dam, J. C., Ras, C., Vinke, J. L., Van Winden, W. A., Van Gulik, W. M., Heijnen, J. J. (2004) MIRACLE: mass isotopomer ratio analysis of U-C-13-labeled extracts. A new method for accurate quantification of changes in concentrations of intracellular metabolites. Biotechnol Bioeng 85, 620–628.

Chapter 8 Analysis of the Citric Acid Cycle Intermediates Using Gas Chromatography-Mass Spectrometry Rajan S. Kombu, Henri Brunengraber, and Michelle A. Puchowicz Abstract Researchers view analysis of the citric acid cycle (CAC) intermediates as a metabolomic approach to identifying unexpected correlations between apparently related and unrelated pathways of metabolism. Relationships of the CAC intermediates, as measured by their concentrations and relative ratios, offer useful information to understanding interrelationships between the CAC and metabolic pathways under various physiological and pathological conditions. This chapter presents a relatively simple method that is sensitive for simultaneously measuring concentrations of CAC intermediates (relative and absolute) and other related intermediates of energy metabolism using gas chromatography-mass spectrometry. Key words: Citric acid cycle, CAC intermediates, GC-MS, metabolomics, mass spectrometry.

1. Introduction In mammalian cells, the citric acid cycle (CAC) is a series of enzyme-catalyzed chemical reactions that result in the generation of reducing equivalents (NADH/NAD+ , FADH2 /FADH) that fuel oxidative phosphorylation (mitochondrial cellular respiration), as well as carbon dioxide (1). The CAC links key metabolic pathways, such as glycolysis (2), gluconeogenesis (3, 4), and anaplerosis/cataplerosis (5–7), and provides precursors for many compounds including fatty acids and some amino acids. This chapter presents a qualitative method that is sensitive for simultaneously measuring concentrations of CAC intermediates (relative and absolute) and other related intermediates of energy metabolism using gas chromatography-mass spectrometry (GC-MS) and stable isotope technologies. This approach has T.O. Metz (ed.), Metabolic Profiling, Methods in Molecular Biology 708, DOI 10.1007/978-1-61737-985-7_8, © Springer Science+Business Media, LLC 2011

147

148

Kombu, Brunengraber, and Puchowicz

been utilized for over two decades in laboratories conducting classical metabolomics, often in parallel with other biochemical analyses that use GC-MS techniques (2, 8, 9). With major advancements in GC-MS technology (increased sensitivity and throughput), database information from the National Institute of Standards and Technology (NIST), and cost-effectiveness, this approach has become more popular to researchers and thus prompted a new generation of scientists interested in metabolomics as a tool for the identification of unexpected correlations between apparently related and unrelated pathways of metabolism (10, 11). We describe a quantification method for assaying concentration profiles of (water-soluble) CAC metabolites as measured against a mixture of stable isotope reference compounds, such as [U-13 C6 ]citrate, [U-13 C4 succinate], and 3-hydroxy[2 H4 ]glutarate (12). Briefly, our approach utilizes the rapid reaction of silylating reagent with alcohols, acids, and amines to form silyl derivatives. Commercially, silylating reagents are available as combinations to accelerate the reaction, as well as to react with the hindered group. We use either a mixture of N,O-bis(trimethylsilyl)trifluoroacetamide (BSTFA) and trimethylchlorosilane (TMCS) to form a TMS derivative or a mixture of N-Methyl-N-(t-butyldimethylsilyl)trifluoroacetamide (MTBSTFA) and t-butyldimethylchlorosilane (TBDMS) to form a TBDMS derivative (see Note 1). The resultant TMS and TBDMS derivatives of these metabolites/intermediates can be performed on plasma, tissue extracts, organ perfusates, or cell media preparations. As previously described (slightly modified method) (12), the assay involves (i) spiking the sample with one or more internal standards, (ii) homogenization with Folch’s solvent or an acid–methanol mixture and then centrifugation, (iii) evaporation of extract and derivatization with either trimethylor t-butyldi-silyl, and (iv) identification of the m/z of each analyte using ammonia-positive chemical ionization (CI) or electron impact ionization (EI) GC-MS. To analyze compounds containing “keto groups,” such as α-ketoglutarate and oxaloacetate, we also describe derivatization of TMS-methoxylamine, which requires a few additional preparatory steps. This chapter does not present mass isotopomer distribution analysis (MIDA) or the use of labeling patterns of intermediates to determine fluxes or turnover, as this approach often requires the use of low isotopomer background derivatizing agents (3, 4).

2. Materials 2.1. Chemicals and Materials

1. Chloroform. 2. Methanol.

Analysis of the Citric Acid Cycle Intermediates

149

3. Glacial acetic acid (≥99.7%). 4. Selected internal standards: [13 C6 ]citric and [13 C4 ]succinic acids (98%) (Isotec, Miamisburg, OH); (RS)-3-hydroxy[2 H4 ]glutarate was prepared as previously described (13). R 5. Regisil + 10% TMCS (TMS; Regis Technologies, Morton Grove, IL). Store in a refrigerator (4◦ C) in a desiccator.

6. N-Methyl-N-(t-butyldimethylsilyl)trifluoroacetamide (MTBSTFA + t-BDMCS; TBDMS; Regis Technologies, Morton Grove, IL). Store in a refrigerator (4◦ C) in a desiccator. 7. 10 N sodium hydroxide solution. Store at room temperature. 8. Methoxylamine–HCl solution (MOX; 100 mM): To prepare MOX solution, weigh 83.52 mg of methoxylamine hydrochloride in a conical tube and add 10 mL of water. Adjust the pH to 8–10 using 10 N sodium hydroxide solution (2). Prepare MOX solution fresh on the day of analysis (Caution: corrosive; avoid contact with skin and eyes; avoid breathing; and wear protective clothing) (see Note 2). 9. Chloroform/methanol mixture (2:1, v/v). This mixture is stable at room temperature and can be used for 1 month. This mixture is used in extraction method A, which is slightly modified from Yang et al. (12). 10. Methanol/water mixture (3:2, v/v). This mixture is stable at room temperature and can be used for 1 month. 11. Acid–methanol mixture: 5% acetic acid in methanol/water (1:1, v/v). This solution is stable at room temperature and can be used for 1 month. This mixture is used in extraction method B (14, 15). 12. Heating block capable of heating up to 70◦ C. 13. Turbovap evaporator (Caliper Life Sciences, Hopkinton, MA). 14. Speedvac vacuum Bellefonte, PA).

concentrator

(Thermo

Scientific,

15. Omni General Laboratory Homogenizer (Kennesaw, GA). 16. Disposable centrifuge tube, polypropylene (15 mL; Fisher Scientific, Fair Lawn, NJ). 17. Disposable culture tube, borosilicate glass (13 × 100 mm; Fisher Scientific). 18. Wide opening glass crimp vial (2 mL; Agilent Technologies, Santa Clara, CA). 19. Glass vial insert (250 μL; Agilent Technologies). 20. Aluminum crimp cap with PTFE/rubber septa (11 mm; Agilent Technologies).

150

Kombu, Brunengraber, and Puchowicz

2.2. GC-MS

1. Mass spectrometer: Agilent 5973 mass spectrometer, linked to a model 6890 gas chromatograph equipped with an autosampler (Agilent Technologies, Santa Clara, CA). 2. Carrier gas: helium (1 mL/min) with a pulse pressure of 40 p.s.i.g. 3. Column: VF-5MS EZ guard capillary column (60 m × 0.25 mm inner diameter) with guard column (10 m) (Varian Technologies, Palo Alto, CA). 4. Chemical ionization gas (CI mode): Ammonia

3. Methods 3.1. Tissue Preparation

3.1.1. Method A: Chloroform/Methanol Extraction Procedure

The tissues are extracted by two methods, either using chloroform/methanol (method A) (12) or using 5% acetic acid in methanol/water (method B) (14, 15). Method A is advantageous for simultaneously analyzing CAC intermediates, as well as cholesterol and fatty acids from the same tissue as the chloroform phase contains these lipids. Method B is specifically useful for CAC intermediates as it completely extracts these water-soluble acids from the tissue. Both methods can be used on tissues such as liver, brain, muscle, and kidney. 1. Weigh about 0.2–0.5 g of powdered frozen (–80◦ C) tissue in a pre-weighed/tarred, 15 mL disposable centrifuge tube previously chilled over dry ice and record the tissue weight. (Caution: dry ice is extremely cold and can cause frost bite. Wear protective gloves and goggles while handling.) 2. After weighing, spike the powdered frozen tissue with the selected internal standards (~50 nM of [13 C6 ]citrate, ~30 nM of [13 C4 ]succinate, and ~30 nM of (RS)-3-hydroxy[2 H4 ]glutarate; see Note 3). 3. Using an Omni homogenizer, homogenize the tissue with 6 mL of pre-cooled (–25◦ C) chloroform/methanol (2:1, v/v) for 5 min on a dry ice–acetone bath (caution: dry ice is extremely cold and can cause frost bite. Wear protective gloves and goggles while handling) (see Note 4). 4. To break the phases, add 2 mL of ice-cold water to the same tube and re-homogenize for 5 min at –25◦ C. 5. Centrifuge the slurry at 670×g for 20 min at 4◦ C. 6. Collect the upper methanol/water phase in a disposable culture tube; if necessary, filter to remove tissue fragments. 7. For complete extraction of analytes, a second extraction is suggested: to the lower organic chloroform phase, add

Analysis of the Citric Acid Cycle Intermediates

151

3 mL of pre-cooled (–20◦ C) methanol/water (3:2, v/v) and vortex for 5 min. 8. Repeat steps 5 and 6 (see Note 5). 9. Combine the two upper methanol/water phase extracts obtained from steps 6 and 8. 10. To the combined extract, add methoxylamine–HCl solution to a final concentration of 200 μM (see Note 6). 11. Adjust the pH to 8.0 with 10 N sodium hydroxide to complete the derivatization process. 12. Evaporate the extract completely under vacuum overnight using a Savant vacuum centrifuge (see Note 7). 13. Add 100 μL of TMS (or TBDMS) to the completely dried residue using a transfer pipette (caution: TMS and TBDMS derivatizing agents are volatile, toxic, carcinogenic, and corrosive. Use protective measures like gloves and hood) (see Notes 8 and 9). 14. Heat the mixture at 70◦ C for 50 min on a heating block; this process yields TMS (or TBDMS) and methoxamates/TMS (or TBDMS) derivatives of the analytes (see Note 10). 15. Allow the sample to cool to room temperature (caution: avoid moisture). 16. Transfer the derivatized mixture to a wide open glass crimp vial with glass vial insert and cap tightly. 17. Inject the mixture on to the GC-MS for analysis (see Note 11; see Section 3.3). 18. For quantification of analytes, prepare stock solutions with known amounts of standards in water. Prepare a serial dilution. Treat the standards the same way the samples are treated. The amounts of selected internal standards need to be the same as that added to the samples. This enables the amounts of analytes in the sample to be quantified as absolute or relative amounts. For the derivatization procedure, follow Section 3.1.2, steps 10–17. 19. Plot a standard curve of the peak areas (standards against internal standards, relative to known amounts of the added standards) (12). 20. Calculate the absolute or relative concentrations of analytes of interest as identified during electron ionization runs (see Section 4). 3.1.2. Method B: Acidified Methanol/Water Extraction Procedure

1. Weigh about 0.2–0.5 g of powdered frozen (–80◦ C) tissue in a pre-weighed/tarred, 15 mL conical “homogenizing” tube previously chilled over dry ice and record the tissue

152

Kombu, Brunengraber, and Puchowicz

weight. (Caution: dry ice is extremely cold and can cause frost bite. Wear protective gloves and goggles while handling.) 2. After weighing, spike the powdered frozen tissue with the selected internal standards (~50 nM of [13 C6 ]citrate, ~30 nM of [13 C4 ]succinate, and ~30 nM of (RS)-3-hydroxy[2 H4 ]glutarate; see Note 3). 3. Using an Omni homogenizer, homogenize the tissue with 5 mL of 5% acetic acid in methanol/water (1:1, v/v) extraction buffer (chilled on ice) for 2 min on ice bath. 4. Centrifuge the homogenate at 670×g for 30 min at 4◦ C. Decant the supernatant into a glass test tube and save on ice and process immediately or freeze the supernatant at –80◦ C until derivatization procedure (see Notes 12 and 13). 5. For the derivatizing procedure, pipette 100–200 μL of the supernatant collected in step 3 or use all of the supernatant collected in step 3 and follow Section 3.1.1, steps 10–20. 3.2. Plasma and Organ Perfusate Preparation

1. Thaw the plasma or perfusate on ice at 4◦ C. 2. Once thawed, immediately aliquot 0.1 mL of the perfusate or plasma and spike with the selected internal standards (~50 nM of [13 C6 ]citrate, ~30 nM of [13 C4 ]succinate, and ~30 nM of (RS)-3-hydroxy-[2 H4 ]glutarate, see Note 3). The reserve can be stored in a deep freezer at –80◦ C for further analysis. 3. To derivatize analytes containing keto groups, add methoxylamine–HCL solution to a final concentration of 200 μM, cap tightly, and heat the mixture at 60◦ C for 3 h. 4. Then add 1 mL of cold acetonitrile/methanol (7:3, v/v) and vortex for 30 s. 5. Centrifuge at 670×g for 20 min at 4◦ C. 6. After centrifugation, collect the supernatant and follow the procedure in Section 3.1.1, steps 12–20.

3.3. GC-MS Assays

1. Following the sample preparation procedures above (see Section 3.1.1, step 16), the derivatized samples can be analyzed either in CI mode (12) or in EI mode (2). The parameters described here are for CI mode and the parameters for EI mode are given in parenthesis. 2. On the Agilent 5973 mass spectrometer linked to a model 6890 gas chromatograph, set the injector temperature at 270◦ C (EI: 250◦ C) and the transfer line at 280◦ C (EI: 300◦ C). 3. Set the ion source and the quadrupole at 150◦ C.

Analysis of the Citric Acid Cycle Intermediates

153

Table 8.1 Chemical and electron ionization mass spectra of the TMS or TBDMS derivatives of CAC and related intermediates PCI-MOX-TMSa Analyte

Retention time

Succinate Fumarate

m/z to monitor

8.55

280

EI-TMSb

EI-TBDMSb

m/z to monitor

m/z to monitor

247

331, 289, 215

9.04

278

245

329, 287, 245

Oxaloacetate

10.46

323

290

–

Malate

10.72

368

335, 245, 233

419, 403, 287

α-ketoglutarate

11.86

337

304, 275

431, 446

Citrate

14.48

481

465, 375, 347

459, 431, 357

Glutamate

–

–

348, 246, 230

432, 330, 272

Glutamine

–

–

347, 245, 229

431, 329, 271

For the internal standards, the m/z of [13 C6 ]citrate yields M+6 citrate, [13 C4 ]succinic acid yields M+4 succinate, and 3-hydroxy-[2 H4 ]glutarate yields 437, 304, and 260 as TBDMS derivatives. a Retention times correspond to the CI method as described in Section 3.1.1 (12). b For EI methods, the retention times are relative according to the GC gradient program (this method uses hydroxylamine instead of MOX) (2) and the m/z of the major fragments for each analyte are given and confirmed by NIST. The m/z ions are selected based on the spectral characterization (fragmentation pattern) of the particular derivative as well as their intensities. The intensites of these ions may differ depending on the mass spectrometer used.

4. The GC temperature program is as follows: start at 80◦ C, hold for 1 min, increase by 10◦ C/min to 320◦ C, hold at 320◦ C for 5 min. (EI: start at 80◦ C, increase by 5◦ C/min to 250◦ C followed by an increase by 50◦ C/min to 300◦ C, hold at 300◦ C for 5 min). 5. Adjust the ammonia pressure to optimize peak areas. 6. The retention times and m/z values monitored using the CI method are listed in Table 8.1 and Fig. 8.1.

3.4. Calculations

1. The raw mass spectrometric areas are used to calculate the analytical parameter (area of analyte)/(area of reference compound). This parameter is not a relative concentration; thus, to calculate relative concentrations, use the following equation: Relative concentration = average[(area of analyte)/ (area of reference compound)]i /average [(area of analyte)/ (area of reference compound)]c where i is the intervention/experimental group and c is the control group.

154

Kombu, Brunengraber, and Puchowicz

Fig. 8.1. Positive chemical ionization chromatograms of metabolites extracted from a liver perfused with 5 mM lactate (adapted from Yang et al. (12)). In addition to m/z spectra, the retention times and relative peak heights are also used to characterize and reference each metabolite. For example, apart from the analytes listed in Table 8.1, the TMS derivatives of related metabolites such as glycine, aspartate, glucose, pyruvate, phosphoenolpyruvate, dihydroxyacetone phosphate, glyceraldehyde-3-phosphate, 2-phosphoglycerate, 3-phosphoglycerate, fructose-6-phosphate, and glucose6-phosphate can be simultaneously analyzed by introducing appropriate ions and retention times in the scan table of the mass spectrometer software (panel A: 1 μL split 10:1; panel B: 2 μL splitless, see Note 13).

2. Absolute concentrations are calculated against a standard curve that contains the same amount of internal standard(s) as added to the samples. Preparation of standard curve: prepare graded amounts of known standards and add a constant amount of internal standard to each of the prepared standards (reference compound). Run the standards on the GC-MS along with the samples under the same conditions. Calculate the ratio of the peak height or area for standards and samples to that of the internal standard. Using the standard curve, the concentrations per sample can be calculated

Analysis of the Citric Acid Cycle Intermediates

155

(9). The mass spectrometer software has the function to build the calibration curves and calculate the concentrations based on the input to the software.

4. Notes 1. The advantage of TBDMS derivatization is that these derivatives are more stable to moisture and reduce the risk of enol silyl ether formation (9, 16). The advantage of the trifluroacetamide groups in these reagents is their volatility, low reactivity, and specificity to react only with alcohols when there is a carboxyl group in the same molecule (16). 2. Instead of MOX, hydroxylamine can be substituted. 3. The estimated amount of internal standards used depends on tissue type. For example, α-ketoglutarate content is greater in liver than in brain. For accuracy, it is very important to add internal standards at the initial processing step, otherwise volume additions or losses would need to be accounted for and the final concentration calculations adjusted. 4. During the 5-min extraction, the tube is partially immersed in acetone kept at –25◦ C by periodic addition of dry ice. 5. The chloroform phase can be used for analyzing fatty acids and cholesterol (12). 6. This step is to protect/stabilize the keto group by derivatizing with MOX. Steps 10 and 11 can be omitted if the analytes of interest do not contain keto groups, as this step is essential for stabilizing the keto group. 7. To shorten the processing time, heat at 50◦ C for 3 h and then dry under nitrogen at 50◦ C using a Turbovap. Take care not to overheat as it will lead to degradation of analytes. The reagents in Section 3.1.1, step 13 are highly moisture sensitive and thus the sample should be completely dry before proceeding. 8. Make a small hole in the cap with a 22 gauge needle to avoid pressure buildup and popping of the cap. 9. TMS and TBDMS derivatizing reagents react readily with glass and plastics. While drawing the reagent using the syringe, it may block it. It is better to use a transfer pipette. If a glass syringe is used, wash immediately with methanol and designate the syringe for these reagents. 10. If moisture or alcohol (protic solvent) is present, the reagents will first react with them to form a white residue.

156

Kombu, Brunengraber, and Puchowicz

Avoid opening the reagent bottle frequently and, once opened, do not use it after a month or two. These reagents are volatile; avoid over heating to prevent complete dryness. Switch on the heating block at least an hour earlier before placing the tubes to get the block equilibrated at 70◦ C. 11. The injection volume and split flow may need to be adjusted according to the intensity of the individual analyte of interest. For example, pyruvate is less intense compared to malate, thus requiring more injection volume without a split flow. 12. To ensure the glass test tube does not break upon freezing, make sure that the volume in the test tube is less that 50% of the height and store on a slight angle in –80◦ C freezer. 13. In some procedures, the first fraction is used for other assays such as for acyl-CoA profiles (14, 15).

Acknowledgments This work was supported, in whole or in part, by National Institutes of Health Roadmap Grant R33DK070291 and Grant R01ES013925. This work was also supported by a grant from the Cleveland Mt. Sinai Health Care Foundation. We acknowledge the Mouse Metabolic Phenotyping Center (MMPC) at Case Western Reserve University where many of these procedures were developed. References 1. Krebs, H. A. (1940) The citric acid cycle and the szent-gyorgyi cycle in pigeon breast muscle. Biochem J 34, 775–779. 2. Des, R. C., Fernandez, C. A., David, F., Brunengraber, H. (1994) Reversibility of the mitochondrial isocitrate dehydrogenase reaction in the perfused rat liver. Evidence from isotopomer analysis of citric acid cycle intermediates. J Biol Chem 269, 27179–27182. 3. Yang, L., Kasumov, T., Kombu, R. S., Zhu, S. H., Cendrowski, A. V., David, F., Anderson, V. E., Kelleher, J. K., Brunengraber, H. (2008) Metabolomic and mass isotopomer analysis of liver gluconeogenesis and citric acid cycle: II. Heterogeneity of metabolite labeling pattern. J Biol Chem 283, 21988–21996.

4. Yang, L., Kombu, R. S., Kasumov, T., Zhu, S. H., Cendrowski, A. V., David, F., Anderson, V. E., Kelleher, J. K., Brunengraber, H. (2008) Metabolomic and mass isotopomer analysis of liver gluconeogenesis and citric acid cycle. I. Interrelation between gluconeogenesis and cataplerosis; formation of methoxamates from aminooxyacetate and ketoacids. J Biol Chem 283, 21978–21987. 5. Reszko, A. E., Kasumov, T., Pierce, B. A., David, F., Hoppel, C. L., Stanley, W. C., Des, R. C., Brunengraber, H. (2003) Assessing the reversibility of the anaplerotic reactions of the propionyl-CoA pathway in heart and liver. J Biol Chem 278, 34959–34965.

Analysis of the Citric Acid Cycle Intermediates 6. Brunengraber, H., Roe, C. R. (2006) Anaplerotic molecules: current and future. J Inherit Metab Dis 29, 327–331. 7. Kasumov, T., Sharma, N., Huang, H., Kombu, R. S., Cendrowski, A., Stanley, W. C., Brunengraber, H.. (2009) Dipropionylcysteine ethyl ester compensates for loss of citric acid cycle intermediates during post ischemia reperfusion in the pig heart. Cardiovasc Drugs Ther 23, 459–469. 8. Beylot, M., Soloviev, M. V., David, F., Landau, B. R., Brunengraber, H. (1995) Tracing hepatic gluconeogenesis relative to citric acid cycle activity in vitro and in vivo. Comparisons in the use of [3-13 C]lactate, [2-13 C]acetate, and alpha-keto[3-13c]isocaproate. J Biol Chem 270, 1509–1514. 9. Schwenk, W. F., Berg, P. J., Beaufrere, B., Miles, J. M., Haymond, M. W. (1984) Use of t-butyldimethylsilylation in the gas chromatographic/mass spectrometric analysis of physiologic compounds found in plasma using electron-impact ionization. Anal Biochem 141, 101–109. 10. Weckwerth, W., Fiehn, O. (2002) Can we discover novel pathways using metabolomic analysis?. Curr Opin Biotechnol 13, 156–160. 11. Katz, J., Wals, P., Lee, W. N. (1993) Isotopomer studies of gluconeogenesis and the krebs cycle with 13c-labeled lactate. J Biol Chem 268, 25509–25521.

157

12. Yang, L., Kasumov, T., Yu, L., Jobbins, K., David, F., Previs, S., Kelleher, J., Brunengraber, B. (2006) Metabolomic assays of the concentration and mass isotopomer distribution of gluconeogenic and citric acid cycle intermediates. Metabolomics 2, 85–94. 13. Chang, H. C., Maruyama, H., Miller, R. S., Lane, M. D. (1966) The enzymatic carboxylation of phosphoenolpyruvate. 3. Investigation of the kinetics and mechanism of the mitochondrial phosphoenolpyruvate carboxykinase-catalyzed reaction. J Biol Chem 241, 2421–2430. 14. Deng, S., Zhang, G. F., Kasumov, T., Roe, C. R., Brunengraber, H. (2009) Interrelations between C4 ketogenesis, C5 ketogenesis, and anaplerosis in the perfused rat liver. J Biol Chem 284, 27799–27807. 15. Zhang, G. F., Kombu, R. S., Kasumov, T., Han, Y., Sadhukhan, S., Zhang, J., Sayre, L. M., Ray, D., Gibson, K. M., Anderson, V. A., Tochtrop, G. P., Brunengraber, H. (2009) Catabolism of 4-hydroxyacids and 4-hydroxynonenal via 4-hydroxy-4phosphoacyl-CoAS. J Biol Chem 284, 33521–33534. 16. Watson, D. (1993) Chemical Derivatization in Gas Chromatography. In: Baugh, P., Ed.. Gas Chromatography: A Practical Approach, Oxford University Press, New York, NY, 133–153.

Chapter 9 Quantification of Pentose Phosphate Pathway (PPP) Metabolites by Liquid Chromatography-Mass Spectrometry (LC-MS) Amber Jannasch, Miroslav Sedlak, and Jiri Adamec Abstract The pentose phosphate pathway plays an important role in several cellular processes including biosynthesis and catabolism of five-carbon sugars and generation of reducing power through NADPH synthesis. Although the pentose phosphate metabolic reaction network has been mapped in substantial detail, the comprehensive quantitative analysis of the rates and regulation of individual reactions remains a major interest for various biofields. Here we describe a simple method for comprehensive quantitative analysis of pentose phosphate pathway intermediates. The method is based on Group Specific Internal Standard Technology (GSIST) labeling in which an experimental sample and corresponding internal standards are derivatized in vitro with isotope-coded reagents in separate reactions, then mixed and analyzed in a single LC-MS run. The use of co-eluting isotope-coded internal standards and experimental molecules eliminates potential issues with ion suppression and allows for precise quantification of individual metabolites. Derivatization also increases hydrophobicity of the metabolites enabling their effective separation using reversed-phase chromatography. Key words: Group Specific Internal Standard Technology (GSIST), isotope coding, metabolomics, pentose phosphate pathway (PPP), sugar derivatives.

1. Introduction The pentose phosphate pathway (PPP) is a critical metabolic pathway for generation of NADPH and for fermentation of five-carbon sugars. Overall, 10 intermediates are involved in the PPP (Fig. 9.1). Whilst glucose-6-phosphate, 6-phosphoglucono-δ-lactone, 6-phosphogluconate, and ribulose5-phosphate represent the oxidative phase of the PPP, erythrose4-phosphate, fructose-6-phosphate, glyceraldehyde-3-phosphate, T.O. Metz (ed.), Metabolic Profiling, Methods in Molecular Biology 708, DOI 10.1007/978-1-61737-985-7_9, © Springer Science+Business Media, LLC 2011

159

160

Jannasch, Sedlak, and Adamec

Fig. 9.1. Intermediates involved in pentose phosphate pathway (PPP).

ribose-5-phosphate, sedoheptulose-7-phosphate, and xylulose-5phosphate are part of the non-oxidative phase. Like any other biochemical study, precise quantification of PPP metabolites is the first and key element for understanding of fundamental cellular processes and for developing successful kinetic models. Recently, several methods for comprehensive quantification of PPP intermediates have been published. Most of the methods are based on separation of metabolites using either gas chromatography (GC) (1–3) or liquid chromatography (LC) (4–6) and their detection by mass spectrometry (MS). A major concern with LC-MS quantification is a well-known phenomenon called ion suppression, in which the ion intensities of specific compounds depend on the co-eluting matrix compositions that affect ionization efficiencies of the analytes. Although the use of corresponding stable isotope standards can solve this issue, only limited numbers of these standards are commercially available and “in-house” or outsourced synthesis is usually very expensive and unrealistic. Another problem related to LC-MS methods is that

Quantification of PPP Metabolites by LC-MS

161

sugars are best separated on anion exchange chromatography, requiring MS-incompatible solvents. To circumvent these obstacles, we have developed a simple method based on Group Specific Internal Standard Technology (GSIST) labeling (7–9). In GSIST, an experimental sample and corresponding internal standards are derivatized in vitro with isotope-coded reagents in separate reactions and then mixed and analyzed in single LC-MS run. Because both the internal standard and corresponding molecule from the experimental sample co-elute, their ionization efficiency is the same, and precise concentrations can be calculated for the ratio between light (sample) and heavy (internal standard of known concentration) forms of the metabolite. Derivatization of sugars also changes the physical properties of the molecules and allows for their effective separation using MS-compatible reversed-phase chromatography (6).

2. Materials 2.1. Cell Growth and Extraction

1. Glucose solution: 50% (w/v) glucose in Milli-Q water. 2. YPD medium: 10 g Bacto yeast extract (BD Diagnostic Systems, Sparks, MD), 20 g Bacto peptone (BD Diagnostic Systems), 40 mL glucose solution, and 960 mL Milli-Q water. Glucose solution must be autoclaved separately. 3. Quenching solution: 60% (v/v) analytical grade methanol; 70 mM HEPES, pH 7.5. 4. Cell wash solution: 60% (v/v) analytical grade methanol. 5. Ethanol solution: 75% (v/v) ethanol in Milli-Q water. 6. Shaker (New Brunswick, Edison, NJ). 7. Nephelo flask with three baffles (300 mL) equipped with a sidearm (BellCo Glass, Vineland, NJ). 8. Klett colorimeter clinical model 800-3 equipped with red filter KS-66 (Scienceware Bel-Art Products, Pequannock, NJ). 9. 50 mL Oak Ridge centrifugation tube FEP (Nalgene, Rochester, NY). 10. Cryostat bath HAAKE Phoenix II P1 (Thermo Fisher Scientific, Waltham, MA). 11. AVANTI J-30I centrifuge equipped with JA-25.50 rotor (Beckman Coulter, Brea, CA). 12. Two microprocessor-controlled 280 Precision Water bathes (Thermo Fisher Scientific, Waltham, MA).

162

Jannasch, Sedlak, and Adamec

2.2. Sample and Standard Derivatization

1. Eppendorf tubes (VWR International, West Chester, PA). 2. Light labeling reagent: Prepare 6 M 12 C6 -aniline solution (pH 4.5) by mixing 5.500 mL of aniline with 2.250 mL of water and 2.250 mL of concentrated hydrochloric acid (HCl). Vortex well and measure pH. If necessary adjust pH with HCl. The solution can be stored at 4◦ C for up to 2 months. 3. Heavy labeling reagent: Prepare 6 M 13 C6 -aniline solution (pH 4.5) by mixing 250 mg of 13 C6 -aniline (Cambridge Isotopes, Andover, MA) with 88 μL of water and 88 μL of concentrated hydrochloric acid (HCl). Vortex well and measure pH. If necessary adjust pH with HCl. The solution can be stored at 4◦ C for up to 2 months. 4. EDC solution: N-(3-dimethylaminopropyl)-N ethylcarbodiimide hydrochloride (200 mg/mL; SigmaAldrich, St. Louis, MO) dissolved freshly in Milli-Q water (see Note 1). 5. Triethylamine (TEA). 6. Standards: 10 mM stock solutions in Milli-Q water: a. Erythrose-4-phosphate (E4P, ≥61%; Sigma-Aldrich) b. Fructose-6-phosphate (F6P, ≥98%; Sigma-Aldrich) c. Glucose-6-phosphate (G6P, 100%; Sigma-Aldrich) d. Glyceraldehyde-3-phosphate (GAP, 260–320 mM; Sigma-Aldrich) e. 6-Phosphogluconate (6PG, ≥98%; Sigma-Aldrich) f. Ribose-5-phosphate (R5P, ≥99%; Sigma-Aldrich) g. Ribulose-5-phosphate (RBU5P, ≥96%; Sigma-Aldrich) h. Sedoheptulose-7-phosphate (S7P, ≥98%; Glycoteam GMbH, Germany) i. Xylulose-5-phosphate (X5P, ≥90%; Sigma-Aldrich) 7. Standard test solutions: Dilute individual standard stock solutions 100× with Milli-Q water (0.1 mM final concentration). 8. Internal standard (IS) working solution: Combine 5 μL of each standard stock solution and 955 μL of Milli-Q water (final concentration CS = 50 μM each) (see Note 2).

2.3. LC-MS Procedure

1. LC-MS system: 1100 series HPLC (Agilent Technologies, Santa Clara, CA) coupled to an Agilent MSD-TOF (time of flight) mass spectrometer (Agilent Technologies) (see Note 3). 2. Analytical column: Zorbax Eclipse XDB-C8 (2.1 × 150 mm, 3.5 μm particle size; Agilent Technologies). 3. Solvent A: 5 mM tert-butylamine (TBA) aqueous solution adjusted to pH 5.0 with acetic acid. 4. Solvent B: 5 mM TBA in 100% acetonitrile.

Quantification of PPP Metabolites by LC-MS

2.4. Metabolite Quantification

163

1. Agilent MassHunter Workstation software – Qualitative Analysis v. B01.03 (Agilent Technologies) (see Note 3). 2. SigmaPlot 11.0 or any other software capable of basic statistical analysis

3. Methods The GSIST approach for simultaneous quantification of PPP metabolites is based on work by Yang et al. (6) and consists of two important segments, including metabolite extraction and metabolite analysis. Whilst metabolite analysis is universal and can be applied to any biological source, the extraction protocol described here was optimized for yeast (10, 11). Experiments with other types of cells, tissues, or whole organisms require modification in the extraction protocol before PPP intermediates can be determined. The list of methods for quenching (arrest) of metabolic activity and subsequent metabolite extraction for prokaryotic microorganisms (bacteria) and eukaryotic microorganisms (yeast and filamentous fungi) can be found in the review by Mashego et al. (12). Plant, animal, and human tissue is generally quickly frozen in liquid nitrogen, then grinded, and metabolites are extracted by organic solvents (13–16). Another important step in this method is derivatization of PPP metabolites in the sample(s) and corresponding standards through the reaction of aniline with carbonyl, phosphoryl, or carboxyl functional groups. The labeling and mixing schema is illustrated in Fig. 9.2. The LC-MS analysis described here assumes the use of an Agilent 1100 series HPLC coupled to an Agilent MSD-TOF (time of flight) mass spectrometer; however, any other LC-MS system can be employed. Precise quantification requires metabolite standards. Although no standard is available for 6-phosphoglucono-δ-lactone, semi-quantification (not described here) can be achieved using 13 C-labeled fructose6-phosphate standard as a structural analogue (6). All other intermediates are commercially available and monoisotopic masses corresponding to their light and heavy derivatized forms are summarized in Table 9.1. Some of the metabolites such as glucose6-phosphate and fructose-6-phosphate share the same molecular mass and cannot be differentiated by MS. For this reason, it is necessary to determine the retention times of individual derivatized standards before analyzing real samples. The concentration of individual intermediates is simply calculated from the intensity

164

Jannasch, Sedlak, and Adamec

Fig. 9.2. Determination of PPP metabolites using GSIST – labeling and mixing schema.

ratios between corresponding light and heavy ions (Fig. 9.3) and usually expressed in moles per gram of cells dry weight. 3.1. Cell Growth and Extraction

1. Inoculate yeast Saccharomyces cerevisiae (ATCC 4124) from an agar plate into 5 mL of YEPD medium and incubate in a shaker at 30◦ C and set to 200 rpm overnight. 2. Transfer overnight culture into 100 mL of YPD media in a 300 mL flask equipped with a sidearm for easy determination of optical density (O.D.) by a Klett colorimeter.

Quantification of PPP Metabolites by LC-MS

165

Table 9.1 Labeling pattern of PPP intermediates m/z value Compound

12 C

Erythrose-4-phosphate

349.09

Fructose-6-phosphate

labeling

13 C

labeling

Labeling pattern

MS species

361.13

Bi-

[M-H]–

334.07

340.09

Mono-

[M-H]–

Glucose-6-phosphate

409.12

421.16

Bi-

[M-H]–

Glyceraldehyde-3-P

319.09

331.13

Bi-

[M-H]–

6-Phosphogluconate

425.11

437.15

Bi-

[M-H]–

Ribose-5-phosphate

379.11

391.15

Bi-

[M-H]–

Ribulose-5-phosphate

304.06

310.08

Mono-

[M-H]–

Sedoheptulose-7-phosphatea

NA

NA

NA

NA

Xylulose-5-phosphate

304.06

310.08

Mono-

[M-H]–

a Sedoheptulose-7-phosphate was not included in the original study (6) and labeling pattern must be obtained experimentally.

Continue incubation in the shaker until optical density reaches ∼500 K.U. (generally 1 A540 unit = 500 K.U.). 3. At this point, add 24 mL of glucose (50% w/v) and continue growth under identical conditions for an additional 3 h. 4. Take two 5 mL aliquots. Filter one aliquot using a preweighed filter paper (WFB ) under vacuum. Wash the filtered cells with 10 mL Milli-Q water and dry the filter with cells at 80◦ C for 5 h (see Note 4). Weigh dry filter with cells (WFA ), subtract weight of the filter, and record the value as “dry weight” (WD ) (see Note 5). 5. Spray another aliquot into a 50 mL centrifugation tube containing 26 mL of quenching solution kept at –45◦ C in a cryostat bath for 3 min (see Note 6). 6. Centrifuge the mixture at 500×g for 5 min with temperature set to –20◦ C (see Note 7). 7. Discard supernatant and resuspend the cell pellet in 5 mL of cell washing solution at –45◦ C. Repeat centrifugation and discard supernatant. Keep the tubes containing the washed pellet at –45◦ C in the cryostat bath (see Note 8). 8. Place the tubes into a 90◦ C water bath, immediately overlay with 5 mL of boiling ethanol solution, vortex, and keep the samples in the water bath for 3 min. 9. Transfer the tubes directly into a –80◦ C freezer.

166

Jannasch, Sedlak, and Adamec

Fig. 9.3. Data analysis and interpretation. (a) Extracted ion chromatograms of 13 C6 -labeled fructose-6-phosphate (m/z = 340.1), indicating the time segment covering the entire chromatographic peak. (b) MS spectra corresponding to selected chromatographic area. Concentration is calculated from intensity ratio of light (IE ) and heavy (IS ) forms of aniline-labeled fructose-6-phosphate.

3.2. Sample Derivatization and Mixing

1. Transfer a 100 μL aliquot from each standard test solution (0.1 mM) into a new Eppendorf tube.

3.2.1. Determination of Retention Times of Individual Metabolites

3. Add 10 μL of light labeling reagent (6 M 12 C6 -aniline solution) and vortex slowly at room temperature.

2. Add 10 μL of EDC solution to each sample (see Note 1).

4. After 2 h, add 2–5 μL TEA (see Note 9). 5. Centrifuge at 13,500×g for 2 min and transfer supernatant into an auto-sampler vial. 6. Analyze each standard by the LC-MS procedure described below. 7. Visualize the results using extracted ion chromatograms and record retention times for individual standards.

Quantification of PPP Metabolites by LC-MS

3.2.2. Sample Labeling

167

1. Remove extracted samples from –80◦ C (Step 8 of Section 3.1) and reduce the volume (≤100 μL) using nitrogen gas bubbling through a needle immersed into the sample at room temperature (at this point, the samples can be stored at –80◦ C until ready to analyze). 2. Pre-weigh one Eppendorf tube for each sample (WTB ). 3. Add 0.5 mL of Milli-Q water to the sample, vortex each sample for 3 min (see Note 10). To repellet solids, centrifuge at 13,500×g for 10 min at 4◦ C. 4. Transfer supernatants to pre-weighed Eppendorf tubes, then reweigh tubes with the samples (WTA ). This value (WD ) will be taken into account during the data evaluation. 5. Transfer 100 μL of each sample into a new Eppendorf tube (see Note 11). 6. In a separate tube, prepare one 100 μL aliquot of IS working solution for each sample. 7. Add 10 μL of EDC solution to both the samples and IS working solution (see Note 1). 8. Add 10 μL of light labeling reagent (6 M 12 C6 -aniline solution) to the samples and vortex slowly at room temperature for 2 h. 9. Add 10 μL of heavy labeling reagent (6 M 13 C6 -aniline solution) to the IS working solution and vortex slowly at room temperature for 2 h (see Note 12). 10. Add 2–5 μL of TEA (see Note 9) and centrifuge at 13,500×g for 2 min. 11. Mix 50 μL of light labeled samples with 50 μL of the heavy labeled internal standards and vortex the mixture for 1 min (see Notes 2 and 13). 12. Transfer to an auto-sampler vial and analyze by the LC-MS procedure described below.

3.3. LC-MS Procedure

1. Set the following HPLC gradient with the flow rate at 0.3 mL/min: a. Set an elution gradient from 5 to 70% Solvent B over 25 min. b. Increase to 100% Solvent B over 2 min. c. Wash the column with 100% Solvent B for 3 min. d. Decrease to 5% Solvent B over 1 min. e. Re-equilibrate the column with 5% Solvent B for 9 min.

168

Jannasch, Sedlak, and Adamec

2. Set the mass spectrometer parameters as follows (see Note 14): a. Use electrospray ionization (ESI) in negative mode. b. Set the mass range to 150–1000 m/z. c. Set the dry temperature and gas flow to 300◦ C and 8 L/min, respectively. d. Set the capillary fragmentor voltages to 4000 and 165 V, respectively. 3. Place vials with sample into auto-sampler with temperature set to 10◦ C. 4. Inject 20 μL for each sample. 3.4. Data Analysis and Quantification

1. Open acquired LC-MS data with Agilent MassHunter Workstation – Qualitative Analysis software. 2. For each metabolite, visualize the results with the Extracted Ion Chromatogram function (Fig. 9.3a). Use m/z of heavy derivatized standards as indicated in Table 9.1. 3. Identify peaks corresponding to individual standards. If more peaks appear for a single m/z, then determine the appropriate peak using retention time. 4. For each compound, select the chromatographic segment covering the entire peak and open the corresponding MS spectra (m/z vs. intensity) in a new panel (Fig. 9.3b). 5. Record intensities of light (IE ) and heavy (IS ) forms of the ions (Fig. 9.3) as indicated in Table 9.1. 6. Determine metabolite concentration in extract: CE [M] = (IE /IS ) × CS [M] 7. Calculate volume of extracted sample: VE [mL] = WTB [g]–WTA [g] 8. Determine amount of metabolite in extract: NE [mol] = (VE /1000) × CE 9. Calculate dry weight of cells: WD [g] = WFB [g]–WFA [g] 10. Normalize the results per gram of cell dry weight: CDW [mol/g] = NE /WD

4. Notes 1. Make a fresh EDC solution in water before the beginning of each experiment. Do not reuse solutions from past experiments.

Quantification of PPP Metabolites by LC-MS

169

2. This concentration can be altered to fit each experiment. The goal is to spike an approximately equal amount of internal standard as real analyte present in each sample. This may need to be determined experimentally. Final concentrations (CS ) of individual IS are used for determination of corresponding metabolite concentrations in sample extract. 3. Any LC-MS system and software capable of data visualization and quantification can be used. 4. Drying time usually depends on the cell type and the cell amount. Typically, the filter is weighted in 1 h intervals until a constant weight is reached. 5. It is recommended to create a calibration curve for the relationship between O.D. (Klett unit) and dry weight ([g]/L). In that case, Step 4 of Section 3.1. is not necessary and dry weight can be determined directly from O.D. 6. Manual sampling is suitable if samples are taken in longer intervals (hours). If samples must be taken in shorter intervals (seconds or minutes), a rapid sampling procedure should be used (17). 7. The centrifugation speed could be increased if the pellet gets disturbed during the discarding of supernatant. However, the centrifugation speed should be kept as low as possible to allow quick pellet resuspension by the washing solution. 8. The rotor must be pre-cooled at –30◦ C in a freezer for at least 2 h. After spinning, the rotor should be placed back at –30◦ C to cool down for the next sample. Tubes containing 60% (v/v) methanol solution must be pre-cooled to –45◦ C in the cryostat bath for 2 h. Quick sample handling is critical because temperature in the samples should not rise above –20◦ C. 9. TEA is added to stop the reaction and increase the stability of derivatized products by adjusting pH to ∼8. 10. Be sure to reconstitute the entire pellet into the water solution. It may take more than the allotted time to do so. 11. Freeze the remaining 400 μL of sample at –80◦ C. It should remain viable for 1 week after the freeze date. 12.

13 C

6 -aniline is supplied as a liquid solution. Transfer the liquid content into a new tube before adding water and HCl.

13. Depending on the number of samples, the appropriate amount of standard solution must be labeled to allow for 50 μL to be mixed with 50 μL of each sample. The reaction can be scaled linearly from the indicated volumes.

170

Jannasch, Sedlak, and Adamec

14. If QQQ mass spectrometer is available selected ion monitoring (SIM) or multiple reaction monitoring (MRM) modes can be used. Due to the differences in instruments, optimal MS conditions must be determined experimentally.

Acknowledgments This work was supported by grants from the US Department of Energy Biomass Program (GO17059-16649) and the National Science Foundation (DBI-0421102). References 1. Cipollina, C., ten Pierick, A., Canelas, A. B., Seifar, R. M., van Maris, A. J. A., van Dam, J. C., Heijnen, J. J. (2009) A comprehensive method for the quantification of the non-oxidative pentose phosphate pathway intermediates in Saccharomyces cerevisiae by GC-IDMS. J Chromatogr B 877, 3231–3236. 2. Koek, M. M., Muilwijk, B., van der Werf, M. J., Hankemeier, T. (2006) Microbial metabolomics with gas chromatography/mass spectrometry. Anal Chem 78, 1272–1281. 3. 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. 4. Luo, B., Groenke, K., Takors, R., Wandrey, C., Oldiges, M. (2007) Simultaneous determination of multiple intracellular metabolites in glycolysis, pentose phosphate pathway and tricarboxylic acid cycle by liquid chromatography-mass spectrometry. J Chromatogr A 1147, 153–164. 5. Wamelink, M. M., Struys, E. A., Huck, J. H., Roos, B., van der Knaap, M. S., Jakobs, C., Verhoeven, N. M. (2005) Quantification of sugar phosphate intermediates of the pentose phosphate pathway by LC-MS/MS: application to two new inherited defects of metabolism. J Chromatogr B 823, 18–25. 6. Yang, W.-C., Sedlak, M., Regnier, F. E., Mosier, N., Ho, N., Adamec, J. (2008) Simultaneous quantification of metabolites involved in central carbon and energy metabolism using reversed-phase liquid

7.

8.

9.

10.

11.

12.

13.

chromatography–mass spectrometry and in vitro 13c labeling. Anal Chem 80, 9508–9516. Yang, W. C., Adamec, J., Regnier, F. E. (2007) Enhancement of the LC/MS analysis of fatty acids through derivatization and stable isotope coding. Anal Chem 79, 5150–5157. Yang, W. C., Regnier, F. E., Adamec, J. (2008) Comparative metabolite profiling of carboxylic acids in rat urine by CEESI MS/MS through positively pre-charged and (2)H-coded derivatization. Electrophoresis 29, 4549–4560. Yang, W. C., Regnier, F. E., Sliva, D., Adamec, J. (2008) Stable isotope-coded quaternization for comparative quantification of estrogen metabolites by high-performance liquid chromatography-electrospray ionization mass spectrometry. J Chromatogr B 870, 233–240. De Koning, W., van Dam, K. A. (1992) A method for the determination of changes of glycolytic metabolites in yeast on a subsecond time scale using extraction at neutral ph. Anal Biochem 204, 118–123. Gonzalez, B., Francois, J., Renaud, M. (1997) A rapid and reliable method for metabolite extraction in yeast using boiling buffered ethanol. Yeast 13, 1347–1355. Mashego, M. R., Rumbold, K., De Mey, M., Vandamme, E., Soetaert, W., Heijnen, J. J. (2007) Microbial metabolomics: past, present and future methodologies. Biotechnol Lett 29, 1–16. Denkert, C., Budczies, J., Weichert, W., Wohlgemuth, G., Scholz, M., Kind, T., Nies-

Quantification of PPP Metabolites by LC-MS porek, S., Noske, A., Buckendahl, A., Dietel, M., et al. (2008) Metabolite profiling of human colon carcinoma – deregulation of TCA cycle and amino acid turnover. Mol Cancer 7, 72. 14. Parab, G. S., Rao, R., Lakshminarayanan, S., Bing, Y. V., Moochhala, S. M., Swarup, S. (2009) Data-driven optimization of metabolomics methods using rat liver samples. Anal Chem 81, 1315–1323. 15. Weckwerth, W., Wenzel, K., Fiehn, O. (2004) Process for the integrated extraction, identification and quantification of metabolites, proteins and RNA to reveal their co-

171

regulation in biochemical networks. Proteomics 4, 78–83. 16. Wu, H., Southam, A. D., Hines, A., Viant, M. R. (2008) High-throughput tissue extraction protocol for NMR- and MSbased metabolomics. Anal Biochem 372, 204–212. 17. Lange, H. C., Eman, M., van Zuijlen, G., Visser, D., van Dam, J. C., Frank, J., de Mattos, M. J., Heijnen, J. J. (2001) Improved rapid sampling for in vivo kinetics of intracellular metabolites in saccharomyces cerevisiae. Biotechnol Bioeng 75, 406–415.

Chapter 10 High-Performance Liquid Chromatography-Mass Spectrometry (HPLC-MS)-Based Drug Metabolite Profiling Ian D. Wilson Abstract The identification of drug metabolites in biofluids such as urine, plasma, and bile, as well as in in vitro systems, is an important step in drug discovery and development. Mass spectrometry, particularly when combined with high-performance liquid chromatography (HPLC-MS), can enable detailed structural information to be obtained on the metabolites of a drug or xenobiotic as a result of metabolism. The successful identification of drug metabolites by HPLC-MS-based techniques requires careful optimisation of a number of factors. First, the chromatographic separation should provide good resolution of the individual xenobiotic metabolites present in the sample. There is also the need to minimise the interference caused by the presence of endogenous metabolites, which may interfere with MS detection. Ideally, untreated samples should be profiled to reduce the likelihood of missing important metabolites due to losses during sample processing, but, depending upon the matrix, some degree of sample cleanup/extraction and concentration of the metabolites may be required using liquid–liquid or liquid– solid extraction. Second, the MS conditions must be carefully selected in order to maximise the potential of detecting the separated metabolites, which may have a very different character to the parent compound. The use of radiolabelled drugs in metabolism experiments greatly aids in the detection (and quantification) of metabolites, directing the investigator towards peaks that need to be characterised by MS. The presence of characteristic isotope patterns from either the incorporation of stable isotopes (e.g. 13 C, 15 N) or naturally occurring isotope patterns from substituents on the molecule (e.g. 35/37 Cl, 79/81 Br) can also provide a useful handle on the drug and its metabolites for the purposes of detection and spectrometric interpretation. This chapter provides guidelines for, and examples of, HPLC-MS-based drug metabolite profiling. Key words: Mass spectrometry, metabolite profiling, metabolite identification, liquid chromatography.

T.O. Metz (ed.), Metabolic Profiling, Methods in Molecular Biology 708, DOI 10.1007/978-1-61737-985-7_10, © Springer Science+Business Media, LLC 2011

173

174

Wilson

1. Introduction Mass spectrometry represents, together with nuclear magnetic resonance (NMR) spectroscopy, a vital component of strategies for the detection and structural elucidation of the biotransformation products of drugs and other xenobiotics. Identification can be performed by obtaining mass spectra via direct analysis of metabolites that have been isolated in a pure form from the biological matrix by the infusion of solutions of the pure metabolite or from desorption from probes inserted into the ion source of the mass spectrometer (MS). However, in modern practice it is more usual to use a hyphenated technique such as highperformance liquid chromatography (HPLC-MS) with the column eluent directly flowing into the mass spectrometer via special interfaces (1). Ideally, in such an analysis, the metabolites are resolved both from each other and from matrix components by the separation technique and are presented to the MS in sufficient quantity to enable a diagnostic mass spectrum to be obtained. HPLC, and the more recently introduced variant of ultra highperformance liquid chromatography (UHPLC or UPLC) (2), provides a high-resolution separation method that can be directly interfaced with the mass spectrometer and that can be optimised to separate the xenobiotic metabolites from each other and from interferences such as endogenous metabolites and dose vehicles (e.g. polyethylene glycol, PEG). For this type of work, a reversedphase separation with gradient elution is generally used, based on either a “generic” gradient for routine/high-throughput profiling or a more optimised “bespoke” separation tuned to maximise the separation for the particular compound under study. The decision on the choice of generic or bespoke separation will depend upon the required outcomes of the study, but, irrespective of these, the generic separation represents a good starting point to obtain metabolic profiles, and this separation can be further optimised to give a more comprehensive profile if required. If the compound is available in a radiolabelled form (14 C and 3 H are the most common labels), online detection (HPLC-RAD) using a suitable radioactivity (or radio) detector can provide a useful means of both highlighting and quantifying the presence of the metabolites for MS (HPLC-MS-RAD) (3–5). The ability of radiodetectors to detect and quantify the compound-related metabolites is also of value in helping to overcome some of the inherent difficulties associated with MS-based approaches. The main difficulties associated with MS are that, in the absence of authentic standards of the metabolite, it can only be used as a qualitative technique. This is because ionisation efficiency is compound dependent and can vary widely between parent and metabolite and indeed metabo-

HPLC-MS-Based Drug Metabolite Profiling

175

lite and metabolite. The presence of a radiolabel that provides an “absolute” measure of the amount of material present in a peak can enable a relative response to be determined and, in extreme circumstances, can highlight the presence of metabolites that have been missed by the MS because the optimal ionisation conditions have not been used. The approach does of course depend upon the retention of the radiolabel in all of the metabolites. Some caution has therefore to be exercised in this respect as placing the radiolabel in a metabolically labile position, or very extensive metabolism that results in the cleavage of large molecules into several parts, may mean that not all of the metabolites retain the label. Clearly, where no label is available, care must be taken not to overinterpret the result of a metabolite profiling study. Having performed HPLC-MS comparison of the mass spectra of the detected metabolites with that of the parent compound, combined with the fragmentation patterns seen in the spectra and knowledge of the potential routes of metabolism, then enables structural elucidation to be attempted. Depending upon the mass spectrometer used, and the amount of material available, a range of MS-based experiments can be undertaken to provide further confidence in the preliminary examination of the spectra. The MS experiments that can be performed, depending upon the mass spectrometer, include techniques such as MS/MS, MSn , MSe , neutral losses (1), with the aim of building up a sufficient body of mass spectral evidence to propose potential metabolite structures. In addition, the data can be searched, either manually or automatically, via bespoke software applications such as mass defect filtering, for the additions (or losses) in mass expected for typical metabolic reactions such as hydroxylations, reductions, oxidations, and conjugations etc (1). In order to obtain a mass spectrum, it is essential to first ionise the molecule. In HPLC-MS, it is most common to use electrospray ionisation (ESI). To obtain the most comprehensive determination of the metabolic profiles, it may be necessary to employ both positive and negative ESI and, depending upon the mass spectrometer used, this may either be performed in one run or require re-running of the sample in a second. In the first instance, the most important information available from the mass spectrum is the molecular ion from which the molecular mass of the metabolite can be obtained. In some cases, depending upon the complexity of the parent compound, it may be possible to derive the structure of the metabolite simply from this information (e.g. addition of 16 mass units to a simple aromatic compound would immediately suggest hydroxylation). If the MS used has the ability to generate accurate mass data (e.g. Time-of-Flight, Orbitrap, and FT-ICR instruments), empirical atomic formulae can be derived, which can also help to confirm or refute potential structures. The next level of interpretation of the spectra comes from looking at

176

Wilson

–80 Da –SO3

100

248

O

Relative Intensity

HN

+O + SO3

F

%

m/z 248 168

249 250

173 212

108 150 167 74 80 93 97 117125

0 50

75

100

125

150

208 175

200

247 251 302 338 230 278290 294 328 346 368382 388 417

225

250

275

300

325

350

375

400

425

455 479 450

475

497

501 m/z

500

m/z Fig. 10.1. Negative ion mass spectrum of a probable hydroxysulphate metabolite of the acetanilide of 2-fluoroaniline showing a fragment ion 80 amu lower in mass (m/z 168) than the parent (m/z 248), consistent with the loss of SO3 characteristic of a sulphate metabolite.

fragmentation patterns obtained from the molecule wherein the loss of diagnostic ions can point directly to structural features. The ions detected in the mass spectrum may show which parts of the xenobiotic remain unchanged by metabolism, and, conversely, which have changed, and may provide strong indications as to what those changes are. An example is given in Fig. 10.1 for a metabolite of 3-fluoroaniline, where the characteristic loss of 80 amu, corresponding to loss of SO3 from the M– H– ion is seen. This provides a clear indication that the metabolite contains a sulphate moiety, strongly suggesting a ring hydroxysulphate as a metabolite of the original aniline. In fact, careful examination of the profile found two other metabolites with the same characteristics representing further positional isomers with essentially the same mass spectrum. This highlights a difficulty with MS in such circumstances in that it may not be possible, based on MS data alone, to obtain a definitive structure without additional spectroscopic data of the type provided by NMR spectroscopy (see Chapter 18) or the synthesis of an authentic standard. Obtaining suitable NMR spectra can be performed in a number of ways, including the semi-preparative isolation of the metabolite or, more efficiently, by online HPLC-NMR (6) (see Chapter 18) or using the multiple hyphenation of HPLC-NMR-MS (7).

HPLC-MS-Based Drug Metabolite Profiling

177

The biological matrix being investigated must also be taken into account when considering metabolite profiling and identification by HPLC-MS. Some matrices, such as urine (or bile), may be injected directly into the HPLC system and, providing the metabolites are present in sufficient amount, profiles are relatively easily obtained. In the case of plasma, or serum, or in vitro metabolism samples, however, direct injection of the sample is more problematic as the proteins present in the matrix will denature and cause the rapid failure of the HPLC column. Protein removal, generally by solvent precipitation, is therefore mandatory before HPLC-MS is undertaken. It is also usually the case that the concentrations of metabolites present in plasma may be very low compared to other sample types, and it may also be necessary to perform some further sample cleanup and concentration (e.g. HPLC with fraction collection). For solid samples such as tissues or faeces, a solvent extraction is also mandatory prior to profiling as these samples cannot be analysed directly by HPLCMS. A difficulty with any extraction procedure is that the recovery of the xenobiotic metabolites is uncertain and can only really be determined when a radiolabel is incorporated into the compound, in which case the recovery can be determined using liquid scintillation counting (LSC). In the absence of a radiolabel, caution should always be exercised in interpreting the metabolite profile obtained from such extracts as significant metabolites may be extracted with different efficiencies (or not at all) leading to a bias in the HPLC-MS data. The procedures described here cover the use of HPLC and UPLC-MS-based metabolite identification exemplified using both generic and “bespoke” reversed-phase gradients with and without the use of radiolabelled xenobiotics.

2. Materials 2.1. Freeze-Drying of Samples or Large Volumes of Sample Extracts

1. Freeze-dryer (e.g. Edwards Modulya 4 K or VirTis Sentry 2.0).

2.2. Solid–Liquid Extraction

1. Biological fluid/sample (e.g. plasma/serum, urine, or bile or in vitro incubation), dry or freeze-dried or dried and powdered faeces.

2. Glass vials with screw top neck. 3. Liquid nitrogen (–196◦ C) or solid carbon dioxide (Cardice, –140◦ C) to freeze the sample.

2. HPLC-grade methanol (MeOH).

178

Wilson

3. HPLC-grade acetonitrile (ACN). 4. Centrifuge (e.g. Megafuge 1.0R, Heraeus instruments, Newport Pugnell, UK). 5. Solvent evaporator or freeze-dryer. 2.3. Plasma or Serum Protein/In Vitro Incubation Precipitation

1. Organic solvent (e.g. HPLC-grade ACN, MeOH). 2. Vortex mixer. 3. Centrifuge. 4. Solvent evaporator or freeze-dryer.

2.4. Liquid Scintillation Counting (LSC)

1. Biological fluid, extract, or HPLC fraction. 2. Scintillant, e.g. Ultima GoldTM (PerkinElmer Life Sciences, Boston, MA). 3. Scintillation Counter (Packard Instruments, Chicago, IL).

2.5. High-Performance Liquid Chromatography (HPLC)

1. HPLC column (appropriate for analyte, e.g. C18 Symmetry, 250 × 4.6 mm, 5 μm particle size, Waters Corporation, Watford, UK or Hichrom RPB, 4.6 × 150 mm, 5 μm particle size, Hichrom Ltd., Theale, UK). 2. HPLC system (e.g. PerkinElmer 200 series HPLC pump fitted with an autosampler, Packard Instruments, Chicago, IL) plus a UV (Jasco UV-VIS 2075 UV 74) or radioactivity detector, if the compound is radiolabelled (e.g. Packard R Radiomatic 500TR series flow scintillation FLO-ONE analyser fitted with a 500 μL liquid flow cell (Packard Instruments)). 3. Ultima-FloTM M scintillant (Packard) for radiolabelled compounds. 4. Mobile phases: HPLC-grade water and ACN or MeOH. 5. Buffers of choice: e.g. HPLC-grade ACN or MeOH and aqueous phase, pH adjusted with analytical-grade formic acid (e.g. 0.1% formic acid, v/v) and/or buffered with analytical-grade ammonium formate (e.g. 10 mM) (see Note 1).

2.6. Ultra-Performance Liquid Chromatography (UPLC)

1. UPLC column (appropriate for analyte, e.g. Acquity UPLCTM BEH C18 column, 2.1 mm × 100 mm, 1.7 μm particle size). 2. AcquityTM UPLC-pump and UV detector (Waters Corporation). 3. Fraction collector capable of collecting samples into 96-well plates or glass vials (e.g. Gilson FC 204). 4. Mobile phases: As for HPLC.

HPLC-MS-Based Drug Metabolite Profiling

2.6.1. Mass Spectrometry

179

1. Mass spectrometer, e.g. Waters Q-TofTM micro or LCT PremierTM orthogonal acceleration time-of-flight mass spectrometer (oa-TOFMS) equipped with an electrospray ion source (Waters Corporation). 2. A LockSprayTM interface to provide accurate mass measurements (Waters). 3. Leucine enkephalin (50 fmol/μL, in 50:50 ACN:H2 O, 0.1% formic acid) (SigmaAldrich).

3. Methods 3.1. Freeze-Drying

Freeze-drying (also termed lyophilisation) can be used to concentrate biofluid samples and reduce large volumes of biofluids, extracts, in vitro incubations, or HPLC fractions to dryness. 1. Biofluids may be freeze-dried directly; however, the removal of organic solvent residues from samples, such as extracts or HPLC fractions, prior to freeze-drying is essential to prevent damage to the freeze-dryer. 2. Freeze small aliquots (ca. 1–5 mL) of biofluids or biofluid extracts in liquid nitrogen (–196◦ C) or solid carbon dioxide (Cardice, –140◦ C) in screw neck glass vials placed at a slight angle. 3. Operate freeze-dryer according to instruction manual. 4. Redissolve dried samples in an appropriate volume of the starting HPLC eluent (see Note 2).

3.2. Solid–Liquid Extraction

Solid–liquid extraction is used to obtain metabolites from solid material (e.g. faeces, tissues), freeze-dried biofluids (e.g. plasma/serum, urine, or bile), or freeze-dried in vitro incubation samples, which cannot be directly injected into the LC-MS system. 1. Homogenise or reduce the sample to a powder (freezedry first if necessary) and mix thoroughly with aliquots of organic solvent (e.g. 1–5 mL MeOH/g). 2. Centrifuge at 4◦ C for 10 min at 1800×g, then carefully remove the supernatant without disturbing the solid material. 3. Repeat the process several times and combine the supernatant fractions. 4. If the compound is radiolabelled, determine the extraction recovery using LSC.

180

Wilson

5. Evaporate the supernatant and redissolve in the starting HPLC eluent (see Note 2). 3.3. Plasma or Serum Protein/In Vitro Incubation Precipitation

Protein precipitation is designed to remove proteins, which would otherwise degrade the chromatographic performance of the system, from the sample while minimising potential losses of metabolites. 1. Protein precipitation is most easily achieved by taking aliquots of plasma/serum or in vitro incubations and mixing with cold (–20◦ C) ACN (1:3, v/v) or MeOH (1:3 v/v), vortexing, and then centrifuging at 4◦ C for 10 min at 1800×g (see Note 3). 2. Transfer the supernatant to a separate vial. 3. Re-suspend the precipitated protein and re-extract with the organic solvent, then repeat the centrifugation step. 4. Combine the supernatants and evaporate the solvent. 5. Reconstitute the sample in the starting HPLC mobile phase (see Note 2). 6. If the compound is radiolabelled, determine the extraction recovery using LSC.

3.4. Liquid Scintillation Counting

The amount of radioactivity present in samples such as biofluids, in vitro incubation media, and sample extracts can be determined by scintillation counting. 1. Take 100 μL of the sample (smaller amounts may be taken if sample size is limited) and place in a glass or plastic scintillation vial. 2. Add 900 μL of water (or make up to 1 mL if less than 100 μL of sample was used). 3. Mix with 10 mL of Ultima GoldTM scintillant. 4. Determine radioactivity present using a Packard Scintillation Counter according to the manufacturer’s instructions.

3.5. HPLC-MS Metabolite Profiling

These instructions assume the use of a suitable liquid chromatograph, a C18-bonded reversed-phase chromatography column, a suitable gradient HPLC system with a UV detector, and a Waters Q-TofTM Micro or LCT Premier orthogonal acceleration time-of-flight mass spectrometer (oa-TOFMS) equipped with an electrospray source capable of operating in both posR itive and negative ion mode and controlled using Waters MassLynxTM software. However, HPLC-MS metabolite profiling and identification can be performed with a wide range of different instruments from different manufacturers and this methodology should be easily modified according to the available instrumentation.

HPLC-MS-Based Drug Metabolite Profiling

181

1. Prior to performing HPLC-MS metabolite profiling, the performance of the HPLC system should be checked to ensure suitability. The flow rate, back pressure, analyte retention, and chromatographic peak shape should be examined and compared with the expected results. 2. Prior to analysis of the samples by HPLC-MS, the performance of the mass spectrometer should be checked to be within the manufacturer’s specifications for factors such as sensitivity and mass accuracy to ensure that valid data will be obtained. 3. Before attempting to detect and characterise metabolites, the chromatographic and mass spectrometric properties of the parent compound itself should be determined so that the subsequent analysis can be optimised both for the separation and detection of biotransformation products. Usually, metabolism will result in the introduction of polar (e.g. hydroxylations) or ionic (e.g. sulphates, glucuronides) groups into the parent compound. As a consequence of the increase in polarity brought about by metabolism in general in gradient reversed-phase chromatographic systems, metabolites will elute before the parent compound. Therefore, the chromatographic retention of the parent drug should be maximised by adjusting the gradient conditions such that it elutes towards the end of the chromatographic run. 4. A suitable starting point (a “generic” gradient) for method development would be HPLC separation on a Hichrom RPB 4.6 × 150 mm (5 μm) column or C18 Symmetry 250 × 4.6 mm (5 μm) column, at a solvent flow rate of 1 mL/min using a combination of 0.1% aqueous formic acid as “solvent A” and ACN-containing 0.1% formic acid as “solvent B”. A gradient profile of 0–5 min at 0% B followed by a linear increase to 50% B from 5 to 45 min and a further linear increase to 90% B at 50 min to wash the column. The column should then be returned to 0% B at 60 min and allowed to re-equilibrate for a further 5 min before the next sample is analysed. 5. The mass spectrometric properties of the parent molecule, such as the optimum ionisation conditions (positive or negative ESI) and fragmentation properties (including the presence of diagnostic fragments or isotope patterns), provide essential information for the detection and characterisation of metabolites enabling metabolic modifications to be detected and interpreted (see refs. 1, 8, 9). 6. Providing that the performance of the HPLC-MS system is acceptable, proceed to develop a suitable method for

182

Wilson

metabolite profiling with the injection of an aliquot of the sample (typically an injection of 10–50 μL is acceptable) (see Note 4). 7. Method development strategies will be affected by the presence or absence of a radiolabel in the parent compound. In the absence of a radiolabel, inject aliquots of the samples (including pre-dose control biofluid samples/preincubation in vitro samples), and, using a linear gradient from low percentages of organic modifier up to 100% over 30–50 min, monitor UV and total ion current traces for evidence of the unchanged parent compound and metabolites (Fig. 10.2). 8. The data can be examined for the presence of ions characteristic of well-known metabolic changes to the parent compound (such as addition of 16 amu to the parent molecular ion indicating hydroxylation), which may be indicative of the presence of a metabolite. This can either be done manually or by using bespoke software (so called data-dependent acquisition, DDA, see refs. 1 and 9). 9. The HPLC-MS data may also be examined for characteristic fragments/isotopic patterns from the parent that might also be expected to be present in metabolites (see Note 5).

Fig. 10.2. Traces showing (a) the UV, (b) total ion current (TIC), and (c) extracted ion chromatograms for the major metabolite (m/z 270) obtained using gradient reversedphase HPLC on rat urine obtained following the administration of 4-bromoaniline (see ref. 12). Chromatographic conditions were Hypersil BDS C18 column (250 × 4.6 mm i.d) (Thermo Scientific, Warrington, UK) packed with 5 μm particles. The mobile phases used to form the gradient were 0.01 M ammonium formate in D2 O (pD 7.0) (solvent A) and ACN (solvent B) at 1 mL/min. The gradient started with 100% A for 10 min increasing, in a linear gradient, to 30% B at 35 min followed by a second linear gradient to 50% B at 45 min.

HPLC-MS-Based Drug Metabolite Profiling

183

10. In addition to increases in mass, the data can be examined for losses characteristic of particular biotransformations (e.g. a fragment showing the loss of 176 amu from the parent compound for glucuronides or 80 amu for a sulphate, observed either directly in the MS spectrum or from a precursor ion in an MS/MS experiment) (see refs. 1, 9, and 10) (Figs. 10.1 and 10.3). 11. If the MS instrument used is capable of generating “exact” masses, then these can be used to provide elemental compositions for putative metabolites, which can aid in metabolite identification by supporting or eliminating metabolic possibilities. 12. MS/MS, MSn , or MSe experiments can be used on putative metabolites to aid in structure elucidation. 13. On the basis of the results obtained from these initial profiling studies, it may be necessary to modify the solvent gradient used for the chromatographic separation to improve the retention or resolution of metabolites. Repeat until the optimum separation is obtained (software such as “Drylab”

100

240

O HN

Relative Intensity

Cl

F

%

+O + C6H8O6

242

m/z 378/380 243

0 50

– 176 Da – C6H8O6

378

380 202 160 244 162 175 222 238 381 292 187 213 256 273 308320 348 376 390 85 97 113 117 147 75 100 125 150 175 200 225 250 275 300 325 350 375 400

436 425 450

478 m/z 475 500

m/z Fig. 10.3. Negative ion mass spectrum of a metabolite of 2-chloro-4-fluoroaniline detected using UPLC-ToF-MS (negative ESI mode) in urine following the administration of the aniline to bile-cannulated rats. The [M-H]– ion at m/z of 378/380 is consistent with N-acetylation, hydroxylation, and glucuronidation to give a hydroxy-2-chloro-4-fluoroacetanilide-Oglucuronide. The presence of the glucuronide is supported by the neutral loss of 176 Da from the parent ion to leave a fragment at 200/202. The ion seen in the spectrum at m/z 240/242 results from co-elution with another metabolite (hydroxy-2-chloro-4-fluoroaniline sulphate), which elutes as a broad peak. The presence of the chlorine isotope pattern is useful for selectively detecting metabolites of the parent 2-chloro-4-fluoroaniline.

184

Wilson

can be used to aid this optimisation step) (see ref. 11) (Fig. 10.4). 14. If step 12 is required, repeat steps 7–11 using the final optimised separation. 15. If a radiolabel is incorporated into the molecule, then the chromatographic optimisation of the metabolite separation can be performed on the basis of resolving the peaks detected using the radiodetector (see ref. 3). This can be performed online with the HPLC-MS or offline as the optimisation of the metabolite separation is based solely on the presence of the radioactive isotope (see Note 6). 16. If a radioactivity detector is not available, then fraction collection can be performed and a chromatogram reconstructed following offline fraction collection and LSC (see Note 7 and ref. 3–5). For good peak resolution using

N

O

N

100 75

100

50 25 0 0

Intensity/cps

S

20 40 Time (mins)

60

75 g

4.0e6

50 b d a

f

% Acetonitrile

Mobile phase B (%)

N

7.5e6

O

N

N

25

e c

0 0.0

10

20

30

40

Time/mins

Fig. 10.4. A generic HPLC gradient separation performed using a Prominence Sil-20A/C auto sampler and a Sil-20 AB binary pump (Shimadzu, Milton Keynes, UK), with the separation performed on a Synergi 150 × 4.6 mm, 4 μm POLARRP 80 Å column (Phenomenex, Macclesfield, UK). MS detection was performed using a Sciex API4000 QTrap mass spectrometer (Applied Biosystems, Warrington, UK) in positive ESI. The initial mobile phase conditions were 90:10:0.1 aqueous:ACN:formic acid (v/v/v). The gradient profile is shown on the figure (also see inset). The flow rate used was 1 mL/min. Metabolite key: a, b = hydroxy-O-glucuronides, d, g = hydroxylated metabolites, e = di-glucuronide, f = parent AZD5438 (structure in inset). See also Fig. 10.5 (from ref. 13, with permission).

HPLC-MS-Based Drug Metabolite Profiling f

100

g

Intensity/ cps

b

75

50

4.0e6

c

d

e

% Acetonitrile

8.2e6

185

25 a

0.0

5

10

0

14

Time/ mins

Fig. 10.5. An optimised UPLC separation for the sample shown in Fig. 10.4 using an Acquity UPLC system (Waters Corporation, Massachusetts, USA), with the separation obtained on an Acquity 2.1 × 100 mm, 1.7 μm column (Waters Corporation). MS detection was performed using a Sciex API4000 QTrap mass spectrometer (Applied Biosystems) in positive ESI. The initial mobile phase was the same as that used for the conventional chromatographic system (see Fig. 10.4 caption), and the gradient profile is shown in the figure. The flow rate employed was 0.66 mL/min. Key, as for Fig. 10.4 (from ref. 13, with permission).

offline fraction collection, a minimum of 2 fractions/min should be collected. 17. Where a radiolabel is available, it is good practice to perform a run on the optimised system, where the total eluent from the column is collected and the recovery of the radiolabel after injection of the sample is determined. This ensures that all of the metabolites are eluted from the column and that the resulting metabolite profile is complete. 18. When an optimised separation of the metabolites based on the radiolabel has been obtained, then steps 7–11 can be performed as for the unlabelled compounds. 3.6. UPLC-MS

As indicated above, the use of UPLC provides a more efficient chromatographic system for metabolite resolution. These instructions assume the use of a Waters Acquity UPLCTM system together with separations performed on a 2.1 × 100 mm BEH C18 column (1.7 μm particle size) (see Note 8) and a Q-tofTM Micro or LCT PremierTM mass spectrometer equipped with an electrospray ion source with a LockSprayTM interface, referenced to leucine enkephalin, to provide accurate mass measurements (50 fmol/μL, in 50:50 ACN:H2 O, 0.1% formic acid). 1. Perform steps 1–3 from Section 3.5. 2. A suitable starting point (a “generic” gradient) for method development for UPLC is based on the use of a 2.1 ×

186

Wilson

100 mm Acquity UPLCTM BEH C18 column (1.7 μm particle size) with the gradient formed from 0.1% (v/v) formic acid in water as solvent A and 0.1% formic acid (v/v) in ACN as solvent B at an eluent flow of 600 μL/min. 3. A suitable gradient profile of 100% A over 0–1 min, rising thereafter via a linear gradient to 50% B at 7 min, 90% B at 15 min, then returning to 100% A and remaining there for 18 min before the next injection. As for HPLC, the UPLC method should be optimised for good retention of the parent compound to ensure retention of the metabolites (Fig. 10.5). 4. When an optimised separation has been obtained, then steps 4–11 from Section 3.5 can be performed until an optimised separation for metabolite characterisation has been obtained. 5. If a radiolabel is present, then method development can also be undertaken using this as a means of detecting the metabolites (see Notes 6 and 7 and refs. 3–5). However, on-flow radiodetection can be problematic with UPLC and fraction collection may be necessary (4 fractions/min). 3.7. Metabolite Structural Elucidation and Identification

Very powerful MS-based techniques may be unable to provide definitive identification directly from samples because the metabolites are present only in trace quantities, ionise poorly, give little fragmentation, or may be one of several possible isomers. In such circumstances, it may be necessary to isolate larger quantities of material for further MS-based studies or for NMR spectroscopy, including HPLC-NMR or HPLC-NMR-MS if available (Fig. 10.6) (see Chapter 18). 1. Larger quantities of material for investigation can be obtained by fraction collection of the required peak(s) using the developed HPLC system and a fraction collector with a number of injections of the sample at the maximum amount on column that does not lead to the unacceptable deterioration of the separation. 2. Following purification of sufficient material for further studies by collection of fractions, these should be freeze-dried to provide a sample for further investigation (see Note 9). 3. If the collection of the metabolite is intended for NMR spectroscopic studies to complement the HPLC-MS investigations, then the sample should be redissolved in a suitable deuterated solvent (see Chapter 18). 4. If MS (and NMR) studies are inconclusive, then it may be necessary to use synthetic methods to obtain a pure standard(s) of the putative metabolite for comparison of HPLC and spectroscopic properties in order to obtain a definitive identification.

HPLC-MS-Based Drug Metabolite Profiling

187

Fig. 10.6. Spectra obtained via HPLC-NMR-MS showing the MS (lower) and 1 H NMR (upper) spectra for an N-oxanillic acid metabolite obtained using gradient reversedphase HPLC on rat urine obtained following the administration of 4-bromoaniline. The NMR spectrum was obtained in stopped flow mode (see Chapter 18). Chromatographic conditions were Hypersil BDS C18 column (250 × 4.6 mm i.d) (Thermo Scientific) packed with 5 μm particles. The mobile phases used to form the gradient were 0.01 M ammonium formate in D2 O (pD 7.0) (solvent A) and ACN (solvent B) at 1 mL/min. The gradient started with 100% A for 10 min increasing, in a linear gradient, to 30% B at 35 min followed by a second linear gradient to 50% B at 45 min (see ref. 12).

4. Notes 1. Avoid non-volatile buffers, such as sodium phosphate, as these can lead to rapid deterioration of the performance of the mass spectrometer. Similarly, avoid pH modifiers, such as trifluoroacetic acid (TFA), as these will tend to suppress the ionisation of the analytes resulting in poor sensitivity (particularly in negative ion mode for TFA). Formic acid and ammonium formate, both of which are volatile, generally provide a good compromise where pH modification of the mobile phase is needed to obtain or optimise retention. 2. Ideally, the sample should be dissolved in the solvent that forms the initial mobile phase used for the HPLC solvent gradient in order to ensure the best possibility for retention on the column. Injection of large volumes of sample

188

Wilson

dissolved in solvents containing a high proportion of organic solvent should be avoided, if at all possible, as this may result in poor retention on the column, low sample loading capacity, or poor peak shape. 3. While both ACN and MeOH can be used to provide efficient removal of proteins, it is generally preferable to use ACN, especially where acyl (or ester) glucuronides may be present. This is because the presence of methanol can result in transesterification reactions whereby the acyl glucuronide is converted to the methyl ester, resulting in the loss of information and the potential for mis-identification of the metabolite. 4. The amount of sample that can be injected onto a HPLC column should be carefully optimised with respect to the column dimensions, to avoid overloading and to provide sufficient material for detection by the various detectors being used (radio, UV and MS). Care should also be taken to ensure that the loop size chosen is compatible with the injection volume. If there is evidence of column overloading the sample volume injected should be reduced. 5. Characteristic losses from the parent molecule can be a useful aid to finding metabolites but this approach cannot be relied upon to find all of the compound-related material. This is because metabolism on the part of the molecule that results in the formation of this characteristic fragment will mean that it is no longer observed in the fragments from such metabolites. Thus, while the presence of the fragment provides good evidence for the presence of a metabolite, the absence of the fragment does not necessarily indicate that a new peak in the sample that was not seen in the pre-dose/control is not a metabolite. Over-reliance on this approach of searching for common fragments alone may lead to metabolites being missed. 6. Two types of radiodetectors are available, using flow cells based on either solid or liquid scintillants. Radiodetectors using solid scintillants may be connected inline with the UV and MS systems. Instruments using liquid scintillants cannot be connected inline because of the need to avoid the introduction of large amounts of the scintillant into the ion source of the mass spectrometer. The use of liquid scintillant-based systems requires flow splitting and careful calibration of the lengths of tubing, etc., to ensure that the relationship between the retention time of peaks in the radiodetector flow cell and ion source of the MS are known. The main reason for using flow cells based on liquid scintillants is higher sensitivity compared to the solid scintillants.

HPLC-MS-Based Drug Metabolite Profiling

189

7. Fraction collection can provide a means of producing radiochromatograms in the absence of an online radiodetector or, alternatively, where the amounts of radioactivity are too low for online detection. Fractions can be collected into vials for conventional determination using scintillation counters or into coated 96-well plates for determination using systems such as “TopCount” (this utilises a yttrium silicate embedded “LumaPlate” for fraction collection and can be used to rapidly determine the radioactivity in each well thereby reducing the overall time taken for determining the profile (see ref. 5) (TopCount NXT and LumaPlates, Perkin Elmer LAS, Buckinghamshire, UK). Fraction collection has the disadvantage that it is relatively slow compared to online monitoring. 8. The use of UPLC provides much more efficient chromatography than conventional HPLC, potentially resulting in much better resolution of metabolite peaks. This improved chromatographic efficiency can be used to advantage in two ways: either reducing run times from 30–50 to 10–15 min without loss of resolution or by obtaining much greater resolution of metabolites (from each other and from endogenous compounds or contaminants such as PEG) in the case of extensive and complex metabolism. 9. If further MS or NMR studies are to be performed on isolated metabolites, then care should be taken to ensure that the collection and subsequent sample handling conditions are suitable for preserving unstable metabolites such as acyl glucuronides (e.g. see Note 3). References 1. Ma, S., Chowdhury, S. K. (2008) Application of liquid chromatography/mass spectrometry for metabolite identification, in (Zhang, D., Zhu, M., Humphreys, W. G., eds.), Drug Metabolism in Drug Design and Development, pp. 319–368. Wiley, New Jersey. 2. Johnson, K. A., Plumb, R. (2005) Investigating the human metabolism of acetaminophen using UPLC and exact mass oa-TOF MS. J Pharm Biomed Anal 39, 805–810. 3. Zhu, M., Zhao, W., Humphreys, W. G. (2008) Applications of liquid radiochromatography techniques in drug metabolism studies, in (Zhang, D., Zhu, M., Humphreys, W. G., eds), Drug Metabolism in Drug Design and Development, pp. 319–368. Wiley, New Jersey.

4. Athersuch, T. J., Sison, R. L., Kenyon, A. S. J., Clarkson-Jones, J. A., McCormick, A. D., Wilson, I. D. (2008) Evaluation of the use of UPLC-tof/MS with simultaneous [14 C]radioflow detection for drug metabolite profiling: application to propranolol metabolites in rat urine. J Pharm Biomed Anal 48, 151–157. 5. Dear, G. J., Patel, N., Kelly, P. J., Webber, L., Yung, M. J. (2006) Topcount coupled to ultra-performance liquid chromatography for the profiling of radiolabeled drug metabolites in complex biological samples. J Chromatogr B 884, 96–103. 6. Lindon, J. C., Nicholson, J. K., Wilson, I. D. (1996) Direct coupling of chromatographic separations to NMR spectroscopy. Prog Nucl Magn Res 29, 1–49.

190

Wilson

7. Scarfe, G. B., Wright, B., Clayton, E., Taylor, S., Wilson, I. D., Lindon, J. C., Nicholson, J. K. (1999) Quantitative studies on the urinary metabolic fate of 2-chloro-4trifluoromethylaniline in the rat using 19 FNMR spectroscopy and directly coupled HPLC-NMR-MS. Xenobiotica 29, 77–91. 8. Duckett, C. J., Lindon, J. C., Walker, H., Abou-Shakra, F., Wilson, I. D., Nicholson, J. K. (2006) Metabolism of 3-chloro4-fluoroaniline in rat using [14 C]radiolabelling, 19 F-NMR spectroscopy, HPLC-MS/MS, HPLC-ICPMS and HPLCNMR. Xenobiotica 36, 59–77. 9. Anari, M. R., Sanchez, R. I., Bakhtiar, R., Franklin, R. B., Baillie, T. A. (2004) Integration of knowledge-based metabolic predicitions with liquid chromatography datadependent tandem mass spectrometry for drug metabolism studies: application to studies on the biotransformation of indinvar. Anal Chem 76, 823–832. 10. Levesen, K., Schiebel, H. M., Behnke, B., Dotzer, R., Dreher, W., Elend, M.,

Thiele, H. (2005) Structure elucidation of phase II metabolites by tandem mass spectrometry: an overview. J Chromatogr A 1067, 55–72. R (2000) Chromatography Optimisa11. Drylab tion Software, version 3.0.09, Molnar-Institut für Angewandte Chromatographie, Berlin, Germany. 12. Scarfe, G. B., Nicholson, J. K., Lindon, J. C., Wilson, I. D., Taylor, S., Clayton, E., Wright, B. (2002) Identification of the urinary metabolites of 4-bromoaniline and 4bromo-[carbonyl-13 C]-acetanilide in the rat. Xenobiotica 32, 325–337. 13. Shillingford, S., Bishop, L., Smith, C. J., Payne, R., Wilson, I. D., Edge, A. M. (2009) Application of high temperature LC to the separation of AZD5438 (4- (1isopropyl-2-menythyl-1 H-imidazol-5-yl)N-4(methylsulphonyl))phenyl]pyrimidin-2amine) and its metabolites: comparison of UPLC and HTLC. Chromatographia 70, 37–44.

Chapter 11 Gas Chromatography-Mass Spectrometry (GC-MS)-Based Metabolomics Antonia Garcia and Coral Barbas Abstract Metabolic fingerprinting, the main tool in metabolomics, is a non-targeted methodology where all detectable peaks (or signals), including those from unknown analytes, are considered to establish sample classification. After pattern comparison, those signals changing in response to a specific situation under investigation are identified to gain biological insight. For this purpose, gas chromatographymass spectrometry (GC-MS) has a drawback in that only volatile compounds or compounds that can be made volatile after derivatization can be analysed, and derivatization often requires extensive sample treatment. However, once the analysis is focused on low molecular weight metabolites, GC-MS is highly efficient, sensitive, and reproducible. Moreover, it is quantitative, and its compound identification capabilities are superior to other separation techniques because GC-MS instruments obtain mass spectra with reproducible fragmentation patterns, which allow for the creation of public databases. This chapter describes well-established protocols for metabolic fingerprinting (i.e. the comprehensive analysis of small molecules) in plasma and urine using GC-MS. Guidelines will also be provided regarding subsequent data pre-treatment, pattern recognition, and marker identification. Key words: Pattern recognition, urine, plasma, multivariate analysis, silylation, volatile compounds, gas chromatography-mass spectrometry.

1. Introduction Gas chromatography coupled to mass spectrometry (GC-MS) has been regarded as the gold standard for analysing many compounds (lipids, drug metabolites, and environmental contaminants), as well as for forensic science. One of the advantages of GC-MS is that identification of detected species is based on both a retention time and a mass spectrum (a compound’s specific fragmentation pattern). Compounds produce reproducible T.O. Metz (ed.), Metabolic Profiling, Methods in Molecular Biology 708, DOI 10.1007/978-1-61737-985-7_11, © Springer Science+Business Media, LLC 2011

191

192

Garcia and Barbas

fragmentation patterns when ionized by a fixed electron voltage (usually –70 eV). Thus, the fragmentation spectra obtained by GC-MS are not instrument dependent and allow for the creation of databases and the sharing of data between users, making the technique particularly valuable. In addition, GC-MS allows for quantitative detection of analytes. In classical methods, measuring either a single compound or a set of chemically related substances (such as short chain fatty acids or amino acids), there are clearly established protocols for sample treatment and for quantification using the appropriate standards. In the last few years (1), a new analytical strategy has evolved for obtaining a global view of the metabolic status of an organism. “Metabo(l/n)omics” is an approach capable of generating a comprehensive data set of metabolites. This is not only a new terminology but a new approach to the analytical problem with very different analytical requirements. In an “omics” methodology, the expectation is that the response pattern(s) of numerous analytes, both known and unknown, is reflective of a biological condition, and the comprehensive nature of the data set enables evaluation of systemic response. The broader scope of the analysis forces compromises in method development (e.g. sample extraction, cleanup, derivatization, chromatography) and requires flexibility in accuracy criteria for specific metabolites. On the one hand, GC-MS is far from the ideal of metabolomics; not only is it limited to compounds that either are volatile or can be made volatile through the derivatization process, but also all nonvolatile compounds must be carefully removed from the sample before analysis, which requires demanding sample treatment. However, on the other hand, it is clear that no single analytical platform is capable of detecting the whole set of metabolites in a biological sample. 1 H-NMR, capable of measuring intact samples and, in principle, ideal, is limited by a bias towards analytes with medium to high concentrations. In contrast, GC-MS is highly sensitive and reproducible and permits working with standard libraries for identification of detected species (2); it is widely used, although only a certain subset of compounds will be analysed. This chapter describes and explains guidelines for metabolite fingerprinting of plasma and urine by GC-MS. It gives experimental details on basic steps such as sample collection, storage, protein precipitation, extraction, concentration, derivatization, data acquisition, raw data processing, and data transformation. In general, procedure descriptions are as straightforward and comprehensive as possible and have been optimized and validated from previously described protocols in our laboratory. With this method, organic acids, amino acids, mono- and di-saccharides, sterols, and other compounds can be detected in the same analysis.

Gas Chromatography-Mass Spectrometry (GC-MS)-Based Metabolomics

193

2. Materials 2.1. Chemicals

1. Standard compounds (Sigma, St. Louis, MO, or Aldrich, Steinheim, Germany): 2-aminobutyric acid, 2-hydroxybutyric acid, 2-hydroxyisobutyric acid, 2ketoglutaric acid, 3-hydroxybutyric acid, 4-hydroxybutyric acid, 4-hydroxyproline, 5-hydroxy-3-indolacetic acid, alanine, asparagine, aspartic acid, cholesterol, citric acid, creatinine, ethylene diamine tetraacetic acid (EDTA) disodium salt, estearic acid, fructose, fumaric acid, glucitol, gluconic acid, glucose, glutamic acid, glutamine, glutaric acid, glyceric acid, glycine, glycolic acid, hexanoic acid, isoleucine, lauric acid, leucine, linoleic acid, malic acid, methyl estearate (IS), methionine, myo-inositol, myristic acid, oleic acid, ornithine, oxalacetate, oxalate, palmitic acid, phenylalanine, pyroglutamate, proline, pyruvic, sarcosine, serine, sodium lactate, succinate, threonine, tryptophan, tyrosine, urea, uric acid, and valine. These standards are of analytical grade except where stated otherwise. Stock solutions of the reference compounds should be prepared at 25 mM either in Milli-Q water or in methanol. Standard solutions should be kept at –80◦ C. Before analysis, the solutions should be thawed and diluted (1:100) with methanol. These compounds are only necessary for confirming the identification of detected metabolites, and the list can be modified as necessary by the researcher. 2. Urease (Sigma-Aldrich, Poole, UK): 600 units/mL (0.0085 g/mL) in Milli-Q water. 3. Pentadecanoic methanol.

acid

(Sigma-Aldrich):

6

mg/mL

in

4. C18:0 methyl ester (Sigma): 1000 ppm stock solution diluted to 10 ppm with heptane. 5. N,O-Bis(trimethylsilyl)trifluoroacetamide (BSTFA) plus 1% trimethylchlorosilane (TMCS; Pierce Chemical Co, Rockford, IL, USA). 6. O-Methoxyamine hydrochloride (Sigma): 15 mg/mL in pyridine. This solution can be stored in the freezer for 2 weeks. 7. Silylation-grade pyridine. 8. Analytical-grade heptane. 9. HPLC-grade methanol. 10. HPLC-grade acetone.

194

Garcia and Barbas

11. HPLC-grade acetonitrile. 12. Ultrapure water can be produced by a Milli-Q Reagent Water System (Millipore, MA, USA) or similar (see Notes 1 and 2). 2.2. Samples

1. Plasma 2. Urine: First-void urine samples are preferred compared to spot urine samples because the influence of lifestyle factors (e.g. diet, physical exertion, stress) on the metabolic urinary profiles is relatively minimized and normalized in the case of first-void urine (3).

2.3. Equipment

1. Centrifuge. 2. Microcentrifuge (Hettich, Tuttlingen, Germany). 3. Speedvac concentrator (Thermo Fisher Scientific, Waltham, MA). 4. 0.22 μm filters (Millipore, Billerica, MA). 5. 3 mL Vacuette plastic tubes containing 4.5 mg of EDTA·K3 for plasma collection (Greiner Bio One, Brazil). 6. 2 mL GC-HPLC glass vials with 200 μL conical narrow opening inserts. 7. Ultrasonic bath. 8. Digitheat 80 L stove (Selecta, Barcelona, Spain). 9. Vortex mixer.

2.4. GC-MS Equipment

1. 3900 Series GC (Varian, Palo Alto, CA). Electronic Flow Control provides constant flow separation. 2. CP-8400 AutoSampler (Varian). 3. Saturn 2110T Series MS ion trap (Varian). D Alternative mass spectrometers, such as quadrupole, time-of-flight, or triple quadrupole, can be used. 4. National Institute of Standards and Technology (NIST) mass spectra library. 5. GC capillary column: VF-5 ms, 30 m × 0.25 mm ID, 0.25 DF, (Varian).

2.5. Software

1. Matlab version 7.0 (MathWorks, Natick, MA). 2. MassTransit version 3.0 (Palisade Corporation, Ringoes, NJ). This is a Microsoft Windows application for reading, displaying, and converting GC-MS data files between different instrument vendor formats. 3. SIMCA P+ version 12 (Umetrics, Umeå, Sweden). This software is used for multivariate analysis of data.

Gas Chromatography-Mass Spectrometry (GC-MS)-Based Metabolomics

195

3. Methods 3.1. Metabolite Fingerprinting in Plasma Samples

3.1.1. Plasma Preparation

Blood plasma is a heterogeneous mixture of lipoprotein particles, proteins, small organic molecules, and ions, which together undergo a variety of possible molecular interactions including metal complexation, chemical exchange processes, micellar compartmentation of metabolites, enzyme-mediated biotransformations, and small molecule–macromolecular binding (4). The low molecular weight compounds, in particular, are chemically diverse and vary widely in concentrations and stabilities and are commonly noncovalently bound to proteins (5). The following analytical procedure is based on two methods previously described (4, 6). The whole process consists of three steps: deproteinization, ketone functional group protection by methoximation, and derivatization to increase the volatility of the metabolites. Deproteinization removes the analytical interference due to proteins; albumin is the most abundant blood protein with a typical concentration of 34–50 g/L (7). After deproteinization, samples should be dried to permit the action of derivatizating agents. Methoximation utilizes CH3 ONH2 in pyridine to stabilize carbonyl moieties by suppressing keto–enol tautomerism and the formation of multiple acetal- or ketal-structures. It also helps to reduce the numbers of derivatives of reducing sugars and generates only two forms of the –N=C< derivative, syn and anti. Silylation involves replacement of active hydrogen atoms with an alkylsilyl group, for example, –SiMe3 . Functional groups that present a problem with volatility – hydroxyl (–OH), carboxyl (–COOH), amine (–NH2 ), thiol (–SH), and phosphate (–PO4 3– ) – can be derivatized by silylation reagents. Importantly, while the (–OH) and (–COOH) groups react simultaneously and very fast, the hydrogen atoms of the (–NH2 ) groups react more slowly (1) and can yield multiple derivatives of some compounds. Silylation thus increases the volatility and thermal stability of polar metabolites (8).

1. Collect fresh blood in the EDTA-treated tubes. 2. Centrifuge anticoagulated blood quickly at 1000×g for 10 min at 4◦ C. 3. Collect the supernatant and freeze the plasma at –80◦ C in 100 μL aliquots. 4. Prior to further manipulation, plasma samples must be thawed, vortex-mixed, and filtered through a 0.22 μm filter before use (see Notes 3–6).

196

Garcia and Barbas

3.1.2. Plasma Deproteinization

1. On ice, pipette 40 μL of plasma to a 650 μL Eppendorf, vortex-mix it with 120 μL of cold acetonitrile for 2 min and let stand for 5 min. Maintain the ratio of acetonitrile to sample at 3:1 to guarantee the precipitation of the proteins. 2. Centrifuge the sample at 15,400×g for 10 min at 4◦ C with the microcentrifuge. 3. Transfer 100 μL of the supernatant to a GC vial equipped with a 200 μL insert and evaporate to dryness by speedvac at 30◦ C. Check to assure the vial is completely dry.

3.1.3. Methoximation

1. Add 10 μL of O-methoxyamine hydrochloride (15 mg/mL) in pyridine to each GC vial and mix vigorously for 1 min. 2. Ultrasonicate the samples three times for 10 s and vortex again for 2 min. 3. Cover the GC vials in aluminium foil and incubate at room temperature for 16 h (see Notes 7 and 8) in the dark.

3.1.4. Derivatization

1. Add 10 μL of BSTFA with 1% TMCS as catalyst to each sample, then vortex-mix for 5 min. 2. Place the GC vials in an oven for 1 h at 70◦ C for silylation. 3. Add 100 μL of heptane containing 10 ppm of C18:0 methyl ester (IS) to each GC vial and vortex-mix for 2 min before GC analysis (see Note 9).

3.2. Metabolite Fingerprinting in Urine Samples

3.2.1. Urine Preparation

Due to the fact that the urine contains metabolic signatures of many biochemical pathways, this biofluid is ideally suited for metabolomic analysis. However, the urine metabolome is highly variable with respect to the composition and quantity of compounds in response to diet, medication, and metabolic state. Therefore, metabolic fingerprinting in urine can be quite challenging. Urease is used to deplete urea, which can cause major chromatographic interference and mask many low-intensity metabolite peaks. Finally, since most compounds identified in urine are also present in blood plasma, the protocol of urine extraction and derivatization is based on that of plasma. 1. Thaw frozen urine samples at room temperature (25 ± 3◦ C) (see Notes 16–18). 2. Add 30 units (50 μL) of urease to 200 μL of urine and incubate at 37◦ C in a stove for 30 min to decompose and remove excess urea.

Gas Chromatography-Mass Spectrometry (GC-MS)-Based Metabolomics

197

3. Add 800 μL methanol to urine samples to precipitate urease and other proteins and add 10 μL pentadecanoic acid (6 mg/mL) as internal standard. 4. Vortex-mix the samples for 5 min and then centrifuge at 15,400×g for 10 min at 4◦ C. 5. Transfer 200 μL of the supernatant to a GC vial and then evaporate to dryness by speedvac at 30◦ C. Check to assure the vial is completely dry. 3.2.2. Methoximation

1. Add 30 μL of O-methoxyamine hydrochloride (15 mg/mL) in pyridine to each GC vial and mix vigorously for 10 min. 2. Ultrasonicate the samples three times for 10 s and vortex again for 2 min. 3. Cover the GC vials in aluminium foil and incubate at room temperature for 16 h in the dark (see Notes 7 and 8).

3.2.3. Derivatization

1. Add 30 μL of BSTFA with 1% TMCS as catalyst to each sample, then vortex-mix for 5 min. 2. Place the GC vials in an oven for 1 h at 70◦ C for silylation. 3. Add 40 μL of heptane to each GC vial and vortex-mix for 10 min before GC analysis (see Note 9).

3.3. GC-MS Analysis

1. Inject 1 μL (splitless) or 2 μL (split ratio of 1:10) of derivatized urine or plasma samples, respectively, into the GC-MS system. 2. Set the helium carrier gas flow rate to 1 mL/min through the column. 3. Set the injector temperature to 250◦ C. 4. Programme the temperature gradient as follows: set initial oven temperature at 50◦ C, hold for 1 min, and then increase temperature at the rate of 3.3◦ C/min to a final temperature of 340◦ C. 5. Set the detector transfer line, trap, and manifold temperatures to 280, 200 and 60◦ C, respectively. 6. Operate the electron ionization source at –70 eV. 7. Operate the mass spectrometer in scan mode only over a mass range of 50–650 m/z at a rate of 5 spectra s–1 (see Notes 10–15).

3.4. Data Analysis

Check the quality of individual chromatograms. In addition to visual inspection, establishing the internal standard signal over a certain value could be a good quality control criterion. The main steps in data analysis are summarized in Fig. 11.1.

198

Garcia and Barbas

DATA ANALYSIS

1,0 0,8 0,6

'c4p' 'c1p' 'c2p' 'c5p'

t[2]

'd1p'

0,0

'c7p' 'd7p'

0

R2X

R2X

R2X Q2

Q2

Q2

'c8p' 'c14p' 'c11p'

Comp[3]

'c15p'

Comp[2]

0,2

10

Comp[1]

0,4

Before multi alignment

'd16p' 'd5p' 'd3p'

After multi alignment

'd19p' 'd15p'

–10 'c3p'

'd8p'

–02

–10

0 t[1]

10

20

MULTIVARIATE ANALYSIS

DATA PRE-TREATMENT Glucose

CONTROL

DIABETIC

MARKER IDENTIFICATION

Fig. 11.1. Scheme of steps in data analysis.

3.4.1. Data Pre-processing

Data can be pre-processed as defined peaks or with the total profile of data points. In any case, the appropriate software should be employed. In our laboratory data treatment was performed as follows: 1. Export each chromatogram with the total ion chromatogram (TIC) signal as TXT format to Excel or ASCII. These last formats can be opened with the MATLAB program. 2. Generate a matrix of data with one column per sample (include all the samples of the metabolomic experiment) and one row per MS signal. 3. Delete the zero signals at the beginning and at the end of the chromatogram. It is very important to keep the same number of data points per sample/column to configure the matrix of data. 4. Simultaneously, export each MS file to MATLAB format with MassTransit (9). Each new file contains the intensity of each m/z fragment detected in any scan per sample. During a standard chromatogram, more than 3500 scans are performed. The new matrix is obtained with the intensity of the sorted m/z fragments in columns and the different samples in rows. 5. The matrix of data is submitted first to baseline correction and then to multialignment by comparing the mass spectra

Gas Chromatography-Mass Spectrometry (GC-MS)-Based Metabolomics

199

in points to be aligned. Finally, normalization by dividing by the internal standard peak is applied. R 1. Load the chromatographic matrix of clean data to Matlab or SIMCA for multivariate data analysis, for example, principal components analysis (PCA), principal least squaresdiscriminant analysis (PLS-DA), or orthogonal projection to latent structure-discriminant analysis (OPLS-DA).

3.4.2. Multivariate Statistical Tools

2. Once samples cluster according to the pathology or problem under study and chemometric models are adequately validated, identify signals accounting for the classification in the loading plot as a variable number. The next step in the analytical process is transforming a variable number in an identified metabolite. Critical to this process is the reliance on both retention time index and spectral databases. Both of these are publicly available through web-based platforms, which have been recently reviewed by Tohge et al. (10) (see chromatogram (TIC) of a plasma sample is shown in Fig. 11.2. Most

3.4.3. Metabolite Identification

kCounts 700

IDENTIFIED COMPOUNDS (A) 32

600 16 500

45

400

31 14

300

22 21

200 1 100

4 5

10

6 7

10

41

50 48

12 23

11

28

24 25

30

33

36 35 38 37

43 42

46

44

49

47

40

19

17

15

kCounts 700

39

18 15

2

29

27

34

13

3 0

26

20

9 8

20

25

30

minutes 40

35

70

54

IDENTIFIED COMPOUNDS (B)

600

*

500 61 400

*

67

59 300

* 60

65 66

200

52 0

51

40

62

45

78

77

*

73 71

63 64 57 55 56 58

*

75

*

53

100

*

*

74 68 69

50

72

76

55

60

65

70 minutes

Fig. 11.2. Plasma fingerprint obtained with GC-MS. For conditions see the text. Numbers correspond to compounds identified that are described in Table 11.1.

50. Unknown 51. Citrate (standard confirmed) 52. Isocitrate (standard confirmed)

24. Glycerate

25. Unknown

26. Serine II (standard confirmed)

48. Ornithine (standard confirmed) 49. Dehydrated citrate

22. Glycine II (standard confirmed)

21. Threonine I (standard confirmed) + proline (standard confirmed)

23. Succinate (standard confirmed)

46. Lysine 47. Xylitol

20. Isoleucine (standard confirmed)

44. Asparagine (standard confirmed) 45. Unknown

19. Unknown + glycerol (standard confirmed)

17. Urea (standard confirmed)

18. Serine I (standard confirmed)

42. Unknown 43. Laurate (standard confirmed)

16. Valine II (standard confirmed)

41. Glutamate (standard confirmed) + phenylalanine II (standard confirmed)

15. 2-Aminobutyrate (standard confirmed)

37. Creatinine (standard confirmed)

11. Oxalate I (standard confirmed)

40. Ketoglutarate (standard confirmed) + phenylpyruvate (standard confirmed)

36. Phenylalanine I (standard confirmed)

10. Oxalate I (standard confirmed)

14. Silica derivative

35. 2,3,4-tri-OH-butyrate I

9. 2-OH-butyric (standard confirmed)

38. 2,3,4-tri-OH-butyrate II

34. Aspartate I (standard confirmed)

8. Glycine I (standard confirmed)

39. Unknown

33. 4-OH-proline (standard confirmed)

7. Unknown

13. 3-OH-butyrate (standard confirmed)

32. Methionine (standard confirmed) + pyroglutamate (standard confirmed) + glutamine (standard confirmed)

6. Alanine I (standard confirmed)

12. Silica derivative

30. Malic acid (standard confirmed) 31. Butylhydroxytoluene

5. Valine I (standard confirmed)

3. Oxalacetic acid

4. Glycolate (standard confirmed)

28. Alanine II (standard confirmed) 29. Aminomalonate

2. Lactate (standard confirmed)

27. Threonine II (standard confirmed)

1. Pyruvate (standard confirmed)

List of identified compounds in human plasma by GC-MS

78. Cholesterol (standard confirmed)

77. Unknown

76. EDTA (standard confirmed)

75. Trimethylsilyl Estearate

74. 11-trans-octadecenoate

73. 11-cis-octadecenoate

72. Linoleate (standard confirmed)

71. Tryptophan (standard confirmed)+5-OHindoleacetic (standard confirmed)

70. Unknown

69. Internal standard – Octadecanoic methyl ester

68. Urate (standard confirmed)

67. Myoinositol (standard confirmed)

66. Palmitate (standard confirmed)

65. Glucose derivative

64. Inositol

63. Gluconate (standard confirmed)

62. Glucose + glucopyranose

61. Unknown

60. Tyrosine (standard confirmed)

59. Glucitol (standard confirmed)

58. Glucose

57. Glucose

56. Fructose II (standard confirmed) + glucose

55. Fructose I (standard confirmed)

54. Myristic acid (standard confirmed)

53. Galactofuranose

Table 11.1 Compounds identified in plasma either by the library or by the corresponding standard in plasma fingerprints 200 Garcia and Barbas

Gas Chromatography-Mass Spectrometry (GC-MS)-Based Metabolomics

201

of the peaks in the chromatograms can be identified as endogenous metabolites by the NIST mass spectra library and many were compared with pure standards, as an additional confirmation. For that purpose, standards prepared as described in Section 2.1 and subjected to the derivatization procedure were analysed either individually or in groups with very different characteristics, and a customized database was developed including retention time and mass spectra. A candidate compound should match both criteria to be confirmed. In the case of isobaric and closely eluting compounds, the standard was also spiked into the sample for confirmation. Identifications are listed in Table 11.1. Finally, a metabolomics experiment should not be a list of compounds increasing or decreasing with the situation under study; therefore, the last step should be the search for the biological interpretation of those changes.

4. Notes 1. Never prepare a new reactant solution during the experiment. Think about the total number of samples, and the total volume of any solution required, and prepare each reactant solution once. 2. Pyridine is a hazardous solvent. Take appropriate precautions. 3. When designing a metabolomics study in humans, control subjects are a main challenge. Study carefully how to select the appropriate controls, matching in age and sex, and also ensure that samples are withdrawn, stored, and processed simultaneously. 4. Plasma samples can be established at withdrawal with citric acid, instead of EDTA. However, this makes the measurement of endogenous citrate impossible. 5. Samples thawed more than twice are not recommended. 6. It is very important to consider variables involved in sample collection, for example, the handling and storage of samples, the types of collection tubes and anticoagulants. The profile may change and disturb the biomarker pattern. Commercially available blood collection tubes contain multiple components that may appear as interfering or confounding peaks during the MS analysis. Silicones are commonly used as lubricants for stoppers or coatings for the internal surface of tubes. Polymeric surfactants, such as polyvinylpyrrolidones or polyethylene glycols, may be added to influence surface wetting (11). For those reasons,

202

Garcia and Barbas

changes in collection tubes during a metabolomics study are never recommended. 7. After the methoximation period, the solution in all the vials must be clear with no solid suspension material. If the sample cannot be analysed immediately, it must not be kept in a refrigerator or freezer but only in a dark place. Otherwise, humidity can hydrolyse it. 8. Total evaporation of solvent prior to methoximation is needed because this reaction will be slower in aqueous media. Minimize the variation between samples with complete drying. 9. During derivatization, some compounds yield different derivatives depending on which acid hydrogen is substituted by the trimethylsilyl group. Each trimethylsilyl group adds +72 Da to the molecule. 10. Use vials with glass insert tubes already included, because the cap closes more tightly. Check if any solvent could be lost during the procedure. 11. In GC-MS-based metabolomics, a method for monitoring the quality of the results and the overall analytical procedure is required. In that way, as the experiment starts, a large volume of a homogeneous sample must be aliquoted to be used in quality control (QC) samples and frozen at –80◦ C. On the day of analysis, one or several aliquots are thawed and deproteinized and derivatized along with the rest of the samples. 12. In normal planning, the operator prepares 12 samples plus 2 quality controls every day. The sequence in the GC could be one injection of blank solution containing 10 ppm IS in heptane (this is to check the blank profile and for conditioning the column to the temperature programme); one QC sample; six unknown samples randomly organized; one QC sample; and finally repeat the sequence except the injection of the blank solution. 13. For standardizing the GC-MS data acquisition process and minimizing biases that cannot be corrected through the use of an internal standard, the metabolomic profiles obtained must be within the linear range of the GC-MS equipment’s operation, and the sequence of analysis must be randomized as much as possible to analyse samples coming from all the groups. In addition, check if the instrument detector is working properly for every analysis day and ensure that all the chromatographic profiles are filtered from any artefacts and adducts. 14. A pre-column of around 10 m length, same diameter, and same film thickness is recommended to preserve the analytical column and the detector.

Gas Chromatography-Mass Spectrometry (GC-MS)-Based Metabolomics

203

15. Running the first 8 min with the filament switched off is recommended when the solvent and any excess of derivatizating reagents appear, in order to extend filament performance in the mass detector. 16. Urine concentrations can be very different, as in the case of streptozotocin diabetic rats and their controls; urinary creatinine may be used as a correction factor. Adjust sample dilution to the same creatinine value before starting the fingerprinting assay. 17. Boric acid, a common preservative in urine collection, interferes with urease activity giving variable urea concentrations and precipitates as insoluble salt during the derivatization process. 18. Urine is commonly treated with urease to reduce the risk of column overloading, peak distortions, or matrix effects (12); urea produces a huge peak at around 19 min more than 5 min wide. Kind et al. (13) observed when comparing urine analysis with and without urease treatment that many other metabolites, including acotinic acid, hypoxanthine, and the tricarboxylic acid (TCA) intermediates, were modified. However, in our experience, alignment when urea is present was impossible, even cutting the section of the peak. Therefore, the risk in using urease treatment is necessary. 19. Looking at the profile, the major component in plasma is glucose with a regulated concentration of 5 mM in humans, in order to provide energy to all organs, especially to the brain. 20. There are software tools that assist in a semi-automated peak identification; however, some limitations for the optimum assignation are chromatographic drift, very similar chemical structures of molecules in biologically significant classes (e.g. mono- or disaccharides), the vast number of peaks of yet unknown origin, the occurrence of peaks that are products of the GC column bleeding, and multiple derivatives of particular metabolites that are produced at different derivatization times.

Acknowledgements The authors acknowledge Joanna Teul for her careful experimental work and funding support from the Comunidad de Madrid, S-GEN-0247-2006 and Ministry of Science and Technology (MCIT) CTQ2008-03779.

204

Garcia and Barbas

References 1. Kanani, H., Chrysanthopoulos, P. K., Klapa, M. I. (2008) Standardizing GC– MS metabolomics. J Chromatogr B 871, 191–201. 2. Pasikanti, K. K., Ho, P. C., Chan, E. C. Y. (2008) Gas chromatography/mass spectrometry in metabolic profiling of biological fluids. J Chromatogr B 871, 202–211. 3. Lenz, E. M., Bright, J., Wilson, I. D., Morgan, S. R., Nash, A. F. (2003) A 1 H NMR-based metabonomic study of urine and plasma samples obtained from healthy human subjects. J Pharm Biomed Anal 33, 1103–1115. 4. Daykin, C. A., Foxall, P. J. D., Connor, S. C., Lindon, J. C., Nicholson, J. K. (2002) The comparison of plasma deproteinization methods for the detection of low-molecularweight metabolites by 1 h nuclear magnetic resonance spectroscopy. Anal Biochem 304, 220–230. 5. Wu, S. L., Amato, H., Biringer, R., Choudhary, G., Shieh, P., Hancock, W. S. (2002) Targeted proteomics of low-level proteins in human plasma by LC/msn: using human growth hormone as a model system. J Proteome Res 21, 253–262. 6. Jiye, A., Trygg, J., Gullberg, J., Johansson, A. I., Jonsson, P., Antti, H., Marklund, S. L., Moritz, T. (2005) Extraction and

7. 8.

9. 10.

11.

12. 13.

GC/MS analysis of the human blood plasma metabolome. Anal Chem 77, 8086–8094. Tietz, N. W. (1986) Textbook of Clinical Chemistry, Saunders, Philadelphia, PA. p. 590. Halket, J. M., Waterman, D., Przyborowska, A. M., Patel, R. K. P., Fraser, P. D., Bramley, P. M. (2005) Chemical derivatization and mass spectral libraries in metabolic profiling by GC/MS and LC/MS/MS. J Exp Bot 56, 219–243. Palisade Corporation web page. http://www.palisade.com. Accesed May 2009. Tohge, T., Fernie, A. R. (2009) Webbased resources for mass-spectrometry-based metabolomics: a user’s guide. Phytochemistry 70, 450–456. Luque-Garcia, J. L., Neubert, T. A. (2007) Sample preparation for serum/plasma profiling and biomarker identification by mass spectrometry. J Chromatogr A 1153, 259–276. Kuhara, T. (2005) Metabolomics: The Frontier of Systems Biology, Springer, Tokyo. Kind, T., Tolstikov, V., Fiehn, O., Weiss, R. H. (2007) A comprehensive urinary metabolomic approach for identifying kidney cancer. Anal Biochem 363, 185–195.

Chapter 12 The Use of Two-Dimensional Gas Chromatography–Time-of-Flight Mass Spectrometry (GC×GC–TOF-MS) for Metabolomic Analysis of Polar Metabolites Kimberly Ralston-Hooper, Amber Jannasch, Jiri Adamec, and Maria Sepúlveda Abstract Metabolites produced by an organism can be quite extensive, and one analytical technique alone is not capable of their comprehensive detection and identification. The majority of environmental metabolomic studies have implemented proton nuclear magnetic resonance (1 H-NMR) spectroscopy with little attention given to mass spectrometry (MS) techniques. In this chapter, an analytical technique is outlined that incorporates two-dimensional gas chromatography–time-of-flight MS (GC×GC–TOF-MS) for the identification and quantification of polar metabolites. Key words: Two-dimensional gas chromatography, time of flight, mass spectrometry.

1. Introduction The use of metabolomics in environmental sciences as a tool to help understand the impacts of different stressors on organisms has substantially increased over the past few years (1–5). As with any scientific field, there are numerous analytical technologies available for metabolome analysis (6). Overall, proton nuclear magnetic resonance (1 H-NMR) spectroscopy has been the most popular choice for environmental metabolomic studies (7–10). Limitations of NMR-based metabolomic techniques include their relatively low sensitivity that results in an examination of a limited T.O. Metz (ed.), Metabolic Profiling, Methods in Molecular Biology 708, DOI 10.1007/978-1-61737-985-7_12, © Springer Science+Business Media, LLC 2011

205

206

Ralston-Hooper et al. Methanol:Chloroform Extraction

Methanol (Polar Metabolites)

Dry Sample

Chloroform (Non-Polar Metabolites) LC/TOF-MS Analysis

See Chapter 15 for more details

Derivatize Sample

GCxGC/TOF-MS Analysis

Metabolite IdentificationMatched against NIST database (SV > 750)

Fig. 12.1. Flowchart of sample preparation steps used for GC×GC–TOF-MS analysis.

number of metabolites (11, 12). Alternatively, mass spectrometry (MS) coupled with a separation technique such as gas chromatography (GC) has been successfully utilized for metabolome analysis in other fields such as human disease and plant metabolism (13, 14) but has not been extensively used for environmental metabolomic studies. For example, GC×GC–TOF-MS has been used successfully for characterizing metabolite profiles of complex biological mixtures (15–17), as well as to study metabolomic profiles of atrazine-exposed invertebrates (18). Since the number of metabolites produced by an organism is quite extensive, one instrument or technique will not be able to detect and identify all metabolites. Using multiple analytical techniques such as GC×GC–TOF-MS and 1 H-NMR can be extremely advantageous. Both these instruments coupled with bioinformatics or multivariate analysis of complex datasets could be beneficial in biomarker identification. In this chapter, a method utilizing MS– metabolomic techniques is outlined (see Fig. 12.1).

2. Materials 2.1. Extraction Solvents

1. Chloroform (>99.9% HPLC grade). 2. Methanol (>99.9% HPLC grade). 3. 18.2 M cm water (see Note 1).

The Use of Two-Dimensional Gas GC×GC–TOF-MS for Metabolomic Analysis

2.2. Derivatization Agents

207

1. Anhydrous pyridine (99.8% purity). 2. O-Methylhydroxylamine hydrochloride (>98% purity Fluka, St. Louis, MO): 20 mg/mL anhydrous pyridine (see Note 2). 3. N-Methyl-N-(tert-butyldimethylsilyl) trifluoroacetamide (MSTFA) (>98.5% GC grade; Thermo Fisher Scientific, Waltham, MA) (see Note 3).

2.3. Equipment

1. Axygen 1.7-mL microtubes (Axygen Scientific, Inc., Union City, CA). 2. Sawtooth stainless steel tissue homogenizer TH-01 (Omni, Marietta, GA). 3. Sonicator (Ultrasonic Power Corporation, Freeport, IL). 4. Sevant SpeedVac Concentrator SPD 131DDA (Thermo Electron Corporation, Milford, MA). R R III and ChromaTOF , 5. GC×GC–TOF-MS Pegasus Leco Corporation, V. 2.32 (Leco Corporation, St. Joseph, MI).

6. First-dimension column: Restek Rtx-5 (10 m, 0.18 mm i.d., 0.2 μm d.f.). 7. Second-dimension column: DB-17 (1.1 m, 0.18 mm i.d., 0.18 μm d.f.). R 8. Statistical Analysis Software, SAS , V. 9.1.3 (http://www. sas.com/technologies/analytics/statistics/stat/index.html).

9. R statistical software, R Development Core Team, V. 2.10.0 (http://www.r-project.org). 10. PLS_ Toolbox Advanced Chemometrics, V. 3.5, R Central, (http://software.eigenvector.com/ Matlab toolbox/3_5/index.html). 11. National Institute of Standards and Technology (NIST) database (http://www.nist.gov/srd/index.htm).

3. Methods 3.1. Sample Preparation

1. Samples that have been stored at –80◦ C are prepared for metabolomic analysis (see Note 4). 2. Weigh samples (either whole organisms or tissues) to the nearest 0.001 g (see Note 5) and add to a clean centrifuge tube. Add 300 μL cold methanol and 150 μL chilled water to each sample.

208

Ralston-Hooper et al.

3. Homogenize samples for 20 s using a sawtooth stainless steel homogenizer. 4. Sonicate the samples using 10 1-s bursts (see Note 6). 5. Add 450 μL of chloroform to each sample. 6. Thoroughly vortex the samples for approximately 30 s. 7. Centrifuge the samples at 3000×g for 20 min. 8. After centrifugation, two distinct solvent phases will be observed (see Note 7). 9. Using a glass transfer pipette, collect each phase and place in a clean centrifuge tube. 10. Evaporate extraction solvents using a SpeedVac. Methanol fractions (polar metabolites) will be prepared for GC×GC– TOF-MS analysis (see Section 3.2), while chloroform fractions (non-polar metabolites) can be prepared for LC– TOF-MS analysis (but will not be discussed in detail here; please see Chapter 15 for more details). 3.2. Derivatization

1. To the methanol fractions, add 30 μL of Omethylhydroxylamine hydrochloride solution per 10 mg of sample (see Note 8). 2. Shake samples for 30 min at 60◦ C. 3. Add 45 μL MSTFA per 10 mg of sample (see Note 8). 4. Allow samples to react for 1 h at 60◦ C. 5. Transfer samples to clean, labeled GC autosampler vials and immediately inject onto the GC×GC–TOF-MS (see Section 3.3 for instrumental conditions).

3.3. GC×GC–TOF-MS Instrumental Conditions

1. Injection volume: 2 μL. 2. Inlet temperature: 280◦ C. 3. Carrier gas: helium. 4. Flow rate: 1 mL/min. 5. Injection type: split mode (ratio = 20). 6. Temperature profile 1: 10◦ C/min from 50 to 300◦ C; 5 min hold at 300◦ C. 7. Temperature profile 2: 10◦ C/min from 75 to 325◦ C; 5 min hold at 325◦ C. 8. Second-dimension separation time: 4 s. 9. Ion source temperature: 200◦ C. 10. Detector voltage: 1600 V. 11. Filament bias: –70 V. 12. Mass range: 50–900 u. 13. Acquisition rate: 50 spectra/s.

The Use of Two-Dimensional Gas GC×GC–TOF-MS for Metabolomic Analysis

3.4. Peak Merging, Alignment, and Normalization

209

1. Chromatographic peaks are merged and aligned using R MSort software. Since retention times and fragment spectra of chromatographic peaks are expected to differ across individual runs, we use “in-house” software to align, match, and compare metabolite peaks from various samples (19). R software include Three major functionalities of MSort peak entry merging, alignment, and contaminant elimination. Multiple peak entries of the same chromatographic R are recognized and peak generated from ChromaTOF merged so that one peak entry represents one chromatographic peak. The merged peaks are then aligned based on their retention times and spectrum similarity of fragment ions. Window sizes used during peak merging and alignment included 1% window for first dimension, 5% for second dimension, and a cutoff correlation coefficient value of 0.95. 2. Data are normalized using R statistical software (19). To allow for multi-experiment analyses, it is important to normalize data to make samples comparable. This normalization step aims to quantitatively filter overall peak intensity variations due to experimental errors such as varying amounts of samples loaded onto the GC×GC–TOF-MS. In order to make all samples comparable, peaks were normalized using a constant mean approach where the total ion current was assumed equal across samples. Normalized peaks were then subjected to statistical significance tests and pattern recognition.

3.5. Statistical Analysis

1. Two-tailed t-tests are used to detect significant differences in metabolite concentration (peak areas) when comparing two groups. If more than two groups are to be compared, analysis of variance (ANOVA) is employed using SAS. 2. Since multiple hypothesis testing was conducted, the associated p-values were adjusted to correct for multiple comparisons using the false discovery rate (FDR) or the expected proportion of errors among the rejected hypothesis using SAS (20). 3. Metabolite pattern recognition is carried out using principal component analysis (PCA) performed with PLS toolbox.

3.6. Metabolite Identification

1. Polar metabolites analyzed via GC×GC–TOF-MS are identified by matching mass spectra against the NIST library resulting in a similarity value (SV). An SV value >750 is considered a positive identification since approximately 75% of collected spectra match the NIST library spectra (17).

210

Ralston-Hooper et al.

4. Notes 1. 18.2 M cm water (which will be referred to as “water” throughout the text) should be used to prepare all solutions unless otherwise noted. 2. O-Methylhydroxylamine hydrochloride is a possible mutagen, so take necessary precautions, e.g., avoid breathing dust and minimize exposure by using the appropriate personal protection equipment (face shield, lab coat, gloves, MSHA/NIOSH-approved respirator). This compound is also hygroscopic, so store in a sealed container in a cool, dry place. For best results, O-methylhydroxylamine hydrochloride solution should be made fresh prior to use. 3. MSTFA is extremely sensitive to moisture and should be handled and stored accordingly. All glassware and syringes, as well as the samples of interest, should be carefully dried because water could interfere with derivatization efficiency. 4. In order to prevent sample degradation, samples must be immediately flash frozen and stored at –80◦ C until analysis. Samples are to be kept on ice during the extraction procedure until Step 7. 5. Sample weights (typically wet weights) are used for sample normalization during the derivatization step prior to GC×GC–TOF-MS analysis. 6. A sonicating water bath may also be used. Place samples in a 37◦ C water bath and sonicate for 3 min. 7. The top phase is methanol and contains polar metabolites. The bottom phase is chloroform and contains non-polar metabolites. 8. The 30 μL of O-methylhydroxylamine hydrochloride solution and 45 μL of MSTFA are added per 10 mg of sample and need to be adjusted according to initial sample weight.

Acknowledgments We would like to thank the Agriculture and Research Program and the Department of Forestry and Natural Resources at Purdue University for supporting the senior author with a 4-year research assistantship towards her PhD. We also thank the Center for the Environment, Purdue University for financially supporting this work.

The Use of Two-Dimensional Gas GC×GC–TOF-MS for Metabolomic Analysis

211

References 1. Viant, M. R. (2007) Metabolomics of aquatic organisms: the new “omics” on the block. Mar Ecol Prog Ser 332, 301–306. 2. Samuelsson, L. M., Larsson, D. G. J. (2008) Contributions from metabolomics to fish research. Mol Biosyst 4, 974–979. 3. Lin, C. Y., Viant, M. R., Tjeerdema, R. S. (2006) Metabolomics: methodologies and applications in the environmental sciences. J Pestic Sci 31, 245–251. 4. Viant, M. R. (2008) Recent developments in environmental metabolomics. Mol Biosyst 4, 980–986. 5. Bundy, J., Davey, M., Viant, M. (2009) Environmental metabolomics: a critical review and future perspectives. Metabolomics 5, 3– 21. 6. Dunn, W. B., Ellis, D. I. (2005) Metabolomics: current analytical platforms and methodologies. Trends Anal Chem 24, 285–294. 7. Ekman, D. R., Teng, Q., Jensen, K. M., Martinovic, D., Villeneuve, D. L., Ankley, G. T., Collette, T. W. (2007) NMR analysis of male fathead minnow urinary metabolites: a potential approach for studying impacts of chemical exposures. Aquat Toxicol 85, 104– 112. 8. Ekman, D., Keun, H., Eads, C., Furnish, C., Murrell, R., Rockett, J., Dix, D. (2006) Metabolomic evaluation of rat liver and testis to characterize the toxicity of triazole fungicides. Metabolomics 2, 63–73. 9. Viant, M. R., Pincetich, C. A., Tjeerdema, R. S. (2006) Metabolic effects of dinoseb, diazinon and esfenvalerate in eyed eggs and alevins of chinook salmon (Oncorhynchus tshawytscha) determined by 1 H NMR metabolomics. Aquat Toxicol 77, 359–371. 10. Samuelsson, L. M., Förlin, L., Karlsson, G., Adolfsson-Erici, M., Larsson, D. G. J. (2006) Using NMR metabolomics to identify responses of an environmental estrogen in blood plasma of fish. Aquat Toxicol 78, 341–349. 11. Keun, H. C., Beckonert, O., Griffin, J. L., Richter, C., Moskau, D., Lindon, J. C., Nicholson, J. K. (2002) Cryogenic probe 13 C NMR spectroscopy of urine for metabonomic studies. Anal Chem 74, 4588–4593. 12. Wang, Y., Bollard, M. E., Keun, H., Antti, H., Beckonert, O., Ebbels, T. M.,

13. 14.

15.

16.

17.

18.

19.

20.

Lindon, J. C., Holmes, E., Tang, H., Nicholson, J. K. (2003) Spectral editing and pattern recognition methods applied to highresolution magic-angle spinning 1 H nuclear magnetic resonance spectroscopy of liver tissues. Anal Biochem 323, 26–32. Houten, S. M. (2009) Metabolomics: unraveling the chemical individuality of common human diseases. Ann Med 41, 402–407. Lisec, J., Steinfath, M., Meyer, R. C., Selbig, J., Melchinger, A. E., Willmitzer, L., Altmann, T. (2009) Identification of heterotic metabolite QTL in Arabidopsis thaliana RIL and IL populations. Plant J 59, 777–788. Hope, J. L., Sinha, A. E., Prazen, B. J., Synovec, R. E. (2005) Evaluation of the dotmap algorithm for locating analytes of interest based on mass spectral similarity in data collected using comprehensive two-dimensional gas chromatography coupled with time-offlight mass spectrometry. J Chromatogr A 1086, 185–192. Welthagen, W., Shellie, R., Spranger, J., Ristow, M., Zimmermann, 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. Mohler, R. E., Dombek, K. M., Hoggard, J. C., Young, E. T., Synovec, R. E. (2006) Comprehensive two-dimensional gas chromatography time-of-flight mass spectrometry analysis of metabolites in fermenting and respiring yeast cells. Anal Chem 78, 2700– 2709. Ralston-Hooper, K., Hopf, A., Oh, C., Zhang, X., Adamec, J., Sepúlveda, M. S. (2008) Development of GC×GC/TOF-MS metabolomics for use in ecotoxicological studies with invertebrates. Aquat Toxicol 88, 48–52. Oh, C., Huang, X., Regnier, F. E., Buck, C., Zhang, X. (2008) Comprehensive twodimensional gas chromatography/time-offlight mass spectrometry peak sorting algorithm. J Chromatogr A 1179, 205–215. Benjamini, Y., Hochberg, Y. (1995) Controlling the false discovery rate: a practical and powerful approach to multiple testing. J R Stat Soc Ser B 57, 289–300.

Chapter 13 LC-MS-Based Metabolomics Sunil Bajad and Vladimir Shulaev Abstract Metabolomics involves qualitative and/or quantitative analysis of hundreds of metabolites in a complex sample. As most of the metabolites are polar in nature (for example, amino acids, nucleotides, carboxylic acids), their chromatographic separation and analysis present a difficult challenge. Recently, hydrophilic interaction chromatography (HILIC) has become a useful tool for analysis of such polar molecules. In this chapter, we present a simple HILIC LC-MS/MS method for targeted multiple reaction monitoring (MRM)-based relative quantitation of hundreds of polar metabolites in a complex sample. The method uses an aminopropyl column with acetonitrile as weak solvent and 20 mM ammonium acetate buffer (pH 9.4) as strong solvent. The method does not use any ion pairing reagent making it suitable for analysis in both positive and negative ionization mode using a simple LC-MS/MS instrument setup. The method has been thoroughly tested, validated, and applied to a variety of samples such as bacteria, yeast, plants, human body fluids, and cell cultures. Key words: Metabolomics, metabolite profiling, LC-MS/MS, chromatography, MRM, HILIC.

1. Introduction Metabolomics aims at quantitative analysis and/or qualitative profiling of large numbers of cellular metabolites. It complements other “omics” disciplines, like proteomics and transcriptomics, to obtain a holistic view of the living systems. Liquid chromatography-mass spectrometry (LC-MS)based metabolomics experiments can be performed using targeted analysis, non-targeted profiling, or metabolic fingerprinting approaches (1, 2). While targeted analysis is based on measuring a set of known metabolites for which chemical structures are identified and authentic standards are available, non-targeted profiling aims at identification and quantitation of all peaks in T.O. Metz (ed.), Metabolic Profiling, Methods in Molecular Biology 708, DOI 10.1007/978-1-61737-985-7_13, © Springer Science+Business Media, LLC 2011

213

214

Bajad and Shulaev

the chromatogram even if the chemical structure of many compounds remains unknown. Similarly, the metabolic fingerprinting approach considers the whole metabolite profile as a fingerprint or pattern, which can be used for sample classification using one of the available multivariate statistical analysis or pattern recognition algorithms. Many cellular metabolites are highly polar in nature and their analysis using chromatography combined with mass spectrometry is a significant analytical challenge. To address this analytical challenge, hydrophilic interaction chromatography (HILIC) was proposed to separate and quantitate various classes of polar metabolites that are poorly retained on the reversed-phase chromatographic matrix (3, 4). HILIC can efficiently separate many classes of highly polar metabolites, including amino acids, nucleotides, nucleosides, nucleotide phosphates, carboxylic acids, and vitamins. The sensitivity and selectivity of the HILIC separation can be greatly enhanced by employing the multiple reaction monitoring (MRM) function of tandem quadrupole mass spectrometers (4, 5). We describe here a procedure for targeted MRM-based relative quantitation of hundreds of polar metabolites in a complex sample. The method uses an aminopropyl column with acetonitrile as weak solvent and 20 mM ammonium acetate buffer (pH 9.4) as strong solvent. The method does not use any ion pairing reagent making it suitable for analysis in both positive and negative ionization mode using a simple LC-MS/MS instrument setup. This method can be used to profile over a hundred cellular metabolites in a single chromatographic separation. The method has been thoroughly tested, validated, and applied to a variety of samples such as bacteria, yeast, plants, human body fluids, and cell cultures.

2. Materials 2.1. Sample Collection and Processing

1. Dry ice. 2. LC-MS-grade methanol. 3. LC-MS-grade water. 4. Benchtop refrigerated (–10◦ C minimum temperature) microcentrifuge (Beckman Coulter, Inc., Brea, CA). The rotor should accommodate up to 2.0 mL microcentrifuge tubes and have a maximum speed of 15,000×g. 5. Ultrasonic cleaner (Fisher Scientific, Pittsburg, PA). 6. Microcentrifuge tubes: 2.0 and 1.7 mL. 7. Plastic falcon tubes: 50 mL. 8. Plastic falcon tubes: 15 mL.

LC-MS-Based Metabolomics

215

9. Uniformly labeled (13 C, 15 N) internal standards: alanine, deoxyadenosine, glutamate, succinate, thymidine, threonine, UMP, AMP, ATP, and TTP (Medical Isotope Inc., Pelham, NH; Sigma, St. Louis, MO). 10. Methanol–water mixture, 50:50 v/v: In a clean glass container, mix 50 mL of LC-MS-grade water and 50 mL of LC-MS-grade methanol. Store at 4◦ C. 11. Quenching solvent: In a clean glass container, mix 40 mL of LC-MS-grade water and 60 mL of LC-MS-grade methanol. Prepare 5 aliquots of 20 mL each in 50 mL plastic tubes. Store at –80◦ C. 12. Extraction buffer: In a clean glass container, mix 20 mL of LC-MS-grade water and 80 mL of LC-MS-grade methanol. Prepare 5 aliquots of 20 mL each in 50 mL plastic tubes. Store at –4◦ C. 13. Internal standard stock solutions (2 mg/mL): Prepare 4 mg/mL solutions of individual internal standards in LC-MS-grade water. Dilute 1:1 with LC-MS-grade methanol. Store 100 μL aliquots at –80◦ C (during preparation, keep the solutions on ice at all times). 14. Internal standard mixture: In a 15 mL plastic tube, mix 0.5 mL each from the stock solutions of alanine, glutamate, threonine, AMP, and UMP and 1.5 mL each from the stock solutions of succinate, deoxyadenosine, thymidine, ATP, and TTP. Store 50 μL aliquots at –80◦ C (during preparation, keep the solutions on ice at all times). 15. Extraction buffer with internal standards (to be prepared on the day of sample processing): For each 1 mL of extraction solvent, add 10 μL of internal standard mixture to 990 μL of extraction solvent. This results in final internal standard concentrations of 1 μg/mL for alanine, glutamate, threonine, AMP, and UMP and 3 μg/mL for succinate, deoxyadenosine, thymidine, ATP, and TTP. 16. Quality control standard: Extraction solvent with internal standards can be used as a quality control sample. 2.2. Hydrophilic Interaction Chromatography (HILIC)

1. Ammonium acetate stock solution: Weigh 38.5 g of ammonium acetate and transfer it to a clean 500 mL cylinder. Make up the volume to 500 mL with LC-MS-grade water. This results in a 1 M ammonium acetate stock solution. Transfer the solution to a clean glass solvent bottle and store at 4◦ C. 2. Ammonium hydroxide stock solution: Add 92 mL of LCMS-grade water to a clean 250 mL glass solvent bottle. Add 100 mL of ammonium hydroxide (25% v/v, see Note 1 for additional information on ammonium hydroxide volume

216

Bajad and Shulaev

calculation) from the original bottle to the water and mix. This results in a 3.26 M ammonium hydroxide stock solution. Transfer the solution to a clean glass solvent bottle and store at 4◦ C. 3. LC-MS-grade acetonitrile. 4. LC-MS-grade water. 5. LC solvent A: In a clean 1 L solvent bottle, add 960 mL of LC-MS-grade water. With a glass pipette, add 20 mL of 1 M ammonium acetate solution and 20 mL of 3.26 M ammonium hydroxide solution and mix. With a glass pipette, remove 50 mL and transfer to a 50 mL falcon tube, then replace with 50 mL of acetonitrile. 6. LC solvent B: LC-MS-grade acetonitrile. 7. Luna NH2 aminopropyl column, 250 mm × 2.1 mm, 5 μm (Phenomenex, Torrance, CA). 8. Luna NH2 aminopropyl guard column, 4 mm × 2.1 mm (Phenomenex, Torrance, CA). 2.3. LC-MS/MS Instrumentation

1. Shimadzu LC10Avp (Shimadzu Scientific Instruments, Columbia, MD), High Performance Liquid Chromatography system equipped with an autosampler capable of operating at 4◦ C. 2. TSQ Quantum Ultra (Thermo Scientific, San Jose, CA) triple quadrupole mass spectrometer.

3. Methods 3.1. LC-MS/MS Analysis

1. Create a LC method with the following gradient: 0 min→ 95% B, 25 min→10% B, 30 min →10% B, 30.1 min → 95% B, 37 min → 95% B. Flow rate →0.3 mL/min. Column temperature→25◦ C (see Note 2). 2. Set the ESI voltage to 4000 V in positive ionization mode or –3500 V in negative ionization mode. 3. Set the sheath gas to 40 psi. 4. Set the auxiliary gas to 10 psi. 5. Set the collision gas to 1.5 mTorr (see Note 3). 6. If the triple quadrupole instrument to be used is a TSQ Quantum Ultra or advanced model from Thermo Scientific (Waltham, MA), then previously optimized MRMs (4) can be used (Table 13.1).

90

112 113 115 116

C3 H8 NO2 + C4 H5 O3 – C5 H14 NO+ C3 H5 O4 – C3 H8 NO3 + C4 H6 N3 O+

C4 H5 N2 O2 +

C4 H3 O4 –

C5 H10 NO2 + C4 H5 O4 – C4 H8 NO3 – C5 H12 NO2 + C4 H10 NO3 + C3 H8 NO2 S+ C6 H4 NO2 – C6 H7 N2 O+ C2 H6 NO3 S– C5 H7 N2 O2 + C5 H15 N4 + C4 H3 O5 – C4 H6 NO4 – C6 H14 NO2 +

Putrescine

Alanine

Uracil

Fumarate

Proline

(Iso)leucine

Aspartate

Oxaloacetate

Agmatine

Thymine

Taurine

Nicotinamide

Nicotinate

Cysteine

Homoserine

Valine

Threonine

Succinate

Cytosine

Serine

Glycerate

Choline

132

132

131

131

127

124

123

122

122

120

118

118

117

106

105

104

101

89

C4 H13 N2 +

Glycine

Acetoacetate

76

C2 H6 NO2 +

Metabolite

parent mass

12 C

Parent ion formulab

Table 13.1 MS/MS parametersa

138

136

135

136

132

126

129

128

125

124

123

122

121

121

119

117

116

109

108

109

105

93

93

78

parent mass

13 C

11

17

12

16

17

16

20

14

27

30

11

12

10

11

11

23

17

13

15

19

21

11

11

16

Collision energy (eV)

86

88

87

72

110

80

80

78

59

44

55

74

73

70

71

70

95

60

75

60

57

44

72

30

product mass

12 C

91

91

90

76

115

80

85

83

61

46

59

76

76

74

74

73

99

62

77

63

60

46

76

31

Product mass

13 C

C5 H12 N+

C3 H6 NO2 –

C 3 H3 O 3 –

C4 H10 N+

C5 H4 NO2 +

SO3 –

C 5 H6 N +

C 5 H4 N –

C 2 H3 S +

C 2 H6 N +

C 4 H7 +

C2 H4 NO2 –

C 3 H5 O 2 –

C 4 H8 N +

C 3 H3 O 2 –

C3 H4 NO+

C 4 H3 N 2 O +

C2 H6 NO+

C 2 H3 O 3 –

C3 H10 N+

C 3 H5 O –

C 2 H6 N +

C4 H10 N+

CH4 N+

Product ion formulac

LC-MS-Based Metabolomics 217

133 133

136 137 137 139 140

146 147 147

C 4 H5 O 5 –

C5 H13 N2 O2 + C 5 H6 N 5 + C4 H10 NO2 S+ C7 H6 NO2 –

C 5 H5 N 4 O +

C 7 H5 O 3 –

C 2 H4 O 5 P –

CH3 NO5 P– C6 H12 N3 O+ C 5 H5 O 5 – C7 H20 N3 +

C5 H11 N2 O3 +

C6 H15 N2 O2 + C5 H10 NO4 + C5 H12 NO2 S+ C 5 H6 N 5 O + C 7 H5 O 4 – C 5 H5 N 4 O 2 + C 5 H3 N 2 O 4 – C6 H10 N3 O2 + C 5 H5 N 2 O 4 – C 4 H7 N 4 O 3 +

Malate

Ornithine

Hypoxanthine

p-Hydroxybenzoate

Acetylphosphate

Carbamoyl-phosphate

Glutamine

Lysine

Allantoin

Dihydroorotate

Histidine

Orotate

Xanthine

2,3-Dihydroxybenzoate

Guanine

Methionine

Glutamate

Spermidine

α-Ketoglutarate

Histidinol

p-Aminobenzoate/ anthranilate

Homocysteine

159

157

156

155

153

153

152

150

148

145

142

136

136

133

C 4 H9 N 2 O 3 +

Asparagine

Adenine

12 C parent mass

Parent ion formulab

Metabolite

Table 13.1 (continued)

163

162

162

160

158

160

157

155

153

153

152

153

150

148

141

141

144

142

143

140

141

138

137

137

parent mass

13 C

11

12

12

13

16

17

18

10

15

15

15

13

11

18

22

22

21

19

16

10

24

12

12

17

Collision energy (eV)

99

113

110

111

110

109

110

133

84

84

84

112

101

95

79

79

93

110

92

90

119

70

115

74

product mass

12 C

102

117

115

115

114

115

114

138

88

89

88

119

105

100

79

79

99

114

98

93

124

74

119

76

Product mass

13 C

C3 H3 N2 O2 +

C4 H5 N2 O2 –

C5 H8 N3 +

C4 H3 N2 O2 –

C4 H4 N3 O+

C6 H5 O2 –

C4 H4 N3 O+

C5 H9 O2 S+

C4 H6 NO+

C5 H10 N+

C4 H6 NO+

C7 H14 N+

C4 H5 O3 –

C5 H7 N2 +

PO3 –

PO3 –

C6 H5 O–

C4 H4 N3 O+

C6 H6 N–

C3 H8 N5 +

C5 H3 N4 +

C4 H8 N+

C4 H3 O4 –

C2 H4 NO2 +

Product ion formulac

218 Bajad and Shulaev

163 166

169 170 171 173 173 173 174 175 175 175 179 179 180 182

C 9 H7 O 3 –

C9 H12 NO2 + C7 H4 NO4 – C 3 H4 O 6 P – C3 H6 O6 P–

C8 H12 NO3 +

C 3 H8 O 6 P –

C6 H13 N4 O2 –

C 7 H9 O 5 –

C 6 H5 O 6 –

C6 H12 N3 O3 –

C 4 H7 N 4 O 4 – C5 H7 N2 O5 –

C7 H15 N2 O3 +

C 9 H7 O 4 –

C6 H11 O6 –

C6 H14 NO5 +

C4 H8 NO5 S– C9 H12 NO3 + C 3 H6 O 7 P – C3 H9 NO6 P+ C 6 H7 O 7 –

Carnitine

Phenylpyruvate

Phenylalanine

Pyridoxine

Glycerol-3-phosphate

Arginine

Shikimate

Trans-aconitate

Citrulline

Allantoate

Carbamoyl-L -aspartate

N-α-acetylornithine

Hydroxyphenylpyruvate

Myo-inositol

Glucosamine

Homocysteic acid

Citrate

3-Phosphoserine

3-Phosphoglycerate

Tyrosine

Dihydroxy-acetonephosphate

Phosphoenolpyruvate

191

186

185

182

167

166

162

C7 H16 NO3 +

Metabolite

Quinolinate

parent mass

12 C

Parent ion formulab

Table 13.1 (continued)

197

189

188

191

186

186

185

188

182

180

179

180

179

180

179

174

178

172

170

173

175

172

169

parent mass

13 C

13

10

15

37

21

10

15

11

14

12

12

13

10

18

12

13

22

38

41

12

28

11

18

Collision energy (eV)

111

88

97

77

80

162

161

107

115

132

132

131

129

93

131

79

134

79

79

122

103

91

103

product mass

12 C

116

91

97

83

80

168

167

114

120

136

135

136

134

99

136

79

142

79

79

128

111

98

107

Product mass

13 C

C5 H3 O3 –

C3 H6 NO2 +

H2 PO4 –

C 6 H5 +

SO3 –

C6 H12 NO4 +

C6 H9 O5 –

C7 H7 O–

C5 H9 NO2 +

C4 H6 NO4 –

C3 H6 N3 O3 –

C5 H11 N2 O2 –

C5 H5 O4 –

C6 H5 O–

C5 H11 N2 O2 –

PO3 –

C8 H8 NO+

PO3 –

PO3 –

C6 H4 NO2 –

C 8 H7 +

C 7 H7 –

C4 H7 O3 +

Product ion formulac

LC-MS-Based Metabolomics 219

195 199 205

C6 H11 O7 –

C 4 H8 O 7 P –

C11 H13 N2 O2 + C 6 H9 O 8 – C5 H10 O7 P– C9 H16 NO5 – C7 H15 N2 O4 S+ C10 H9 O6 – C9 H13 N2 O5 + C5 H10 O8 P– C10 H13 N2 O5 – C9 H11 N2 O6 – C9 H14 N3 O5 + C10 H17 N2 O3 S+ C10 H14 N5 O3 + C10 H13 N4 O4 + C6 H12 O9 P– C6 H15 NO8 P+ C6 H15 NO8 P+ C12 H17 N4 OS+ C10 H14 N5 O4 + C10 H14 N5 O4 + C10 H13 N4 O5 +

2-Dehydro-D-gluconate

Gluconate

Erythrose-4-phosphate

Tryptophan

Inosine

Deoxyguanosine

Adenosine

Thiamine

Glucosamine-6-P

Glucosamine-1-P

D-hexose-phosphate

Deoxyinosine

Deoxyadenosine

Biotin

Cytidine

Uridine

Thymidine

D-rib(ul)ose-5-phosphate

Deoxyuridine

Prephenate

Cystathionine

Pantothenate

Deoxyribose-phosphate

269

268

268

265

260

260

259

253

252

245

244

243

241

229

229

225

223

218

213

209

193

C 6 H9 O 7 –

Metabolite

D-glucarate

parent mass

12 C

Parent ion formulab

Table 13.1 (continued)

279

278

278

277

266

266

265

263

262

255

253

252

251

234

238

235

230

227

218

215

216

203

201

199

parent mass

13 C

14

15

27

17

15

15

40

12

20

18

12

19

15

48

11

15

11

19

33

16

16

17

15

12

Collision energy (eV)

137

152

136

122

126

162

79

137

136

227

112

200

125

79

113

91

134

146

79

85

146

97

129

103

product mass

12 C

142

157

141

128

132

168

79

142

141

237

116

208

130

79

117

97

138

153

79

89

155

97

134

107

Product mass

13 C

C5 H5 N4 O+

C5 H6 N5 O+

C5 H6 N5 +

C6 H8 N3 +

C6 H8 NO2 +

C6 H12 NO4 +

PO3 –

C5 H5 N4 O+

C5 H6 N5 +

C10 H15 N2 O2 S+

C4 H6 N3 O+

C8 H10 NO5 –

C5 H5 N2 O2 –

PO3 –

C4 H5 N2 O2 +

C6 H3 O–

C4 H8 NO2 S+

C7 H16 NO2 –

PO3 –

C4 H5 O2 –

C9 H8 NO+

H2 PO4 –

C5 H5 O4 –

C4 H7 O3 –

Product ion formulac

220 Bajad and Shulaev

300 307 308

C9 H12 N2 O8 P–

C9 H15 N3 O7 P+ C10 H18 N3 O6 S+ C10 H19 O7 P2 – C10 H16 N2 O8 P+ C9 H15 N3 O8 P+ C9 H14 N2 O9 P+ C10 H11 N5 O6 P– C10 H15 N5 O6 P+ C6 H13 O12 P2 – C12 H21 O11 –

Guanosine

dUMP

dCMP

Riboflavin

Orotidine-phosphate

Xanthosine-5-phosphate

GMP

IMP

dGMP

AMP

Thiamine-phosphate

Trehalose

Fructose-1,6-bisphosphate

dAMP

cyclic-AMP

UMP

CMP

dTMP

Geranyl-PP

Reduced glutathione

N-acetyl-glucosamine-1phosphate

C12 H18 N4 O4 PS+ C10 H15 N5 O7 P+ C10 H15 N5 O7 P+ C10 H14 N4 O8 P+ C10 H15 N5 O8 P+ C10 H14 N4 O9 P+ C10 H12 N2 O11 P– C17 H21 N4 O6 +

284

C10 H14 N5 O5 + C10 H13 N4 O6 + C8 H15 NO9 P–

6-Phospho- D-gluconate

377

367

365

364

349

348

348

345

341

339

332

328

325

324

323

313

308

285

275

C6 H12 O10 P–

Metabolite

Xanthosine

parent mass

12 C

Parent ion formulab

Table 13.1 (continued)

394

377

375

374

359

358

358

357

353

345

342

338

334

333

333

323

318

317

316

308

295

294

281

parent mass

13 C

24

16

11

19

19

36

21

13

18

28

21

31

12

16

17

18

19

16

16

32

20

33

11

Collision energy (eV)

243

323

97

152

137

135

136

122

179

97

136

134

97

112

81

79

162

112

195

79

153

135

97

product mass

12 C

255

332

102

157

142

140

141

128

185

97

141

139

102

116

86

79

167

116

200

79

158

140

97

Product mass

13 C

C12 H12 N4 O2 +

C9 H12 N2 O9 P–

C5 H5 O2 +

C5 H6 N5 O+

C5 H5 N4 O+

C5 H3 N4 O+

C5 H6 N5 +

C6 H8 N3 +

C6 H11 O6 –

H2 PO4 –

C5 H6 N5 +

C5 H4 N5 –

C5 H5 O2 +

C4 H6 N3 O+

C5 H5 O+

PO3 –

C5 H8 NO3 S+

C4 H6 N3 O+

C5 H8 O6 P–

PO3 –

C5 H5 N4 O2 +

C5 H3 N4 O+

H2 PO4 –

Product ion formulac

LC-MS-Based Metabolomics 221

386

C9 H14 N3 O10 P2 – C5 H12 O14 P3 – C15 H23 N6 O5 S+

dCDP

dGTP

ATP

dATP

UTP

CTP

TTP

dUTP

dCTP

5-Methyl-THF

FMN

7,8-Dihydrofolate

GDP

Folate

IDP

dGDP

APS

ADP

UDP

CDP

dTDP

S-adenosyl-L-methionine

PRPP

C10 H15 N2 O11 P2 – C9 H14 N3 O11 P2 – C9 H13 N2 O12 P2 – C10 H14 N5 O10 P2 – C10 H13 N5 O10 PS– C10 H14 N5 O10 P2 – C10 H13 N4 O11 P2 – C19 H20 N7 O6 + C10 H14 N5 O11 P2 – C19 H22 N7 O6 + C17 H20 N4 O9 P– C20 H26 N7 O6 + C9 H15 N3 O13 P3 – C9 H14 N2 O14 P3 – C10 H16 N2 O14 P3 – C9 H15 N3 O14 P3 – C9 H14 N2 O15 P3 – C10 H15 N5 O12 P3 – C10 H15 N5 O13 P3 – C10 H15 N5 O13 P3 –

385

C14 H21 N6 O5 S+

S-adenosyl-L-homocysteine

506

506

490

483

482

481

467

466

460

455

444

442

442

427

426

426

426

403

402

401

399

389

381

C15 H27 O7 P2 –

trans, trans-farnesyl diphosphate

Metabolite

parent mass

12 C

Parent ion formulab

Table 13.1 (continued)

516

516

500

492

491

491

476

475

480

472

463

452

461

437

436

436

436

412

411

411

414

394

395

399

396

parent mass

13 C

28

21

27

33

32

31

25

28

19

19

30

19

16

14

25

20

24

15

29

29

13

18

25

19

21

Collision energy (eV)

159

408

159

159

159

159

159

159

313

213

178

159

295

159

159

346

134

305

159

159

250

291

159

136

79

product mass

12 C

159

418

159

159

159

159

159

159

328

218

185

159

309

159

159

356

139

314

159

159

260

296

159

141

79

Product mass

13 C

HO6 P2 –

C10 H12 N5 O9 P2 –

HO6 P2 –

HO6 P2 –

HO6 P2 –

HO6 P2 –

HO6 P2 –

HO6 P2 –

C15 H17 N6 O2 +

C5 H10 O7 P–

C7 H8 N5 O+

HO6 P2 –

C14 H11 N6 O2 +

HO6 P2 –

HO6 P2 –

C10 H13 N5 O7 P–

C5 H4 N5 –

C9 H10 N2 O8 P–

HO6 P2 –

HO6 P2 –

C10 H12 N5 O3 +

C5 H9 O10 P2 –

HO6 P2 –

C5 H6 N5 +

PO3 –

Product ion formulac

222 Bajad and Shulaev

522 565 579 602 606 613

C10 H15 N5 O14 P3 –

C15 H23 N2 O17 P2 –

C15 H21 N2 O18 P2 – C16 H24 N5 O15 P2 – C10 H16 N5 O17 P4 –

C17 H26 N3 O17 P2 –

C20 H33 N6 O12 S2 + C21 H26 N7 O14 P2 – C21 H28 N7 O14 P2 –

C21 H36 N7 O13 P2 S+ C21 H27 N7 O17 P3 – C21 H29 N7 O17 P3 – C21 H37 N7 O16 P3 S+ C27 H34 N9 O15 P2 + C23 H39 N7 O17 P3 S+ C24 H41 N7 O17 P3 S+ C25 H41 N7 O18 P3 S+ C24 H39 N7 O19 P3 S+ C25 H41 N7 O19 P3 S+ C27 H43 N7 O20 P3 S–

ITP

GTP

UDP-D-glucose

UDP-D-glucuronate

UDP-n-acetyl-D-glucosamine

Oxidized glutathione

Dephospho-CoA

910

868

854

852

824

810

786

768

744

742

688

664

662

937

893

878

877

848

833

813

789

765

763

709

685

683

633

623

612

604

594

580

532

517

parent mass

13 C

a Reproduced from Bajad et al. (4) with permission. Copyright Elsevier Science. b Charge on the ion indicates ionization mode used for the metabolite. c The chemical formulas for product ions have not been confirmed for all metabolites.

3-hydroxy-3-methylglutarylCoA

Succinyl-CoA

Malonyl-CoA

Acetoacetyl-CoA

Propionyl-CoA

Acetyl-CoA

FAD

CoA

NADPH

NADP+

NADH

NAD+

Guanosine 5 -diphosphate, 3 -diphosphate

588

507

C10 H14 N4 O14 P3 –

Metabolite

ADP-D-glucose

parent mass

12 C

Parent ion formulab

Table 13.1 (continued)

43

38

26

34

33

28

24

37

34

18

25

31

19

33

26

22

22

24

23

23

19

Collision energy (eV)

408

361

347

345

317

303

348

261

408

620

348

408

540

231

385

504

346

403

323

424

409

product mass

12 C

418

376

361

360

331

316

358

272

418

635

358

418

555

239

394

514

356

412

332

434

419

Product mass

13 C

C10 H12 N5 O9 P2 –

C15 H25 N2 O6 S+

C14 H23 N2 O6 S+

C15 H15 N2 O5 S+

C14 H25 N2 O4 S+

C13 H23 N2 O4 S+

C10 H15 N5 O7 P+

C11 H21 N2 O3 S+

C10 H12 N5 O9 P2 –

C15 H20 N5 O16 P3 –

C10 H15 N5 O7 P+

C10 H12 N5 O9 P2 –

C15 H20 N5 O13 P2 –

C8 H11 N2 O2 S2 +

C9 H11 N2 O11 P2 –

C10 H13 N5 O13 P3 –

C10 H13 N5 O7 P–

C9 H13 N2 O12 P2 –

C9 H12 N2 O9 P–

C10 H12 N5 O10 P2 –

C10 H11 N4 O10 P2 –

Product ion formulac

LC-MS-Based Metabolomics 223

224

Bajad and Shulaev

7. To add MRMs for additional metabolites, the auto MRM generation tool of Xcalibur software, common to Thermo Scientific Mass Spectrometers, can be used. (If the instrument is from another manufacturer, then optimization of MRMs is required, see Note 4.) 8. Run the representative sample or standard mixture of metabolites several times in positive and negative ionization mode with 60 MRMs in each run to obtain the retention times for the metabolites of interest. 9. Once the retention times are obtained, divide the run time into four to five time segments. In a time segment, insert the MRMs of the metabolites eluting in that particular time segment. Use 50–100 ms transition time/SRM and a scan width of 1 m/z (see Note 5). 10. Insert MRMs corresponding to the internal standards in the same time segments as the corresponding 12 C metabolites. Use alanine, glutamate, deoxyadenosine, UMP, and AMP in positive ionization mode and succinate, threonine, thymidine, ATP, and TTP in negative ionization mode. 3.2. Extraction of Metabolites

1. Centrifuge 50 mL of Escherichia coli culture (optical density A650 of ∼ 0.3) at 5000×g for 4 min. 2. Remove the supernatant and immediately add 300 μL of cold (–75◦ C) extraction buffer. 3. After mixing, keep on dry ice for 15 min. 4. Transfer to a 1.7 mL microcentrifuge tube and centrifuge at 14,000 ×g for 3 min at 4◦ C. 5. Collect supernatant in a fresh tube and then repeat extraction with 200 μL of ice cold methanol-water mixture, 50–50 v/v. 6. Perform a third round of extraction with water bath sonication for 15 min in ice-cold water. 7. Combine the extracts.

3.3. Sample Analysis

1. Label the LC vials and keep them on wet ice for at least 10 min. 2. Centrifuge the processed extracts at 10,000×g for 5 min at 4◦ C before transferring them to the LC vials to remove particulate matter. 3. Transfer the samples to LC vials, then place the vials in the LC autosampler set at 4◦ C (it is important to keep the samples at 4◦ C or below at all times before and during analysis as some sensitive metabolites such as NADH, NADPH, and ATP degrade over time). We have found that samples are stable for 24 h in an autosampler set at 4◦ C.

LC-MS-Based Metabolomics

225

4. Create a sample analysis sequence with the first run as a blank followed by a quality control standard followed by the experimental samples. It is important to run a blank sample before each sequence to obtain reproducible retention times. If the sequence is stopped in the middle of the sequence, then a blank injection is required to reequilibrate the column in order to obtain the same retention times as obtained before stopping the sequence. Else, the first sample will show very different chromatography. 5. Insert an appropriate number of blank samples in between the experimental samples to avoid carryover (a blank after every three samples is good practice even for very complex samples). The sequence should end with a quality control sample followed by a blank sample (see Note 6). A representative chromatogram is shown in Fig. 13.1. 3.4. Data Analysis

1. Using one representative data file, create a data processing method which includes assignment of name, expected retention time, internal standard, and peak integration parameters to each metabolite with the Xcalibur quantitative method. 2. The method should also include automated results output to Microsoft Excel for further data processing and graphing. 3. Using this method, process all the raw data in a batch mode. 4. This generates a report in Microsoft Excel with each metabolite as a row and metabolite peak area, internal standard peak area, metabolite/internal standard peak area ratio, and retention time as columns. 5. Further processing can be done in Microsoft Excel as required, for example, to calculate the relative ratios of metabolite peak areas in control and experimental samples, as well as the mean, standard deviation, relative standard deviation, etc. 6. Representative relative quantitation data obtained by analyzing stationary phase and exponential phase E. coli cultures is shown in Fig. 13.2.

4. Notes 1. The concentration of ammonium hydroxide (MW: 35 g; specific gravity: 0.9 g) in the bottle is 25% v/v. 2. The gradient can be changed appropriately if a column with different dimensions than indicated in this method is

Bajad and Shulaev

Adenosine

Lysine

4.0x105

2.0x105

Glutamate

Tryptophan

6.0x105

Succinyl-CoA

Glucosamine-6P

8.0x105

AMP

1.0x106

Glutamine

226

X 10

0.0

6.0x104

25

30

35

NAD

Trehalose

Hexose-phosphate

4.0x104

2.0x104

20

Citrate

Thymidine

8.0x104

15

Dihydroorotate

10

5

(a)

ATP

dGTP

0.0

(b)

5

10

15

20 25 Retention time (min)

30

35

Fig. 13.1. A representative chromatogram obtained by injecting metabolite standards in positive ionization mode (top panel ) and negative ionization mode (bottom panel). Reproduced from Bajad et al. (4) with permission. Copyright Elsevier Science.

used. For example, a 150 mm × 2.1 mm, 3 μm column can also be used with careful modification of the gradient. Lowering solvent B to 80% at the start of the gradient will result in faster elution of nucleotide phosphates, CoAs, and organic acids. However, nucleotide bases/nucleosides are not sufficiently retained and co-elute with lipids. So, a

LC-MS-Based Metabolomics

227

2

P-Value >0.01 1

0

−1

−2

−3

Histidine Phenylalanine cyclic-AMP Tryptophan (Iso)leucine Lysine S-adenosyl-L-methionine Tyrosine Methionine NAD+ TMP Phosphoenolpyruvate NADP+ Arginine ATP ADP Citrate dTDP Aspartate AMP TTP alpha-ketoglutarate FAD 3-Phosphoglycerate GMP CDP Threonine Oxidized Glutathione UMP CTP CMP Riboflavin Ornithine Glutamate D-rib(ul)ose-5-P UTP Acetyl-CoA Proline Valine Alanine Citrulline Glutamine Fumarate D-Hexose-P UDP-N-Acetyl-D-glucosamine UDP-D-glucose Malate Glucosamine-6-Phosphate 2,3-Dihydroxybenzoic Acid Succinate Fructose-1,6-bis-P IMP

Log10([Stationary culture signal]/[Exponential culture signal])

P-Value <0.01

Compounds

Fig. 13.2. Representative data obtained using the described LC-MS/MS method. The figure shows differences in the metabolome of stationary phase and exponentially growing Escherichia coli cells. Reproduced from Bajad et al. (4) with permission. Copyright Elsevier Science.

lower starting organic percentage should be used only if the focus of the analysis is quantitation of phosphorylated or acidic compounds. Methanol cannot be substituted for acetonitrile as it produces broad peak shapes. 3. These parameters are for the TSQ Quantum Ultra. For other instruments, source optimization is required which can be done by using a mixture of standard metabolites. 4. Purchasing all the metabolite standards and creating MRMs is costly and time consuming. Therefore, we suggest optimization of instrument specific parameters for published MRMs using an experimental sample. For example, on an ABI 4000 triple quadrupole or advanced model, the declustering potential can be simultaneously optimized for 100 s of MRMs by running samples at three different declustering potentials and choosing the value that gives the best signal.

228

Bajad and Shulaev

5. Without dividing the LC run time into time segments, only 60 metabolites can be monitored. Hence, the LC run time needs to be divided into time segments, each containing MRMs for the metabolites eluting in that time segment (4). The best approach is to run the representative sample several times with 60 MRMs in each run to obtain the retention times for the metabolites of interest. Once the retention times are obtained, it is easy to divide the run time into segments and insert the appropriate MRMs. 6. For a new analytical column, it is important to wash the column with 100% acetonitrile for 30 min to remove the shipping solvent (hexane). During this washing step, do not connect the column outlet to the MS inlet. After this step, condition the column in the starting solvent and then run two blank samples before starting the actual sample analysis. References 1. Halket, J. M., Waterman, D., Przyborowska, A. M., Patel, R. K., Fraser, P. D., Bramley, P. M. (2005) Chemical derivatization and mass spectral libraries in metabolic profiling by GC/MS and LC/MS/MS. J Exp Bot 56, 219–243. 2. Shulaev, V. (2006) Metabolomics technology and bioinformatics. Brief Bioinform 7, 128–139. 3. Tolstikov, V. V., Fiehn, O. (2002) Analysis of highly polar compounds of plant origin: combination of hydrophilic interaction chro-

matography and electrospray ion trap mass spectrometry. Anal Biochem 301, 298–307. 4. Bajad, S. U., Lu, W., Kimball, E. H., Yuan, J., Peterson, C., Rabinowitz, J. D. (2006) Separation and quantitation of water soluble cellular metabolites by hydrophilic interaction chromatography-tandem mass spectrometry. J Chromatogr A 1125, 76–88. 5. Bajad, S., Shulaev, V. (2007) Highly-parallel metabolomics approaches using LC-MS for pharmaceutical and environmental analysis. Trends Analyt. Chem 26, 625–636.

Chapter 14 Capillary Electrophoresis–Electrospray Ionization-Mass Spectrometry (CE–ESI-MS)-Based Metabolomics Philip Britz-McKibbin Abstract Metabolomics is a rapidly emerging field of functional genomics research whose aim is the comprehensive analysis of low molecular weight metabolites in a biological sample. Capillary electrophoresis–electrospray ionization-mass spectrometry (CE–ESI-MS) represents a promising hyphenated microseparation platform in metabolomics, since a majority of primary metabolites are intrinsically polar. CE–ESI-MS offers a convenient format for the separation of complex mixtures of cationic, anionic, and/or zwitterionic metabolites, as well as their isobaric/isomeric ions without complicated sample handling. Moreover, online sample preconcentration with desalting is readily integrated during separation prior to ionization, where the migration behavior and ionization response of metabolites can be predicted based on their fundamental physicochemical properties. Herein, we describe recent developments in CE–ESI-MS with emphasis on practical protocols necessary for realizing reliable analyses as applied to targeted metabolite profiling and untargeted metabolomic studies in various biological samples. Key words: Metabolite profiling, metabolomics, CE–ESI-MS, biological samples, sample pretreatment, metabolite identification.

1. Introduction Since its first introduction by Smith et al. (1) over 20 years ago, capillary electrophoresis–electrospray ionization-mass spectrometry (CE–ESI-MS) has developed into a promising orthogonal analytical platform for the high-resolution separation of complex mixtures of polar metabolites. Due to the low flow rates (nL/min) and high voltage (30 kV) requirements in CE, there has been significant research aimed at developing compatible interface designs in order to ensure stable electrospray processes T.O. Metz (ed.), Metabolic Profiling, Methods in Molecular Biology 708, DOI 10.1007/978-1-61737-985-7_14, © Springer Science+Business Media, LLC 2011

229

230

Britz-McKibbin

without compromising detector sensitivity and/or separation performance (2). To date, most commercial CE–ESI-MS instruments utilize a stainless steel coaxial sheath liquid as a robust ESI interface, where a makeup liquid flow and nebulizer gas mix with the effluent (i.e., analyte, background electrolyte) exiting the distal end of the capillary outlet. Unlike conventional liquid chromatography (LC)–ESI-MS interfaces, several parameters are required for optimization of ion responses in CE–ESI-MS, such as cone voltage, sheath liquid flow rate, capillary alignment, and nebulizer gas pressure (3). However, as separation and ionization processes are effectively decoupled in CE–ESI-MS, independent optimization of conditions for resolution in CE and ionization efficiency in ESI-MS can be tuned without significant loss in sensitivity due to post-capillary dilution effects (4). Indeed, online sample preconcentration with desalting is effectively integrated in CE prior to ionization, which allows for direct analysis of complex biological samples with minimal sample pretreatment while improving concentration sensitivity (5–7). As analyte migration behavior in CE can be modeled based on intrinsic physicochemical properties of a metabolite (8), unambiguous identification among isomeric/isobaric ions is feasible in silico via their characteristic relative migration time (RMT) (9). Nevertheless, with growing interest in CE–ESI-MS-based metabolomic applications (10, 11), there is urgent need for the development of standardized protocols for reliable analysis of polar metabolites, which includes sample pretreatment, data pre-processing, and metabolite identification/quantification strategies. One of the major limitations of CE–ESI-MS is the lack of migration time reproducibility due to variations in electroosmotic flow (EOF) (CV > 10%) that serves as a natural electrokinetic pumping mechanism. The development of CE–ESIMS has also lagged behind LC-MS due to the large number of experimental variables that can influence separation performance, ionization efficiency, and long-term robustness. Although various non-covalent and covalently modified capillary coating procedures have been reported to better control EOF for improved absolute migration time reproducibility in CE-MS (12), these methods have limitations related to coating desorption and ion suppression effects, higher operating costs, and/or longer capillary conditioning times. Alternatively, normalization of migration times using an internal standard(s) can be performed (i.e., RMT) to improve method precision or data alignment software compatible with the peak profiles in CE can be used for processing differential metabolomic datasets (13). Indeed, a recent inter-laboratory validation for analysis of amino acids in plant extracts demonstrated that CE–ESI-MS has significant advantages in terms of higher sample throughput, lower operating costs, and reduced sample handling relative to GC-MS (14).

CE–ESI-MS-Based Metabolomics

231

Although the selectivity of CE–ESI-MS under acidic and alkaline conditions for the resolution of polar metabolites is similar to LC-MS when using reversed-phase and hydrophilic interaction chromatography, respectively (15), there are certain applications where CE–ESI-MS allows for broader coverage of metabolites that differ widely in their polarity, such as simultaneous resolution of polar amino acids and surface-active, long-chain acylcarnitines relevant to expanded newborn screening (6). Herein, standardized protocols are described in detail to aid in the successful transition of new researchers interested in CE–ESI-MS for targeted metabolite profiling and comprehensive metabolomics as applied to the analysis of polar metabolites in complex biological samples, including whole blood, plasma, cell lysates, and urine.

2. Materials 2.1. CE–ESI-MS System with Coaxial Sheath Liquid Interface

1. Agilent CE system (see Note 1) equipped with an XCT 3D ion trap mass spectrometer, an Agilent 1100 series isocratic pump, and a G16107 CE–ESI-MS coaxial sheath liquid sprayer interface (Agilent Technologies Inc., Waldbronn, Germany). 2. Flexible open-tubular, fused-silica capillaries (Polymicro Technologies, Inc., Phoenix, AZ) with total length of 80 cm and internal diameter of 50 μm were prepared (see Note 2) using a ShortixTM diamond capillary cutter (Sigma-Aldrich, Inc., St. Louis, USA).

2.2. Solvents and Chemicals

1. Solvents and buffer reagents (Sigma-Aldrich, Inc.). 1a. Acetonitrile (ACN) 1b. Ammonium acetate 1c. Ammonium bicarbonate 1d. Ammonium hydroxide 1e. Concentrated formic acid 1f. Methanol (MeOH) 1g. Sodium chloride (NaCl) 1h. Sodium hydroxide (NaOH) 1i. Sodium dihydrogen phosphate (NaH2 PO4 ) 2. Chemical standards (>95% purity; Sigma-Aldrich, Inc.). 2a. O-Acetyl- L-carnitine (C2) 2b. Adenosine (A) 2c. L-Alanyl-L-alanine (diAla)

232

Britz-McKibbin

2d. L -Arginine (Arg) 2e. Atenolol (At) 2f. L -Carnitine (C0) 2g. L -Carnosine (Carn) 2h. L -Citrulline (Cit) 2i. 3-Choloro-L-tyrosine (ClTyr) 2j. Cystathionine (Cyst) 2k. Dopamine (DopN) 2l. Estradiol-3-glucuronide (E2-3G) 2m. Estradiol-3-sulfate (E2-3S) 2n. Estriol-3-glucronide (E3-3G) 2o. Estriol-3-sulfate (E3-3S) 2p. Estrone-3-glucuronide (E1-3G) 2q. Estrone-3-sulfate (E1-3S) 2r. L -Glutamine (Gln) 2s. L -Glutamic acid (Glu) 2t. Glutathione – oxidized (GSSG) 2u. Glutathione – reduced (GSH) 2v. Guanosine (G) 2w. L -Histidine (His) 2x. 3-Hydroxy-L-tryptophan (OHTrp) 2y. L -Leucine (Leu) 2z. L -Isoleucine (Ile) 2aa. L -Allo-isoleucine (allo-Ile) 2ab. L -Lysine (Lys) 2ac. L -Methionine (Met) 2ad. 3-Methyl adenosine (MeA) 2ae. N-Methyl-L-aspartic acid (MeAsp) 2af. 3-Methyl-L -histidine (MeHis) 2ag. O-Myristoyl-L-carnitine (C14) 2ah. 3-Nitro-L-tyrosine (NTyr) 2ai. L -Ornithine (Orn) 2aj. O-Palmitoylcarnitine (C16) 2ak. p-Aminobenzoic acid (PABA) 2al. L -phenylalanine (Phe) 2am. L -Proline (Pro) 2an. 5-Oxo-L-proline or pyroglutamic acid (5-oxo-Pro) 2ao. Serotonin (Sero)

CE–ESI-MS-Based Metabolomics

233

2ap. Trans-4-hydoxy-L-proline (4-OH-Pro) 2aq. Tryptamine (TyrN) 2ar. L-Tryptophan (Trp) 2as. L-Tyrosine (Tyr) 2at. L-Valine (Val) 3. Acylcarnitine standards (Larodan Fine Chemicals Inc., Malmö, Sweden). 3a. Propionyl-L-carnitine HCl (C3) 3b. Butyryl-L-carnitine HCl (C4) 3c. Octanoyl-L-carnitine HCl (C8) 4. Dilute individual stock solutions of metabolites (10 mM) in 1:1 methanol/water and prepare all reduced thiol standards (e.g., GSH) daily due to their intrinsic lability. 5. Degas all solutions and metabolite standards by sonication, then store at 4◦ C prior to use. 6. Quality control mixture: 20 μM of Ala, Leu, Phe, Tyr, Glu, Asp, Lys, His, and Arg; 50 μM diAla. 2.3. Equipment

1. Barnstead EASYpure II LF ultrapure water system (Dubuque, USA). 2. Polypropylene centrifuge tubes (0.5 and 1.5 mL; VWR International, Inc., Toronto, Canada). R 3. 3-kDa Nanosep centrifugal filters (Pall Life Sciences, Inc., Michigan, USA).

4. Disposable lancets (Unistik 3; Owen Munford Ltd, Georgia, USA). 5. Grade 903 Protein Saver Card filter paper and 1/8th-in.diameter hole puncher (Whatman, Inc., New Jersey, USA). 6. Data processing and statistical analysis software: Excel 2007 (Microsoft, Inc., Redmond, WA), MATLAB 2008 (The Mathworks, Inc., Natick, MA), and Igor 5.0 (Wavemetrics, Inc., Lake Oswego, OR).

3. Methods Metabolite profiling by CE–ESI-MS is typically performed under either acidic (pH 1.8) or alkaline (pH > 8.5) background electrolyte (BGE) conditions in conjunction with positive-ionor negative-ion-mode ESI-MS in order to resolve and detect complex mixtures of cationic and anionic metabolites, respectively. This not only ensures that weakly ionic or zwitterionic metabolites are adequately ionized but also avoids the pH range at

234

Britz-McKibbin

which the EOF tends to be more variable (pH ≈ 4–8). Thus, neutral metabolites (e.g., androgens) are not analyzed by CE–ESIMS unless using a surface-modified capillary suitable for capillary electrochromatography (CEC) (16) or partial-filling micellar electrokinetic chromatography (17), which are hybrid separation techniques more difficult to implement routinely. In some cases, complementary ionization sources can be coupled to CE in order to improve the detection of certain classes of metabolites, such as atmospheric photoionization (18). Nevertheless, a common constraint of all liquid-infused MS platforms is the lack of a universal ion source for gas-phase desorption (19), since ionization efficiency in ESI is highly dependent on the intrinsic physicochemical properties of metabolites with relative response factors that can vary up to three orders of magnitude (5). Given the selectivity constraints of the separation technique and ion source, CE–ESI-MS is primarily amenable to the analysis of polar and weakly ionic metabolites, which comprise a majority (>60%) of known metabolites or degradation products associated with primary metabolism (20). The sample pretreatment procedure for processing a variety of biological specimens is described in detail below, where only 10 μL of sample is typically required for injection (≈ 0.1 μL injected) using commercial CE–ESI-MS instruments. However, special precaution is needed when analyzing labile metabolites (e.g., thiols) in complex biological samples due to oxidation artifacts that can occur during sample collection, pretreatment, storage, and/or analysis (7). In most cases, internal standards are included in all samples in order to improve assay reproducibility in terms of quantification (i.e., integrated relative peak area) and migration time (i.e., RMT) performance. In addition, recovery standards can also be used to assess method accuracy after sample pretreatment, as well as identify sources of bias during large-scale batch analyses. Data pre-processing of large-scale metabolomic datasets requires adequate noise filtering, data alignment/normalization, and data exploration methods to reveal biologically significant features relevant to the experimental design, which can be facilitated with integrated data analysis software. However, a major challenge in untargeted metabolomics remains the identification of a large fraction of unknown metabolites that do not correspond to known candidates listed within public databases (e.g., KEGG, Human Metabolome Database, MetLin) despite access to accurate mass, isotopic composition, and fragmentation information by ESI–TOF-MS/MS (21). Indeed, this dilemma is considerable when less than 10% of total metabolite peaks (i.e., features) detected by hyphenated separation MS techniques can be quantified due to limited access to authentic standards (22). In this context, CE separations offer orthogonal qualitative information for unambiguous identification among

CE–ESI-MS-Based Metabolomics

235

putative candidates based on their characteristic migration behavior that can be predicted from the intrinsic properties of metabolites (9). Several examples of targeted analyses of cationic or anionic analytes relevant to metabolite profiling (e.g., expanded newborn screening), as well as a comprehensive metabolomic strategy for biomarker discovery (e.g., oxidative stress), are illustrated in the subsequent figures. 3.1. Sample Pretreatment of Dried Blood Spots

1. Collect human blood samples using a finger-prick method via disposable lancets and spot them on a Grade 903 Protein Saver Card and dry overnight. 2. Punch out a 3.2-mm (1/8 in.) disk (≈ 3.4 μL) manually from each dried blood spot (DBS) with a hole puncher into a 0.5-mL centrifuge tube that contained 100 μL of icecold 1:1 MeOH:H2 O with the internal standard dialanine (DiAla, 100 μM). 3. Extract the DBS under sonication for 10 min. 4. Filter the resulting extract solution through a 3-kDa R Nanosep centrifugal filter at 150×g at 4◦ C for 10 min prior to analysis (see Note 3). 5. Dilute the filtrate 1:1 with an aqueous ammonium acetate solution (400 mM, pH 7.0) to produce the final sample solution used for analysis (200 mM ammonium acetate, 25% MeOH, 50 μM dialanine). 6. In most cases, filtered DBS samples are analyzed directly by CE–ESI-MS without sample pretreatment steps such as chemical derivatization or solvent/reagent evaporation (see Note 4).

3.2. Sample Pretreatment of Human Plasma and Red Blood Cell Lysates

1. Collect human blood samples using a finger-prick method (see above) or via a venous catheter inserted in the antecubital vein, kept patent using a saline (0.9%, w/v) solution. 2. Immediately place each blood sample on ice and subsequently centrifuge at 20×g at 4◦ C for 5 min to fractionate plasma from erythrocytes. 3. Transfer the plasma supernatant and remove plasma proR teins using a 3-kDa Nanosep centrifugal filter at 150×g for 10 min prior to dilution and subsequent analysis (see Note 3). 4. Wash the red blood cells (RBCs) with phosphate buffered saline (10 mM NaH2 PO4 , 150 mM NaCl, pH 7.4, 4◦ C), then vortex and centrifuge at 70×g at 4◦ C for 1 min to isolate RBCs. 5. Repeat the washing steps three times until the supernatant is clear without any evidence of premature hemolysis.

236

Britz-McKibbin

6. Hemolyze 200 μL of RBCs by adding 600 μL of pre-chilled deionized water, then centrifuge at 70×g at 4◦ C for 1 min to sediment cell debris. 7. Next, filter 100 μL of RBC lysate with a 3-kDa filter at 150×g at 4◦ C for 10 min to remove excess hemoglobin. 8. Dilute all filtered protein-free RBC lysates twofold in BGE containing 50 μM diAla and 50 μM HEPES as internal standards for positive and negative ESI, respectively, and store frozen at –80◦ C. Thaw only once prior to analysis (see Note 5). 3.3. Sample Pretreatment of Human Urine

1. Collect human urine mid-stream as a first void morning sample and store at 4◦ C and normalize to creatinine to correct for urine dilution effects (see Note 6). 2. Centrifuge the urine samples at 70×g at 4◦ C for 2 min to remove particulate matter, then dilute the samples 10fold using deionized water containing 50 μM diAla and 50 μM HEPES as internal standards when using CE–ESIMS with positive (acidic BGE) and negative (alkaline BGE) ion modes, respectively.

3.4. Capillary Conditioning, Buffer Preparation, and CE–ESI-MS Operation

1. Prior to first use, condition fused silica capillaries installed in the coaxial sheath liquid interface (see Note 7) for 15 min each with MeOH, 1 M NaOH, 1 M formic acid, de-ionized H2 O and then 60 min with acidic or alkaline BGE. 2. Perform several trial runs with the newly conditioned capillary using a quality control test mixture as the sample and the desired BGE until stable currents and reliable ion signals are attained (see Note 8). 3. Once conditioning is complete, perform a 10-min prerinse/flush of the capillary with BGE prior to each separation. 4. High-purity (>98%) volatile buffer reagents (SigmaAldrich, Inc.) used for CE separations were prepared on a weekly basis, filtered, and sonicated prior to use. 5. In general, 1.0 M formic acid, pH 1.8 (diluted from concentrated formic acid), can be used as the acidic BGE for cationic metabolites under positive-ion-mode ESI-MS, whereas 50 mM ammonium acetate or ammonium bicarbonate, pH 8.5–9.5 (adjusted with ammonium hydroxide), can be used as the alkaline BGE for anionic metabolites under negative-ion-mode ESI-MS (see Note 9). 6. Perform CE separations at 20◦ C with an applied voltage of 30 kV unless otherwise indicated (see Note 10).

CE–ESI-MS-Based Metabolomics

237

7. The sheath liquid consisting of 1:1 MeOH:H2 O with 0.1% (v/v) formic acid (used with acidic BGE) or 5 mM ammonium acetate (used with alkaline BGE) is supplied by the 1100 series isocratic pump at a flow rate of 10 μL/min. 8. Nitrogen is used as both a nebulizing and a drying gas supplied at 6 psi and 10 L/min, respectively, whereas helium at 6 × 10–6 mbar is used as a damping gas for the ion trap. 9. Perform all ESI-MS analyses using a ± 4 kV cone voltage in positive- or negative-ion mode at a temperature of 300◦ C unless otherwise stated. 10. Record MS data within a range of 50–750 m/z using an ultrascan mode of 26,000 m/z per second (see Note 11). 3.5. Targeted Profiling of Cationic Metabolites

1. Introduce all samples to the capillary after a two- or tenfold dilution in 100 mM ammonium acetate, pH 7.0, using a low-pressure hydrodynamic injection by first injecting a sample for 75 s at 50 mbar (see Note 12) followed by a 60 s injection of the acidic BGE (1 M formic acid, pH 1.8) at 50 mbar prior to voltage application (see Note 13). 2. Perform all separations for cationic metabolites using unmodified fused-silica capillaries under normal polarity by CE with positive-ion-mode ESI-MS. 3. In most cases, all analyses should be performed in triplicate using a fresh BGE reservoir (inlet/anode) for each run. Data should be processed (e.g., peak integration) using extracted ion electropherograms (EIE) based on the m/z for each target metabolite (M+H+ ). 4. Perform a pre-rinse with acidic BGE for 10 min after each run in order to re-condition the capillary. 5. Figure 14.1a shows a schematic that depicts the coaxial sheath liquid interface configuration used in CE with an acidic BGE under positive-ion-mode ESI-MS conditions for the analysis of cationic metabolites, which elute in order of their apparent charge density (i.e., charge–size ratio) (see Note 14). 6. Figure 14.1b depicts the simultaneous analysis of polar amino acids (e.g., Arg) and surface-active acylcarnitines (e.g., palmitoyl- L-carnitine, C16) under a single elution condition by CE–ESI-MS relevant to expanded newborn screening of inborn errors of metabolism as derived from filtered DBS extracts (6), which also allows for the resolution of isomeric (e.g., Ile, allo-Ile) and isobaric (e.g., 4-OH-Pro) co-ion interferences (see Note 15). 7. Figure 14.2a highlights the interference-free time window (≈ 7–12 min) for quantification of low-abundance cationic

238

Britz-McKibbin

(a)

(b)

N+ H

C16

OO

O

O

6

H2N

6

Isomeric/Isobaric Ions

0.8

C14 0.6 0.4

Ile 0.2

4 2

Leu IS

Arg

NH3+

C0

8

x106

Ion Response

C2

O-

N H

1.2 1.0

O H

1.4x10

NH2+

C4

4 - OH - Pro Pro Cit allo - Ile Met

C16

C3

Arg

C8

Phe Tyr Orn

0 10

Val

11

12

13

14

15

16

5 - oxo - Pro

0.0 12

13

14

15

16

17

18

Time (min)

Fig. 14.1. (a) High-efficiency metabolite separations by CE in free solution and multivariate ion desorption model in ESIMS, which allows for (b) isomeric resolution of a wide variety of polar metabolites in filtered DBS extracts with minimal sample handling (reproduced from ref. 5 (a) and Ref. 6 (b) with permission from ACS).

metabolites from filtered RBC lysates under acidic conditions (5), where excess sodium ions (Na+ ) and reduced glutathione (GSH) represent major co-ion interferences that are fully resolved (see Note 16). 3.6. Targeted Profiling of Anionic Metabolites

1. Introduce all samples to the capillary using a low-pressure hydrodynamic injection by first injecting a sample prepared in 10 mM HCl/5 mM ammonium acetate, pH 2.0, for 75 s

CE–ESI-MS-Based Metabolomics

239

(a) O

C4-DC-M

Gln

C5-3-OH

H2N

allo-Ile

O O

2.0x107

OH

GSH

Isomeric Ions: 232 m/z

Salt Front

IS

C4-Butyryl

3

Reduced Glutathione

4

1.5

Na

x10

Apparent Ion Response

O O

2

At

C4-Isobutryl

1

1.0 0 12.0

12.5

13.0

MeA

0.5

Thea Orn

Carn MeHis

Sero DopN Cyst TyrN PABA

ClTyr

OHTrp Cit

NTyr

G MeAsp

0.0 6

8

10

12

14

Migration Time (min)

(b)

O

17-OH-Preg-S

HO

OH H H

Gal

H

C6-Gly H

OH O OH

3-OH-PA

N

HO

O

OH

Glucuronides

Apparent Ion Response

50x103

Sulfates

E23G

Positional Isomers

40

E13S

E13G

30

E217G

E33S

E33G

20

E23S

10

6.0

6.2

6.4

6.6

6.8

7.0

7.2

7.4

Migration Time (min)

Fig. 14.2. Representative 2D chemical structures of diverse classes of (a) cationic and (b) anionic metabolites and their isomers amenable to CE–ESI-MS and a series of extracted ion electropherograms showing direct analysis of cationic metabolites derived from red blood cell lysates without ion suppression effects (Na+, GSH)∗ 21 and anionic estrogen glucuronide/sulfate conjugates in urine (reproduced from ref. (5) (a) with permission from ACS, whereas (b) represents unpublished data).

at 50 mbar followed by a 60 s injection of the alkaline BGE at 50 mbar prior to voltage (see Notes 12 and 13). 2. Perform all separations for anionic metabolites using unmodified fused-silica capillaries under normal polarity by CE with negative-ion-mode ESI-MS detection (see Note 17).

240

Britz-McKibbin

3. Figure 14.2b highlights the direct analysis of several estrogen (e.g., E1, E2, E3) sulfate (S) and glucuronide (G) conjugates spiked in urine without chemical derivatization or desalting steps, which also allows for resolution of positional isomers, such as E2 -3-G and E2 -17-G (see Note 18). 4. Various classes of weakly and strongly acidic metabolites can also be analyzed by CE–ESI-MS, including organic acids, nucleotides, acylglycines, steroid conjugates, and sugar derivatives. However, the ionization efficiency of small polar metabolites tends to be poor in ESI-MS that compromises concentration sensitivity (5) notably under negativeion-mode detection due to ionization suppression effects caused by co-ion electrolytes used in the BGE (e.g., acetate) (23). 5. Due to the strong EOF under alkaline conditions in CE, both cations and anions can be simultaneously analyzed; however, in practice, the migration time window is too narrow for adequate resolution of cationic metabolites from background interferences to allow for reliable quantification (see Section 3.5). 3.7. Comprehensive Metabolomics for Biomarker Discovery

1. Perform sample pretreatment and CE–ESI-MS analysis for cationic and anionic metabolites as described in previous sections. 2. Neutral metabolites and poorly responsive/low-abundance polar metabolites are undetected by this method without selective chemical derivatization (23) or the use of a complementary hyphenated separation platform, such as LC–ESIMS or GC–EI-MS. 3. Perform data analysis by first manually processing each EIE at nominal mass resolution (Δm/z = 1) over a total range of 50–750 m/z under both positive and negative-ion modes in CE–ESI-MS (see Note 19). 4. Perform noise filtering and data normalization (see Note 20) prior to multivariate analysis by processing each putative metabolite signal as a single-paired variable denoted as m/z:RMT, which facilitates data alignment among replicates and experimental conditions. 5. Most of the data pre-processing procedures described above can be more efficiently performed with new advances in integrated software for CE–ESI-MS (13), although manual inspection of data quality is still strongly advised. 6. Perform unsupervised data exploration and pattern recognition of pre-processed metabolomic datasets (see Note 21) by principal component analysis (PCA) and hierarchal cluster analysis (HCA) using commercial multivariate software, such

CE–ESI-MS-Based Metabolomics

241

as MATLAB. These methods are particularly useful for classification and differentiation of biologically relevant metabolites in complex datasets. 7. Figure 14.3a depicts a 3D heat map of dynamic changes in intra-cellar metabolism of erythrocytes as a result of a 52-min acute bout (time interval: 45–82 min) of exhaustive exercise for a healthy volunteer (e.g., standardized ergometer cycling), where each segment represents a unique metabolite defined by its characteristic m/z:RMT, which are sequenced using a HCA algorithm (24). Various groupings of metabolites can be deduced from their characteristic time-dependent normalized peak area pattern from 0 h (preexercise) until 6 h (post-exercise/recovery). 8. Figure 14.3b illustrates a PCA (i.e., control and trial) and differential PCA (i.e., control–trial, refer to inset) scores plot, which allows for clear identification of putative biomarkers of oxidative stress associated with the study (24), two of which were subsequently identified (see Note 22) as oxidized glutathione (GSSG) and L-carnitine (C0).

Heat Map:Control Pr e- exercise Exhaustive exercise

(b)

Post-exercise (Recovery)

241 m/z: 4 132 m/z: 0 .953 RMT 1.054 RMT

204 m/z: 0.983 RMT

*

–15

–10

0 –5

* *

* *

307 m/z: 1.321 RMT

-2 -4 -6

5

-8

307 m/z: 1.321 RMT

4

-10 -12

2

241 m/z: 1.054 RMT

204 m/z: 0.983 RMT

0

–3

2 –2

162 m/z: 0 .929 RMT

–4

Time (min)

(c)

0

6 PC2

*

Metabolite ( m/z : RMT)

Hierarchical Cluster Analysis

PC1 (68.3%)

8

308 m/z: 1.447 RMT

2 162 m/z: 0 .929 RMT

10

PC2 (24.5%)

(a)

12

7 132 m/z: 0 .953 RMT

17

22

PC1

Control

NAC

308 m/z: 1.447 RMT

180 150 120

X2

Concentration (μM)

90 60

GSSG

30 0 0

100

200

300

C0

80

60

X6

40

20

0 0

100

200

300

Time (min)

Fig. 14.3. Differential metabolomics strategy using CE–ESI-MS for the discovery of biomarkers of exercise-induced oxidative stress and the assessment of the efficacy of nutritional intervention (e.g., NAC pretreatment) to delay the onset of fatigue, where (a) 3D heat maps of global perturbations in erythrocyte metabolism, (b) differential PCA for detection of putative biomarkers of oxidative stress, and (c) unknown metabolite identification and quantification, such as oxidized glutathione (GSSG) and L-carnitine (C0).

242

Britz-McKibbin

9. Figure 14.3c shows that GSSG and C0 can serve as early and late-stage biomarkers of oxidative stress, respectively, based on their unique time-dependent concentration profiles (24).

4. Notes 1. Alternative CE systems can also be coupled with other MS instruments using various sheath and sheathless interface designs; however, analyses of large sample sizes can be challenging without full system integration/automation. Indeed, there is increasing use of CE in conjunction with time of flight (TOF)/MS in metabolite profiling applications due to the improved robustness, mass resolution, and cost effectiveness of recent commercial instruments (25). 2. The polyimide coating at the distal end of the capillary (≈ 2–3 cm) that is directed toward the ion source was removed by burning it with a match and subsequent cleaning using a Kimwipe wetted with a few drops of methanol (MeOH). This capillary was then installed into the coaxial sheath liquid interface to generate a bare fused-silica tip that served as the effective emitter in the ion source. 3. Ultrafiltration is the deproteinization method of choice when processing whole blood or RBC lysate samples without oxidation artifacts, which are generated with protein denaturation (e.g., oxygenases, hemoglobin) with acid or organic solvent precipitation (7). 4. The solvent of the dried blood spot extract can also be evaporated under a gentle stream of N2 and reconstituted in 20 μL of sample solution in order to improve concentration sensitivity, which can provide an additional 10-fold sample enrichment prior to CE–ESI-MS. 5. Blood collection and sample workup should be processed within about 2 h, while blood specimens are kept on ice (4◦ C) at all times using degassed/pre-chilled solutions in order to avoid premature hemolysis and oxidation artifacts. 6. Creatinine is determined under acidic BGE conditions with positive-ion mode by CE–ESI-MS. The creatinine concentration is then used to normalize all other reported urinary metabolite concentration levels (i.e., μmol/mmol creatinine), which corrects for dilution effects when using singlespot morning urine samples. 7. Reproducible alignment of the capillary emitter into the ion source was facilitated by the Agilent coaxial sheath liquid ESI-MS interface, where the distal end of the bare

CE–ESI-MS-Based Metabolomics

243

fused-silica capillary was allowed to protrude from the sprayer by about 0.1 mm in order to minimize postcapillary dilution effects. 8. The average capillary lifespan is about 3 weeks when analyzing complex biological samples by CE–ESI-MS. In most cases, capillary replacement is made evident by an unstable capillary current and/or increasing noise in the total ion electropherogram that cannot be improved with subsequent conditioning or rinsing with BGE. The incorporation of a daily quality control test run is strongly recommended to ascertain capillary condition. Also, periodic cleaning of sprayer assembly and ion source is recommended to improve long-term performance. 9. In some cases, a small fraction of organic solvent may be added to the BGE (e.g., 15% (v/v) acetonitrile or methanol) in order to improve the solubility of hydrophobic analytes or suppress micellar formation of surface-active analytes, such as long-chain acylcarnitines. 10. In all cases, the current generated during voltage application in CE should be kept under 50 μA in order to minimize Joule heating effects and improve ion signal stability. 11. These conditions have been found to be appropriate for comprehensive metabolite profiling of various classes of metabolites when using full-scan ion monitoring in CE– ESI-MS. However, optimization of coaxial sheath liquid, ion source, and/or MS scanning conditions is recommended for targeted metabolite profiling to lower detection limits since ionization efficiency in ESI-MS is highly dependent on solute physicochemical properties. 12. The first long sample injection plug (≈ 10% of capillary length) allows for online sample preconcentration of lowabundance metabolites directly in capillary during electromigration prior to ionization based on differences in co-ion electrolyte mobility and pH at the sample and BGE interface. Up to a 50-fold improvement in concentration sensitivity can be realized by this method without compromising separation efficiency or resolution when using conventional instrumentation (9). 13. The second injection sequence is used to displace the original sample plug within the capillary past the electrode interface at the inlet (anode), which is required to avoid CEinduced oxidation artifacts when analyzing low micromolar levels of oxidized glutathione (GSSG) in the presence of excess reduced glutathione (GSH) (7). 14. Under these conditions, neutral metabolites (e.g., urea) co-migrate with the suppressed EOF (>20 min), whereas strongly acidic anions (e.g., chloride) migrate out of the

244

Britz-McKibbin

capillary at the inlet upon voltage application. The RMT of an ion can be accurately predicted in CE based on its characteristic absolute mobility (μo ) and thermodynamic pKa as derived from its putative chemical structure (9), which provides a novel strategy for identification of unknown metabolites complementary to ESI-MS. 15. Stereoselective analysis of allo-Ile from dried blood spot extracts allows for specific diagnosis of maple syrup urine disease without false positives associated with total Leu, whereas resolution of 4-OH-Pro prevents misdiagnosis of hydroxyprolinemia, a benign disorder. 16. A variety of classes of metabolites can be analyzed by this method, including amino acids, (biogenic) amines, peptides, acylcarnitines, and nucleosides, where the relative response factor of recently identified/novel metabolites can be predicted based on their fundamental physicochemical properties in CE–ESI-MS (5). 17. Note that separations performed under reverse polarity (inlet/cathode) conditions for anionic metabolites with cationic polymer-coated capillaries can lead to corrosion of the stainless steel electrode (i.e., ESI spray needle) in CE–ESI-MS with poor long-term robustness (26). Although platinum-based electrodes have been suggested as a solution to this problem, our protocol uses inexpensive, unmodified, bare fused-silica capillaries under normal polarity with a conventional stainless steel spray needle. 18. In this case, E2 -17-G migrates with a longer migration time (i.e., higher negative mobility) under alkaline conditions (pH > 9) relative to E2 -3-G due to its weakly acidic (unmodified) phenol moiety. Similarly, native estrogens can also be analyzed by CE–ESI-MS due to their weakly acidic phenol (pKa ≈ 10.2) functionality. 19. Each EIE was manually integrated with a minimum signal threshold over background noise of S/N ≈ 10 in order to avoid signal artifacts. Also, subsequent data pre-processing was performed to eliminate redundant signals originating from the same ion, including isotope contributions, fragment ions, salt adducts, as well as common ions detected under both ± ESI-MS conditions for unequivocal quantification of unique metabolite signatures. 20. Due to potential signal artifacts when analyzing complex biological samples, quantification of metabolite features in the total ion electropherogram was performed for peaks above a signal/noise threshold of >10. All integrated peak areas were then normalized to the internal standard. Data alignment was performed manually by

CE–ESI-MS-Based Metabolomics

245

labeling each metabolite feature as a paired variable, namely mass/charge:relative migration time (m/z:RMT) that had a corresponding normalized peak area, which allowed for easy comparison of datasets among different replicates and/or samples despite differences in apparent migration times caused by EOF. 21. All multivariate analyses were pre-processed as standardized datasets (X-matrix), where mean-centered responses for all metabolites were normalized to the inverse of the standard deviation determined at each experimental condition (e.g., time variable), which provided an effective way to scale different ion responses having different absolute values and ranges. 22. Unambiguous metabolite identification was realized by a series of experiments involving the acquisition/interpretation of multi-stage mass spectra when using a 3D ion trap mass analyzer, which was followed by an extensive search for candidate metabolites on public databases. Lead metabolite candidates were rationally selected based on several criteria, including equivalent nominal mass, characteristic fragments/isotope patterns as well as relevant intrinsic charge states deduced by their putative chemical structure. Once a series of lead candidates were selected, a comparison of their measured (or predicted) RMTs by CE and/or spiking of filtered RBC lysate samples with authentic standards (if available) was performed for unambiguous identification.

Acknowledgments The author wishes to acknowledge funding support from National Science and Engineering Research Council of Canada, Premier’s Research Excellence Award, as well as a Japan Society for Promotion of Science – Invitation Fellowship.

References 1. Smith, R. D., Barinaga, C. J., Udseth, H. R. (1988) Improved electrospray ionization interface for CE-MS. Anal Chem 60, 1948–1952. 2. Maxwell, J. E., Chen, D. D. Y. (2008) Twenty years of interface development for capillary electrophoresis–electrospray

ionization–mass spectrometry. Anal Chim Acta 627, 25–33. 3. Nilsson, S. L., Bylund, D., Jornten-Karlsson, M., Petersson, P., Markides, K. W. (2004) A chemometric study of active parameters and their interaction effects in a nebulized sheath–liquid electrospray interface for

246

4.

5.

6.

7.

8. 9.

10. 11. 12.

13.

14.

15.

Britz-McKibbin capillary electrophoresis–mass spectrometry. Electrophoresis 25, 2100–2107. Mokaddem, M., Gareil, P., Belgaied, J. -E., Varenne, A. (2009) New insight into suction and dilution effects in CE coupled to MS via an ESI interface II – dilution effect. Electrophoresis 30, 1692–1697. Chalcraft, K. R., Lee, R., Mills, C., BritzMcKibbin, P. (2009) Virtual quantification of metabolites by CE–ESI-MS: predicting ionization efficiency without chemical standards. Anal Chem 81, 2506–2515. Chalcraft, K. R., Britz-McKibbin, P. (2009) Newborn screening of inborn errors of metabolism by CE–ESI-MS: a second-tier method with improved specificity and sensitivity. Anal Chem 81, 307–314. Lee, R., Britz-McKibbin, P. (2009) Differential rates of glutathione oxidation for assessment of cellular redox status and antioxidant capacity by CE-MS: an elusive biomarker of oxidative stress. Anal Chem 81, 7047–7056. Hruska, V., Janos, M., Gas, B. (2006) Simul 5 – free dynamic simulator of electrophoresis. Electrophoresis 27, 984–991. Lee, R., Niewzcas, L., Britz-McKibbin, P. (2007) Integrative metabolomics for characterizing unknown low-abundance metabolites by CE-MS with computer simulations. Anal Chem 79, 403–415. Ramautar, R., Somsen, G. W., Jong, G. Jd. (2009) CE-MS in metabolomics. Electrophoresis 30, 276–291. Monton, M. R. N., Soga, T. (2007) Metabolome analysis by CE-MS. J Chromatogr A 1168, 237–246. Huhn, C., Ramautar, R., Wuhrer, W., Somsen, G. W. (2010) Relevance and use of capillary coatings in capillary electrophoresis– mass spectrometry. Anal Bioanal Chem 396, 297–314. Sugimoto, M., Hirayama, A., Ishikawa, T., Robert, M., Baran, R., Uehara, K., Kawai, K., Soga, T., Tomita, M. (2010) Differential metabolomics software for CEMS data analysis. Metabolomics 6, 27–41. 10.1007/s11306-009-0175–1. Williams, B. J., Cameron, C. J., Workman, R., Broeckling, C. D., Sumner, L. W., Smith, J. T. (2007) Amino acid profiling in plant cell cultures: an inter-laboratory comparison of CE-MS and GC-MS. Electrophoresis 28, 1371–1379. Buscher, J. M., Czernik, D., Ewald, J. C., Sauer, W., Zamboni, N. (2009) Crossplatform comparison of methods for quantitative metabolomics of primary metabolism. Anal Chem 81, 2135–2143.

16. Kato, M., Onda, Y., Sekimoto, M., Degawa, M., Toyo’oka, T. (2009) A CEC–ESI-MS method for simultaneous analysis of charged and neutral constituents of a hepatocarcinoma cell metabolome. J Chromatogr A 1216, 8277–8282. 17. Amundsen, L. K., Kokkonen, J. T., Soren, H. (2008) Comparison of partial filling MEKC analyses of steroids with use of ESI-MS and UV spectrophotometry. J Sep Sci 31, 803–813. 18. Mol, R., Jong, G. D., Somsen, G. W. (2005) On-line capillary electrophoresis–mass spectrometry using dopant-assisted atmospheric pressure photoionization: setup and system performance. Electrophoresis 26, 146–154. 19. Nordstrom, A., Want, E., Northen, T., Lehtio, J., Siuzdak, G. (2008) Multiple ionization mass spectrometry strategy used to reveal the complexity of metabolomics. Anal Chem 80, 421–429. 20. Ohashi, Y., Hirayama, A., Ishikawa, T., Nakamura, S., Shimizu, K., Ueno, Y., Tomita, M., Soga, T. (2008) Depiction of metabolome changes in histidine-starved Escherichia coli by CE–TOFMS. Mol Biosyst 4, 135–147. 21. Kind, T., Fiehn, O. (2006) Metabolomic database annotations via query of elemental compositions: mass accuracy is insufficient even at less than 1 ppm. BMC Bioinform 7, 234–244. 22. Soga, T., Ohashi, Y., Ueno, Y., Naraoka, H., Tomita, M., Nishioka, T. (2003) Quantitative metabolome analysis using capillary electrophoresis mass spectrometry. J Proteome Res 2, 488–494. 23. Yang, W. C., Regnier, F. E., Adamec, J. (2008) Comparative metabolite profiling of carboxylic acids in rat urine by CE– ESI MS/MS through positively pre-charged and (2)H-coded derivatization. Electrophoresis 29, 4549–4560. 24. Lee, R., West, D., Phillips, S. M., Britz-McKibbin, P. (2010) Differential metabolomics for quantitative assessment of oxidative stress with strenuous exercise and nutritional intervention: thiol-specific regulation of cellular metabolism with N-acetylL-cysteine. Anal Chem 82, 2959–2968. 25. Staub, A., Schappler, J., Rudaz, S., Veuthey, J. -L. (2009) CE–TOF/MS: fundamental concepts, instrumental considerations and applications. Electrophoresis 30, 1610–1623. 26. Soga, T., Igarashi, K., Ito, C., Mizobuchi, K., Zimmermann, H.-P., Tomita, M. (2009) Metabolomic profiling of anionic metabolites by CE-MS. (2009) Anal Chem 81, 6165–6174.

Chapter 15 Liquid Chromatography-Mass Spectrometry (LC-MS)-Based Lipidomics for Studies of Body Fluids and Tissues Heli Nygren, Tuulikki Seppänen-Laakso, Sandra Castillo, Tuulia Hyötyläinen, and Matej Orešicˇ Abstract In this paper, analytical methodologies for the global profiling of lipids in serum and tissue samples are reported. The sample preparation is based on a modified Folch extraction, and the analysis is carried out with ultrahigh-performance liquid chromatography combined with mass spectrometry (UPLC-MS). For further identification, MSn mass spectrometry is carried out utilizing an LTQ-Orbitrap mass spectrometry as the detector. Such a system affords determination of accurate masses and is thus a highly useful tool for lipid identification. The repeatability of the analysis proved to be good, with relative standard errors for spiked samples being between 4.51 and 10.44%. The throughput of the methodology described here is over 100 samples a day. Key words: Lipidomics, liquid chromatography, mass spectrometry.

1. Introduction Lipids are a broad group of naturally occurring molecules, which include fats, waxes, sterols, monoglycerides, diglycerides, triglycerides, phospholipids, and other hydrophobic or amphiphilic small molecules. Lipids originate entirely or in part from two distinct types of building blocks: ketoacyl and isoprene groups (1). They are a functionally as well as a structurally very diverse group of compounds in part due to the many possible variations of lipid building blocks and the different ways of noncovalent linkage (2).

T.O. Metz (ed.), Metabolic Profiling, Methods in Molecular Biology 708, DOI 10.1007/978-1-61737-985-7_15, © Springer Science+Business Media, LLC 2011

247

248

Nygren et al.

Lipids play important roles in biological systems as signaling or energy storage molecules and as major constituents of cellular membranes (3). Tight control of membrane lipid composition is of central importance for the maintenance of normal cellular physiology, and its dysregulation may affect, e.g., membrane fluidity as well as topology, mobility, or activity of membranebound proteins. Even minor changes of lipid composition may affect the membrane properties (4) and thus the physiology of the cell. To be able to detect such changes, sensitive and accurate platforms are needed to measure the lipids in biological systems. Global characterization of lipids in biological samples, i.e., lipidomics, remains a challenging task. In recent years, lipidomics has emerged owing to the advances in mass spectrometry-based analytical technologies, which allow detection and quantitation of hundreds of intact lipid molecular species in parallel (5, 6). These global lipidomics approaches include shotgun lipidomics, which uses direct infusion of lipid extracts into the mass spectrometer (7, 8), as well as liquid chromatography coupled to mass spectrometry (LC-MS) (9, 10). Both approaches have their own advantages and disadvantages. In both methodologies, sample preparation is a critical step and should be fast, repeatable, and nonselective. The shotgun approach is advantageous because it is a relatively simple and rapid way to profile the lipids in crude lipid extracts. However, ion suppression is the major complication in this methodology (11, 12). To avoid ion suppression, more careful sample pre-treatment, using, e.g., fractionation, is often needed, increasing the total analysis time. The advantage of LCMS-based methodologies, on the other hand, is potentially higher sensitivity as well as the ability, by using a non-targeted strategy, to detect and identify novel lipids. However, with conventional LC-MS methods, the analysis time can be relatively long. Moreover, the LC-MS methods can suffer from carryover effects, which must be taken into account in the method development. It should also be noted that due to the extremely large number of different lipids, coelution cannot be totally avoided. Coelution of lipids can be reduced by using novel ultrahigh-pressure liquid chromatography such as UPLCTM , which allows highly efficient separation in a very short analysis time (13). Here we describe a UPLC-MS-based global lipidomics platform, with applications to serum and tissue sample profiling. As an integral component of the platform, identification of lipids using UPLC-MSn is also described. In the identification stage, an LTQOrbitrap is used as the detector because this system provides outstanding mass accuracy and mass resolution. The ability to detect accurate masses allows the unequivocal compositional and structural elucidation of the compounds.

LC-MS-Based Lipidomics

249

2. Materials 2.1. Samples

1. Samples: 10 μL serum/plasma or cell concentrate (centrifugation residue, e.g. macrophages, islets, adipocytes); 5 mg of heart, liver, muscle or fat tissue (e.g. white or brown adipose tissue).

2.2. Standards and Chemicals

The internal standard mixture contains compounds from several different lipid classes: phosphatidylcholines (PC), a phosphatidylethanolamine (PE), a ceramide (Cer), a phosphatidylserine (PS), a phosphatidic acid (PA) as well as mono-, di-, and triacylglycerols (MG, DG, and TG, respectively). Lipids are denoted by their molecular composition as follows: :/< Number of carbon atoms in the second fatty acid moiety>:. For example, the abbreviation PC(17:0/17:0) indicates a phosphatidylcholine comprising two C17 fatty acids with no double bonds. 1. Standard mixture 1A added to plasma and cell samples before extraction: PC(17:0/0:0), PC(17:0/17:0), PE (17:0/17:0), PG(17:0/17:0)[rac], Cer(d18:1/17:0), PS(17:0/17:0), and PA(17:0/17:0) (Avanti Polar Lipids, Inc., Alabaster, AL, USA) and MG(17:0/0:0/0:0)[rac], DG(17:0/17:0/0:0)[rac] and TG(17:0/17:0/17:0) (Larodan Fine Chemicals, AB, Malmö, Sweden). The concentration of each standard is approximately 0.2 μg/sample. 2. Standard mixture 1B added to fat tissue samples (after homogenization) or other tissue samples (before homogenization): PC(17:0/0:0), PC(17:0/17:0), PE(17:0/17:0), and Cer(d18:1/17:0) (Avanti Polar Lipids, Inc.) and TG(17:0/17:0/17:0) (Larodan Fine Chemicals). The concentration of each standard is 0.5–1 μg/sample. 3. Standard mixture 2 added after extraction: PC (16:1/0:0D3 ), PC(16:1/16:1-D6 ), and TG(16:0/16:0/16:0-13 C3) (Larodan Fine Chemicals). The concentration of each standard is 1 μg/tissue sample and 0.1–0.2 μg/serum or cell sample. 4. Extraction solvents: mixture of HPLC-grade chloroform/methanol (2:1, v/v).

2.3. UPLC-MS Systems

1. Acquity Ultra Performance LCTM (UPLC; Waters Corporation, Wexford, Ireland). 2. Sample organizer for automatic sampling.

250

Nygren et al. R 3. LC column: Acquity UPLC BEH C18 1.7 μm, 2.1 mm × 100 mm column (Waters Corporation).

4. Solvent A: 1% 1 M NH4 Ac and 0.1% HCOOH in water. 5. Solvent B: AcCN/2-propanol (1:1, v/v), 1% 1 M NH4 Ac, and 0.1% HCOOH. 6. Quadrupole-time-of-flight (Q-Tof) Premier mass spectrometer (Waters Corporation). 7. LTQ-Orbitrap mass spectrometer (Thermo Scientific Corp., San Jose, CA). 2.4. Other Instrumentation

1. Ultrasonication bath (Fritsch, Laborette 17, Idar-Oberstein, Germany). 2. Retsch Mixer Mill MM400 homogenizer (Retsch GmbH, Haan, Germany). 3. Multiskan EX instrument (Thermo Scientific, Inc., Waltham, MA, USA) for spectrophotometric determination. 4. Micro BCATM Protein Assay Kit (Pierce, Rockford, IL, USA). 5. Vortexer. 6. Centrifuge.

2.5. Data Processing Software

1. MZmine 2 software for peak detection, alignment, and normalization is used to process raw MS data (http://mzmine.sourceforge.net/). Supported data formats are mzML (version 1.0 and 1.1), mzXML (versions 2.0, 2.1, and 3.0), mzData (versions 1.04 and 1.05), NetCDF (no MSn support), and Thermo RAW (only on Windows with Thermo Xcalibur version 2).

3. Methods The UPLC-MS-based lipidomics platform consists of multiple steps (Fig. 15.1), including sample preparation, separation and detection, and finally data analysis. Total lipid extracts are obtained using a modified Folch extraction and the extracts are analyzed by UPLC-MS in positive- (ESI+) and/or negative (ESI–)-ion mode. The choice of ionization may depend on the biological questions asked. To avoid potential bias, it is important that the samples are randomized prior to sample preparation and analysis. In large-scale analyses, it is common that the samples are analyzed in multiple batches of several hundred samples (14, 15).

LC-MS-Based Lipidomics

Tissue sample 5 mg

Cell concentrate 10 µl

Addition of PBS-buffer/NaCl Homogenisation

251

Serum sample 10 µl Addition of NaCl

Addition of internal standards Liquid extraction CHCl3:MeOH (2:1) Centrifugation Collection of lower phase Addition of external standards

Protein content determination

UPLC-QTof-MS

UPLC-LTQ-Orbitrap: MS/MS

Data analysis: MZmine integration, alignment of the data, identification, normalization

Fig. 15.1. Typical UPLC-MS lipidomics sample analysis workflow.

Peaks are detected, aligned, and calibrated with internal standards in specialized software packages such as MZmine 2. The identification of lipids may be performed automatically based on an internal database of mass-to-charge ratio (m/z) and retention time values (16). However, both identified and unidentified lipids are usually included in further data analyses. The most important peaks, either for validation purposes if known or for de novo identification if unknown, need to be characterized further by mass spectrometry (UPLC-MS/MS or UPLC-MSn ). To minimize carryover effects, relatively strong solvent is used as the UPLC eluent. No significant carryover was noticed, proved by blank analyses after real samples. The repeatability of the analysis proved to be good, with RSD of the spiked standards varying from 4.51% (TG standard) to 10.44% (PE standard) (n = 10). 3.1. Sample Preparation

1. Serum/plasma: a. Dilute 10 μL of serum/plasma with 10 μL of 0.15 M (0.9%) sodium chloride and add internal standard mixture 1A such that the concentration of internal standard is 0.2 μg/10 μL serum/plasma. b. Extract serum/plasma lipids using 100 μL of HPLCgrade chloroform/methanol (2:1, v/v). c. Vortex the mixture for 2 min, then allow to stand for 30 min.

252

Nygren et al.

d. Centrifuge the sample for 3 min at 7800×g. e. Collect the lower phase and add 10–20 μL of internal standard mixture 2 such that the concentration of internal standard is 0.1–0.2 μg/10 μL serum/plasma. 2. Cells: a. Dilute 10 μL of cell concentrate with 100 μL of PBS buffer or 0.15 M NaCl (0.9%) and add standard mixture 1A such that the concentration of internal standard is 0.2 μg/10 μL serum/plasma. b. Homogenize in an ultrasonicating bath. c. Remove 5 μL of homogenized cell suspension for determination of protein content. Dilute this aliquot with PBS buffer for use in the Micro BCATM Protein Assay Kit and determine protein concentration on a Multiskan EX instrument or equivalent. The lipid concentrations in cells will be normalized with the measured protein content. d. Extract cellular lipids by combining 20 μL of the homogenate with 100 μL HPLC-grade chloroform/ methanol (2:1, v/v). e. Vortex the mixture for 2 min and allow to stand for 30 min. f. Centrifuge the sample for 3 min at 7800×g. g. Collect the lower phase and add 10–20 μL of internal standard mixture 2 such that the concentration of internal standard is 0.1–0.2 μg/10 μL cell concentrate. 3. Tissues: a. Add 10–20 μL of internal standard mixture 1B and to 5 mg of heart, liver, or muscle sample such that the concentration of internal standard is 0.5–1 μg/5 mg of sample and homogenize with the extraction solvent (HPLC-grade chloroform/methanol, 2:1, v/v) and three grinding balls (Ø 3 mm) for 2 min (20 Hz) with mixer mill homogenizer. b. Add 50 μL of 0.15 M NaCl (0.9%) to the homogenate, vortex the mixture for 2 min, and allow to stand for 30 min. c. Centrifuge the sample for 3 min at 7800×g. d. Collect the lower phase and add 10–20 μL of internal standard mixture 2 such that the concentration of internal standard is 1 μg/5 mg tissue sample. 4. Adipose tissue: a. Add 100 μL of 0.15 M (0.9%) sodium chloride or PBS buffer to 5 mg of white or brown adipose tissue and homogenize with three grinding balls (Ø 3 mm) for 2 min (20 Hz) with mixer mill homogenizer.

LC-MS-Based Lipidomics

253

b. Dilute 10 μL of the homogenate with 10–40 μL of 0.15 M NaCl (0.9%), add 10–20 μL of internal standard mixture 1B such that the concentration of internal standard is 0.5–1 μg/5 mg of sample, and combine with 100–400 μL of HPLC-grade chloroform/methanol (2:1, v/v), vortex the mixture for 2 min, and allow to stand for 30 min. c. Centrifuge the sample for 3 min at 7800×g. d. Collect the lower phase and add 10–20 μL of internal standard mixture 2 such that the concentration of internal standard is 1 μg/5 mg tissue sample. 1. The UPLC gradient is as follows: set to start at 35% B to reach 80% B in 2 min, 100% B in 7 min and remain there for 7 min. The total run time including a 4-min reequilibration step at initial conditions is 18 min. An example chromatogram is shown in Fig. 15.2.

3.2. Lipidomics Profiling Platform (UPLC-MS)

2. Set the temperature of the column to 50◦ C. 3. Set the temperature of the sample organizer to 10◦ C. 4. Set the flow rate to 0.400 mL/min. 5. Inject 2 μL of lipid extract (chloroform phase). 6. Lipid compounds are detected by a Q-Tof mass spectrometer using electrospray ionization in positive- or negative-ion mode. 7. Collect data in continuum mode using extended dynamic range over a mass range of m/z 300–1200 with a scan duration of 0.2 s (see Note 1). 8. Process data using the MZmine 2 software (17, 18) (see Note 2) or equivalent:

Phospholipids

Triacylglycerols

%

100

Lysophospholipids

0 1.00

2.00

3.00

4.00

5.00

6.00

7.00

8.00

9.00

10.00

11.00

12.00

Time (min)

Fig. 15.2. Total ion chromatogram (positive ion mode) from UPLC-MS-based lipidomics analysis of serum (Waters QTof Premier as mass analyzer) showing major lipid fractions.

254

Nygren et al.

a. Data processing includes alignment of peaks, peak integration, normalization, and identification. b. Lipids are identified using an internal spectral library or with tandem mass spectrometry (see below and Fig. 15.3). A

866.81738

100

[M+NH4]+

Δ M = +0.3 ppm

90

Relative Abundance

80 70 867.82080

60 50 40 30

868.82434

20 10 0 865

B

866

867

868 m/z

870

871

579.50

100

Neutral loss of RCOOH+NH3

90

Relative Abundance

869

+

[M+H-270]

80 70 60 50 40 30 20 10 0

Relative Abundance

C

160 150 140 130 120 110 100 90 80 70 60 50 40 30 20 10 0

250

300

350

400

450

500

550 600 m/z

650

700

750

800

850

[C17H32OH]+ Δ M = –1.6ppm

253.25217

354.30007 235.24130

220 240 260 280 300 320 340 360 380 400 420 440 460 m/z

Fig. 15.3. UPLC LTQ-Orbitrap lipidomics analysis of human serum extract in positive-ion mode. (a) Orbitrap highresolution mass spectrum at retention time 9.2 min, showing the peak of internal standard TG (17:0/17:0/17:0). (b) MS2 spectrum of the peak with m/z 866.8, showing the product ion with m/z 579.6. (c) Orbitrap high-resolution MS3 spectrum, showing the product ion of protonated aldehyde.

LC-MS-Based Lipidomics

255

c. The data is normalized using one or more internal standards representative of each class of lipid present in the samples; the intensity of each identified lipid is normalized by dividing it with the intensity of its corresponding standard and multiplying it by the concentration of the standard. All monoacyl lipids except cholesterol esters, such as monoacylglycerols and monoacylglycerophospholipids, are normalized with PC (17:0/0:0), all diacyl lipids except ethanolamine phospholipids are normalized with PC (17:0/17:0), all ceramides with Cer (d18:1/17:0), all diacyl ethanolamine phospholipids with PE (17:0/17:0), and TG and cholesterol esters with TG (17:0/17:0/17:0). Other (unidentified) molecular species are calibrated with PC (17:0/0:0) for retention time <300 s, PC (17:0/ 17:0) for retention time between 300 and 410 s, and TG (17:0/17:0/17:0) for higher retention times. In the case of cell samples, lipids are further normalized by dividing the normalized lipid concentration with the protein content of the sample. 3.3. Lipidomics Profiling Platform for Lipid Identification (UPLC-MSn )

1. Chromatographic separation is carried out as described above (see Section 3.2). 2. MSn fragmentation is performed on an LTQ-Orbitrap mass spectrometer using electrospray ionization in positive- or negative-ion mode. 3. Calibrate the instrument externally according to the instructions of manufacturer. 4. Acquire MS2 and MS3 data using either low-resolution (LTQ) or high-resolution Orbitrap up to target mass resolution R = 60,000 (full-width, half maximum at m/z = 400). 5. Use normalized collision energies of 30 and 35% in MS2 and MS3 experiments, respectively (see Note 3).

4. Notes 1. The method parameters are optimized for the specific analytical system used. The primary aim of the described method is rapid screening of lipid molecular species across a broad range of lipid classes. While use of positive-ion mode is described, the same platform can also be applied using electrospray ionization in negative-ion mode, which will lead to better sensitivity for specific phospholipid classes such as

256

Nygren et al.

phosphatidylinositols, phosphatidylserines, and phosphatidic acids. The method can be transferred to a different LC-MS system. Ideally, such system should have MS/MS and/or accurate mass capabilities. 2. The updated version of the open source MZmine 2 software, including tutorials, is available at http:// mzmine.sourceforge.net/. The users are recommended to join the developer’s mailing list, where specific questions can be asked from the user’s and developer’s community. The data processing parameters need to be optimized for a specific analytical system used, in order to account for different peak shapes and lengths as well as different mass spectrometer resolution. Differential profiling of multiple samples requires steps such as peak detection, alignment (matching of peaks across multiple samples), and normalization using internal standards. 3. In general, the MS/MS or MSn experiments can be performed on the same instrument where the LC-MS data were acquired (e.g., Q-Tof or ion trap instruments) or, as in our case, on separate instruments. For the latter, it is essential that the chromatographic conditions are identical and that the retention times across the two platforms are validated for the standard compounds.

Acknowledgments This project was supported by the EU-funded project ETHERPATHS (FP7-KBBE-222639, http://www.etherpaths.org/). References 1. Fahy, E., Subramaniam, S., Brown, H. A., Glass, C. K., Merrill, A. H. Jr., Murphy, R. C., Raetz, C. R. H., Russell, D. W., Seyama, Y., Shaw, W., Shimizu, T., Spener, F., van Meer, G., VanNieuwenhze, M. S., White, S. H., Witztum, J. L., Dennis, E. A. (2005) A comprehensive classification system for lipids. J Lipid Res 46, 839–862. 2. Yetukuri, L., Ekroos, K., Vidal-Puig, A., Oresic, M. (2008) Informatics and computational strategies for the study of lipids. Mol Biosyst 4, 121–127. 3. Alberts, B., Johnson, A., Lewis, J., Raff, M., Roberts, K., Walter, P. (2007) Molecular

4.

5.

6. 7.

Biology of the Cell, Garland Science, New York, NY. Niemelä, P. S., Ollila, S., Hyvönen, M. T., Karttunen, M., Vattulainen, I. (2007) Assessing the nature of lipid raft membranes. Plos Comp Biol 3, e34. Oresic, M., Hänninen, V. A., Vidal-Puig, A. (2008) Lipidomics: a new window to biomedical frontiers. Trends Biotechnol 26, 647–652. Wenk, M. R. (2005) The emerging field of lipidomics. Nat Rev Drug Discov 4, 594–610. Schwudke, D., Liebisch, G., Herzog, R., Schmitz, G., Shevchenko, A. (2007)

LC-MS-Based Lipidomics

8.

9.

10.

11.

12.

Shotgun lipidomics by tandem mass spectrometry under data-dependent acquisition control. Methods Enzymol 433, 175–191. Han, X., Gross, R. W. (2005) Shotgun lipidomics: electrospray ionization mass spectrometric analysis and quantitation of cellular lipidomes directly from crude extracts of biological samples. Mass Spectrom Rev 24, 367–412. Houjou, T., Yamatani, K., Imagawa, M., Shimizu, T., Taguchi, R. (2005) A shotgun tandem mass spectrometric analysis of phospholipids with normal-phase and/or reversephase liquid chromatography/electrospray ionization mass spectrometry. Rapid Commun Mass Spectrom 19, 654–666. Laaksonen, R., Katajamaa, M., Päivä, H., Sysi-Aho, M., Saarinen, L., Junni, P., Lütjohann, D., Smet, J., Coster, R. V., Seppänen-Laakso, T., Lehtimäki, T., Soini, J., Oresic, M. (2006) A systems biology strategy reveals biological pathways and plasma biomarker candidates for potentially toxic statin-induced changes in muscle. Plos One 1, e97. Carrasco-Pancorbo, A., Navas-Iglesias, N., Cuadros-Rodríguez, L. (2009) From lipid analysis towards lipidomics, a new challenge for the analytical chemistry of the 21st century. Part I: modern lipid analysis. Trends Anal Chem 28, 263–278. Haynes, C. A., Allegood, J. C., Park, H., Sullards, M. C. (2009) Sphingolipidomics: methods for the comprehensive analysis of sphingolipids. J Chromatogr B 877, 2696–2708.

257

13. Nordstrom, A., O’Maille, G., Qin, C., Siuzdak, G. (2006) Nonlinear data alignment for UPLC−MS and HPLC−MS based metabolomics: quantitative analysis of endogenous and exogenous metabolites in human serum. Anal Chem 78, 3289–3295. 14. Bijlsma, S., Bobeldijk, I., Verheij, E. R., Ramaker, R., Kochhar, S., Macdonald, I. A., vanOmmen, B., Smilde, A. K. (2006) Largescale human metabolomics studies: a strategy for data (pre-) processing and validation. Anal Chem 78, 567–574. 15. Oresic, M., Simell, S., Sysi-Aho, M., NäntöSalonen, K., Seppänen-Laakso, T., Parikka, V., Katajamaa, M., Hekkala, A., Mattila, I., Keskinen, P., Yetukuri, L., Reinikainen, A., Lähde, J., Suortti, T., Hakalax, J., Simell, T., Hyöty, H., Veijola, R., Ilonen, J., Lahesmaa, R., Knip, M., Simell, O. (2008) Dysregulation of lipid and amino acid metabolism precedes islet autoimmunity in children who later progress to type 1 diabetes. J Exp Med 205, 2975–2984. 16. Yetukuri, L., Katajamaa, M., Medina-Gomez, G., Seppanen-Laakso, T., Vidal-Puig, A., Oresic, M. (2007) Bioinformatics strategies for lipidomics analysis: characterization of obesity related hepatic steatosis. BMC Syst Biol 1, e12. 17. Katajamaa, M., Oresic, M. (2005) Processing methods for differential analysis of LC/MS profile data. BMC Bioinform 6, 179. 18. Katajamaa, M., Miettinen, J., Oresic, M. (2006) Mzmine: toolbox for processing and visualization of mass spectrometry based molecular profile data. Bioinformatics 22, 634–636.

Chapter 16 Electrospray Ionization Tandem Mass Spectrometry (ESI-MS/MS)-Based Shotgun Lipidomics Giorgis Isaac Abstract In the past decade, many new strategies for mass spectrometry (MS)-based analyses of lipids have been developed. Lipidomics is one of the most promising research fields to emerge as a result of these advances in MS. Currently, mass spectrometric analysis of lipids involves two complementary approaches: direct infusion (shotgun lipidomics) and liquid chromatography coupled to MS. In this chapter, I will demonstrate the approach of shotgun lipidomics using electrospray ionization tandem MS for the analysis of lipid molecular species directly from crude biological extracts of tissue or fluids. Key words: Electrospray, lipids, mass spectrometry, neutral loss, precursor scan, shotgun lipidomics, tandem mass spectrometry.

1. Introduction Electrospray ionization mass spectrometry (ESI-MS) is a highly sensitive method that has been widely used for the selective characterization and quantification of lipid molecular species from various biological samples (1). In the traditional analytical approach to lipid analysis using gas chromatography, chemical derivatization is often required prior to analysis to increase the volatility of lipid species. In contrast, ESI is a mild ionization technique that generates intact molecular ions from solution with relatively little fragmentation and no derivatization is required. In the past decade, many new strategies for MS-based analyses of lipids have been developed. Currently, mass spectrometric analysis of lipids involves two complementary approaches: direct T.O. Metz (ed.), Metabolic Profiling, Methods in Molecular Biology 708, DOI 10.1007/978-1-61737-985-7_16, © Springer Science+Business Media, LLC 2011

259

260

Isaac

infusion and liquid chromatography (LC) coupled to MS. The term “shotgun lipidomics” is increasingly being used to describe the direct infusion approach (2). In this chapter, I will demonstrate the approach of shotgun lipidomics using ESI tandem MS for the analysis of lipid molecular species directly from crude biological extracts of tissue or fluids. For LC-MS-based lipidomics, refer to Chapter 15. In shotgun lipidomics, intact total lipid extracts are infused directly into a MS via the electrospray ion source, and lipid species are subsequently identified and quantified by tandem MS using lipid class-specific precursor ion scan (PIS) and constant neutral loss scan (NLS), reviewed in (2, 3). For quantification purposes, a cocktail of non-naturally occurring internal standards representative of the different lipid classes is added to the lipid extracts. Alternatively, the internal standards can be added before lipid extraction. The internal standards are used to correct for multiple experimental factors such as possible variations in MS parameters, which can affect ionization and ion count for accurate quantification. Figure 16.1 shows a flowchart for quantitative analysis of lipid molecular species using ESI-MS/MS.

Biological Sample

Lipid extraction Addition of internal standards

Dissolve in infusion solvent ESI - MS full scan

Ionized lipid species Product ion scan

Identify common fragment ion or neutral loss of lipid class Precursor or neutral loss scan

Lipid molecular species of a particular class

Data analysis including background subtraction, isotopic overlap correction and internal standard normalization

Fig. 16.1. Flow chart of quantitative analysis of lipid molecular species in a biological sample by electrospray ionization tandem mass spectrometry.

Electrospray Ionization Tandem Mass Spectrometry

261

In ESI-MS, biological sample extracts can be infused directly into the MS, or indirectly via an LC system. Whether to use direct infusion MS or LC-MS depends on the complexity of the biological sample and the interest in detecting minor lipid species. Direct infusion, or analysis of each analyte in the presence of all other analytes, has the advantage of simplicity and application to a wide range of lipid species. Direct infusion of liquids is performed with a syringe pump or an autosampler operating at microliters per minute flow rates. The composition of the sample entering the instrument is constant, so ratios of internal standards to compounds of interest are constant throughout the analysis, making quantification straightforward. The main drawbacks of direct infusion are potential difficulties in resolving isobaric compounds, especially in the absence of tandem MS techniques, and a risk of ion suppression that may lead to decreased sensitivity, especially in the analysis of minor lipid species (3). After sample introduction, ionization technology is necessary for transfer of lipid molecular species from liquid into gas-phase ions to be separated by a mass analyzer operating under vacuum. There are several types of ion sources for creating charged species. The choice of ionization technique depends on the type of analyte, sample preparation, separation technique, and compatibility with the available mass analyzer. ESI is a soft ionization technique commonly used for complex lipid profiling and produces intact or pseudomolecular gas-phase ions from molecules in solution, including protonated species [M+H]+ , alkali metal ion adducts such as [M+Li]+ and [M+Na]+ , and ammonium adducts [M+NH4 ]+ in positive-ion mode, or deprotonated species [M-H]– , acetate adducts [M+CH3 COO]– , chlorine adducts [M+Cl]– , and bromine adducts [M+Br]– in negativeion mode, depending on the infusion solvent system used. In full scan mode, the MS provides information on the mass-tocharge ratios (m/z) of the ions formed. Figure 16.2a shows full scan MS spectra of phosphatidylcholine (PC), lysophosphatidylcholine (LPC), phosphatidylethanolamine (PE), and lysophosphatidylethanolamine (LPE) standard mixtures detected as [M+H]+ and [M+Na]+ species. The identities of the lipid species shown in Fig. 16.2a are given in Table 16.1. The ions produced from the full scan mode can be selected for collisioninduced dissociation (CID), with most lipid classes producing one or more characteristic fragment ions. These characteristic fragments can be used to determine the presence of individual molecular species within each class through PIS or constant NLS after direct infusion (1). A number of mass analyzers are available for lipidomics, including quadrupole, ion trap, time of flight, ion cyclotron resonance, and sector instruments, all separate charged species according to their m/z ratio. The most widely used instrument

262

Isaac

A

B

m/z 782.6 [M+H]+ 18:2/18:2 PC ESI-MS/MS

m/z 184.1 [C5H15O4PN]+

Fig. 16.2. (continued)

Electrospray Ionization Tandem Mass Spectrometry

O

O

C H3C(H2C)15

263

O

P

O

O HO

H3C(H2C)15

O

NH3

H

O

+

m/z 720.6 [M+H] 17:0/17:0 PE ESI-MS/MS

O O

H3C(H2C)15

O H

O

(CH2)15CH3

NL 141

m/z 579.6

14:0/14:0 PC

18:3/18:3 PC

D 17:0/17:0 PC

18:1 LPC

14:0 LPC 18:0 LPC 18:0/18:0 PC

26:0 LPC

23:0/23:0 PC

Fig. 16.2. Representative MS spectra in positive-ion mode of the PC, LPC, PE, and LPE standard mixtures. (a) Full scan MS showing both [M+H]+ and [M+Na]+ species. (b) Product ion scan of m/z 782.6 corresponding to the [M+H]+ ion for 18:2/18:2 PC molecular species. Abundant product ion corresponding to the phosphocholine head group is shown at m/z 184.1 (c) Product ion scan of m/z 720.6 corresponding to the [M+H]+ ion for 17:0/17:0 PE molecular species. Abundant product ion corresponding to the neutral loss of the head group ethanolamine (141) is shown at m/z 579.6. (d) Precursor ion scan of m/z 184.1 gives more simplified spectra showing only the [M+H]+ ions that produced 184.1 upon CID. The identity of the lipid species is given in Table 16.1.

264

Isaac

Table 16.1 The identity of lipid standard mixtures from Fig. 16.2a Lipid standard

[M+H]+

[M+Na]+

14:0 LPC

468.3

490.3

15:0 LPCa

482.3

504.3

16:0 LPC

496.2

518.2

17:0 LPC

510.1

532.2

18:1 LPC

522.2

544.2

18:0 LPC

524.2

546.2

26:0 LPC

636.3

658.2

14:0/14:0 PC

678.3

700.3

14:0/16:0 PC

706.3

728.3

14:0/18:0 PC

734.3

756.4

17:0/17:0 PC

762.4

784.4

18:3/18:3 PC

778.4

800.2

18:2/18:2 PC

782.2

804.2

18:1/18:1 PC

786.3

808.4

18:0/18:0 PC

790.3

812.3

23:0/23:0 PC

930.5

992.4

14:0 LPE

426.2

448.2

18:0 LPEa

482.3

504.3

16:0/16:0 PE

692.2

714.2

17:0/17:0 PE

708.2

730.2

18:0/18:0 PE

748.4

770.4

a Isobaric lipid species.

for shotgun lipidomics is the triple quadrupole (Q1-q2-Q3), which is a tandem MS instrument capable of PIS and NLS (Fig. 16.3). Triple quadrupole instruments are currently available from ABI Sciex (Foster, City CA), Thermo Scientific (San Jose, CA), Waters Corporation (Milford, MA), and Agilent Technologies (Santa Clara, CA). In a triple quadrupole instrument, the middle, field-free quadrupole focuses and transmits almost all ions and may be used as a collision cell for CID. The fragmentation is performed by collision of a selected ion with nitrogen or an inert gas such as helium or argon. As a result of the collision, the internal energy of the ion may be increased by conversion of kinetic energy into internal energy (4, 5). The choice of collision gas, its pressure, and the collision energy affect the degree of fragmentation. In addition to structural information, tandem MS provides higher sensitivity, specificity, and reduced chemical background by selecting a molecular ion of interest from a number of ions present in the mass spectrum. Selection of scan modes specific for

Electrospray Ionization Tandem Mass Spectrometry Q1

q2

265

Q3

Detector

Scanning

CID

Select

Fig. 16.3. Schematic representation of the precursor ion scan in a triple quadrupole mass spectrometer. The quadrupoles Q1 and Q3 are operated as mass analyzers, whereas q2 operates as a collision cell.

the different lipid classes, as described below, allows resolution of lipid molecular species in a complex mixture. Product ion scans, PIS, and constant NLS are the three most common tandem MS scan modes used during shotgun lipidomics. The product ion scan is the most common type of scan performed by tandem MS instruments. During a product ion scan, an ion of interest produced by ESI with a given m/z value is selected in the first mass analyzer (Q1). The selected ion is passed into the collision cell (q2) and fragmented by collision with gas molecules. The product ions are then analyzed with the second mass analyzer (Q3). Fragment ions arising from the molecular ion are detected and recorded (Fig. 16.2b, c). The method is valuable for characterization of unknown lipid and other metabolite structures and has also been used to confirm the identity of known lipid molecular species (6–8). The PIS (Fig. 16.3) is based on independent operation of two mass analyzers positioned inside a tandem MS on each side of the collision cell. In the PIS mode, the first mass analyzer (Q1) of a triple quadrupole MS is scanning, and the second analyzer (Q3) is set to transmit a constant m/z ion. Ions from the first mass analyzer are recorded only when they produce, in the collision cell, a fragment of the specific m/z at which the second analyzer has been set. A PIS can selectively detect precursor ions corresponding to the molecular species in a lipid class, because those species produce the same characteristic head group-derived fragment ion upon CID. For example, a product ion scan of the [M+H]+ species of PC, LPC, and sphingomyelin (SM) yields an abundant characteristic fragment ion at m/z 184.07 Da corresponding to the protonated phosphocholine head group, [C5 H15 O4 PN]+ . A product ion scan of m/z 782.6 corresponding to the [M+H]+ ion for 18:2/18:2 PC molecular species is shown in Fig. 16.2b (lipids are generally represented as X:a/Y:b, where “X” and “Y” represent the number of carbons in the fatty acid chains and “a” and “b” represent the number of double bonds in the fatty acid chains). As shown in Figs. 16.2d and 16.4a, PC, LPC, and SM molecular species can be selectively analyzed as PIS of 184.07 Da

266

Isaac 38:4 PC A

18:0 LPC 34:2 PC

14:0 LPC I.S.

40:6 PC 14:0/14:0 PC I.S.

20:4 LPC

23:0/23:0 PC I.S. 22:6 LPC

38:4 PE

B

42:8 PE

36:2 PE

40:6 PE

34:1 PE

Fig. 16.4. (a) Precursor ion scan of m/z 184 from a rat plasma extract showing PC, LPC, and SM protonated molecular species. Internal standards (IS; 1 μM) of 14:0 LPC (m/z 468.3), 17:0 LPC (m/z 510.3), 14:0/14:0 PC (m/z 678.6), and 23:0/23:0 PC (m/z 930.7) were added to the plasma lipid extract. (b) Neutral loss scan of 141 from the bovine liver total lipid extract (0.25 μg/mL) showing PE molecular species.

in the presence of other lipid classes. Likewise, a PIS can detect a group of complex lipids containing a common fatty acid, because those species can produce the same fatty acyl fragment upon CID (9, 10). Table 16.2 shows the commonly used PIS utilizing characteristic fragments of specific lipid classes.

– + +

Sulfatides

Monogalactosyl diacylglycerols

Digalactosyl diacylglycerols

PIS m/z 184.07 NLS 141.02 PIS m/z 153.0 NLS 189.04 NLS 87.03 NLS 185.01 PIS m/z 153.0 NLS 115.0 PIS m/z 241.01 NLS 277.06 PIS m/z 320.98 PIS m/z 400.94 PIS m/z 480.91 NLS FA+NH3 NLS FA+NH3

[M+H]+ [M+H]+ [M-H]– [M+NH4 ]+ [M-H]– [M+H]+ [M-H]– [M+NH4 ]+ [M-H]– [M+NH4 ]+ [M-H]– [M-H]– [M-H]– [M+NH4 ]+ [M+NH4 ]+ [M+NH4 ]+

PIS m/z 264.27 PIS m/z 96.96 NLS 179.08 NLS 341.13

[M+H]+ [M-H]– [M+NH4 ]+ [M+NH4 ]+

NLS FA+NH3 PIS m/z 369.35

MS/MS mode

Major adduct ion

C12 H23 O10 N

C6 H13 O5 N

[C6 H12 O14 P3 ]– [C6 H13 O17 P4 ]– C16 H35 O2 Na C18 H37 O2 Na C20 H35 O2 Na [C27 H45 ]+ [C18 H34 N]+ [HSO4 ]–

[C6 H11 O11 P2 ]–

[C6 H10 O8 P]– C6 H16 O9 PN

[C3 H6 O5 P]– H6 O4 PN

C3 H5 NO2 P C3 H8 O6 PN

[C3 H6 O5 P]– C3 H12 O6 PN

C2 H8 O4 PN

[C5 H15 O4 PN]+

PIS or NL chemical formula

(18)

(18)

(6, 17)

(16)

(15)

(11)

(11)

(14)

(14)

(14)

(1)

(1)

(1)

(1)

(1)

(1)

References

a These scan modes represent neutral loss mass as fatty acid chain plus ammonia (RCOOH + NH ). As an example, the chemical formula of the neutral loss scan of (16:0+NH ), 3 3 (18:1+NH3 ), and (20:4+NH3 ) is provided for diacylglycerols, triacylglycerols, and cholesteryl esters, respectively.

+

Ceramides with d18:1 long-chain base

–

Phosphatidylinositol bisphosphates

+

–

Phosphatidylinositol monophosphates

Cholesteryl esters

– +

Phosphatidylinositols

+

– +

Phosphatidic acids

Triacylglycerols

– +

Phosphatidylserines

–

– +

Phosphatidylglycerols

+

+

Phosphatidylethanolamine

Phosphatidylinositol trisphosphates

+

Phosphatidylcholines Sphingomyelins

Diacylglycerols

Polarity

Lipid class analyzed

Table 16.2 Summary of the commonly used specific PIS and constant NLS when using the infusion solvent system chloroform/methanol/10.5 mM ammonium acetate (30/66.5/3.5)

Electrospray Ionization Tandem Mass Spectrometry 267

268

Isaac

During a constant neutral loss scan (NLS), both Q1 and Q3 are scanned in parallel with a constant specific mass difference between the two analyzers. Like the PIS, it is selective for a particular functional group and can also be performed directly by a triple quadrupole instrument. A constant NLS is used when the characteristic common fragment for a group of related compounds is uncharged. For example, a product ion scan of PE and LPE species produces dominant fragment ions due to head group neutral losses of 141.02 Da (C2 H8 O4 PN). A product ion scan of m/z 720.6 corresponding to the [M+H]+ ion for 17:0/17:0 PE molecular species is shown in Fig. 16.2c. As shown in Fig. 16.4b, PE and LPE species can be selectively analyzed as constant NLS of 141.02 Da in the presence of other lipid classes. In this example of constant NLS, when Q1 scans at m/z 744 Da, Q3 transmits m/z 603 Da. As Q1 moves to m/z 745 Da, Q3 moves in conjunction with m/z 604 Da, and as Q1 moves to m/z 746 Da, Q3 moves to m/z 605 Da, etc. A signal is obtained at the detector only when the ion is transmitted by Q1 fragments with the loss of mass of interest. Like PIS, NLS has been applied to identify phospholipid molecular species within a class (1) or a group of complex lipids containing a common neutral loss of the fatty acid chain (11). Table 16.2 shows the commonly used specific constant NLS utilizing characteristic fragments of specific polar lipid classes.

2. Materials 2.1. Chemicals

1. HPLC-grade methanol and chloroform (see Note 1). Prepare the lipid extraction solvent by combining two parts chloroform with one part methanol. 2. MS-grade ammonium acetate (Sigma-Aldrich, St. Louis, MO). 3. Lipid standards: PC (14:0/14:0, 14:0/16:0, 14:0/18:0, 17:0/17:0, 18:0/18:0, 18:1/18:1, 18:2/18:2, 18:3/18:3, 23:0/23:0), LPC (14:0, 15:0, 16:0, 17:0, 18:0, 18:1, 26:0), PE (16:0/16:0, 17:0/17:0, 18:0/18:0), and LPE (14:0, 18:0) (Avanti Polar Lipids, Alabaster, AL, USA). 4. Total bovine liver lipid extract (Avanti Polar Lipids). 5. Rat plasma (Bioreclamation, Inc., Westbury, NY). 6. Dry ice. 7. Liquid nitrogen. 8. Nitrogen gas source.

Electrospray Ionization Tandem Mass Spectrometry

2.2. Equipment

269

1. TSQ Vantage triple quadrupole MS (Thermo Scientific, San Jose, CA). 2. Barnstead Nanopure Infinity water system (Thermo Scientific, Waltham, MA). 3. Syringe pump (Harvard Apparatus, Holliston, MA). 4. Syringes (Hamilton, Reno, Nevada) of various volumes (see Note 2). 5. Glass vials and tubes for extraction, agitation, and centrifugation with Teflon-covered liner screw caps or siliconized, low-retention, individually wrapped and sterilized Eppendorf tubes (Bio Plas, Inc., San Rafael, CA) (see Note 3). 6. Centrifuge (Eppendorf 5424, Westbury, NY). 7. 8-mm crimp-top, conical-bottom, 700-μL glass vials with 8-mm natural caps and Teflon septa (MicroLiter Analytical Supplies, Inc., Suwanee, GA). 8. 2-mL glass vials with Teflon-lined caps (Thermo Scientific, Rockwood, TN). 9. 8-mL glass vials with Teflon-lined caps (Thermo Scientific). 10. 100-mL volumetric flask.

3. Methods 3.1. Infusion Solvent Preparation

1. Dissolve 2.312 g ammonium acetate in a 100-mL volumetric flask and dilute to mark with distilled water to make a final concentration of 300 mM. 2. Mix chloroform, methanol, and 300 mM ammonium acetate aqueous solution in a volume ratio of 30/66.5/3.5, respectively. The concentration of ammonium acetate in the final infusion solvent is 10.5 mM.

3.2. Lipid Standard Preparation

1. For lipid standards received in chloroform at 2.5 mg/mL, add 40 μL of each standard to 2-mL glass vials.

3.2.1. Liquid Standards

2. Evaporate the chloroform under a constant nitrogen gas stream and then add 1 mL of the infusion solvent to prepare 0.1 mg/mL stock solutions. 3. Prepare working standard solutions by diluting the stock solutions to 1 nmol/mL with the infusion solvent based on the molecular weights of the individual lipid standards. 4. Store all stock and working standard solutions in Teflonlined, screw cap glass vials at –80 and –20◦ C, respectively, until analysis.

270

Isaac

3.2.2. Powder Standards

1. For lipid standards received in powder form, prepare 0.1 mg/mL stock solutions by weighing the standards (0.50 ± 0.01 mg) and dissolving in 5 mL of infusion solvent in 8-mL Teflon-lined, screw cap glass vials. 2. Prepare working standard solutions by diluting the stock solutions to 1 nmol/mL with the infusion solvent based on the molecular weights of the individual lipid standards. 3. Store all stock and working standard solutions in Teflonlined, screw cap glass vials at –80 and –20◦ C, respectively, until analysis.

3.3. Total Bovine Liver Lipid Extract Preparation

1. The total bovine liver lipid extract was received at a concentration of 2.5 mg/mL in chloroform. 2. Add 40 μL of the bovine liver lipid extract to a 2-mL glass vial. 3. Evaporate the chloroform under a constant nitrogen gas stream and then add 1 mL of infusion solvent to prepare a 0.1 mg/mL stock solution. 4. Prepare a working bovine liver lipid extract of 0.25 μg/mL by diluting 5 μL of the above stock solution with 1995 μL of the infusion solvent in a 2-mL glass vial. 5. Store all stock and working standard solutions in Teflonlined, screw cap glass vials at –80 and –20◦ C, respectively, until analysis.

3.4. Lipid Extraction 3.4.1. Plasma Total Lipid Extraction

1. Add 25 μL of rat plasma to a sterile siliconized 0.6-mL Eppendorf tube. 2. Add 200 μL of cold (–20◦ C) chloroform/methanol (2:1). 3. Vortex the sample for 30 s at room temperature. 4. Allow the sample to stand for 5 min at room temperature followed by vortexing for 30 s. 5. Centrifuge the sample at 13,000×g for 5 min at room temperature to separate precipitated protein. 6. Pierce the protein disc with a pipette tip, collect the lower organic phase, and transfer it to a new 700-μL glass vial (see Note 4). 7. Discard the protein interface and upper aqueous phase. 8. Evaporate the organic phase under vacuum or constant nitrogen gas stream and store at –80◦ C. 9. Immediately prior to analysis, dissolve the evaporated lipid extract in a final volume 20 times that of the original volume of plasma using the infusion solvent.

Electrospray Ionization Tandem Mass Spectrometry

271

10. Add synthetic lipid internal standards 14:0 LPC, 17:0 LPC, 14:0/14:0 PC, and 23:0/23:0 PC to make a final concentration of 1 μM. 11. Centrifuge the sample at 350×g for 5 min to remove any particulates prior to MS analysis (see Note 5). 3.4.2. Tissue Lipid Extraction

1. When possible, freeze fresh tissue samples (brain, liver, heart, etc) in liquid nitrogen. 2. Grind the frozen tissue in a dry ice-chilled mortar and pestle. 3. Transfer the ground tissue to glass tubes with Teflon-lined screw caps. 4. Homogenize the tissue with chloroform/methanol (2:1) to a final volume 20 times the weight of the original tissue sample (e.g., 0.5 g in 10 mL of solvent). 5. Agitate the sample for 30 min in an orbital shaker at room temperature. 6. Centrifuge the homogenate at 9000×g for 10 min, recover the supernatant, and transfer it to a new glass tube with glass pipette. 7. Wash the supernatant with 0.2 volumes (e.g., 2 mL for 10 mL) of water. 8. Vortex the sample for 30 s and then centrifuge at 1000×g for 5 min to separate the two phases. 9. Remove and discard the upper phase and then rinse the interface one or two times with methanol/water (1:1), without mixing the whole preparation. 10. After centrifugation at 350×g for 3 min, remove and discard the upper phase. 11. Evaporate the lower organic phase containing lipids under vacuum or a constant nitrogen gas stream, then store at –80◦ C. 12. Immediately prior to analysis, reconstitute the lipid extracts in infusion solvent to any desired final volume depending on the weight of the original tissue sample and the amount of material that would be infused to the mass spectrometer. 13. Centrifuge the sample at 350×g for 5 min to remove any particulates prior to MS analysis.

3.5. Mass Spectrometric Analysis

1. ESI-MS/MS analyses are performed on a tandem quadrupole mass spectrometer equipped with an electrospray ion source and operated in both positive- and negative-ion mode (see Note 6). 2. Load sample solutions in a gastight 250-μL Hamilton Syringe and deliver them to the mass spectrometer by a

272

Isaac

syringe pump through a 100-cm capillary transfer line of 150 μm internal diameter and 360 μm outer diameter at a flow rate of 0.3–0.5 μL/min. The construction of the transfer line and the flow rate may differ depending on the ion source and mass spectrometer being used (see Note 7). 3. Flush the capillary with infusion solvent after each sample infusion to prevent sample to sample cross-contamination (see Note 8). 4. For full scan MS, set the ion spray voltage to 2.2 kV and the capillary temperature to 200◦ C. Collect a 1 min period of signal in profile mode. Figure 16.2a shows [M+H]+ and [M+Na]+ full scan MS spectra of PC, LPC, PE, and LPE standard mixtures. The identity of the lipid species is given in Table 16.1. The ions produced from the full scan mode can be selected for product ion scan. 5. For tandem MS, set the collision gas pressure to 1.5 mTorr. Use collision energies of 40 eV for LPC, PC, SM and 35 eV for LPE and PE (see Note 9). Reference collision energies for the different lipid classes are provided by Han et al. (2). Collect 1–6 min of signal in profile mode. 6. A representative product ion scan of protonated 18:2/18:2 PC (m/z 782.6) and protonated 17:0/17:0 PE (m/z 720.6) from the standard mixture is shown in Fig. 16.2b, c, respectively. 7. Representative PISs of 184.1 are shown in Figs. 16.2d and 16.4a from the standard mixture and rat plasma lipid extract, respectively, showing the main expected lipid species of PC, LPC, and SM. 8. A representative constant NLS of 141 is shown in Fig. 16.4b from the bovine liver lipid extract in the presence of other complex lipid classes. 9. Isobaric species (lipid species with the same m/z value) from different lipid classes can be resolved by PIS and NLS. In contrast, it is impossible to identify isobaric species from full scan MS. For example, the standard lipid species 15:0 LPC and 18:0 LPE have the same chemical formula C23 H48 NO7 P and hence produce the same m/z at 482.32 [M+H]+ and 504.31 [M+Na]+ , respectively (Fig. 16.2a and Table 16.1). PIS of 184.1 only produces the parent ion 15:0 LPC, and NLS 141 produces only the parent ion 18:0 LPE. 10. Data processing and quantification is performed using the open source lipid mass spectrum analysis (LIMSA) software. LIMSA is a Microsoft Excel add-on that can perform peak finding, integration, assigning, isotope correction,

Electrospray Ionization Tandem Mass Spectrometry

273

and quantification with internal standards in an imported MS spectrum (12, 13). It is available as a free download at http://www.helsinki.fi/science/lipids/software.html.

4. Notes 1. Chloroform is a carcinogen and necessary precautions should be taken to limit human exposure. Always work in a fume hood. 2. All volume measurements for organic solvents should be carried out using glass syringe needles. Do not use plastic pipette tips for organic solvent volume measurements to avoid the danger of introducing polymers and any source of contamination leaching from the plastic pipette tips. 3. Containers should be resistant to organic solvents. Never use rubber, cork, polyethylene, or Parafilm. 4. Pierce the protein disc carefully and try to exclude the contamination of the lower organic phase by the upper phase containing salt and other water-soluble materials. 5. If the 700-μL glass vial does not fit into the centrifuge or the vacuum concentrator, place it in 1.5-mL Eppendorf tube prior to centrifugation or vacuum concentration. 6. Other triple quadrupole mass spectrometers (such as ABI Sciex API 3000, 4000, 5000, 5500; Waters Quattro Premier XE, ACQUITY TQD, and Agilent 6400 series) capable of performing CID (PIS and constant NLS) and equipped with electrospray ionization can be used. 7. Optimum infusion flow rates should be optimized depending on the ion source and mass spectrometer being used. 8. To save time and facilitate the flushing, push the syringe by hand. 9. The fragmentation of each lipid molecular species depends on the applied collision energy and needs to be optimized depending on each lipid class and type of mass spectrometer used.

Acknowledgments The author would like to thank Dr. Ruth Welti from the Kansas Lipidomics Research Center for her review of and constructive

274

Isaac

comments on the manuscript. Portions of this work were supported by NIH grant DK071283 and performed in the EMSL, the Environmental Molecular Sciences Laboratory, a national scientific user facility located at Pacific Northwest National Laboratory (PNNL) and supported by the US Department of Energy (DOE) Office of Biological and Environmental Research. PNNL is operated by Battelle for the DOE under Contract No. DEAC06-76RLO-1830.

References 1. Brügger, B., Erben, G., Sandhoff, R., Wieland, F. T., Lehmann, W. D. (1997) Quantitative analysis of biological membrane lipids at the low picomole level by nanoelectrospray ionization tandem mass spectrometry. Proc Natl Acad Sci USA 94, 2339–2344. 2. Han, X., Gross, R. W. (2005) Shotgun lipidomics: electrospray ionization mass spectrometric analysis and quantitation of cellular lipidomes directly from crude extracts of biological samples. Mass Spectrom Rev 24, 367–412. 3. Isaac, G., Jeannotte, R., Esch, S. W., Welti, R. (2007), in (Setlow, J. K., ed.), New Mass Spectrometry Based Strategies for Lipids. vol. 28. New York, NY, USA. 4. Hoffmann, E. D. (1996) Tandem mass spectrometry: a primer. J Mass Spectrom 31, 129–137. 5. Kuksis, A., Myher, J. J. (1995) Application of tandem mass spectrometry for the analysis of long-chain carboxylic acids. J Chromatogr B 671, 35–70. 6. Hsu, F. F., Bohrer, A., Turk, J. (1998) Electrospray ionization tandem mass spectrometric analysis of sulfatide. Determination of fragmentation patterns and characterization of molecular species expressed in brain and in pancreatic islets. Biochim Biophys Acta 1392, 202–216. 7. Hsu, F. F., Turk, J. (2000) Structural determination of sphingomyelin by tandem mass spectrometry with electrospray ionization. J Am Soc Mass Spectrom 11, 437–449. 8. Moreau, R. A., Doehlert, D. C., Welti, R., Isaac, G., Roth, M., Tamura, P., Nun˜ez, A. (2008) The identification of mono-, di-, tri-, and tetragalactosyldiacylglycerols and their natural estolides in oat kernels. Lipids 43, 533–548. 9. Ekroos, K., Chernushevich, I. V., Simons, K., Shevchenko, A. (2002) Quantitative profiling of phospholipids by multiple precursor

10.

11.

12.

13.

14.

15.

16.

ion scanning on a hybrid quadrupole timeof-flight mass spectrometer. Anal Chem 74, 941–949. Han, X., Yang, J., Chang, H., Ye, H., Gross, R. W. (2004) Toward fingerprinting cellular lipidomes directly from biological samples by two-dimensional electrospray ionization mass spectrometry. Anal Biochem 330, 317–331. Murphy, R. C., James, P. F., McAnoy, A. M., Krank, J., Duchoslav, E., Barkley, R. M. (2007) Detection of the abundance of diacylglycerol and triacylglycerol molecular species in cells using neutral loss mass spectrometry. Anal Biochem 366, 59–70. Haimi, P., Uphoff, A., Hermansson, M., Somerharju, P. (2006) Software tools for analysis of mass spectrometric lipidome data. Anal Chem 78, 8327–8331. Haimi, P., Chaithanya, K., Kainu, V., Hermansson, M., Somerharju, P. (2009), in (Armstrong, D., ed.), InstrumentIndependent Software Tools for the Analysis of MS-MS and LC-MS Lipidomics Data. vol. 580. New York, NY, USA. Wenk, M. R., Lucast, L., Di Paolo, G., Romanelli, A. J., Suchy, S. F., Nussbaum, R. L., Cline, G. W., Shulman, G. I., McMurray, W., De Camilli, P. (2003) Phosphoinositide profiling in complex lipid mixtures using electrospray ionization mass spectrometry. Nat Biotechnol 21, 813–817. Liebisch, G., Binder, M., Schifferer, R., Langmann, T., Schulz, B., Schmitz, G. (2006) High throughput quantification of cholesterol and cholesteryl ester by electrospray ionization tandem mass spectrometry (ESI-MS/MS). Biochim Biophys Acta 1761, 121–128. Merrille, A. H. J., Sullards, M. C., Allegood, J. C., Kelly, S., Wang, E. (2005) Sphingolipidomics: high-throughput, structure-specific, and quantitative analy-

Electrospray Ionization Tandem Mass Spectrometry sis of sphingolipids by liquid chromatography tandem mass spectrometry. Methods 36, 207–224. 17. Isaac, G., Pernber, Z., Gieselmann, V., Hansson, E., Bergquist, J., Mansson, J. E. (2006) Sulfatide with short fatty acid dom-

275

inates in astrocytes and neurons. FEBS J 273, 1782–1790. 18. Welti, R., Wang, X., Williams, T. D. (2003) Electrospray ionization tandem mass spectrometry scan modes for plant chloroplast lipids. Anal Biochem 314, 149–152.

Chapter 17 Processing and Analysis of GC/LC-MS-Based Metabolomics Data Elizabeth Want and Perrine Masson Abstract Data processing forms a crucial step in metabolomics studies, impacting upon data output quality, analysis potential and subsequent biological interpretation. This chapter provides an overview of data processing and analysis of GC-MS- and LC-MS-based metabolomics data. Data preprocessing steps are described, including the different software available for dealing with such complex datasets. Multivariate techniques for the subsequent analysis of metabolomics data, including principal components analysis (PCA) and partial least squares discriminant analysis (PLS-DA), are described with illustrations. Steps for the identification of potential biomarkers and the use of metabolite databases are also outlined. Key words: GC-MS, LC-MS, metabolomics, metabolite, alignment, multivariate, PCA, PLS-DA.

1. Introduction Mass spectrometry (MS) technologies, often coupled with metabolite separation via liquid chromatography (LC) or gas chromatography (GC), offer high sensitivity and reproducibility for metabolomics studies, together with quantitative metabolite analyses. Data generated through metabolomics studies can yield important metabolic insights into disease onset and progression, mechanisms of drug toxicity, or growth and ageing (1–3). Advances in sample separation techniques and MS instrumentation in metabolomics studies have resulted in the generation of large, complex datasets, which in turn has led to a demand for improved data analysis approaches, including preprocessing and advanced chemometric approaches (4). However, this complexity T.O. Metz (ed.), Metabolic Profiling, Methods in Molecular Biology 708, DOI 10.1007/978-1-61737-985-7_17, © Springer Science+Business Media, LLC 2011

277

278

Want and Masson

means that the data must undergo stringent processing steps in order to afford meaningful interpretation. The purpose of this chapter is to provide an overview of both data preprocessing and multivariate analysis in MS-based metabolomics studies. The use of principal components analysis (PCA) and partial least squares discriminant analysis (PLS-DA) in metabolomics studies is illustrated using a biological dataset. Metabolite characterization is also discussed, through the application of high mass accuracy measurements, fragmentation and database consultation.

2. Materials Key to data processing and analysis is the ability to distinguish genuine biological variation and metabolic changes from analytical interferences. It is therefore essential that data processing is sufficiently robust to enable the researcher to analyze and interpret metabolomics data properly. Most MS manufacturers

2.1. Data Preprocessing Software

Table 17.1 Commercial software for processing of MS metabolomics data Software

Manufacturer

Capabilities

Link

Bluefuse

BlueGnome

MS and NMR data: filtering, feature detection, alignment, multivariate data analysis

http://www.cambridge bluegnome.com/

Markerlynx

Waters

LC-MS data: feature detection, alignment, PCA

http://www.waters.com

MarkerView

Applied Biosystems

LC-MS data: feature detection, alignment, PCA

https://products.applied biosystems.com

MassHunter

Agilent Technologies

LC-MS data: detection/ extraction, alignment

http://www.chem. agilent.com

Metabolic Profiler

Bruker Daltonic

MS and NMR data: data bucketing (retention time, m/z values), PCA

http://www.brukerbiospin.com/ metabolicprofiler.html

Metabolyzer

Metabolon

Automated chromatographic peak alignment tools. Used as part of proprietary data processing system in commercial studies

http://www.metabolon. com/services/ technology.php

metAlign

PlanResearch International

http://www.metalign. wur.nl/UK/

Phenomenome Profiler

Shimadzu/ Phenomenome Discoveries

LC-MS, GC-MS data: filtering, baseline correction, feature detection, alignment MS data: feature detection, alignment, statistical analysis

Processing and Analysis of GC/LC-MS-Based Metabolomics Data

279

now offer software for metabolomics data analysis, with options for feature detection and alignment, as well as visualization and evaluation (Table 17.1). As the metabolomics field is growing, so are the number of available software preprocessing packages, and there is also freeware available for MS metabolomics data processing (Table 17.2). For processing with freeware, data will need to be converted into the appropriate format, e.g. netCDF, mzXML or mzData, which can often be achieved using the manufacturer’s own software. Freeware is available for file conversion (e.g. trapper for conversion of MassHunter (Agilent) raw data (.d directories) into mzXML files (http://tools.proteome center.org/wiki/index.php?title=Software:trapper); and mass wolf for conversion of Masslynx (Waters) (.raw) files into mzXML files (http://tools.proteomecenter.org/wiki/index.php?title= Software:massWolf)). 2.2. Requirements for Multivariate Analyses

All multivariate analyses shown in this chapter were performed using SIMCA-P (11.5) software (Umetrics, Umea, Sweden). Other programmes such as MATLABTM package (Mathworks, Natick, MA; http://www.mathworks.co.uk/) and R (http://www.r-project.org/) also offer similar functions for multivariate data analysis.

3. Methods 3.1. Data Preprocessing

LC-MS and GC-MS data will be stored as (instrument specific) raw data files, the size of which will depend on the instrument, the scan rate employed and other parameters. The complex, multidimensional MS metabolomics datasets need careful treatment, as the data preprocessing steps employed will affect the potential for metabolite identification, as well as subsequent quantification capabilities and biological interpretation. Metabolomics datasets must be extensively preprocessed and converted into organized data matrices prior to multivariate analysis and subsequent visualization of important (discriminatory) metabolites. Key factors in data preprocessing are the following: a. Initial raw data quality b. Choice of preprocessing software c. Preprocessing parameter settings (including parameter optimization (see Note 1)) d. Data normalization method An example of MS metabolomics data preprocessing workflow is shown in Fig. 17.1, which can be divided broadly into

Data type

GC-MS

GC–FID

LC-MS, GC-MS

GC-MS, CE-MS

GC-MS

LC-MS, GC-MS, CE-MS

LC-MS, GC-MS

GC-MS, LC-MS

LC-MS, GC-MS

LC-MS

LC-MS, GC-MS

Software

AMDIS

Chrompare

COMSPARI

MathDAMP

Metabolite Detector

MET-IDEA

MSFACTs

mSPECS

MZMine/MZMINE2

OPEN MS

XCMS/XCMS2

Filtering, feature detection, alignment, visualization, highresolution feature extractor MSn analysis

Editing, normalization, alignment, feature extraction (multiple algorithms)

Filtering, feature detection, alignment, normalization, visualization. MZmine 2 can process unit mass resolution and exact mass resolution data (e.g. FTMS), including fragmentation data (continuum or centroid)

Management and editing of mass spectral libraries and associated information. Capabilities for the automatic management of libraries, plus automatic calculation of formulas and masses

Automated import, alignment and comparison of feature lists or raw chromatograms. Allows rapid visualization and interrogation of data

Chromatogram compression, peak detection, chromatogram deconvolution, alignment, compound integration/ quantification Ion intensity data extraction – outputs list of m/z retention time values

Direct data comparison (no feature detection), binning, baseline subtraction, smoothing, normalization

Visualization capabilities for investigating differences between two runs

Feature list and raw chromatogram comparison, data normalization

Automatically extracts pure component mass spectra from highly complex GC-MS data files. Uses these purified spectra to search a mass spectral library

Capabilities

Table 17.2 Freeware for the processing and management of MS metabolomics data

http://masspec.scripps. edu/xcms/xcms.php

http://open-ms.source forge.net/OpenMS.php

http://mzmine.source forge.net/

http://mspecs.tu-bs.de/

http://www.noble.org/ plantbio/sumner/ msfacts/index.html

http://www.noble.org/ plantbio/sumner/ met-idea.html

http://metabolite detector.tu-bs.de/

http://mathdamp.iab. keio.ac.jp/

http://sourceforge.net/ projects/comspari/

http://www.chrom pare.com/

http://www.amdis.net/ index.html

Link

280 Want and Masson

Processing and Analysis of GC/LC-MS-Based Metabolomics Data

281

Fig. 17.1. Metabolomics data preprocessing workflow, outlining the main steps.

six main steps as detailed below (Sections 3.1.1, 3.1.2, 3.1.3, 3.1.4, 3.1.5 and 3.1.6). 3.1.1. Peak Picking (Detection)

Accurate peak detection and peak matching between samples are crucial to enable the comparison of metabolites between different samples, as well as for precise quantitative metabolite analysis (if required). Therefore, the initial data preprocessing step for both LC-MS and GC-MS data concentrates on distinguishing metabolite features from other peaks and background noise in the sample/chromatogram. This is termed peak picking or peak detection. For the purpose of this chapter, a metabolite feature is defined as a “mass-to-charge ratio/retention time pair” (m/z_RT pair). Depending on the software used, isotopes, fragments, adducts and dimers may be reported for LC-MS data, the latter two particularly where electrospray ionization (ESI) is used. Fragments are also observed in GC-MS data due to the electron ionization (EI) process. There is also the potential for artefact formation during GC-MS sample preparation, through incomplete derivatization, possible analyte conversion, by-product formation or degradation of the final product(s). LC-MS and GC-MS data can be simplified through the detection and grouping of some of these features, as described in Section 3.1.2. Depending on the data being analyzed, additional steps that may be included at this peak picking stage are the following: 1. Baseline correction. Baseline variation across samples in a batch may be due to instrumentation problems, external

282

Want and Masson

environmental sources or the separation process. Baseline drift can result from mobile phase interferences, column temperature changes or contaminant build up on the column. A repeatable drift between chromatographic runs indicates that the problem is most likely mobile phase related. In GC-MS analyses, variability in carrier gas flow rate can cause baseline drift. A baseline correction step can be employed to eliminate such background drift and may aid in the introduction of a threshold to obtain noise reduction (see Note 2). 2. Noise removal/reduction. In LC-MS analyses, solvent impurities or solvent clusters formed in the ESI source may complicate the data, while in GC-MS analyses, peaks known to be artefacts from the analytical process (e.g. siloxanes from the sample vial septum) may be present. Therefore, in order to reduce spectral complexity, a separate noise removal/filtering step may be implemented prior to the peak picking/detection stage (see Note 2). 3. Smoothing. It may be necessary to smooth the data, and some software contains an optional smoothing stage in data processing, such as a moving average filter or a Savitzky – Golay filter. However, smoothing may not be necessary if the data are not noisy or if the input data are in centroid form. Smoothing may affect the final feature intensity output and therefore care is needed when using a smoothing step if quantitation is required. 3.1.2. Deconvolution

Ideally, a single detected peak will equate to a single metabolite feature; however, this is not usually the case, particularly with soft ionization techniques such as ESI, which is used commonly in LC-MS studies. Realistically, metabolite features detected from LC-MS experiments can comprise isotopes, adducts (e.g. Na+ , K+ in ESI+ mode; Cl– in ESI– mode), fragments (e.g. loss of water) and dimers. Fragments will also be observed in GC-MS spectra and, although ultimately useful for metabolite identification, may complicate the metabolomics data. Multiple metabolite features originating from the same molecule may also make quantitative analysis problematic. A further challenge in LC-MS and GC-MS metabolomics studies (where complex mixtures are usually analyzed) is the co-elution of two or more analytes, resulting in overlapping chromatographic peaks with similar retention times and perhaps overlapping isotope patterns. This can make the extraction of pure components and their corresponding mass spectra difficult, in turn hampering unambiguous metabolite identification. Deconvolution approaches are therefore needed to a. assign different ions/metabolite features to the same metabolite and

Processing and Analysis of GC/LC-MS-Based Metabolomics Data

283

b. address the problem of extracting pure component signals in cases of incomplete chromatographic separation Deconvolution is used widely in GC-MS data preprocessing, and specific software exists, e.g. the Automated Mass Spectral Deconvolution and Identification System (AMDIS) software available from The National Institute of Standards and Technology (NIST). Deconvolution algorithms often utilize the fact that different fragments from the same molecule have the same retention time. Algorithms based on a model peak approach may be used, where ions with similar peak shapes are extracted. These algorithms may also assume high correlation between multiple chromatographic profiles within a batch, considering them subject to the same biological and analytical variation. However, in a particular biological system, several metabolites may be subject to the same regulatory mechanisms, meaning that their levels will be highly correlated. This may make separating biological variability and analytical variation difficult, which is needed for successful deconvolution. It is important to consider noise at this point (Section 3.1.1) to enable the detection of small signals. Isotope pattern detection may be included already in the peak detection step of an algorithm and can be performed by 1. fitting a generic isotope pattern model to the raw signal/ pattern matching with raw data and 2. grouping detected features with suitable m/z differences However, this is not a feature in all software and so further processing with different software or perhaps using additional inhouse scripts may be required. For example, a GC-MS add-on for XCMS (5), “Flagme,” detects fragments and uses the fragment intensities to calculate an overall integrated intensity for the parent ion. 3.1.3. Alignment (Chromatographic or Peak)

A typical metabolomics study may involve the comparison of two or more sample classes such as healthy vs. disease populations or control vs. dosed animals, with each group comprising tens or hundreds of individual samples. Retention time drifts between samples must be accounted for, as detected metabolite features must be aligned correctly across all samples in a batch in order to match the corresponding features between the multiple GC/LCMS runs. During LC separation, retention time drifts can arise due to (i) variations in temperature, pressure and mobile phase (composition and flow rates), (ii) changes in the column stationary phase (saturation/degradation), or (iii) sample matrix effects (due to variations in sample composition, e.g. solvent and salts) (6). Importantly, the use of ultra-performance (UP) LC and other sub 2 μm particle column technologies has meant that retention time shifts and analyte co-elution are reduced. Experimental sources of variability in GC-MS analyses are similar and include

284

Want and Masson

(i) column ageing, (ii) temperature changes and (iii) changes in the stationary phase and experimental conditions. Therefore, retention time correction issues are similar in both GC-MS and LC-MS metabolomics studies. Deviations may also be observed in the m/z dimension, but these are generally smaller, and with the advent of lock mass calibration correction methods, they are smaller still. The overall change in RT and m/z is called warp. Retention time shifts can be corrected through 1. chromatographic profile alignment prior to peak detection (profile or chromatogram alignment) and 2. matching signal peaks/metabolite features after peak detection (peak alignment). Automated retention time correction is particularly important in non-targeted profiling studies, due to data size and complexity. The retention time correction approach used will depend on which software is employed and thus may rely on solely the retention time data or may incorporate m/z information. 3.1.4. Peak Integration

Detected peaks must then be integrated to enable the comparison of relative metabolite abundances, in order to be able to elucidate differences between the sample groups. Both peak height and peak area can be used and may be software dependent. Peak area is often used as it is more robust than peak height alone. Several peak integration methods have been developed and are often coupled to the peak detection process.

3.1.5. Normalization

Normalization procedures enable more accurate metabolite feature matching to be achieved and can permit quantitation between samples. Normalization entails the removal of unwanted changes in ion intensities between sample runs, thus reducing the systematic error. These changes may be due to technical or analytical variation, or alterations in sample or column composition, and are dependent on the sample type, batch size and instrumentation. For example, ion intensities may change due to source contamination, column degradation or sample degradation. Quality control procedures for monitoring such changes are detailed in Section 3.2.3.3. Metabolomics data normalization should therefore enable biological variation to be observed more clearly, but in reality this can be challenging due to the chemical diversity of metabolites in biological samples, resulting in varying extraction recoveries or differing ionization responses during MS analysis. Normalization can be performed through the following: 1. The addition of one or more internal standards to the sample (prior to extraction) or external standards (after extraction). Limitations include ion suppression, challenges in the choice of standards and which particular standard to use to normalize each endogenous metabolite for an untargeted

Processing and Analysis of GC/LC-MS-Based Metabolomics Data

285

metabolomics study. Further, this approach is not sufficiently versatile when there are very pronounced deviations from the ideal, i.e. severe cases of baseline drift and changes in peak shapes. 2. The utilization of statistical models to obtain optimal scaling factors for each sample based on the whole dataset, e.g. normalization by intensity unit norm or median (7, 8) or the maximum likelihood method (9). 3.1.6. Output of Results

Results must be output in a format suitable for subsequent multivariate statistical analysis and visualization of features that discriminate sample groups. Typically, this is in the form of a data matrix with the metabolite features (m/z_RT pairs) in rows and the samples in columns, with corresponding feature intensity reported (as height or area) for each detected feature (Section 3.2.1).

3.2. Multivariate Analysis

The output table obtained during the data preprocessing step is a large matrix of dimensions N (samples) ∗ K (variables), with N being mostly small compared to K (typically tens to hundreds of samples vs. thousands of metabolite features (variables)). These features are often highly correlated and information is likely to be found in combinations of features rather than in individual ones. Analysing the variables one by one using univariate statistics may lead to important information being overlooked. Moreover, spurious results can be produced if differences between groups are tested using all variables independently (there is an increased risk of false positives due to the high number of variables). Therefore, as metabolomics data are multivariate, multivariate techniques should be used to analyze the multiple variables simultaneously. This increases the power of detecting meaningful information in the data and facilitates the identification of deviating samples. However, certain characteristics of metabolomics data, such as the high number of variables compared to the number of samples, high correlation between variables, and noisy, incomplete data, are problematic for classical statistics methods, e.g. multiple linear regression and linear discriminant analysis. Appropriate multivariate statistics techniques are needed to handle these characteristics and analyze the multiple variables simultaneously in order to extract comprehensive information from the large data table and to subsequently visualize and interpret it. These techniques can be used to overview a data table (e.g. to identify relationships between observations, relationships between variables, deviating observations, trends in the data), to model differences between groups of observations and classify unknown observations, and to link two blocks of variables. They are usually divided into unsupervised and supervised methods. Unsupervised methods are used to analyze the data without any a priori

3.2.1. Introduction

286

Want and Masson

information concerning sample groups/classes and are therefore a good way of revealing groups or clustering in the data. Typical unsupervised techniques applied to metabolomics data include principal components analysis (PCA) (10–12), independent component analysis (ICA), multi-dimensional scaling (MDS), hierarchical clustering, k-means clustering and self-organising maps (SOMs). Conversely, in supervised analysis, information related to the dataset (e.g. sample classes, biological/experimental parameters, metadata such as clinical chemistry) is used to construct a predictive model from a training dataset, this model being then used to predict class or parameter value for observations not included in the training set. Typical supervised techniques applied to metabolomics data include partial least squares (PLS) regression (13), orthogonal PLS (OPLS) (14–16), partial least squares discriminant analysis (PLS-DA), random forests, support vector machines (SVMs) and k-nearest neighbour algorithm (kNN). The choice of some techniques over others depends on the objective sought in the data analysis and on the type of dataset. This chapter does not aim to review all these techniques; it focuses on how to practically apply some of them to MS-based metabolomics datasets. The working examples used in this chapter are XCMS output datasets obtained from UPLC-MS analysis of rat serum samples from toxicological metabolomics studies. The study objectives are to investigate possible differences between animal groups in terms of metabolic profiles and discover biomarkers of toxic effects. Two projection methods, PCA and PLS-DA, are frequently used for this purpose and so are illustrated in this chapter. These techniques can handle incomplete and noisy data containing more variables than samples, as well as highly correlated variables. 3.2.2. Data Scaling

After data preprocessing but before performing PCA or PLSDA, MS data are usually centred, i.e. variable averages are subtracted from the data to centre it on 0. This is essential to remove the influence of the variable averages and focus on the variation among the data. The question that arises next is: Should the data be scaled? If scaling is not performed, focus is on metabolite features with high intensities. This is because variables with high intensities are likely to show higher variance than variables with low intensities, and PCA projects onto the direction of maximum variance. As the range of metabolite classes and concentrations in biofluids and tissues is large (e.g. from millimolar to nanomolar or below), MS feature intensities can vary from tens to hundreds of thousands of counts (arbitrary units dependent on detector). Thus, the most relevant metabolites in a particular study may not necessarily be the most abundant. In order to give equal weights to the variables, unit variance (UV) scaling is often applied after centring the data. Here, each variable is divided by its standard

Processing and Analysis of GC/LC-MS-Based Metabolomics Data

287

deviation, resulting in a variance of 1 for all metabolite features. All features then become equally important in PCA. However, the drawback is that background features (noise) are inflated as well and are thus given the same importance as metabolite features. This is not a problem if the preprocessing step removed the noise efficiently (Section 3.1.1). If this is not the case, Pareto scaling might be a good alternative. Pareto scaling, often applied for spectroscopic data, is similar to UV scaling, the difference being that the square root of the standard deviation is used as the scaling factor rather than the standard deviation itself. Compared to no scaling, this reduces the relative importance of metabolites with high intensities by decreasing large fold changes more than small ones. Additionally, the impact of MS noise is reduced compared to UV scaling. 3.2.3. Unsupervised Approaches: PCA

After the scaling step, PCA can be performed on the data. PCA is the most widely used unsupervised chemometric technique for the analysis of MS-based metabolomics data. It is a useful tool to obtain an overview of the large datasets, visualize similarities and differences between observations and gain information about the metabolite features responsible for the observed patterns. PCA fits a model that approximates the data as well as possible with only a few uncorrelated (orthogonal) principal components (PCs), these components explaining most of the variance in the data. This allows for the reduction of data dimensionality, while retaining maximum information. Each PC is a linear combination of the original variables and explains the maximum amount of variance possible, not accounted for by the previous PCs.

3.2.3.1. Scores and Loadings

Conversion of the data into PCs results in a scores matrix and a loadings matrix. Scores represent the coordinates of the samples in the PCA model; loadings define the contribution of the original variables to form the scores. The directions are the same in both scores and loadings spaces so that an interesting pattern identified in the scores space can be interpreted by looking at the corresponding direction in the loadings space. The example in Fig. 17.2 shows PC1 vs. PC2 scores and loadings plots of data from the UPLC-MS analysis of 26 biological samples from a metabolomics study involving two groups of animals. In the scores plot (Fig. 17.2a), each point represents one sample from the study (circles=group 1, triangles=group 2). There is clear separation between the two groups of animals along the first PC, which explains the most variance in the data (55%). This indicates that the metabolic profiles from the two groups are different. The specific metabolite features responsible for this differentiation can be identified using the corresponding loadings plots (Fig. 17.2b), in which each point represents one feature. Variables far from the origin play a crucial role on the model, while variables around

288

Want and Masson A

200

B

G1 G2

M2 M3 0.2

0

p2

t2

M1 0

–0.2

–200

–600 –400 –200

0

t1

200

400

600

–0.2

0

0.2

p1

Fig. 17.2. PCA of 26 rat serum UPLC-MS metabolic profiles from a toxicological study. (a) PC1 vs. PC2 scores plot; (b) PC1 vs. PC2 loadings plots. There are two sample groups: G1 and G2. Metabolite feature M1 is higher in G2 than in G1, whereas metabolite features M2 and M3 are higher in G1 than in G2.

the origin have little or no influence. Here, features with highpositive x-coordinate in loadings plots, such as M1, have higher intensities in group 2 than in group 1. Conversely, features at the far left, such as M2 and M3, have higher intensities in group 1 than in group 2. Hence, scores plots can be used to investigate the degree of similarity between samples, and loadings plots to interpret these observations in terms of metabolic profiles (see Note 3). The scaling of the data (Section 3.2.2) has a large impact on the PCA model and therefore on the scores and loadings plots, as illustrated in Fig. 17.3. Here it can be seen that while the two experimental groups are separated in the scores plots using all three scaling methods, the plots differ, with a similar pattern seen with the centred data with no scaling and the Pareto-scaled data plots. Interpretation of loadings plots can be challenging with UV scaling, due to the density of metabolite features plotted (Fig. 17.3b2). 3.2.3.2. Detection of Outliers

PCA scores plots are also useful for the detection of outliers in the data. Outliers are observations that are extreme or do not fit the model. In PCA, strong outliers are found outside of the 95% Hotelling’s T2 confidence ellipse in the scores plot (17), e.g. sample 9 in Fig. 17.4. If they are far from the ellipse, they have a large effect on the model as they may require a whole PC to account for themselves. It is important to examine closely the outliers and investigate the features responsible for their outlier status to try to determine if they are due to experimental/analytical issues or if they reflect real metabolic/biological differences between a particular animal and the rest of the group. One should not forget though that the ellipse is a 95% confidence interval, so 5 observations out of 100 are expected to be outside the ellipse. Moderate outliers are observations for which the residual (distance to the

Processing and Analysis of GC/LC-MS-Based Metabolomics Data

A1

B1

400000

t2

289

G1 G2

0.2

p2

0

–400000

0

–0.2

–500000

0

500000

–0.4

t1

–0.2

–0

0.2

0.4

p1

A2

B2 G1 G2

0.04

0

p2

t2

40

0

–40

–80

–40

0

40

80

–0.04 –0.02

t1

0

0.02

p1

A3

B3 G1 G2

0.2

0

p2

t2

2000

0

–2000

–0.2

–4000

–2000

0

t1

2000

4000

–0.1

0

0.1

p1

Fig. 17.3. Effect of different data scaling on the PCA of UPLC-MS metabolic profiles of 36 rat serum samples. Column A: PC1 vs. PC2 scores plot, column B: corresponding loadings plots. Top (1): centred data (no scaling), Middle (2): centred data with UV scaling, Bottom (3): centred data with Pareto scaling.

model) is large, but compared to strong outliers, they do not have a large impact on the model. However, they do not fit the model very well, so care is needed when drawing conclusions for these observations from the model. 3.2.3.3. Assessment of Data Quality

Scores plots can also be used to interrogate trends detected in the data, which may be attributed to either analytical or technical variation. When analysing large batches of samples using GC/LCMS (typically around 100 samples – but this largely depends on the sample type), differences observed between the first and the

Want and Masson 2000

G1 G2 1000 t2

290

0

Sample 9

–1000 –2000 –10000

–5000

0 t1

5000

10000

15000

20000

Fig. 17.4. Effect of an outlier sample on the PCA scores plot (PC1 vs. PC2) of serum metabolic profiles from a metabolomics study with 19 animals divided into two groups. Sample 9 is an outlier and takes up the first principal component; separation between the two groups of rats is seen along the second PC.

last injection of the batch might be due to analytical changes (e.g. variation in GC/LC conditions, column degradation, mass spectrometer source contamination) or to metabolite decay in the samples over time rather than to real biological differences (Section 3.1.3). It is useful to check in the scores plot that the distribution of the points (samples) in the plot is not linked with the sample injection order during the analytical run. Quality control (QC) samples can be extremely helpful for this purpose and this approach is becoming more and more widespread in MSbased metabolomics studies (18, 19). This QC approach consists of preparing a sample representative of the sample batch by combining equal aliquots from each sample in the study. This QC sample is then injected periodically throughout the analytical run (typically every 6–10 injections) to assess analytical reproducibility. In the PCA scores plot, multiple injections of the QC sample should be tightly clustered (Fig. 17.5a) and show no/minimal drift over time. Injecting the QC sample several times (typically 5–15 injections) before starting the analysis of the actual samples – to condition the analytical column – is also a common practice in metabolomics studies, as the first injections are often not reproducible, particularly with a new chromatographic column. Therefore, observing a drift in the first conditioning QC injections in the PCA scores plot is common. The clustering of the last conditioning QC injections with the periodical QC injections throughout the analytical run ensures that the system had achieved stability prior to analysing the samples of interest. These conditioning samples can then be excluded from further analysis. If the injection order of the QCs analyzed within the batch of samples can be clearly identified in the PC1 vs. PC2 scores plot (Fig. 17.5b), it indicates that there has been a change in the analytical platform during the run (e.g. increase/decrease in metabolite feature intensities, retention time shifts) or in the QC

Processing and Analysis of GC/LC-MS-Based Metabolomics Data

A

B 2000

400

t2

QCs

0

0

–400

QCs

ion or

G1 G2

Inject

800

G1 G2

der

4000

1200

t2

291

–2000 –800 –1200 –1200

–800

–400

0

t1

400

800

1200

–4000

–6000 –4000 –2000

0

2000

4000

6000

t1

Fig. 17.5. PC1 vs. PC2 scores plots of metabolic profiles from two different studies modelled by PCA. In addition to rat serum samples, multiple injections of the QC samples during the run are also shown (conditioning QC injections have been removed). (a) QC injections are tightly clustered and no time trend can be seen on the plot. (b) The order of the injections of the QC sample can be clearly identified in the plot and is following the second PC. This indicates a drift during the analytical run.

sample itself (e.g. metabolite decay, sample evaporation leading to changes in metabolite concentrations). If this is the case, a new model can be constructed including only the QC samples in order to identify the metabolite feature(s) responsible for the observed drift. Sometimes, this drift can be corrected, either by preprocessing the data again with refined alignment step parameters (if the drift is due to retention time shift(s)) or by normalizing the data (Section 3.1.5). Alternatively, the features responsible for the drift can be excluded as if they are not reproducible in the QC samples, then they will not give reliable results throughout the run and so could not be used as discriminatory markers. Therefore, some researchers retain only metabolite features with a coefficient of variation (CV) of <20 or 30% in the QC injections. 3.2.4. Supervised Approaches: PLS-DA and OPLS-DA

PCA is a good starting point for the analysis of multivariate data, providing an overview of similarities and differences between samples. When a clear separation between the investigated groups is seen (e.g. Fig. 17.2a), it certifies that there is a real separation between the groups, as the separation is seen despite no class data being included in the algorithm. However, principal components model the largest variation(s) in the dataset and these directions may not coincide with the maximum separation between groups. Other directions might be more pertinent for discriminating between groups of samples. For this purpose, supervised multivariate approaches, such as PLS-DA and OPLS-DA, can be used. In this kind of approach, the algorithm takes into account sample class information and tries to construct a model that predicts this membership from the data. PLS-DA is a classification method based on a regression technique called PLS, often regarded as a regression extension of PCA. PLS models the association between the data table X and a

292

Want and Masson

matrix of responses Y, containing sample information (responses can be either categorical or continuous, e.g. age, growth, composition, chemistry parameters). While PCA is a projection of maximum variance in X, PLS is a maximum covariance model of the relationship between X and Y. In PLS-DA, the Y matrix defines class membership. Hence, PLS-DA attempts to construct a model that separates the different groups of samples on the basis of their X variables (metabolite features). Similar to PCA, X scores plot can be investigated to look at similarities/differences between samples. The PLS-DA model can be interpreted by considering the PLS weights for the X and Y variables. The interpretation of these weights, which gives information about how the variables combine to form the scores, is similar to the PCA loadings. Metabolite features highly correlated with the Y matrix (i.e. with the class separation) have high X weights. The X and Y weights are often plotted in the same graph, which gives the possibility to assess the relationship between X and Y. PLS-DA works well for separated homogenous classes, but PLS-DA models are negatively affected by systematic variation in the X matrix that is not related to the Y matrix, such as high within-class variance. When the direction separating the classes in PLS-DA is a combination of components, models might be difficult to interpret as between-class and within-class variations are mixed. A modification of the PLS method, named OPLS (orthogonal PLS), can be used to facilitate the interpretation. OPLS separates the variation in the data matrix X into two parts: one part correlated to Y and one part orthogonal to Y. The OPLS model thus comprises two modelled variations, the Y-predictive and the Y-orthogonal components. This partitioning facilitates the interpretation of the model, as linear between-class and within-class variations are separated. When performing OPLS-DA with two classes for instance, all the information relevant for the explanation of the membership are summarized in one component. OPLS-DA makes the interpretation of the model more straightforward while maintaining the predictive ability of PLS-DA (see Note 4). Figure 17.6 shows the PCA, PLS-DA and OPLS-DA scores plot of the same metabolomics dataset including two experimental groups. Separation is seen between the two groups of animals by PCA (a), with a small overlap of the two classes. This separation is well defined using PLS-DA (b) and follows a direction combining the first and second components. Using OPLS-DA (c), the first component (predictive component) separates the two classes, facilitating the identification of metabolite features differentially expressed between both groups. 3.2.5. Model Validation

When creating a model, it is important to determine how well this model reflects the data and to what extent the conclusions based on this model can be relied on. For this purpose, it is necessary

Processing and Analysis of GC/LC-MS-Based Metabolomics Data

A

4000

G1 G2

2000

t2

293

0

–2000

–4000 –6000

B

–4000

–2000

0 t1

2000

4000

4000

G1 G2

2000

t2

6000

0

–2000

–4000 –6000

C

–4000

–2000

0 t1

2000

4000

4000

G1 G2

2000

to

6000

0

–2000

–4000 –6000

–4000

–2000

0 tp

2000

4000

6000

Fig. 17.6. PC1 vs. PC2 scores plot of 33 metabolic profiles of rat serum samples modelled by (a) PCA, (b) PLS-DA, (c) OPLS-DA. Separation between the two groups of animals is seen with PCA, which indicates that there are differences in metabolic profiles between the groups. The separation is amplified using PLS-DA as the algorithm is taking into account the class membership to construct the model. Using OPLS-DA, the predictive component separates the two groups, which facilitates the interpretation of the model.

to check 1) the goodness of fit to the data and 2) the predictive ability of the model. Assessing the goodness of fit is essential as it indicates how well the model fits the observations. However, by increasing the number of PCs in a model, the fit of the data will improve. Therefore checking only the goodness of fit is not sufficient; it is crucial to also assess the predictive power of the model. This can be done using either data included to construct

294

Want and Masson

the model (cross-validation) (see Note 5) or a new set of observations (test set) (see Note 6) (20, 21). There is a trade-off between the goodness of fit and the predictive ability. By increasing the number of PCs, noise gets modelled by the PCs, which decreases the predictive ability of the model. This decrease in predictive power is characteristic of an over-fitted model. Therefore the predictive capacity of a model is often used as a criterion to determine the optimal number of components. A good model should fit well the data and be predictive of new data. 3.3. Metabolite Identification 3.3.1. Databases

3.3.2. Metabolite Identification Approaches

Whilst data preprocessing tools (Section 3.1) and ensuing multivariate analysis approaches (Section 3.2) become more sophisticated and robust, structural characterization of discriminatory metabolites remains a significant challenge. Consultation with metabolite databases forms the initial step in metabolite characterization. These databases include HumanCyc (http://biocyc.org) (22), the Human Metabolome Database (HMDB; http://www.hmdb.ca/) (23), Kyoto Encyclopedia of Genes and Genomes (KEGG; http://www.genome.jp/kegg/ligand.html) (24), METLIN (http://metlin.scripps.edu/) (25) and the NIST database (http://www.nist.gov/srd/index.htm) (26). Although incomplete, these databases contain information regarding thousands of endogenous and drug metabolites, including MS spectra and in some cases MS/MS spectra. Importantly for metabolite identification, the HMDB contains experimental and predicted metabolite 1 H- and 13 C-NMR data, and the Spectral Database for Organic Compounds (SDBS) possesses NMR, MS and IR spectra for organic compounds – useful information, as often MS alone is not sufficient for unambiguous metabolite identification. The NIST (26) and the Wiley databases focus on EI data; the NIST database contains >220,000 spectra of nearly 200,000 unique compounds (often with chemical structures) and >224,000 Kovats retention index values for ∼22,000 compounds, while the Wiley database contains ∼400,000 EI mass spectra and >180,000 chemical structures. Together these two databases form an important resource for metabolite identification using GC-MS. The GOLM open access database (27) at the Max Planck Institute of Molecular Plant Physiology focuses on EI data and also acts as a GC-MS data repository. Well-characterized metabolites may be identified through database searches, and if samples are analyzed using highresolution MS, many candidates can be excluded at this stage. For unambiguous metabolite identification, co-chromatography and comparison of MS/MS data with the authentic compound are necessary (28, 29). In the case of an unknown molecule, de novo

Processing and Analysis of GC/LC-MS-Based Metabolomics Data

295

identification is required – a significant challenge given the often trace quantities of some metabolites and limited sample amount. Elemental composition can be obtained using a combination of the following: 1. Accurate mass determinations using high-resolution FTMS or Orbitrap instrumentation 2. High accurate tandem mass measurements (e.g. on a Q-TOF, LTQ-FT or LTQ-Orbitrap) for structural characterization However, despite recent advances, the lack of comprehensive MS libraries and databases often hinders metabolite identification based solely on MS information. Ultimately, a combination of technologies will be required for metabolite identification. These include high-sensitivity capillary NMR, chemical modification for functional group identification and finally independent compound synthesis for verification.

4. Notes 1. Parameter optimization. It is crucial to optimize the data preprocessing parameters for each metabolomics study, as every dataset will differ. Parameter optimization is not straightforward and can be time consuming, as some input parameters that need to be selected for peak alignment are not always simple to determine. Key peak picking parameters include the signal-to-noise (S/N) threshold, peak width parameters and peak shape parameters. The optimum values for settings such as peak width should be easy to determine and in general are constant for all chromatograms within a dataset. Conversely, parameters that define real chromatographic peaks vs. noise or window sizes in which peaks in two chromatograms are considered the same are more difficult to determine. Criteria have to be set for determining the alignment quality, which can be subjective. Some studies use control samples that are analyzed both spiked with known compounds and unspiked to check the analytical system and the method, as well as evaluating peak alignment quality. 2. Baseline correction, noise reduction and smoothing. Correct use of baseline correction and noise reduction tools is essential. Different GC-MS and LC-MS technologies may require specific parameter settings and algorithms, which should be optimized by each vendor, or perhaps the instrument operator. Therefore, as a rule, smoothing and baseline correction are best performed by vendor and system-specific software applications.

296

Want and Masson

3. Scores and loadings plots. Scores and loadings plots are 2D representations of the scores and loadings space. It is important to keep in mind how much variation is explained by the plotted PCs. When many principal components are necessary to explain all the variations in the data, precaution is needed when interpreting the plots. 4. Predictive performance. It is important to note that OPLS and PLS have equivalent prediction performance: an OPLS model that uses 1 predictive component and n orthogonal components is equivalent in terms of prediction quality to a PLS model using n+1 components. However, the model interpretation is easier with OPLS. 5. Cross-validation. The basic idea of cross-validation is to exclude part of the data, construct a new model on remaining data, predict the omitted data with this model and compare the predictions with the actual values, this being done until all data have been removed once. The sum of the squared differences between predicted and observed values can be used as a measure of the predictive power of the model. 6. Model validation. A good way of validating a model, if the number of samples is sufficient, is to split the dataset into two parts: a training set and a test set (the latter containing typically at least one-third of the samples). A model is constructed and optimised using only the training set. Then, the developed model is used to predict the samples in the test set, which have not influenced the model. The goodness of fit on the test data is an indicator of the predictive ability of the model.

Acknowledgements The authors would like to acknowledge Dr. Timothy Ebbels for valuable discussions during the preparation of this chapter. EW would like to acknowledge Waters Corporation for funding. Perrine Masson would like to acknowledge Servier Laboratories Ltd. for funding. References 1. Nicholson, J. K., Connelly, J., Lindon, J. C., Holmes, E. (2002) Metabonomics: a platform for studying drug toxicity and gene function. Nat Rev Drug Discov 1, 153–161. 2. Fiehn, O. (2002) Metabolomics – the link

between genotypes and phenotypes. Plant Mol Biol 48, 155–171. 3. Nicholson, J. K., Lindon, J. C. (2008) Systems biology: metabonomics. Nature 455, 1054–1056.

Processing and Analysis of GC/LC-MS-Based Metabolomics Data 4. Trygg, J., Holmes, E., Lundstedt, T. (2007) Chemometrics in metabonomics. J Proteome Res 6, 469–479. 5. Smith, C. A., Want, E. J., O’Maille, G., Abagyan, R., Siuzdak, G. (2006) XCMS: processing mass spectrometry data for metabolite profiling using nonlinear peak alignment, matching, and identification. Anal Chem 78, 779–787. 6. Burton, L., Ivosev, G., Tate, S., Impey, G., Wingate, J., Bonner, R. (2008) Instrumental and experimental effects in LC-MS-based metabolomics. J Chromatogr B Anal Technol Biomed Life Sci 871, 227–235. 7. Scholz, M., Gatzek, S., Sterling, A., Fiehn, O., Selbig, J. (2004) Metabolite fingerprinting: detecting biological features by independent component analysis. Bioinformatics 20, 2447–2454. 8. Wang, W., Zhou, H., Lin, H., Roy, S., Shaler, T. A., Hill, L. R., Norton, S., Kumar, P., Anderle, M., Becker, C. H. (2003) Quantification of proteins and metabolites by mass spectrometry without isotopic labeling or spiked standards. Anal Chem 75, 4818– 4826. 9. Oresic, M., Clish, C. B., Davidov, E. J., Verheij, E., Vogels, J., Havekes, L. M., Neumann, E., Adourian, A., Naylor, S., van der Greef, J., Plasterer, T. (2004) Phenotype characterization using integrated gene, protein and metabolite profiling. Appl Bioinform 3, 205–217. 10. Yeung, K. Y., Ruzzo, W. L. (2001) Principal components analysis for clustering gene expression data. Bioinformatics 17, 763–774. 11. Jolliffe, I. T. (2002) Principal Components Analysis, 2nd edn, Springer, New York, NY. 12. Ivosev, G., Burton, L., Bonner, R. (2008) Dimensionality reduction and visualization in principal components analysis. Anal Chem 80, 4933–4944. 13. Barker, M., Rayens, W. (2003) Partial least squares for discrimination. J Chemom 17, 166–173. 14. Bylesjo, M., Rantalainen, M., Cloarec, O., Nicholson, J. K. (2006) OPLS discriminant analysis: combining the strengths of PLSDA and SIMCA classification. J Chemom 20, 341–351. 15. Trygg, J., Wold, S. (2002) Orthogonal projections to latent structures (O-PLS). J Chemom 16, 119–128. 16. Trygg, J. (2002) O2-PLS for qualitative and quantitative analysis in multivariate calibration. J Chemom 16, 283–293. 17. Jackson, J. E. (2003) A User’s Guide to Principal Components, Wiley–Interscience, New York, NY.

297

18. Zelena, E., Dunn, W. B., Broadhurst, D., Francis-McIntyre, S., Carroll, K. M., Begley, P., O‘Hagan, S., Knowles, J. D., Halsall, A., HUSERMET Consortium, Wilson, I. D., Kell, D. B. (2009) Development of a robust and repeatable UPLC-MS method for the long-term metabolomics study of human serum. Anal Chem 81, 1357–1364. 19. Gika, H. G., Macpherson, E., Theodoridis, G. A., Wilson, I. D. (2008) Evaluation of the repeatability of ultra-performance liquid chromatography–TOF-MS for global metabolic profiling of human urine samples. J Chromatogr B Anal Technol Biomed Life Sci 871, 299–305. 20. Brereton, R. G. (2006) Consequences of sample sizes, variable selection, model validation and optimisation for predicting classification ability from analytical data. Trends Anal Chem 25, 1103–1111. 21. Anderssen, E., Dyrstad, K., Westad, F., Martens, H. (2006) Reducing over-optimism in variable selection by cross-model validation. Chemom Intell Lab Syst 84, 69–74. 22. Romero, P., Wagg, J., Green, M. L., Kaiser, D., Krummenacker, M., Karp, P. D. (2005) Computational prediction of human metabolic pathways from the complete human genome. Genome Biol 6, R2. 23. Wishart, D. S., Tzur, D., Knox, C., Eisner, R., Guo, A. C., Young, N., Cheng, D., Jewell, K., Arndt, D., Sawhney, S., Fung, C., Nikolai, L., Lewis, M., Coutouly, M. A., Forsythe, I., Tang, P., Shrivastava, S., Jeroncic, K., Stothard, P., Amegbey, G., Block, D., Hau, D. D., Wagner, J., Miniaci, J., Clements, M., Gebremedhin, M., Guo, N., Zhang, Y., Duggan, G. E., Macinnis, G. D., Weljie, A. M., Dowlatabadi, R., Bamforth, F., Clive, D., Greiner, R., Li, L., Marrie, T., Sykes, B. D., Vogel, H. J., Querengesser, L. (2007) HMDB: the human metabolome database. Nucleic Acids Res 35, D521–D526. 24. Ogata, H., Goto, S., Sato, K., Fujibuchi, W., Bono, H., Kanehisa, M. (1999) KEGG: Kyoto Encyclopedia of Genes and Genomes. Nucleic Acids Res 27, 29–34. 25. Smith, C. A., O’Maille, G., Want, E. J., Qin, C., Trauger, S. A., Brandon, T. R., Custodio, D. E., Abagyan, R., Siuzdak, G. (2005) METLIN: a metabolite mass spectral database. Ther Drug Monit 27, 747–751. 26. Babushok, V. I., Linstrom, P. J., Reed, J. J., Zenkevich, I. G., Brown, R. L., Mallard, W. G., Stein, S. E. (2007) Development of a database of gas chromatographic retention properties of organic compounds. J Chromatogr A 1157, 414–421.

298

Want and Masson

27. Kopka, J., Schauer, N., Krueger, S., Birkemeyer, C., Usadel, B., Bergmüller, E., Dörmann, P., Weckwerth, W., Gibon, Y., Stitt, M., Willmitzer, L., Fernie, A. R., Steinhauser, D. (2005) [email protected]: the Golm Metabolome Database. Bioinformatics 21, 1635–1638. 28. Weckwerth, W., Morgenthal, K. (2005) Metabolomics: from pattern recognition to

biological interpretation. Drug Discov Today 10, 1551–1558. 29. Saghatelian, A., Trauger, S. A., Want, E. J., Hawkins, E. G., Siuzdak, G., Cravatt, B. F. (2004) Assignment of endogenous substrates to enzymes by global metabolite profiling. Biochemistry 43, 14332–14339.

Chapter 18 Nuclear Magnetic Resonance (NMR)-Based Drug Metabolite Profiling Eva M. Lenz Abstract The identification of drug metabolites in biofluids such as urine, plasma and bile is an important step in drug discovery and development. Proton nuclear magnetic resonance (1 H-NMR) spectroscopy can provide detailed information regarding the structural transformation of a compound as a consequence of metabolism. However, successful identification of drug metabolites by 1 H-NMR spectroscopy is generally compromised by the presence of endogenous metabolites, which can obscure the signals of the drug metabolites in question. Hence, sample clean-up and separation of the metabolites from the biofluid matrix is crucial. This is generally achieved by extraction of the biofluid, solid-phase extraction (SPE), high-performance liquid chromatography (HPLC) or any combination of these. Apart from 1 H, other NMR-active nuclei, such as 19 F, can provide a useful handle for metabolite profiling, provided they are not naturally present in the biofluid. Successful studies have shown that the presence of a fluorine-handle on the drug and its metabolites can provide additional qualitative and quantitative data by 19 F-NMR spectroscopy. This chapter provides guidelines and examples of NMR-based drug metabolite profiling. Key words: 1 H- and 19 F-NMR spectroscopy, SPE, HPLC, drug metabolites.

1. Introduction Proton nuclear magnetic resonance (1 H-NMR) spectroscopy is a very useful structural tool to assess the biotransformations of drugs, provided the drug metabolites are isolated from the biological matrix and present in concentrations of >50 μM when using conventional probes (although with new CryoProbe technology, detection limits are in the nanomolar region). A prerequisite of successful structural determination of metabolites by 1 H-NMR spectroscopy is the full structural characterisation of T.O. Metz (ed.), Metabolic Profiling, Methods in Molecular Biology 708, DOI 10.1007/978-1-61737-985-7_18, © Springer Science+Business Media, LLC 2011

299

300

Lenz

G G

A)

G

B) S

S

G

S

C

hippurate

P urea 7.5

7.0

6.5

6.0

5.5

ppm

2.15

2.10 ppm

creatinine

C)

9.5

9.0

8.5

8.0

7.5

7.0

6.5

6.0

5.5

5.0

4.5

4.0

3.5

3.0

2.5

2.0

1.5

1.0

ppm

Fig. 18.1. The 1 H-NMR spectrum of urine from a healthy human volunteer 3–6 h after administration of paracetamol (dose = 13.7 mg/kg, i.e. 1 g in total) without any sample preparation. The 1 H-NMR spectrum was acquired with water suppression (with noesypr1d), showing the characteristic resonances from the major urinary paracetamol metabolites, the paracetamol glucuronide (G) and the sulphate (S). The regions of interest are shown in the insets: (a) representing the aromatic region, as well as the diagnostic anomeric proton at δ1 H 5.15 of the paracetamol glucuronide moiety, and (b) representing the N-acetyl groups. The whole 1 H-NMR spectrum of the urine collected 3–6 h post-dose is shown in (c). Key: G, glucuronide conjugate; S, sulphate conjugate; C, cysteinyl conjugate and P, parent.

the parent molecule itself (see Note 1). According to metabolite concentration or amount extracted, several NMR experiments can be carried out to aid the structural identification of the metabolite (1). In the simplest case, drug metabolites can be detected and partially identified directly by 1 H-NMR spectroscopy of the biofluid without further sample workup, as in the example of paracetamol (APAP), for which a body of literature evidence exists (e.g. 2). The diagnostic resonances of APAP are visible in the 1 H-NMR spectra of urine and plasma, despite the presence of endogenous signals (Fig. 18.1). However, in most cases, drug metabolite identification by 1 H-NMR spectroscopy is obscured by the presence of endogenous metabolites (as shown in Fig. 18.2), as well as exogenous contaminants, such as solvent or dosing vehicles. Exogenous contaminants comprise solvent signals (e.g. ACN, MeOH) and pH modifiers (e.g. acetic and formic acid). This can be a challenge in hyphenated HPLC–NMR, as solvent concentrations are present in vast excess, resulting in large 1 H-NMR signals that require suppression to enable the drug metabolite to be digitised and detected, as shown in Fig. 18.3. Other typical contaminants are polyethylene glycol (PEG, from dosing solutions) or glycerol (contained in microsomal solutions), which can be present in high concentrations (Fig. 18.4; ref. 3).

301

water

TSP

Nuclear Magnetic Resonance (NMR)-Based Drug Metabolite Profiling

A

B

C

9.5

9.0

8.5

8.0

7.5

7.0

6.5

6.0

5.5

5.0

4.5

4.0

3.5

3.0

2.5

2.0

1.5

1.0

ppm

Fig. 18.2. 1 H-NMR spectra of typical biofluids, showing the wealth of signals from the endogenous metabolites contained. Biofluid profiles comprise complex mixtures of glucose, amino acids, organic acids, lipids and proteins. Whilst the urine spectrum is characterised by sharp peaks from small molecular weight components such as creatinine, urea, citrate, amino acids and glucose (amongst many other components), the plasma profile is characterised by a rolling baseline deriving from the protein macromolecules contained, embedding amino and organic acids, and additional broad signals from lipids. Amino acids, bile acids and lipids are the major contributors to the bile profile. Successful metabolite identification by 1 H-NMR spectroscopy relies on the removal of the interfering biological matrices. Key: (a) a human urine sample, (b) human plasma and (c) rat bile. All acquired with water suppression (with noesypr1d) and containing TSP as internal reference standard.

Unfortunately, these agents are UV silent, which can make sample isolation quite difficult (see Note 2). Endogenous contaminants are generally co-eluting biofluid constituents, which have similar polarity as the analyte in question. Again, some are UV silent, such as bile acids, hence during the HPLC-UV separation, these can co-elute unnoticed, as shown in Fig. 18.5. Therefore, every effort must be made to extract the metabolites of interest from the biological matrix in order to minimise the problem of signal overlap with either endogenous or interfering material such that the structures of the metabolites can

302

Lenz

ACN 13C

* *

satellites*

0.55% of ACN–peak

A

8.5

8.0

7.5

7.0

6.5

6.0

5.5

5.0

4.5

4.0

3.5

3.0

2.5

2.0

2.5

2.0

1.5

1.0

13C

ppm

ppm

satellites

B water

8.5

8.0

7.5

7.0

6.5

6.0

5.5

5.0

4.5

4.0

3.5

3.0

2.5

2.0

1.5

1.0

ppm

Fig. 18.3. These 1 H-NMR spectra represent a typical HPLC–NMR separation, using D2 O and ACN as mobile phase, and highlight the problem solvent signals can pose. The spectra are acquired in stop-flow mode, without solvent suppression. (a) showing the dominant ACN peak and its 13 C satellites. Double-solvent suppression (b) achieves attenuation of the ACN peak (to below the 13 C satellite level) and, hence, enables digitisation and detection of the drug metabolite signals. The drug metabolite was derived from a 48-h pooled urine sample from a human volunteer, which was extracted first by step-gradient SPE, where it eluted in the 50% MeOH fraction. The dried SPE fraction was then further separated by hyphenated HPLC–NMR (the experimental conditions are detailed in ref. (11)).

be fully elucidated. However, there is no single way to achieve drug metabolite separation and identification by 1 H-NMR spectroscopy due to the differences in biofluid matrices and the concentration and chemical properties of the metabolites (4, 5). Hence, metabolite isolation typically involves extraction, protein precipitation, solid-phase extraction or HPLC separation to obtain a pure enough drug metabolite sample to result in successful identification by NMR spectroscopy. In addition, several extraction steps might be required in order to isolate the metabolite in question from the biofluid matrix. Perhaps the least used extraction method is liquid–liquid extraction, as the more polar metabolites might not partition into the organic phase (see Note 3). Hence, generally, chromatographic separations such as SPE or HPLC are used for the isolation of metabolites from the biological matrix. HPLC is a high-resolution chromatographic method enabling the separation and isolation of drug metabolites from endogenous metabolites. However, co-elution can still be an issue with metabolites of similar polarity and, hence, elution characteristics. Therefore, retention characteristics have to be assessed

Nuclear Magnetic Resonance (NMR)-Based Drug Metabolite Profiling

303

A

4.4

4.2

4.0

3.8

3.6

3.4

3.2

3.0

2.8

ppm

B

3.8

4.4

4.2

4.0

3.8

3.6

3.4

3.7

3.2

3.0

3.6

2.8

ppm

ppm

Fig. 18.4. 1 H-NMR spectra of the typical exogenous UV-silent contaminants (a) glycerol, often contained in the medium of microsomal preparations, and (b) PEG, often used as a dosing vehicle. These components co-extract or co-elute undetected but can obscure large sections of the NMR spectrum and even pose a dynamic range problem when present in high concentrations compared to the drug metabolite in question.

with the parent compound, prior to loading of the biofluid or the biofluid extract, in order to ascertain maximum retention of the drug metabolites. In general, HPLC is coupled with UV or DAD detection; however, when dealing with a radiolabelled compound, radiodetection (HPLC–RAD) greatly facilitates the identification of drug metabolites, provided the label is preserved in the metabolites. Drug metabolite isolation and identification can be achieved off-line (e.g. by SPE or HPLC; see Notes 4 and 5) or, if the concentrations of the drug metabolites are sufficiently high (e.g. >50 μM), on-line by HPLC–NMR or HPLC–SPE–NMR spectroscopy, which enables the separation and structural characterisation by 1 H-NMR spectroscopy in a single step. Alternatively, 19 F-NMR spectroscopy has been successfully utilised (6–8), as it provides a sensitive and specific NMR ‘handle’ and does not

acetate

Lenz

MeOH–d4

304

taurine

C18

C19

C21

*

*

* * taurocholic acid 4.0

3.5 21

OH 18 19

HO

H

3.0

2.5

2.0

1.5

1.0

0.5

ppm

1

ppm

CONHCH2CH2HSO3

taurocholic acid

H

OH

biliary drug metabolite*

9

8

7

6

5

4

3

2

Fig. 18.5. 1 H-NMR spectrum of an isolated biliary drug metabolite contaminated with taurocholic acid, which co-eluted undetected during the HPLC separation as it is UV-silent. In this case, the presence of the taurocholic acid signals did not perturb metabolite identification, although the cholesterol-backbone occupies a large section of the aliphatic region (0.7–2.2 ppm). Note: this HPLC separation was carried out with acetic acid buffer, hence the additional signal in the spectrum from acetate.

suffer from biological contamination and solvent interferences. 19 F-NMR spectroscopy is 83% as sensitive as 1 H-NMR spectroscopy and, with its large spectral width (∼200 ppm), the signals are generally nicely dispersed. Assuming that no fluorinated solvent or buffer is used, the observed signals solely derive from the drug and its metabolites, providing a metabolite profile and information on the relative amounts and concentrations of the metabolites present, prior to their isolation. The metabolite profile can also be assessed quantitatively against an internal standard, if required, as shown in Fig. 18.6. For HPLC–NMR spectroscopy, due to the spectral simplicity, it can be used as an additional or even sole detector in order to identify metabolites during the HPLC separation (e.g. Fig. 18.7). When used in conjunction with off-line fractionation, each fraction can be examined for 19 F content, giving additional evidence as to whether the fractions contain a drug metabolite (Fig. 18.8, see Note 6). This chapter provides guidelines and examples of NMR-based drug metabolite profiling.

Nuclear Magnetic Resonance (NMR)-Based Drug Metabolite Profiling

305

Is 24–48h

C Is 8–24h

B 0–8h

Is

A

Fig. 18.6. 19 F-NMR profile of neat rat urine samples collected at (a) 0–8 h, (b) 8–24 h and (c) 24–48 h intervals following a single i.p. dose of 50 mg/kg of 2-trifluoromethylacetanilide. Addition of an internal standard (Is) enabled the quantification of the metabolites contained (experimental details described in ref. (8)). Reprinted from (8), with permission from Elsevier.

2. Materials 2.1. Sample Preparation, Extraction and Isolation 2.1.1. Freeze-drying of Samples or Large Volumes of Sample Extracts

1. Freeze-drier (e.g. Edwards Modulya 4K or VirTis Sentry 2.0).

2.1.2. Liquid–Liquid Extraction

1. Biological fluid (i.e. aqueous samples, generally acidified to suppress ionisation).

2. Glass vials with screw-top neck. 3. Liquid nitrogen (–196◦ C) or solid carbon dioxide (Cardice, –140◦ C) to freeze the sample.

2. Water-immiscible solvents (e.g. HPLC-grade chloroform; Fisher Scientific, Loughborough, UK).

306

Lenz

A)

C)

B) UV-chromatogram rows

A time

* transfer-time

* *

F2 retention time/min.

Fig. 18.7. The pseudo-2D contour plot of an HPLC separation of a rat urine concentrate monitored by 19 F-NMR spectroscopy. The pseudo-2D contour plot shows the individual metabolites of 2-trifluormethylacetanilide (dosed at 50 mg/kg) represented by their CF3 -handles (a). The relationship between the pseudo-2D experiment and the chromatogram is schematically depicted (b), enabling the assessment of their exact chromatographic retention times. The HPLC separation was then repeated in stop-flow mode, where the chromatographic separation is halted at the retention times of interest. Once the metabolite fractions are trapped inside the NMR flow-cell, 1 H-NMR spectra can be acquired for structural identification of the metabolites (c) (the experimental details described in ref. (8)). Reprinted from (8), with permission from Elsevier.

3. Separating funnel. 4. Solvent evaporator or water bath. 2.1.3. Solid–Liquid Extraction

1. Biological fluid (e.g. urine or bile), dry or freeze-dried, or faeces. 2. Organic solvent (e.g. HPLC-grade methanol (MeOH) or acetonitrile (ACN); Fisher Scientific). 3. Centrifuge (e.g. Megafuge 1.0R; Heraeus Instruments, Newport Pugnell, UK). 4. Solvent evaporator, water bath and/or freeze-drier.

2.1.4. Plasma or Serum Protein Precipitation

1. Organic solvent (e.g. HPLC-grade ACN; Fisher Scientific). 2. Centrifuge. 3. Solvent evaporator, water bath and/or freeze-drier.

Nuclear Magnetic Resonance (NMR)-Based Drug Metabolite Profiling

A0013 100

47.59

Channel 1 An1 4.09e5

A0146 3.94

100

Parent drug

A

307

7.02 9.48 11.11 13.39 13.86 17.60

An1 1.00e6

46.49 47.45

31.61 37.33

B 21.95

43.97

32.69 23.89 25.42

45.94 27.56

19.47

39.20

%

%

18.69

3.94

19.80

55.18

42.76 28.12

42.06

49.04

54.72

33.57

59.83

29.92

34.25

50.06

52.92 50.55

41.45

59.86

0

Time

10.00

20.00

30.00

40.00

50.00

0

60.00

Time

5.00

10.00 15.00 20.00 25.00 30.00 35.00 40.00 45.00 50.00 55.00 60.00

**

C

*

FA Metabolite*

water impurities

9.5

9.0

8.5

8.0

7.5

7.0

6.5

6.0

5.5

5.0

4.5

4.0

3.5

3.0

2.5

D

2.0

1.5

1.0

ppm

19F

111

112

113

114

115

116

117

118

119

120

121

122

123

124

ppm

Fig. 18.8. An example of an off-line HPLC separation of a drug metabolite from 400 mL of human urine. The fluorinecontaining drug was metabolised extensively (28 metabolites eluting over 55 min, as determined by HPLC-MS (data not shown)). The isolated HPLC fractions were subsequently analysed by 19 F-NMR for ‘ metabolite content’ and identified by 1 H-NMR for structural identification. Shown are (a) HPLC method development with parent compound (eluting at 48 min at 30% ACN), (b) the HPLC-UV trace of an aliquot (80 μL) of urine concentrate collected into 96-well plates, and the 1 Hand 19 F-NMR spectra (c and d) of a metabolite isolated by HPLC. Key: FA, formic acid, contained in the mobile phase.

2.1.5. Precipitation of Bile Salts

1. Acid or acidified water (e.g. formic acid, HPLC-grade acidified water at pH 2; Fisher Scientific). 2. Centrifuge (for small volumes, such as 2 mL aliquots, e.g. Eppendorf Centrifuge 5417C, Hamburg, Germany). 3. Organic solvent (HPLC-grade MeOH or ACN; Fisher Scientific) for extraction of the precipitate, in case metabolites or parent drug have also precipitated. 4. Solvent evaporator, water bath and/or freeze-drier. 5. Alternatively, samples may be freeze-dried and extracted with organic solvent as described in Section 3.1.3.

2.1.6. Solid-Phase Extraction (SPE) Chromatography

1. 1–3 mL SPE columns of suitable packing material (e.g. R Waters Oasis HLB or Whatman C18 packing). 2. Organic solvent (e.g. HPLC-grade MeOH; Fisher Scientific).

308

Lenz

3. Aqueous phase (e.g. HPLC-grade water at pH 2, acidified with HCL or formic acid; Fisher Scientific). 4. Solvent mixtures to enable step-gradient elution of the analytes (e.g. MeOH:aqueous, 0:100, 20:80, 40:60, 60:40, 80:20, 100:0% (vol/vol); or even finer steps, e.g. 10% MeOH). 5. Vacuum manifold to aid elution of the chromatographic fractions. 2.1.7. High-Performance Liquid Chromatography (HPLC)

1. HPLC column (appropriate for analyte, e.g. Prodigy 5 μm ODS; Phenomenex, UK; BDS Pursuit 3 μm PFP; Varian, Netherlands). 2. HPLC pump and UV detector (e.g. Agilent Technologies, West Lothian, UK) or radiodetector if compound is radiolabelled (e.g. 14 C, Berthold radiodetector, Harpenden, UK, equipped with solid cell). 3. Fraction-collector or glass vials for manual collection. 4. Mobile phases: HPLC-grade water and ACN (Fisher Scientific). 5. Buffers of choice: e.g. HPLC-grade ACN and aqueous phase, pH adjusted with analytical-grade formic acid (e.g. 0.1% formic acid, v/v; Fisher Scientific) and/or buffered with analytical-grade ammonium formate (e.g. 10 mM; Fisher Scientific) (see Note 7).

2.2. NMR Spectroscopy 2.2.1. NMR Spectroscopy

1. NMR tubes (ideally of small volume not to dilute sample unnecessarily, e.g. 1-, 2.5- or 3-mm tubes, or the use of Shigemi tubes) and corresponding NMR probes. 2. Suitable deuterated NMR solvents (e.g. D2 O, MeOH-d4 , ACN-d3 , dimethyl sulphoxide (DMSO-d6 ); Goss Scientific, Nantwich, UK).

2.2.2. Hyphenated HPLC–NMR

1. HPLC–NMR system comprised of an integrated HPLC pump allowing stop-flow analysis (e.g. Bruker Agilent HPLC system, operated via HystarTM software) and an LC– NMR flow probe (e.g. 1 H/19 F dual tunable (LCDXI), with 60 μL cell volume). 2. HPLC solvents: D2 O (Eurisotop, Giv-sur-Yvette, France) and ACN (Pestanal Grade, Riedel-De-Haën, Sigma-Aldrich, Dorset, UK), or ACN-d3 (99.8 atom %D(it says on the bottle); Sigma-Aldrich). The most suitable acid to adjust the pH of the mobile phase is formic acid-d2 (Goss Scientific, Nantwich, UK).

Nuclear Magnetic Resonance (NMR)-Based Drug Metabolite Profiling

2.2.3. Hyphenated HPLC–SPE–NMR

309

1. Dedicated HPLC–SPE system (e.g. Bruker Agilent (HPLC) Spark Prospekt 2 (SPE) system) with Hystar SPE cartridges (Spark, Netherlands). 2. Dedicated LC–SPE flow probe (e.g. 1 H/19 F dual tunable (LCDXI), with 30 μL cell volume). 3. Non-deuterated mobile-phase constituents for initial HPLC separation: water and ACN (Pestanal Grade, Riedel-DeHaën, Sigma-Aldrich); pH adjusted with formic acid. 4. Deuterated ACN-d3 (99.8 at.% D; Sigma-Aldrich) for elution of SPE cartridge.

3. Methods 3.1. Sample Preparation, Extraction and Isolation 3.1.1. Freeze-drying

Freeze-drying is an ideal way to concentrate biofluid samples and to reduce large volumes of biofluid extracts and off-line collected SPE or HPLC fractions (see Note 8). 1. Ensure complete evaporation of organic solvents from sample prior to freeze-drying. This applies to biofluids extracted with organic solvents such as MeOH. 2. Freeze small aliquots (∼1–5 mL) of biofluids or biofluid extracts in liquid nitrogen (–196◦ C) or solid carbon dioxide (Cardice, –140◦ C) in screw-neck glass vials placed at a slight angle. 3. Operate freeze-dryer according to instruction manual.

3.1.2. Liquid–Liquid Extraction

1. Liquid–liquid extraction is based on the partition coefficient of an analyte between two immiscible phases. This is probably the least used extraction method in drug metabolite studies, as metabolites generally exhibit a range of polarities; hence, the extraction efficiency can be inferior. 2. Where appropriate, biofluids are pH adjusted (e.g. acidified prior to extraction in order to suppress ionisation of the metabolites by drop-wise addition of an acid while measuring the pH of the biofluid). 3. Add equal volumes of the acidified biofluid and chloroform into a separating funnel. Mix vigorously, then gently release pressure. To avoid inhalation of chloroform, this procedure must be carried out in a fume hood.

310

Lenz

4. Allow the phases to separate and carefully decant and collect the chloroform phase, then the aqueous phase. The chloroform phase ought to contain the drug and its metabolites. 3.1.3. Solid–Liquid Extraction

1. This refers to solid material (e.g. rodent faeces) or freezedried biofluid samples (e.g. urine or bile). 2. If large quantities of human urine are to be extracted, subaliquot the samples into smaller volumes (e.g. 5–10 mL) and freeze-dry. 3. Extract the material successively in aliquots of organic solvent (e.g. 1–5 mL MeOH), mix thoroughly, centrifuge at +4◦ C for 10 min at 1800×g, then carefully remove the supernatant. Repeat the process several times and combine the supernatant fractions. 4. Evaporate the supernatant solvent.

3.1.4. Plasma Protein Precipitation

1. This can be achieved twofold, either via protein crashing or ultrafiltration (see Note 9). 2. Protein crashing requires aliquots of plasma to be combined with ACN (1:3, v/v), mixed and centrifuged at +4◦ C for 10 min at 1800×g. 3. Collect and combine the supernatants and evaporate the ACN.

3.1.5. Precipitation of Bile Salts

1. This is a useful approach to remove excess salt in bile samples, as precipitated bile salts can block the SPE or HPLC columns when large quantities (e.g. dog bile or pooled samples) are to be loaded. Mix aliquots of bile with acid or acidified water (∼1:1, v/v) (see Note 10). 2. Centrifuge the precipitate for 5 min at 8000×g and collect the supernatant. Repeat process several times and collect and combine the supernatants. 3. Freeze-dry supernatant extracts prior to separation by SPE or HPLC, if required, and reconstitute in the initial mobile phase. 4. To avoid loss of metabolites, additionally extract the precipitate–pellet with organic solvent (e.g. MeOH), evaporate the solvent and combine with the supernatant. 5. As an alternative approach, especially when dealing with large volumes of bile (e.g. dog bile), the sample can be freeze-dried and then extracted with an organic solvent (see Section 3.1.3)

3.1.6. SPE Chromatography

1. This section assumes a reversed-phase column (e.g. C8 or C18 ).

Nuclear Magnetic Resonance (NMR)-Based Drug Metabolite Profiling

311

2. Prepare elution solvents, such as acidified water (e.g. pH 2, ∼100 mL), MeOH (∼100 mL) and stock solutions of MeOH:acidified water mixtures with increasing MeOHcontent, to allow step-gradient elution (e.g. 20% MeOH or finer, e.g. 10% MeOH (v/v), ∼20 mL total volume each). 3. Acidify sample to pH 2 and centrifuge at +4◦ C for 10 min at 1800×g, if necessary (in case of precipitation). 4. Condition the column/cartridge before loading the sample, according to the manufacturer’s requirements. For a 3-mL C18 column, typical sorbent equilibration conditions are detailed below: (a) Elute column with MeOH (∼3 mL). (b) Follow by elution with acidified water (e.g. pH 2) (∼3 mL). 5. Make up the parent drug in an aqueous solvent acidified to pH 2. Prior to loading of the biofluid sample, ensure that the parent drug is retained on the column and elutes in the higher MeOH-fractions (e.g. 80% MeOH). 6. Once retention of parent compound is ensured, load acidified sample (onto the freshly equilibrated column). 7. Collect unretained material (known as eluate). 8. Elute with acidified water and collect fraction (known as acid wash). 9. Elute with a step gradient (e.g. 10–100% or 20–100% MeOH, v/v) and collect every fraction separately. 10. Evaporate MeOH, reconstitute in water, readjust pH of the fractions back to pH 7 and freeze-dry fractions prior to NMR analyses. 3.1.7. HPLC

1. This section assumes description of HPLC in one of the earlier chapters (here, the procedure using a reversed-phase column is described as an example). For reversed-phase columns, conditioning of the HPLC column-packing material with the mobile phase (e.g. to ensure elution of potential contaminants from previous analyses) is required prior to loading of the biofluid sample or extract. This generally means elution with, e.g. ACN and aqueous phase (both buffered with, e.g. 0.1% formic acid, v/v, or 10 mM ammonium formate), followed by a steep gradient, e.g. 10–80% ACN over 20 min. 2. Develop a suitable HPLC method to maximise retention of the parent compound and to ensure good peak shape. 3. Equilibrate the HPLC column well with the initial mobile phase, prior to loading of the sample.

312

Lenz

4. Inject small volumes of biofluid or biofluid extract and collect fractions of interest into individual glass vials. 5. Combine fractions and evaporate solvent prior to freezedrying and NMR analysis. 3.2. NMR Spectroscopy

3.2.1. 1 H-NMR Spectroscopy

These instructions assume the use of a Bruker 600-MHz NMR spectrometer; however, they should be easily adapted to other instruments. As for every NMR experiment, acquisition of spectra involves locking to the solvent and tuning and matching of each sample, followed by shimming to achieve optimum line-shape. 1. A prerequisite of structural determination of metabolites is the full structural characterisation of the parent molecule itself. Hence, ideally, the parent drug should be dissolved in the same solvent or a similar solvent system as the metabolites in order to avoid differences in chemical shifts. Once the parent spectrum has been fully assigned, spectral modifications as a consequence of metabolism can be observed and interpreted on the isolated metabolites of the chromatographic fractions with more confidence (see ref. 1). 2. Depending on the amount of metabolite extracted, several NMR experiments can be carried out to aid the structural identification of the metabolite. However, owing to the typical low concentration of the metabolites, the choice of NMR experiments can be limited. Generally, 1 H-NMR experiments involve solvent suppression. Typical 1D 1 HNMR experiments include zgpr, noesypr1d (for singlesolvent suppression) and lc1pnf2 (for double-solvent suppression), whilst typical 2D NMR experiments comprise COSY (cosyphpr) and TOCSY (mlevphpr). Heteronuclear experiments include 1 H–13 C HSQC (hsqcetgpsisp2) and HMBC (hmbcgplpndqf). 3. Dissolve the isolated metabolite in the smallest amount of suitable solvent possible (e.g. 0.2–0.5 mL). Ideally, this would be the same solvent as the parent itself, in order to compare chemical shift changes as a result of biotransformation (see Notes 1 and 11). 4. If solvent suppression is required, define the exact frequency of the solvent peak to be suppressed and set as the offset frequency (O1). Ensure that the spectral width is sufficiently large to capture all the signals. Typical values are sw = 20 ppm and td = 64 k, resulting in an acquisition time of 2.7 s. Allow a further 2.3 s for d1 between successive scans, i.e. ensuring a pulse repletion time of ∼5 s. Typically, the solvent signal itself (e.g. the water contained in D2 O) or the water peak contained in the organic

Nuclear Magnetic Resonance (NMR)-Based Drug Metabolite Profiling

313

solvent (e.g. MeOH-d4 or ACN-d3 ) might require suppression; hence, set the O1 onto the solvent peak and apply a pulse sequence such as zgpr or noesypr1d. For suppression with zgpr, set the power level for suppression (pl9) to approximately 70 dB. For noesypr1d, a mixing time of 100 ms (delay d8) is recommended and pl9 of ∼60– 70 dB for the suppression during the relaxation delay. For noesypr1d, the measurement of the 90◦ pulse width is required. 5. Small amounts of isolated metabolite (e.g. 10 μg and below) can require long acquisition times (e.g. 1–4 k scans); hence, temperature control is advisable. 3.2.2. 19 F-NMR Spectroscopy

1. This assumes that the compound dosed or administered contains a fluorine-handle, which is unlikely to be lost as a consequence of metabolism. 2. Initially, the whole biofluid can be examined to obtain a 19 F-NMR metabolite profile, revealing the number and relative concentrations of the metabolites. For this purpose, the biofluid sample may require concentrating by freeze-drying and reconstituting in a smaller volume of a suitable solvent (e.g. D2 O or MeOH-d4 ). 3. Fluorine atoms are quite sensitive to structural modifications due to metabolism (approx. up to eight bonds from the fluorine atom), causing changes in the chemical shift compared to the parent peak (Fig. 18.6; ref. 8). 4. Biofluid fractions (isolated metabolites) can be ‘screened’ for fluorine in addition to the 1 H-NMR spectrum. Although 19 F-NMR spectroscopy does not provide structural information, it gives additional evidence that the metabolite is contained in the sample (e.g. Fig. 18.8). 5. The acquisition of a 19 F-NMR spectrum involves setting up an experiment for 19 F detection (e.g. rpar 19 F or 19 FCPD) and tune and match the sample on the 19 F channel. Fluorine spectra are run with either pulse programs zg (if 1 H decoupling is not required, i.e. for a CF3 group) or zgfhigqn (for 1 H decoupling). 6. Again, acquire a spectrum of the parent compound first. This allows optimisation of the correct O1 and pulse width measurements. Typically, set O1 to –100 ppm (for monofluoro) or –60 ppm (for CF3 groups), sw = 100 ppm and td = 128 k. For 19 F-NMR experiments, pulse widths smaller than 90◦ can be used (e.g. zg30). For 1 H decoupling, the decoupling power pl12 needs to be measured or calculated (in edprosol) to give a 90◦ decoupling pulse of 80 μs (pcpd2). Cpdpg2 should be set to waltz 16.

314

Lenz

7. Depending on metabolite concentrations, lengthy acquisition times can be required; hence temperature control is advisable. 3.2.3. HPLC–NMR Spectroscopy

1. These instructions assume the use of a Bruker NMR spectrometer equipped with a dedicated HPLC system (e.g. Bruker Agilent HPLC system) operated via Bruker’s HystarTM software. This set-up also includes a dedicated LC flow probe (e.g. 3-mm 1 H/19 F dual-tunable LCDXI, with 60 μL cell volume) and a DAD-UV detector. 2. This description assumes that the transfer time between the UV detector and the flow probe is calibrated so that the peak of interest is reliably eluted into the NMR flow-probe. 3. Again, the HPLC method has to be developed and optimised with the parent compound to ensure retention of the metabolites. 4. The mobile phase typically consists of D2 O (instead of H2 O) and ACN (not necessarily deuterated), both buffered with 0.1% formic acid-d2 (v/v). 5. It is not atypical to load the entire biofluid extract to capture all the metabolites contained therein (see ref. 9 and Note 12). Hence, the biofluid extract or freeze-dried biofluid requires reconstitution in a very small volume (e.g. 100 μL) of the initial mobile phase (see Note 13). 6. During the HPLC gradient, the metabolites can be collected either in time-slice mode (in timed intervals, e.g. every 10 s) or in stop-flow mode, halting the pump at given UV peaks. 7. Once the peak is captured in the NMR flow cell, the ‘captured fraction’ requires tuning and matching, as well as reshimming and defining of the frequencies for suppression of the mobile-phase peaks (i.e. the residual water in the D2 O and the ACN peak). Double-solvent suppression is achieved with the lc1pnf2 pulse sequence set-up via rpar LC1D12. Generally, O1 is set on the larger peak (generally the ACN peak) and O2 on the residual water peak. Typical values are mixing time (d8) of 100 ms, 70 dB for pl9 (for suppression of the larger peak, defined by O1) and 80 dB for pl21 (for the smaller solvent peak, defined by O2). This is followed by ‘rga’ (automatic receiver gain adjustment) prior to acquisition of the spectra (as described in Section 3.2.1). Alternatively, this can be run in automation, rpar LC1D12, xaua (au-prog au_lc1d). 8. Again, as the acquisitions can be lengthy (1–8 k scans), depending on metabolite concentration, temperature control is advisable.

Nuclear Magnetic Resonance (NMR)-Based Drug Metabolite Profiling

315

9. Post-acquisition, the individual peaks can be collected straight from the LC probe outlet if further purification or analysis (e.g. MS) is required. 10. If the metabolites contain a fluorine handle, each chromatographic fraction can be additionally screened by 19 FNMR spectroscopy once captured inside the LC-probe. This gives supporting evidence that the metabolite in question is indeed drug related. Here, the solvent signals are invisible; hence, no solvent suppression is required. The acquisition is as in Section 3.2.2. 11. Alternatively, if the metabolites are sufficiently concentrated in the biofluid sample (e.g. rodent urine) and enough biofluid sample is available, then 19 F-NMR can be utilised in the on-flow mode in order to identify the exact retention times of the metabolites during the HPLC separation, as shown in Fig. 18.7. This takes advantage of the specificity of fluorine and the lack of interfering solvent signals. 12. For on-flow HPLC–NMR, a pseudo-2D experiment is acquired (see Note 14). This comprises the acquisition of 1D experiments (representing the F2 axis) over the period of the chromatographic separation (represented by F1). In order to mirror the chromatographic separation, the residence times of the fractions in the flow cell are limited; hence, the quality of corresponding NMR spectra is typically compromised. Typically, the flow rate is reduced (e.g. 0.5 mL/min); however, the data acquisition by NMR has to be accelerated to collect sufficient data of the fractions flowing through the flow cell. Typical NMR parameters, therefore, include a reduced number of scans (e.g. 16–24 scans per increment, a spectral width of ∼60 ppm, 32 k data points, an acquisition time of ∼1 s and a very short relaxation delay, e.g., of 1 s). The required pulse program is lc2 (without decoupling, ideal for CF3 groups) or lc2pg (if 1 H decoupling is required). The duration of the separation has to be ‘covered’ by collecting sufficient increments (in eda). Hence, calculate the time required to acquire the 1D spectrum and enter the required number of increments in F1 for the full chromatographic separation to complete. When the pseudo-2D data set is acquired (or even during the acquisition), the pseudo-2D matrix is converted with xf2. This achieves Fourier transformation in F2 only (as F1 represents a time axis). 13. After completion of the on-flow experiment, the exact retention times of the metabolites, based on the presence of the 19 F signals in the pseudo-2D matrix, can be evaluated.

316

Lenz

The separation is then repeated, stopping the separation on the signals of interest (in stop-flow mode), capturing the metabolite inside the flow cell and acquiring 1 H-NMR spectra as described in Section 3.2.1. 3.2.4. HPLC–SPE–NMR Spectroscopy

1. These instructions assume the use of a Bruker NMR spectrometer equipped with a dedicated HPLC–SPE system (Bruker Agilent Spark Prospekt 2-system) operated via Bruker’s HystarTM software. This set-up also includes a dedicated LC flow probe (e.g. 3-mm, 1 H/19 F dual-tunable LCDXI, with 30 μL cell volume) and a DAD-UV detector. 2. Here, the HPLC separation is developed as in conventional HPLC. The mobile phase is not required to be deuterated. Again, method development with the parent compound is required before loading of the sample, in order to ensure optimum retention of the metabolite peaks. 3. In SPE–NMR, it is imperative that the SPE cartridges are pre-equilibrated, prior to loading of the HPLC fractions/peaks, in order to avoid ‘breakthrough’. This requires knowledge of the number of expected metabolites to be captured. 4. Once the HPLC fractions of interest are identified, they are transferred onto the SPE cartridges, a process during which the fractions are diluted with water in order to ensure retention. 5. Several trapping steps can be carried out, i.e. the HPLC separation can be repeated three times and the fractions/peaks are trapped onto the same cartridge every time (see ref. 10). 6. The trapped SPE fractions are dried with nitrogen gas and finally eluted with a single organic solvent, typically ACNd3 , into the LC–SPE–NMR probe head, where 1 H- or 19 FNMR experiments can be acquired.

4. Notes 1. Ideally, the parent drug ought to be dissolved in the same solvent as the metabolites in order to avoid differences in chemical shifts that may occur in different solvents. However, in some cases, this is not possible, and a suitable solvent should be used that dissolves the parent fully. Generally speaking, the metabolites are more polar than the parent and might be soluble in D2 O and MeOH-d4 . Both

Nuclear Magnetic Resonance (NMR)-Based Drug Metabolite Profiling

317

of these solvents can be easily freeze-dried or evaporated if further experiments with the metabolite are required. 2. Due to the high concentrations of these agents, several repeat chromatographic ‘washing steps’ are recommended, where the analyte is loaded onto the column (SPE or HPLC) and several wash steps (elutions with the aqueous phase) are performed. 3. In order to achieve partitioning into the organic phase, pH adjustment, e.g. acidification of the biofluid prior to liquid– liquid extraction, ensures ion suppression/protonation. 4. SPE is a low-resolution chromatographic method and a range of column-packing materials and sizes are commercially available to suit the properties of the drug and its metabolites. Optimisation of retention has to be carried out with drug parent prior to loading of the biofluid matrix. 5. SPE allows large volumes of biological fluid (e.g. 20 mL) to be eluted, the fractions of which can subsequently be dried and reconstituted in small volumes of suitable solvent prior to NMR analysis or further HPLC separation. In many cases, owing to the low concentration of the metabolites excreted (urine, bile) or contained (plasma), it might be necessary to extract the total volume of the biofluid. This can be a problem in human studies when dealing with large quantities, e.g. several litres of urine collected over 24–48 h, containing the metabolites in microgram quantities. Here, samples can be initially cleaned up and separated by SPE chromatography. This allows, provided the metabolites have a range of polarities, separation of the metabolites from the majority of endogenous metabolites and also enables concentration of the metabolites. 6.

19 F-NMR

signals can be prone to chemical shift differences, mainly due to solvent or pH differences. Thus, the exact chemical shift values are not always compatible between the overall metabolite profile and the isolated metabolites.

7. Avoid buffers that give rise to NMR signals, such as acetic acid (acetate signal). Opt for ammonium formate or formic acid instead of the acetic acid equivalents. Formate has the advantage of evaporating during the freeze-drying process. For hyphenated HPLC–NMR, deuterated formic acid is generally used (see Section 2.2.2). 8. Freeze-drying and solvent evaporation are generally carried out several times during the sample preparation stages, in order to produce a dry isolated extract (ideally containing only the analyte/metabolite in question) which can then be dissolved in a suitable deuterated solvent for NMR analysis.

318

Lenz

9. Protein crashing has the advantage that the sample will be mixed with a single solvent (ACN), rather than being exposed to a filter that contains glycerol. For ultrafiltration, the filter has to be cleaned several times with water to get rid of the glycerol. Hence, due to the higher contamination risk, the ultrafiltration process is not discussed here. 10. Bile salt precipitation can lead to some loss of metabolites, which can also co-precipitate in the process. Hence, it might be advisable to dilute the sample and inject in smaller aliquots, with plenty of aqueous wash cycles prior to reloading. For SPE, the frit sitting on the surface of the sorbent bed can be punctured. 11. Ensure that the solvent signals do not obscure important signals in the spectrum. This can be assessed with the parent compound prior to the analysis of the metabolites. 12. As a precaution, the entire HPLC separation should be collected (post-NMR flow probe), in case the sample elutes in the void volume. 13. Make sure the injection loop is sufficiently large for the volume injected. Alternatively, the sample can be injected in small volumes (e.g. in 20 μL aliquots) in successive intervals, whilst keeping the mobile phase at 0% organic for ∼5–10 min, to ensure refocusing of the sample on the column head (see ref. 9). 14. Optimise the spectral parameters first with parent compound, i.e. the definition of the correct centre of spectrum (O1) and the spectral width and the correct pulse sequence (i.e. the need for proton decoupling or not). In addition, measure (or generously estimate) the total chromatographic run-time with the parent compound, as sufficient increments for F1 in eda will have to be defined to cover the separation of the biofluid sample. References 1. Braun, S., Kalinowski, H.-O., Berger, S. (1998) 150 and More Basic NMR Experiments, Wiley-VCH Verlag GMBH, 69469 Weinheim, second expanded edition, Germany. 2. Bales, J. R., Nicholson, J. K., Sadler, P. J. (1985) Two-dimensional proton nuclear magnetic resonance “maps” of acetaminophen metabolites in human urine. Clin Chem 31, 757–762. 3. McCormick, A. D., Slamon, D. L., Lenz, E. M., Phillips, P. J., King, C. D., McKillop, D., Roberts, D. W. (2007) In vitro metabolism of a tricyclic alkaloid (M445526) in human

liver microsomes and hepatocytes. Xenobiotica 37, 972–985. 4. Moffat, A. C., Jackson, J. V., Moss, M. S., Widdop, B. (eds.) (1986) Clarke’s Isolation and Identification of Drugs in Pharmaceuticals, Body Fluids, and Post-mortem Material, 2nd edn, The Pharmaceutical Press (publications division of The Pharmaceutical Society of Great Britain, 1 Lambeth High Street, London SE1 7JN). 5. Holzgrabe, U., Wawer, I., Diehl, B. (1999) NMR Spectroscopy in Drug Development and Analysis, Wiley-VCH Verlag GmbH, 69469 Weinheim, Germany.

Nuclear Magnetic Resonance (NMR)-Based Drug Metabolite Profiling 6. Lindon, J. C., Nicholson, J. K., Wilson, I. D. (1995) The development and application of coupled HPLC–NMR spectroscopy. Adv Chromatogr 36, 315–382. 7. Lindon, J. C., Nicholson, J. K., Wilson, I. D. (1996) Direct coupling of chromatographic separations to NMR spectroscopy. Prog Nucl Magn Res 29, 1–49. 8. Tugnait, M., Lenz, E. M., Hofmann, M., Spraul, M., Wilson, I. D., Lindon, J. C., Nicholson, J. K. (2003) The metabolism of 2-trifluoromethyl aniline and its acetanilide in the rat by 19f NMR monitored enzyme hydrolysis and 1 H/19 F HPLC–NMR spectroscopy. J Pharm Biomed Anal 30, 1561– 1574. 9. Lenz, E. M., D’Souza, R. A., Jordan, A. C., King, C. D., Smith, S. M., Phillips, P. J.,

319

McCormick, A. D., Roberts, D. W. (2007) HPLC–NMR with severe column overloading: fast-track metabolite identification of urine and bile samples from rat and dog treated with [14 C]-ZD6126. J Pharm Biomed Anal 43, 1065–1077. 10. Exarchou, V., Fiamegos, Y. C., van Beek, T. A., Nanos, C., Vervoort, J. (2006) Hyphenated chromatographic techniques for the rapid screening and identification of antioxidants in methanolic extracts of pharmaceutically used plants. J Chromatogr A 1112, 293– 302. 11. Martin, P. D., Warwick, M. J., Dane, A. L., Hill, S. J., Giles, P. B., Lenz, E. (2003) Metabolism, excretion, and pharmacokinetics of rosuvastatin in healthy adult male volunteers. Clin Ther 25, 2822–2834.

Chapter 19 Nuclear Magnetic Resonance (NMR)-Based Metabolomics Hector C. Keun and Toby J. Athersuch Abstract Biofluids are by far the most commonly studied sample type in metabolic profiling studies, encompassing blood, urine, cerebrospinal fluid, cell culture media and many others. A number of these fluids can be obtained at a high sampling frequency with minimal invasion, permitting detailed characterisation of dynamic metabolic events. One of the attractive properties of solution-state metabolomics is the ability to generate profiles from these fluids following simple preparation, allowing the analyst to gain a naturalistic, largely unbiased view of their composition that is highly representative of the in vivo situation. Solutionstate samples can also be generated from the extraction of tissue or cellular samples that can be tailored to target metabolites with particular properties. Nuclear magnetic resonance (NMR) provides an excellent technique for profiling these fluids and is especially adept at characterising complex solutions. Profiling biofluid samples by NMR requires appropriate preparation and experimental conditions to overcome the demands of varied sample matrices, including those with high protein, lipid or saline content, as well as the presence of water in aqueous samples. Key words: NMR spectroscopy, urine, plasma, tissue, cellular extracts, culture media.

1. Introduction Solution-state nuclear magnetic resonance (NMR) spectroscopy is an efficient profiling tool that can generate information-rich spectra that have proven success in the study of metabolism and related research and clinical areas (1–3). NMR spectroscopy possesses many general features that make it highly fit for purpose as a metabolic profiling technique. It can operate in a largely untargeted fashion, i.e. any sufficiently abundant molecule containing the nuclei of interest will be detected. As for biological NMR in general, 1 H-NMR spectroscopy provides in principle access to T.O. Metz (ed.), Metabolic Profiling, Methods in Molecular Biology 708, DOI 10.1007/978-1-61737-985-7_19, © Springer Science+Business Media, LLC 2011

321

322

Keun and Athersuch

many metabolites, given the ubiquitous presence of these nuclei in biochemistry. Other NMR-visible nuclei such as 13 C, 31 P and 15 N can also be exploited to a lesser extent and provide complementary information. The resonances produced are very sensitive to structure, so many compounds can be identified in spectral data simultaneously and the fine structure of the signals in the frequency domain provides strong structural clues for the identification of unknown metabolites. In addition, routinely available technical extensions, such as multidimensional NMR, give further detailed structural information. Where stable isotopes are used to target one or several metabolic pathways, it is possible to distinguish isotopomers and isotopologues and thus gain information on biotransformation and pathway flux. Crucially, the data generated are amenable to quantitative interpretation across a wide dynamic range. Despite being relatively insensitive compared to other forms of spectroscopy and mass spectrometry, NMR remains highly competitive in many applications, due in part to the quality of the structural information it can provide. However, the most unique feature of NMR spectroscopy is that it can analyse the ‘native’ tissue or biofluid with minimal preparation and in an essentially non-destructive manner. This has many important consequences: NMR spectroscopy tends to be more analytically reproducible and robust than other platforms (4, 5); it also allows the measurement of metabolites non-invasively in vivo. These characteristics in turn make NMR spectroscopy a useful tool in translational research, i.e. taking results from the bench into practical use in the field or clinic (1, 6). One of the key advantages of metabolic profiling as a means of biomarker discovery (and metabolic biomarkers themselves) over genomic and proteomic counterparts is that metabolites are a defined chemical entity irrespective of species, genotype, localisation and biological matrix. In principle therefore, analytical procedures should be more translatable and the particular properties of NMR spectroscopy enhance that advantage. There are two general problems faced in biofluid NMR that have led to a preferential use of two pulse sequences for metabolic profiling, namely the 1D nuclear Overhauser effect spectroscopy with presaturation (NOESYpresat) (7) and the 1D Carr–Purcell– Meiboom–Gill (CPMG) (8) sequences. The first issue is adequate suppression of the solvent resonance. Although very effective solutions using excitation sculpting (9) exist, presaturation continues to be the most simple and prevalent option, applied during the relaxation delay. Field inhomogeneity decreases suppression efficiency, and so a single increment of the standard NOESY pulse sequence is typically used, rather than the single pulse experiment, to reduce contributions from regions of the active volume that

Nuclear Magnetic Resonance (NMR)-Based Metabolomics

323

experience an incomplete 90◦ pulse, thus reducing the residual water resonance. The second common experimental issue in biofluid NMR is the discrimination between metabolites of low molecular weight (typically <1000 Da) and macromolecular species in the sample (e.g. proteins, lipoproteins). Macromolecules produce broad resonances, due to reduced rotational diffusion (‘tumbling’) and short T2 relaxation times, which confound spectral interpretation. Commonly, T2 editing via the CPMG experiment is used to reduce the contribution of high molecular weight species in the resulting spectra. This is the experiment of choice for a number of important biofluids, including serum or plasma. An example of the use of these experiments is shown in Fig. 19.1. In contrast to routine NMR experiments for other purposes (e.g. structure elucidation in synthetic chemistry laboratories, protein structure and phosphorylation studies, process chemistry), the goals of metabolomic analysis require highly consistent procedures for collection, preparation, storage and analysis of all samples. Any differences in protocol, even if very minor, can potentially produce confounding systematic variation within and between analytical batches, particularly in long-term studies where sample collection may be conducted across several sites and

(a)

(b)

Fig. 19.1. Comparison of human serum spectra acquired using (a) NOESYpresat and (b) CPMG pulse sequences. The most striking difference is the reduced contribution in the CPMG spectra from broad background resonances attributable to high-abundance serum proteins, a result of the spin echo used to edit protons with short transverse (T2 ) relaxation times.

324

Keun and Athersuch

over many years. These studies are already prone to unavoidable sources of bias, and therefore to maximise the chances of observing biologically relevant changes, particular care should be taken to randomise the order of preparation and analysis and to record and review factors such as freeze/thaw cycles and time to analysis that could influence spectral data (e.g. see (10, 11)).

2. Materials 2.1. Sample Preparation Reagents

1. Na2 HPO4 , 99+% ACS, anhydrous. 2. NaH2 PO4 , 99%, anhydrous. 3. 3-(Trimethylsilyl)propionic-2,2,3,3-d4 acid sodium salt, 98 at.% D (Sigma-Aldrich, Poole, UK). 4. Sodium azide (NaN3 ). 5. D2 O. 6. Water, LC-MS grade. 7. Chloroform, AnalaR grade 99%. 8. Methanol, LC-MS grade. 9. Acetonitrile, LC-MS grade. 10. CDCl3 , ‘100%’ grade, 99.96 at.% D, contains 0.03% (v/v) TMS (Sigma-Aldrich, Poole, UK). 11. Methanol-d4 , 99.8%.

2.2. Sample Preparation Equipment

1. TissueLyser sample disruptor/homogeniser Crawley, UK, or similar).

(Qiagen,

2. 5-mm stainless steel beads (Qiagen or similar). Note: The efficiency of the sample disruption/homogenisation is influenced by numerous factors including type and quantity of sample and bead composition and size. Therefore, optimal homogenisation conditions should be derived by the analyst as part of method development. For further information, see the instrument manufacturer recommendations. 3. 2-mL reaction tubes (Greiner Bio-One, Stonehouse, UK, or similar). Note: Avoid the use of narrow tapered 1.5-mL sample tubes in combination with larger homogeniser beads as they may become jammed resulting in incomplete homogenisation or sample loss. 4. Plain blood collection tubes (Becton Dickinson, Oxford, UK, or similar).

Nuclear Magnetic Resonance (NMR)-Based Metabolomics

325

5. Lithium–heparin blood collection tubes (Becton Dickinson or similar). 6. Concentrator plus rotary evaporator (Eppendorf, Cambridge, UK, or similar). 7. LyoLab 3000 sample freeze drier (Heto Holten, Camberley, UK, or similar). 8. E-LAB2 rotary pump (Edwards, Crawley, UK, or similar). 9. Eppendorf 1.5 mL (Eppendorf or similar). 10. Riplate, 1-mL 96-well sample plate, with sealing mats (Ritter, Schwabmünchen, Germany, or similar). 11. Harrier 18/80 refrigerated centrifuge (Sanyo, Loughborough, UK, or similar). 12. Cell scraper for 75 cm2 vessels, 18-mm blade (Becton Dickinson or similar). 13. Nanosep centrifugal filter device, 10 kDa molecular weight cut-off (Pall Life Sciences, Portsmouth, UK, or similar). 2.3. NMR Equipment

NMR experiments in metabolic profiling applications can be (and are) conducted on instruments across a wide magnetic field strength range. For the purpose of these protocols, we assume the use of a 600-MHz (14.1 T) magnet. These instruments provide a good compromise in terms of cost benefit. 1. 600-MHz Avance NMR spectrometer (Bruker Biospin, Karlsruhe, Germany, or similar, Varian, Jeol, etc.). 2. FI TXI 600 SB 5 mm with Z-gradient flow injection NMR probe (Bruker Biospin or similar). 3. TXI 600-MHz S3 5-mm XYZ gradient tube probe (Bruker Biospin or similar). 4. Gilson 215 flow injection system (Gilson, Inc., Middleton, USA, or similar) with ICONNMR (Bruker Biospin or similar). 5. B-ACS60 tube NMR autosampler (Bruker Biospin or similar). 6. Gilson 215 sample preparation robot (Gilson, Inc. or similar) with SampleTrack (Bruker Biospin or similar).

3. Methods The methods described here assume the use of 600-MHz 1 HNMR spectroscopy with a 5-mm tube NMR probe and describe analyses for the most common biological samples. However, these may be easily adapted for many other applications.

326

Keun and Athersuch

3.1. Method A: Urine (Also Suitable for Low-Protein Samples in General, Aqueous Cell Extracts)

1. Sample collection. a. Ideally, prepare sample collection vessels containing 100 μL of 0.02% NaN3 (w/v) as preservative. b. Where appropriate, samples can be prefiltered using a 0.2μm filter and syringe to remove cellular material during collection. c. Store at –40◦ C or below and transport on dry ice. 2. Sample preparation. a. Defrost sample at room temperature and vortex to mix. b. Combine 400 μL of sample and 200 μL of urine buffer (0.2 M Na2 HPO4 , 0.043 M NaH2 PO4 , 1 mM TSP, 3 mM NaN3 , 80% H2 O and 20% D2 O) and mix (see Note 1). c. Centrifuge sample for 10 min at >10,000×g. d. Transfer 550 μL of sample to a 5-mm NMR tube, taking care not to disturb any pelleted material (see Note 2). 3. General instrument set-up (Steps a–f are readily automated, see Note 3). a. Set probe temperature (e.g. 300 K), insert sample and wait for temperature equilibration (∼5 min). b. ‘Lock’ the instrument to the D2 O resonance. c. Tune and match the probe. d. Adjust shims to optimise spectral lineshape. Half-height linewidth of ∼1 Hz should be readily obtainable on samples with low protein content. e. Using a single pulse experiment with presaturation, determine the 90◦ pulse length and optimise the spectrometer frequency offset to minimise the residual solvent resonance. f. Optimise the receiver gain to remove digitisation errors while not exceeding the dynamic range of the receiver. g. Select suitable recycle delay (RD) for total recycle time to be of the order of 5∗T1 and for suitable water suppression. Typical parameters at 600 MHz 1 H frequency might be an RD of 2 s and a total acquisition time (AQ) of 2.73 s recorded into 64 K complex data points to give a spectral width of ∼12 kHz. h. Presaturation pulse power should be the minimum required to achieve the necessary reduction in the water resonance, e.g. equivalent of 25 Hz bandwidth. 4. Specific pulse sequence optimisation. a. Change to ‘NOESYpresat’ pulse sequence (RD90◦ -3 μs-90◦ -tm -90◦ -AQ) for more effective solvent

Nuclear Magnetic Resonance (NMR)-Based Metabolomics

327

suppression. A typical mixing time (tm ) is 100 ms (see Note 4). b. Reoptimise the receiver gain if required. c. Accumulation of 128 scans should provide an adequate signal-to-noise (s/n) ratio in ∼11 min. 5. Spectral processing. a. Typically, application of 0.3 Hz exponential linebroadening transformation and zero filling to at least 64 K data points if fewer are recorded is beneficial prior to Fourier transformation (FT). b. After FT, reference the chemical shift scale to TSP (see Note 5) and apply zero- and first-order phase correction and a global linear baseline correction. Examples of typical urine NMR spectra are shown in Fig. 19.2. 3.2. Method B: Plasma/Serum (Also Suitable for Other High-Protein Samples, e.g. Cyst Fluid)

1. Sample collection (also see (10)). a. Collect blood into heparinised (plasma) or plain (serum) collection tubes. b. Centrifuge within a defined time interval and note clot contact time (ideally <30 min). c. Store at –40◦ C or below and transport on dry ice. 2. Sample preparation (see Note 6). a. Defrost samples on ice and mix. b. Combine 200 μL of sample and 400 μL of saline (0.9% (w/v) NaCl) in 90:10 H2 O/D2 O in a sample tube and mix. c. Centrifuge for 10 min at >10,000×g at 4◦ C to remove suspended debris. d. Transfer 550 μL of the supernatant to a 5-mm NMR tube taking care not to disturb any pelleted material. 3. General instrument set-up. a. See Method A, Step 3. 4. Specific pulse sequence optimisation. In addition to the 1D ‘NOESYpresat’ (Method A, Step 4), the CPMG pulse sequence (RD-90◦ -(τ -180◦ -τ ) n -AQ) with presaturation can be used for suppression of macromolecular signals on the basis of T2 editing. Optimise τ and n to achieve desired suppression of macromolecular signals; τ is often kept fixed and n varied. Typical parameters for blood serum/plasma are τ =400 μs and n=300 giving a total mixing time of ∼240 ms (see Note 7). Other possible profiling experiments include J-resolved and diffusion-edited pulse sequences (see Beckonert et al. (13)). 5. Spectral processing. As for Step 5 of Method A, with the exception that spectral calibration is often via the glucose

328

Keun and Athersuch

TMA

(a) taurine

succinate

creatine + creatinine

citrate

creatine

3-hydroxypropionic acid creatinine DMA acetate

lysine

(b) creatinine

2-OG

2-OG

TMAO

creatinine

hippurate trans-aconitate

glycine N-methylnicotinamide + N-methylnicotinic acid

lactate alanine

4

3

2

1

0

Fig. 19.2. Typical 600-MHz 1 H NOESYpresat NMR spectra of urine from (a) mouse and (b) rat. Clear species differences can be seen in the relative concentrations of metabolites in these urinary profiles. Key: DMA, dimethylamine; DMG, dimethylglycine; MA, methylamine; TMA, trimethylamine; TMAO, trimethylamine-N-oxide. Reproduced from Bollard et al. (12).

alpha-anomeric doublet resonance at 5.233 ppm, a process that can be automated (14). 3.3. Method C: Tissue Extracts

1. Sample collection. a. Tissues should be snap frozen using liquid nitrogen and the time to freezing controlled to minimise the effects of ischemia, etc.

Nuclear Magnetic Resonance (NMR)-Based Metabolomics

329

b. Store at –80◦ C or below and transport on dry ice. 2. Sample extraction (see Note 8). a. Weigh frozen tissue into an Eppendorf or glass vial (a minimum of 20 mg). b. Add cold CHCl3 /MeOH (2:1) solution (300 μL assuming a 20–30 mg sample) and manually grind the tissue or preferably use a tissue homogeniser. c. Add an equivalent volume (300 μL) of HPLC-grade water and vortex mix. Keep the homogenate on ice. d. Centrifuge the homogenate for 10 min at >10,000×g. e. The lower organic (CHCl3 ) and upper aqueous (methanol/water) phases should be clearly separated by an insoluble interface. Carefully isolate the two layers by pipetting the aqueous (upper) phase and organic (lower) phase into separate clean glass vials. f. Repeating the extraction with the same volume of solvents and pooling the second set of fractions with the first will increase recovery. g. Remove the organic solvents from the samples using a speed vacuum concentrator/rotary evaporator or a stream of nitrogen. h. Freeze-dry the aqueous phase to remove residual water. i. Perform a ‘blank’ extraction procedure on an empty sample tube to allow a baseline spectrum to be obtained (allows identification of contamination during preparation). 3. NMR sample preparation. a. Reconstitute the aqueous extract in 600 μL of reconstitution buffer (0.2 M Na2 HPO4 , 0.043 M NaH2 PO4 , 1 mM TSP, 3 mM NaN3 and 100% D2 O). b. Transfer to a sample tube and centrifuge for 5 min at >10,000×g). c. Pipette 550 μL into an NMR tube. d. Reconstitute the organic extract in 700 μL of CDCl3 (containing 0.03% (v/v) TMS). e. Transfer to a sample tube and centrifuge for 5 min at >10,000×g. f. Pipette 600 μL into an NMR tube. 4. General instrument set-up. a. See Method A, Step 3. 5. Specific pulse sequences. a. For the aqueous sample, the NMR acquisition can proceed as for urine or blood serum (Method B, Step 4) with

330

Keun and Athersuch

either NOESYpresat or CPMG sequences used. Residual protein signal can be suppressed by the use of the CPMG sequence with a short T2 relaxation time (e.g. 64 ms). For either experiment, an increased number of scans (e.g. 256) is recommended. b. For the organic sample the residual solvent resonance is small and hence a single pulse experiment can be used. 3.4. Method D: Cell Extracts and Culture Media

1. Sample generation. a. For a monolayer culture of mammalian cells, typically >106 cells are required, with 4–5 × 106 giving adequate concentrations of metabolites. Cells can be cultured in 75-cm2 flasks with 12 mL of media, with each flask yielding a single biological replicate (see Notes 9 and 10). 2. Sample collection (also see (15)). a. Remove cultures from the incubator and aspirate the media. b. If desired, retain the media in a sterilised tube, centrifuge (4◦ C, 4 min, 150×g) to pellet dead cells and freeze supernatant for later analyses. c. Wash the cells (add–distribute–remove) using 1 mL of cold (4◦ C) PBS to remove residual media. Discard, wash and repeat. d. Lyse cells and quench metabolism by adding 1 mL cold methanol (4◦ C) to the culture vessel. e. After 2 min, detach cellular material from the culture vessel using a cell scraper. f. Transfer the resulting suspension of cellular material to an Eppendorf or glass tube and dry down the sample using a solvent evaporator. 3. Sample preparation. a. The resulting cell pellet can be extracted using the same procedure as for whole tissue (Method C, Step 2) without the need for grinding or sonication. b. The residual pellet after extraction can be used for sample normalisation (see Note 11). 4. NMR sample preparation. a. Aqueous and organic cell extracts can be prepared as for tissue extracts, (Method C, Step 3) but for mass-limited samples, it may be advisable to dilute the reconstitution buffer two to fivefold. b. For culture media, mix 550 μL of sample with 50 μL of 0.2 % (w/v) TSP in 100% D2 O before transferring to a 5-mm NMR tube.

Nuclear Magnetic Resonance (NMR)-Based Metabolomics

331

Fig. 19.3. Typical 600-MHz 1 H CPMG spin-echo NMR spectrum of an aqueous intracellular extract obtained from cultured primary rat hepatocytes. Key: GPC, glycerophosphocholine; PC, phosphocholine; GLY, glycogen; GLU, glucose; GSH, glutathione. Adapted from Ellis et al. (16).

5. General instrument set-up. a. As for Method A, Step 3. 6. Specific pulse sequences. a. As for Method C, Step 5. The culture media can be treated similarly to the aqueous cell extract (see Note 12). An example of a typical cell extract spectrum is shown in Fig. 19.3.

4. Notes 1. The pH of samples can be finely adjusted by the addition of small volumes (5–10 μL) of 1 M HCl or 1 M NaOH. Probes for measuring pH directly in NMR tubes can also be obtained. 2. Many sample types including urine can be conveniently stored frozen in NMR tubes down to –40◦ C prior to analysis at the expense of one freeze/thaw cycle. Breakages can be avoided by ensuring the NMR tube is tightly capped, inverted so that the sample is at the head of the tube against the cap and the tube placed horizontally during freezing. 3. Where resources allow, the use of cooled sample holders will reduce the opportunity for changes in sample composition/degradation during prolonged wait times in well plate racks or in a sample carousel. 4. In standard implementations of NOESYpresat acquisitions, presaturation occurs during the mixing time (tm ) in addition to the prescan delay (RD). Mixing time presaturation

332

Keun and Athersuch

does not have a significant impact on spectral quality, but should be kept constant across an analytical batch. 5. Several chemical shift reference substances are used in solution-state NMR of aqueous samples. Most commonly used is 3-(trimethylsilyl)propionic-2,2,3,3-d4 acid sodium salt (TSP) (17), although 4,4-dimethyl-4-silapentane-1sulfonic acid (DSS) and 4,4-dimethyl-4-silapentane-1ammonium trifluoroacetate (DSA) (18) are also used. An additional shift reference that can be used in combination with one invariant with pH can be used to accurately determine sample pH (1,1-difluoro-1trimethylsilanyl methylphosphonic acid, DFTMP) (19). 6. Broad background resonances attributable to protein can also be reduced by deproteinisation via methanol precipitation or MW filtering (e.g. 10-kDa filter, see Section 2.2). 7. In NMR experiments using spin echoes, the optimal τ value will vary with field strength. Typically, at 600 MHz (14.1 T), a τ value between 300 and 400 μs will be optimal. This value can be determined on a single representative sample. Long CPMG sequences can induce heating in the sample and cause broadening of peaks. Increasing the number of dummy scans (e.g. 16–32) can remedy this problem by improving thermal equilibration at the start of the acquisition. 8. Solvents used for sample extraction and reconstitution should be checked for purity prior to use to minimise the inadvertent introduction of contaminants. For organic extraction buffers, a larger aliquot should be dried down under N2 and then any remaining material reconstituted in the NMR solvent (e.g. CDCl3 ). 9. Cell cultures should be harvested at a similar level confluence. 10. Profiles of media, media with growth supplements added and a dummy incubation (no cells) should accompany cell culture experiments to aid spectral assignment, indicate any changes in media independent of cellular activity and help identify sources of unwanted contamination (e.g. nonsterile media). 11. The insoluble interface of the liquid–liquid extraction procedure can be used for estimating sample protein. Most commonly, the bicinchoninic acid (BCA) assay is used – a colorimetric method assay with a wide linear range and detection limit of around 5 μg/mL that can be conducted easily in 96-well plate format. Calibrated protein measurements are a useful measure for the normalisation of cell culture materials and can be used to approximate cell number.

Nuclear Magnetic Resonance (NMR)-Based Metabolomics

333

12. Organic solvents are commonly found at relatively high concentrations in cell culture media samples when used as the solvent for chemical exposure experiments. Solvent signals in NMR spectra can often be successfully suppressed by the use of multiple presaturation or excitation sculpting pulse sequences. Alternatively, the sample may be dried down using a solvent evaporator, but this may present problems where the solvent has a high boiling point (e.g. DMSO), requiring severe conditions that may compromise sample quality.

Acknowledgements Development of the protocols described and the preparation of the manuscript was funded in part by the European Union FP6 carcinoGENOMICS project (contract number PL037712) and a Cefic Long-range Research Initiative (LRI) Innovative Science Award. The authors wish to acknowledge Dr. Olaf Beckonert, Dr. James Ellis and Dr. Orla Teahan (Imperial College London) for their useful help and advice. References 1. Keun, H. C., Athersuch, T. J. (2007) Application of metabonomics in drug development. Pharmacogenomics 8, 731–741. 2. Lindon, J. C., Nicholson, J. K. (2008) Spectroscopic and statistical techniques for information recovery in metabonomics and metabolomics. Annu Rev Anal Chem 1, 45–69. 3. Coen, M., Holmes, E., Lindon, J. C., Nicholson, J. K. (2008) NMR-based metabolic profiling and metabonomic approaches to problems in molecular toxicology. Chem Res Toxicol 21, 9–27. 4. Keun, H. C., Ebbels, T. M. D., Antti, H., Bollard, M. E., Beckonert, O., Schlotterbeck, G., Senn, H., Niederhauser, U., Holmes, E., Lindon, J. C., Nicholson, J. K. (2002) Analytical reproducibility in H-1 NMR-based metabonomic urinalysis. Chem Res Toxicol 15, 1380–1386. 5. Dumas, M. E., Maibaum, E. C., Teague, C., Ueshima, H., Zhou, B., Lindon, J. C., Nicholson, J. K., Stamler, J., Elliott, P., Chan, Q., Holmes, E. (2006) Assessment of analytical reproducibility of 1H NMR spec-

6.

7.

8.

9.

10.

troscopy based metabonomics for large-scale epidemiological research: the INTERMAP study. Anal Chem 78, 2199–2208. Bundy, J. G., Iyer, N. G., Gentile, M. S., Hu, D. E., Kettunen, M., Maia, A. T., Thorne, N. P., Brenton, J. D., Caldas, C., Brindle, K. M. (2006) Metabolic consequences of P300 gene deletion in human colon cancer cells. Cancer Res 66, 7606–7614. Neuhaus, D., Ismail, I. M., Chung, C. W. (1996) ‘FLIPSY’ – a new solvent-suppression sequence for nonexchanging solutes offering improved integral accuracy relative to 1d NOESY. J Magn Reson Ser A 118, 256–263. Meiboom, S., Gill, A. (1958) Modified spin-echo method for measuring nuclear relaxation times. Rev Sci Instrum 29, 688–691. Hwang, T. L., Shaka, A. J. (1995) Water suppression that works: excitation sculpting using arbitrary wave-forms and pulsed-field gradients. J Magn Reson 112, 275–279. Teahan, O., Gamble, S., Holmes, E., Waxman, J., Nicholson, J. K., Bevan, C., Keun, H. C. (2006) Impact of analytical bias

334

11.

12.

13.

14.

Keun and Athersuch in metabonomic studies of human blood serum and plasma. Anal Chem 78, 4307– 4318. Griffin, J. L., Nicholls, A. W., Daykin, C. A., Heald, S., Keun, H. C., Schuppe-Koistinen, I., Griffiths, J. R., Cheng, L. L., RoccaSerra, P., Rubtsov, D. V., Robertson, D. (2007) Standard reporting requirements for biological samples in metabolomics experiments: mammalian/in vivo experiments. Metabolomics 3, 179–188. Bollard, M. E., Keun, H. C., Beckonert, O., Ebbels, T. M., Antti, H., Nicholls, A. W., Shockcor, J. P., Cantor, G. H., Stevens, G., Lindon, J. C., Holmes, E., Nicholson, J. K. (2005) Comparative metabonomics of differential hydrazine toxicity in the rat and mouse. Toxicol Appl Pharmacol 204, 135–151. Beckonert, O., Keun, H. C., Ebbels, T. M. D., Bundy, J. G., Holmes, E., Lindon, J. C., Nicholson, J. K. (2007) Metabolic profiling, metabolomic and metabonomic procedures for NMR spectroscopy of urine, plasma, serum and tissue extracts. Nat Protoc 2, 2692–2703. Pearce, J. T., Athersuch, T. J., Ebbels, T. M., Lindon, J. C., Nicholson, J. K., Keun, H. C. (2008) Robust algorithms for automated

15.

16.

17.

18.

19.

chemical shift calibration of 1d 1H NMR spectra of blood serum. Anal Chem 80, 7158–7162. Teng, Q., Huang, W. L., Collette, T. W., Ekman, D. R., Tan, C. (2009) A direct cell quenching method for cell-culture based metabolomics. Metabolomics 5, 199–208. Ellis, J. K., Chan, P. H., Doktorova, T., Athersuch, T. J., Cavill, R., Vanhaecke, T., Rogiers, V., Vinken, M., Nicholson, J. K., Ebbels, T. M. D., Keun, H. C. (2010) Effect of the histone deacetylase inhibitor trichostatin a on the metabolome of cultured primary hepatocytes. J Proteome Res 9, 413– 419. Pohl, L., Eckle, M. (1969) Sodium 3trimethylsilyltetradeuteriopropionate, a new water-soluble standard for 1H-NMR. Angew Chem Int Ed Engl 8, 381. Nowick, J. S., Khakshoor, O., Hashemzadeh, M., Brower, J. O. (2003) DSA: a new internal standard for NMR studies in aqueous solution. Org Lett 5, 3511–3513. Reily, M. D., Robosky, L. C., Manning, M. L., Butler, A., Baker, J. D., Winters, R. T. (2006) DFTMP, an NMR reagent for assessing the near-neutral ph of biological samples. J Am Chem Soc 128, 12360–12361.

Chapter 20 Slow Magic Angle Sample Spinning: A Non- or Minimally Invasive Method for High-Resolution 1 H Nuclear Magnetic Resonance (NMR) Metabolic Profiling Jian Zhi Hu Abstract High-resolution 1 H magic angle spinning nuclear magnetic resonance (NMR), using a sample spinning rate of several kilohertz or more (i.e., high-resolution magic angle spinning (hr-MAS)), is a wellestablished method for metabolic profiling in intact tissues without the need for sample extraction. The only shortcoming with hr-MAS is that it is invasive and is thus unusable for non-destructive detections. Recently, a method called slow MAS, using the concept of two-dimensional NMR spectroscopy, has emerged as an alternative method for non- or minimally invasive metabolomics in intact tissues, including live animals, due to the slow or ultra-slow sample spinning used. Although slow MAS is a powerful method, its applications are hindered by experimental challenges. Correctly designing the experiment and choosing the appropriate slow MAS method both require a fundamental understanding of the operation principles, in particular the details of line narrowing due to the presence of molecular diffusion. However, these fundamental principles have not yet been fully disclosed in previous publications. The goal of this chapter is to provide an in-depth evaluation of the principles associated with slow MAS techniques by emphasizing the challenges associated with a phantom sample consisting of glass beads and H2 O, where an unusually large magnetic susceptibility field gradient is obtained. Key words: High-resolution 1 H-NMR metabolomics, tissues, organs, live animals, slow magic angle spinning, magic angle turning, magnetic susceptibility, line broadening, molecular diffusion.

1. Introduction Magnetic resonance imaging (MRI) is widely used for clinical diagnosis of malignancies in various tissues and organs (1, 2). However, MRI is mainly useful for detecting malignancies when tumors have already developed to a relatively large size, e.g., a few T.O. Metz (ed.), Metabolic Profiling, Methods in Molecular Biology 708, DOI 10.1007/978-1-61737-985-7_20, © Springer Science+Business Media, LLC 2011

335

336

Hu

hundreds of micrometers or more depending on the sensitivity of the spectrometer and the methods used. Thus, MRI is not an effective method for early diagnosis. Since biochemical changes in the diseased tissues precede tumor formation, early diagnosis can be achieved if molecular-level information is obtained. These detailed biochemical changes can, in principle, be measured using high-resolution nuclear magnetic resonance (NMR) methods through ex vivo analysis of chemical extracts of excised tissues (3). The ex vivo process usually starts with lysing the cells, followed by extracting the cell lysate with organic solvents. Finally, standard high-resolution, liquid-state NMR can be used to analyze the extracted molecular entities. Although impressive spectral resolution can be obtained, standard ex vivo methods involve extensive sample preparation and are therefore prone to artifacts induced by incomplete sample extraction and fractionation, as well as sample degradation during this lengthy process (4). Like solids, tissues and cells cannot be directly analyzed by standard liquid-state NMR spectroscopy due to the line broadenings induced by the variation of local magnetic field gradients at the compartment boundaries in cells and tissues (5), and to a lesser extent by the residual homonuclear dipolar interactions and residual chemical shift interactions. However, when spinning the sample about an axis at the magic angle (54◦ 44 ) and using a sample spinning rate of several kilohertz or more, all of the line broadenings can be effectively averaged out, resulting in a high-resolution 1 H-NMR spectrum. High-resolution magic angle spinning (hr-MAS) 1 H-NMR has been successfully applied to analyze intact cells and tissues from organs such as brain, lung, kidney, heart, and muscle (4, 6–12), and the method has been reviewed recently in The Handbook of Metabonomics and Metabolomics (13). With hr-MAS, high-spectral-resolutionapproaching liquid-state NMR has been achieved. The major advantage of hr-MAS 1 H-NMR over other methods for tissue sample analysis is that there is minimal sample preparation and thus less artifacts and better correlation with in vivo techniques. However, the large centrifugal force associated with the fast sample spinning rates destroys the tissue structure and even some cell types (5), which makes the method unusable for nondestructive metabolic profiling or localized spectroscopy in live animals. Hence, it is important to develop alternative methods where the spinning speed can be reduced. However, the spinning rate cannot be arbitrarily reduced. A problem with traditional MAS at slow sample spinning is that it gives rise to numerous, overlapping spinning sidebands (SSBs), which renders analysis of the spectra difficult or impossible. Fortunately, in solid-state NMR, many methods have been developed to overcome this problem. A few of these methods have the potential to be extended to biological fluid samples. These methods include 2D phase-adjusted spinning sidebands (PASS) (14)

Slow Magic Angle Sample Spinning

337

and 2D phase-corrected magic angle turning (PHORMAT) (15). It has been shown (16–18) that nearly sideband-free isotropic spectra could be obtained with PASS in a variety of excised intact organs at spinning speeds as low as 43 Hz and with a spectral resolution comparable to or better than the resolution obtained with fast MAS (19). 1 H PASS using a sample spinning rate of less than 100 Hz has been successfully utilized to study live cells directly from agar growth plate (20) and intact food seeds such as oil seeds. While the frequency used in 1 H PASS is already two orders of magnitude lower than the frequencies employed in hr-MAS, it is still too large to keep larger size soft tissues and organs undamaged. Furthermore, in PASS the magnetization is constantly present in the transverse plane, and the first signal is observed after one rotor period (16). This means that the amplitude of the signal is reduced as a result of the decay of the magnetization during this period, which is governed by the spin–spin relaxation time T2 (16). Therefore, serious signal attenuation occurs when the spinning rate is comparable to or less than (1/T2 ). Moreover, spectral distortions may occur if the T2 values of the different spectral lines are not the same. For example, in brain tissue, where T2 values of the metabolites are of the order of 100–450 ms (21), we found that PASS was limited to spinning speeds of 10–20 Hz or larger. Additionally, the homonuclear J-coupling may complex the spectrum if the spinning frequency becomes comparable to the J-coupling constant because use of the 180 pulses will not refocus the homonuclear J-coupling during the evolution dimension of the PASS experiment. In contrast, T2 attenuation is avoided in a PHORMAT experiment. It has been demonstrated (22) that PHORMAT, applied with a 1 Hz spinning speed, produces spectra of excised rat liver with a resolution approaching that obtained from PASS or hrMAS methods. Because of the ultra-slow sample spinning used, 1 H PHORMAT represents one of the practical ways for obtaining high-resolution 1 H spectroscopy of the metabolites in live whole animals. In fact, a high-resolution metabolite spectrum of whole mouse has been successfully acquired at 2T field using a sample spinning rate of 1.5 Hz (23). Recently, a localized 1 H-PHORMAT, namely the LOCMAT experiment, has been developed for obtaining localized metabolite spectra inside a live mouse (24, 25). The potential application of 1 H PASS and PHORMAT/LOCMAT in metabolomics and the directions for further improving the methods have been extensively discussed recently (26) and will not be duplicated in this article. The focus of the current article is on understanding the operation principles of the methods, in particular the details of line narrowing at the presence of molecular diffusion, that have not yet been reported in detail previously. Understanding these principles is critically important both for correctly running the slow MAS experiment and for choosing the right type of slow MAS method.

338

Hu

2. Methods 2.1. The Basic Principles of a 1 H PHORMAT Experiment

The basic principle of a PHORMAT experiment acting on the magnetic susceptibility field is best illustrated with the prototype magic angle turning (MAT) pulse sequence (15) given in Fig. 20.1. The pulses labeled as r1 , r2 , r3 , and r4 are synchronized to the one-third of the rotor period T. p1 , p2 , and p3 are projection pulses, which project either the cos(ι ) or the sin(ι ), where ι = 1,2,3, component of the procession magnetization during the corresponding t1 /3 period to the z-axis. A free induction decay (FID) is acquired following the last 90◦ read out pulse (r4 ). With proper phase cycling (15, 22) of the projection pulses p1 , p2 , and p3 and the receiver, it is possible to obtain FID(t2 , t1 ) = exp(−i(1 (t1 /3) + 2 (t1 /3) + 3 (t1 /3))FID(t2 ) [1] The precession angles 1, 2 , and 3 can be written in the usual way as time integrals of the resonant angular frequency ω(r, t) = ωiso + γ Bz (r, t), where ωiso is the isotropic frequency of interest and Bz (r, t) is the susceptibility field at the observation point r in the rotor frame, which is a function of time due to sample rotation: 1 = ωiso t1 /3 + 2 = ωiso t1 /3 + = ωiso t1 /3 +

= ωiso t1 /3 +

r1

p1

r2

t1/3

Φ1 0

T /3+t1 /3

[2b]

0

p2

Φ2 T/3

Fig. 20.1. The prototype MAT experiment.

+ T /3)dt

γ Bz (r, t)dt

2T /3 t1 /3

[2a]

γ Bz (r, t)dt

t1 /3

[2c] γ Bz (r, t + 2T /3)dt

r3

t1/3 L

γ Bz (r, t)dt

γ Bz (r, t 0 2T /3+t1 /3

0

T /3

3 = ωiso t1 /3 +

t1 /3

p3

r4

t1/3 L

Φ3 2T/3

t2/3 L

T

Slow Magic Angle Sample Spinning

We obtain

1 + 2 + 3 = ωiso t1 +

t1/3

0

γ [Bz (r, t) + Bz (r, t + T /3)

339

[3]

+Bz (r, t + 2T /3)] dt The term (Bz (r, t) + Bz (r, t + T /3) + Bz (r, t + 2T /3)) inside the integral of Eq. [3] represents a summation of the susceptibility field at the observation point r located in the rotor frame at three chosen times that are exactly one-third of rotor period apart or that are exactly 120◦ apart around the circle of rotation. If the axis of rotation is at the magic angle, i.e., at an angle of 54◦ 44” relative to the main field B0 , such a rotation will place B0 along three perpendicular axes, i.e., the x, y, and z. Since the effect of sample rotation relative to the magnetic field is equivalent to keeping the sample static while the magnetic field rotates, we found that it is convenient to adapt the concept of rotating the magnetic field to find the value of the summation in the integrant of Eq. [3]. In the following, we will prove that the summation (Bz (r, t) + Bz (r, t + T /3) + Bz (r, t + 2T /3)) is zero for a biological tissue sample of arbitrary shape so that a high-resolution 1 H-NMR metabolite spectrum free from the magnetic susceptibility-induced line broadening is obtained. To reach this goal, the analytical equation Bz (r, t) must be obtained first. 2.2. Analytical Equations of Basic Magnetic Susceptibility Fields

A biological tissue consists mainly of two types of structures, i.e., a cellular matrix, which may be approximated by a combination of many spheres, and a venous system, which can be treated as cylinders. All other shapes can then be constructed using a combination of these two basic geometries. To simplify the discussion, we only treat the susceptibility field created from both a sphere and an infinitely long straight cylinder. The results are then generalized to an arbitrary shape.

2.2.1. Spherical Geometry

Considering a sphere with radius R containing homogenous material with relative permeability μi surrounded by another infinite material with relative permeability of μe (μ = 1 + χ, where χ is the corresponding magnetic susceptibility), it follows from ref. (27). that the field inside the sphere is homogeneous and is parallel to the external magnetic field B0 with the field due to the susceptibility difference given in cgs units (28) by Bi = 8π

μi − μe B0 μi + 2μe

[4]

and the susceptibility field outside the sphere is a dipolar field with the component along B0 , denoted as the z-component, given by

340

Hu

μi − μe 3 3 cos2 θ − 1 R B0 (Be )z = 4π μi + 2μe r3

[5]

where θ is the angle between r, which is a vector joining the origin of the sphere and the point of observation with a length equal to r, and the external field B0 . 2.2.2. Cylindrical Geometry

Considering an infinitely long cylinder with radius R, containing homogenous material with relative permeability μi surrounded by another infinite material with relative permeability of μe , it follows from refs. (27, 29–33) that the field inside the cylinder is homogeneous and is parallel to the external magnetic field B0 with the field due to the susceptibility difference given in cgs units by Bi =

4π μi − μe 3 cos2 θ − 1 B0 3 μi + μ e

[6]

and the component of the susceptibility field along B0 outside the cylinder is given by (Be )z = 4π

μ i − μe R 2 sin2 θ cos 2φB0 , (r ≥ R) μi + μ e r 2

[7]

In Eqs. [6] and [7], r is the distance from the point of interest to the cylinder axis, θ is the angle between the cylinder axis and the external field B0 , and ϕ is the angle between the vector r and a plane containing both B0 and the cylinder axis. 2.3. The Effect of Magic Angle Turning on the Susceptibility Field Outside a Magnetized Sphere

All the terms in Eq. [5] are independent of sample rotation except the angle θ, which is the angle between the external main field B0 , i.e., the z-axis, and the vector jointing the observation point and the origin of the laboratory frame. Using the spherical angles θ and φ, the unit vector along r expressed in the Cartesian frame is r = sin(θ)cos(ϕ)i + sin(θ)sin(ϕ)j + cos(θ)k

[8]

where i, j, and k are unit vectors along the x-, y-, and z-directions, respectively. We will adapt the right-hand rotation with the initial field direction along z, i.e., the k-direction with the initial direction cos between the vector r and B0 denoted by cos(θ1 ) = cos(θ)

[9]

After the first 120◦ rotation about the magic angle, B0 is rotated to the x-direction, and the direction cos between the vector r and B0 is now cos(θ2 ) = sin(θ)cos(ϕ)

[10]

Slow Magic Angle Sample Spinning

341

The second successive 120◦ rotation about the magic angle places B0 in the y-direction, and the direction cos between the vector r and B0 becomes cos(θ3 ) = sin(θ)sin(ϕ)

[11]

Bz (r, t) + Bz (r, t + T /3) + Bz (r, t + 2T /3) 3 μi − μe 3 3 cos2 θi − 1 = 4π R B0 μi + 2μe r3 1

= 4π

μi − μe B0 μi + 2μe r 3 R3

3

3 cos2 θi − 1

[12]

1

=0 Equation [3] becomes 1 + 2 + 3 = ω

iso t1

[13]

Equation [13] means that the susceptibility field produced by a magnetized sphere is averaged to zero by the technique of MAT. 2.4. The Effect of Magic Angle Turning on the Susceptibility Field Due to a Cylinder Geometry

Due to the factor of 3 cos2 θ − 1 (see Eq. [6]), the susceptibility field inside the cylinder is averaged to zero by MAT for the same reason as described above, i.e., 1 + 2 + 3 = ωiso t1 . For the field outside the cylinder, we have to find the three sets of (θ ι , ϕ ι ), ι =1,2,3, corresponding to each of the 120◦ rotations about an axis at the magic angle so that we can use Eq. [7] to perform the MAT average. Recall that in Eq. [7], θ is the angle between the main field B0 and the cylinder axis, and ϕ is the angle between the vector r and the plane containing both B0 and the cylinder axis, where r is perpendicular to the cylinder axis (see Fig. 20.2). Since the magic angle forms a cone about the B0 -direction, a general choice is to place the cylinder axis along the z-axis in the laboratory frame, and the initial B0 lies in the z–x plane, i.e., along the z1 -axis. A rotation (ψ) about the z1 -axis will place the magic angle (MA) in a general place in space. In the final frame denoted by x2 –y2 –z2 , each 120◦ rotation about the magic angle axis will rotate the B0 from x2 to y2 then to z2 . The task to find the three sets of (θ ι , ϕ ι ) thus becomes a task to find the three θ ι angles between the cylinder axis and each of the x2 , y2 , z2 axes, respectively, and the ϕ ι angles between the vector r and the plane containing both the cylinder axis and one of the corresponding axes in the frame x2 –y2 –z2 , respectively. Using the right-hand rotation, two successive rotations are needed to rotate the laboratory frame x–y–z to the destination frame x2 –y–z2 . The

342

Hu

z

Magic Angle

B0 Ψ θ

z1 z2

β

y2 y1

φ

y

r

x2 x x1 Fig. 20.2. The coordinate used to perform the susceptibility field average in a cylindrical geometry by MAT. The axis of the cylinder is along the z-direction.

first rotation is θ about the laboratory frame y-axis and the resultant frame is x1 –y1 –z1 . The corresponding rotation matrix is ⎤ cos(θ) 0 sin(θ) ⎥ ⎢ R1 = ⎣ 0 1 0 ⎦ −sin(θ) 0 cos(θ) ⎡

[14]

The second rotation is about the z1 -axis and the resultant frame is x2 –y2 –z2 . The related rotation matrix is ⎤ cos(ψ) −sin(ψ) 0 ⎥ ⎢ R2 = ⎣ sin(ψ) cos(ψ) 0 ⎦ 0 0 1 ⎡

The rotation matrix for the combined rotation is ⎤ ⎡ cos(ψ)cos(θ) −sin(ψ) cos(ψ)sin(θ) ⎥ ⎢ R = R2 R1 = ⎣ sin(ψ)cos(θ) cos(ψ) sin(ψ)sin(θ) ⎦ −sin(θ) 0 cos(θ)

[15]

[16]

The cylinder axis, i.e., the laboratory frame z-axis, expressed in the final frame x2 –y2 –z2 is ⎡ ⎤ 0 ⎢ ⎥ z = R ⎣ 0 ⎦ = cos(ψ) sin(θ)i + sin(ψ) sin(θ)j + cos(θ)k [17] 1 where i, j, and k are unit vectors along x2 -, y2 -, and z2 -axes, respectively.

Slow Magic Angle Sample Spinning

343

It follows from Eq. [17] that the three θ ι (ι =1,2,3) angles have the following relationship: cos(θ1 ) = cos(ψ)sin(θ) cos(θ2 ) = sin(ψ)sin(θ) cos(θ3 ) = cos(θ)

[18]

The vector r, expressed in the final frame x2 –y2 –z2 , is ⎤ cos(ϕ) ⎥ ⎢ r =R ⎣ sin(ϕ) ⎦ = (cos(ψ) cos(θ) cos(φ) − sin(ψ) sin(φ))i 0 ⎡

+ (sin(ψ) cos(θ)cos(φ)+ cos(ψ) sin(φ))j + (− sin(θ) cos(φ))k [19] In order to find the three ϕ ι angles, we need to obtain the three normals ni , i = 1,2,3, where n1 is the normal to the plan formed by x2 and the cylinder axis and n2 is the normal to the plan formed by y2 and the cylinder axis and so on. We will define [20] n1 = x1 i + y1 j + z1 k Since n1 • i = 0 and n1 • z = 0, we obtain x1 = 0, sin(ψ) sin(θ)y1 + cos(θ)z1 = 0,

[21]

y12 + z12 = 1 We obtain

cos(θ) n1 = ± 1/2 j cos2 (θ) + sin2 (ψ)sin2 (θ) sin(ψ)sin(θ) − 1/2 k cos2 (θ) + sin2 (ψ)sin2 (θ)

[22]

Similarly, we obtain

cos(θ) n2 = ± 1/2 i cos2 (θ) + cos2 (ψ)sin2 (θ) cos(ψ)sin(θ) − 1/2 k cos2 (θ) + cos2 (ψ)sin2 (θ)

[23]

n3 = ± [−sin(ψ)i + cos(ψ)j ]

[24]

344

Hu

The ± sign in Eqs. [22]–[24] is arbitrary because the normal to a plane can differ by 180◦ . i, j, and k are unit vectors along the corresponding axis in the x2 –y2 –z2 frame. The equation to determine the three φ ι , ι =1,2,3, angles become cos( π2 − φ1 ) = r • n1 cos( π2 − φ2 ) = r • n1 cos( π2 − φ3 ) = r • n1

[25]

It follows that sin(φ1 ) = ±

(sin ψ cos ϕ+cos ψ sin ϕ cos θ ) (cos2 (θ )+sin2 (ψ)sin2 (θ ))1/2

sin(φ2 ) = ±

(cos ψ cos ϕ−sin ψ sin ϕ cos θ ) (cos2 (θ )+cos2 (ψ)sin2 (θ ))1/2

[26]

sin(φ2 ) = ±sin(ϕ) Finally, with Eqs. [18] and [26] we are able to perform the time average in Eq. [3] for the susceptibility outside a cylindrical geometry: Bz (r, t) + Bz (r, t + T /3) + Bz (r, t + 2T /3) μi − μe R2 2 B0 sin θi cos 2φi μi + μe r 2 3

= 4π

1

= 4π

μi − μe B0 μi + μe r 2 R2

3

[27] (1 − cos2 θi )(1 − 2 sin2 φi )

1

=0 Once again Eq. [3] becomes 1 + 2 + 3 = ωiso t1

[28]

The results given in Eqs. [3], [27], and [28] indicate that MAT is also able to average out the susceptibility field outside a cylindrical geometry. Since all other shapes can be constructed using a combination of the basic spherical and cylindrical geometries, based on the distribution rules of integration, we conclude that MAT is able to average the susceptibility field of an arbitrary sample shape to zero. This result indicates that in the absence of translational molecular diffusion, a high-resolution isotropic metabolite NMR spectrum should be obtained along the evolution dimension of the PHORMAT experiment regardless of the sample spinning rate used. However, translational molecular diffusion is present in a biological fluid object. In the following, we will investigate how molecular diffusion affects the spectral resolution along the isotropic dimension of the PHORMAT experiment.

Slow Magic Angle Sample Spinning

2.5. The Theory of Translational Molecular Diffusion at a Magnetic Gradient Field

345

It is well known (34–36) that the time dependence of magnetization in the presence of isotropic molecular diffusion for an isolated, one-half spin is described by the Block–Terrey equation (37): Mx i + My j ∂M (r, t) =γ M (r, t) × B(r, t) − ∂t T2 Mz − M 0 − k + D0 ∇ 2 M (r, t) T1

[29]

where i, j, and k are unit vectors along the x-, y-, and z-directions in the laboratory reference frame, respectively, γ is the gyromagnetic ratio of the nuclei, T1 and T2 are the NMR spin-lattice and spin–spin relaxation times, respectively, and D0 is the isotropic diffusion coefficient. B(r,t) = B0 k + b(r,t) is the vector sum of the magnetic field at the observation point (r,t), where B0 k is the main external magnetic field and b(r,t) is an additional magnetic field generated by the susceptibility variations in the sample. Since it is generally true that B0 >> |b(r,t)|, only the component of b(r,t) along the main field direction, i.e., bz (r,t), needs to be considered. Equation [29] may be simplified and expanded into its Cartesian terms to yield Mx ∂Mx (r, t) = γ (B0 + bz (r, t))My (r, t) − + D0 ∇ 2 Mx (r, t) ∂t T2 [30] ∂My (r, t) My = −γ (B0 + bz (r, t))Mx (r, t) − + D0 ∇ 2 My (r, t) ∂t T2 [31] Mz − M 0 ∂Mz (r, t) =− + D0 ∇ 2 Mz (r, t) ∂t T1

[32]

Equations [30] and [31] are recombined to yield ∂m(r, t) m(r, t) = −iγ (B0 + bz (r, t))m(r, t) − + D0 ∇ 2 m(r, t) ∂t T2 [33] where m(r, t) = Mx (r, t) + iMy (r, t) is the transverse magneti√ zation and i = −1 is the complex unit. Defining m(r, t) = m(r, t) × exp(−(iω0 + T12 )t) with ω0 = γ B0 yields ∂m(r, t) = −iγ bz (r, t)m(r, t) + D0 ∇ 2 m(r, t) ∂t

[34a]

Mz − M 0 ∂Mz (r, t) =− + D0 ∇ 2 Mz (r, t) ∂t T1

[34b]

346

Hu

Solving Eqs. [34a] and [34b] under an arbitrary field Bz (r,t) is difficult. Instead, the method of cumulative expansion (35) is used. Using the fact that ∇ 2 M (r, 0) = 0, we obtain ∇ 2 Mz (r, 0) = 0 and ∇ 2 m(r, 0) = 0. Using ∇ 2 m(r, 0) = 0 to replace ∇ 2 m(r, t) in Eq. [34a] and integrating with respect to t, the first cumulative solution of Eq. [34a] is

t

m(r, t) = m(r, 0) exp −iγ

bz (r, t )dt

[35]

0

Substituting Eq. [35] into the right-hand side of Eq. [34a] and again integrating with respect to t, the second cumulative solution of Eq. [34a] is

t

m(r, t) =m(r, 0) exp −i ⎡

γ bz (r, t )dt

0

⎤ 2 γ t ∇b (r, t )dt } dt z 0 0 ⎦ ×⎣ t t 2 {−iD } + exp 0 0 dt γ 0 ∇ bz (r, t )dt =m(r, 0)exp(−iψ(r, t)) × exp (−bD0 ) + exp −iϕ(r, t) [36] t where ψ(r, t) = 0 γ bz (r, t )dt is the usual phase accumulation of the magnetization in the presence of the field bz (r, t ) and exp {−D0

t

t

ϕ(r, t) = {D0

t

dt γ

0

0

∇ 2 bz (r, t )dt

is a phase accumulation due to the combination of diffusion and the second derivative of the field bz (r, t ). In case ∇ 2 bz (r, t) = 0, which is valid for most pulsed field gradient experiments, Eq. [36] is simplified to m(r, t) = m(r, 0) exp(−iψ(r, t)) × exp (−bD0 ) t b= k2 (r, t )dt 0

k(r, t ) = γ

t

G(r, t )dt

[37a] [37b] [37c]

0

where G(r, t) = ∇bz (r, t) is the gradient of the field. By inserting Eq. [37] into Eq. [34a] to carry out the third cumulative expansion, it is trivial to prove that Eq. [37] is the exact solution of Eq. [34a] provided that ∇ 2 bz (r, t) = 0, i.e., a constant gradient field. Under a constant field gradient G(r, t) = G0 , it follows from

Slow Magic Angle Sample Spinning

347

Eqs. [37a], [37b], and [37c] that the magnetization following a single pulse 900 − acq(t) is 1 2 2 3 [38] m(r, t) = m(r, 0) exp(−iψ(r, t)) × exp − D0 γ G0 t 3 and 0 the magnetization following a CPMG pulse sequence 90 − τ − 1800 − τ − n − acq(t) is

1 2 2 2 m(r, t) = m(r, 0) exp(−iψ(r, t)) × exp − D0 γ G0 (2τ ) (nτ ) 12 [39] where the effective gradient field defined in ref. (36), i.e., the π pulse changes the sign of the gradient prior to it, is used to perform the integration in Eq. [37c]. These results are consistent with those reported previously (38, 39). Similarly, using the fact that ∇ 2 Mz (r, 0) = 0, it can be seen from Eq. [34b] that ∇ 2 Mz (r, t) = 0 is satisfied when t = 0. Consequently, the longitudinal magnetization Mz (r, t) is independent of molecular diffusion. This means that the T1 values can still be faithfully measured even at the existence of the pulse field or susceptibility field gradients.

2.6. The Effect of Molecular Diffusion to the PHORMAT Experiment

It is known from Eqs. [36] and [37] that if ∇ 2 bz (r, t) = 0, the effect of diffusion on the FID is an attenuation of the signal by a factor of exp (−bD0 ). This means that when working on a pulse sequence, we can treat the phase part and the amplitude part separately. It follows from the principles given above and those given in the original PHORMAT article (15) that the phase part created by the magnetic susceptibility field from both the sphere and the cylindrical geometries is averaged to zero and only the isotropic contribution is left along the evolution dimension (t1 ). The resultant FID observed by the PHORMAT sequence is FID(t2 , t1 ) = exp(j ωiso t1 )F2 (t2 )

[40]

where F2 (t2 ) is the FID along the acquisition dimension that is broadened by the magnetic susceptibility field. The results in Eq. [40] mean that the isotropic dimension of the PHORMAT experiment is free from susceptibility broadening. In the following, we will examine the attenuation factor of the PHORMAT experiment due to diffusion under a constant field gradient. Working backward and ignoring the triple-echo sequence because its length is usually smaller than the evolution

348

Hu t1/3

t1/3

t1/3

t2

RF T/3

T/3

Effect Gradient

Fig. 20.3. The effective field gradient in a PHORMAT experiment at a constant gradient field G0 . The pulses are all π /2 pulses.

increment in a biological application (22), the effective field gradient for both the (+) and the (–) pulse sequences is simplified and the result is pictured in Fig. 20.3. It follows from Eq. [37] that by performing the first-time integration over the effective gradient using Eq. [37c], one finds k(r, t ). b is obtained by performing the second-time integration over the square of k(r, t )(see Eq. [37b]) and the resultant b for the evolution dimension is b=

1 2 2 2 γ G0 t1 T 27

[41]

The b value along the acquisition dimension is the same as that described in Eq. [38], i.e., b=

1 2 2 3 γ G 0 t2 3

[42]

With diffusion incorporated, Eq. [40] becomes 1 FID(t2 , t1 ) = exp − D0 γ 2 G02 t12 T exp −jωiso t1 27 [43] 1 2 2 3 exp − D0 γ G0 t2 F2 (t2 ) 3 Including the attenuation due to spin-lattice relaxation time along the evolution dimension, the modified version of Eq. [43] is −t1 1 exp(−jω FID(t2 , t1 ) = exp − 27 D0 γ 2 G02 t12 T exp −2 T3T t ) iso 1 1 exp − 13 D0 γ 2 G02 t23 F2 (t2 ) [44] The detailed evaluation of Eq. [44] in the presence of large magnetic susceptibility field using the phantom sample of glass beads + H2 O is given in Section 3.

Slow Magic Angle Sample Spinning

2.7. 2D PASS Experiment at a Magnetic Susceptibility Field and at the Existence of Molecular Diffusion

349

It follows from the original 2D PASS report (14) that the powderaveraged signals as a function of pitch θ without molecular diffusion are given by FID(θ, t2 ) =

∞

a(k) exp( − ikθ)exp i(ωiso + kωr )t2

k =−∞

[45]

× exp[−(T + t2 )/T2 ] where a (k) is the powder-averaged sideband amplitude of order k in a conventional MAS experiment, T is the rotor period, and T2 is the spin–spin relaxation time constant. For the 2D PASS experiment, θ is incremented between 0 and 2π and is related to the positions of the five π pulses according to the set of PASS equations given in the original report (14). The PASS experiment is named by the total number of θ steps. For example, PASS-16 means that there are a total of 16 increments along the θ dimension. Equation [45] means that in a 2D PASS experiment, the sidebands are separated according to the order of sideband. T2 produces a uniform attenuation of the FID that is θ independent. In order to simplify the discussion, the concept of constant field gradient, i.e., ∇ 2 Bz (r, t) = 0, is again used. The effective gradient in a 2D PASS experiment is given in Fig. 20.4. It follows from Eqs. [37c] and [37b] that the b factor corresponding to Fig. 20.4b along the evolution dimension θ is 2 b = γ 2 G02 T 3 (1 − t5 )3 + (2t5 − t4 − 1)3 + (t3 − 2t2 + 2t1 )3 3 +(t2 − 2t1 )3 + t13 =γ 2 G02 T 3 f (θ)

[46a] t2

(a) 0

t1

t2

t3

t4

t5

T

t2 (b)

0

t1

t2

t3

t4

t5

T

Fig. 20.4. The effective field gradient in a 2D PASS experiment at a constant gradient field G0 . (a) Pulses sequence, where the initial pulse is a π /2 pulse and the remaining five are π pulses. (b) The effective gradient in the existence of a constant gradient G0 when the effect of the π pulses is considered.

350

Hu

2 (1 − t5 )3 + (2t5 − t4 − 1)3 + (t3 − 2t2 + 2t1 )3 3 +(t2 − 2t1 )3 + t13 [46b] where t1 –t5 are expressed in units of T and for each θ there is a unique set of t1 –t5 (14). In order to obtain Eq. [46a], the PASS condition, i.e., Eq. [29] in ref. (14), was used. When the five π pulses are equally spaced, we have f (θ) =

b=

1 2 2 3 γ G0 T 108

[46c]

A plot of the function f (θ) versus θ for both PASS-16 and PASS-32 is provided in Fig. 20.5. Figure 20.5 indicates that f(θ) is only a function of θ. This is because the same set of time delays (t1 –t5 ) is found for the same value of θ, which is independent of the total number of evolution increments along the θ dimension in a 2D PASS experiment (14). The amplitude part in Eq. [37] for the acquisition dimension t2 is described again by that of Eq. [38]. Putting it all together, Eq. [45] becomes FID(θ, t2 ) =exp(−(T /T2 )) ⎫ ⎧ (k) exp(−ikθ)exp(−D γ 2 G 2 T 3 f (θ))× ⎪ ⎪ a ⎪ 0 ∞ ⎨ ⎬ 0 ⎪ exp i(ωiso + kωr )t2 exp − 13 D0 γ 2 G02 t23 × ⎪ ⎪ ⎪ ⎭ ⎩ exp(−t /T ) k=−∞ ⎪ 2

2

[47] According to the theoretical prediction given in Eq. [47], the spin–spin relaxation time T2 causes a fixed attenuation of the total signal according to exp(−(T /T2 )), where T denotes the rotor period, but this term does not affect the quality of the 2D PASS in separating the spinning sidebands by order. Molecular diffusion also causes signal attenuation; however, the attenuation factor is θ dependent, i.e., according to the factor exp(−D0 γ 2 G02 T 3 f (θ)).

f(θ)

0.0 0.03 0.0 0.02 0.0 0.01 0.0 0.00 0

PASS-32 PASS-16 0

10

20

30

40

Pitch (θ) Fig. 20.5. The relationship between f(θ) and θ for PASS-32 and PASS-16.

Slow Magic Angle Sample Spinning

351

The detailed evaluation of the θ-dependent term in the presence of large magnetic susceptibility field using the sample of glass beads + H2 O will be given in Section 3.

3. Model Experiments and Typical Results

3.1. 1 H T1 and T2 Experiments

Unless otherwise specified, all 1 H-NMR experiments were performed on a Varian–Chemagnetics 300-MHz Infinity spectrometer, with a proton Larmor frequency of 299.982 MHz. A standard Chemagnetics CP/MAS probe with a 7.5-mm pencil-type spinner system was used. In order to spin at low frequencies, the rotor was equipped with a flat drive tip, and an airflow restriction was used in the driver channel. The spinning rate was controlled using a commercial Chemagnetics MAS speed controller under the automated control mode. By marking the rotor with three evenly spaced marks, the frequency stability was better than ±0.3 Hz at a spinning rate from 1 to 200 Hz. Spinning rates higher than 200 Hz were obtained after removing the airflow restriction in the driver channel and by replacing the flat drive tip with a standard tip. Glass beads with diameters of 210–250 μm were used as a model tissue sample. The glass beads were loaded into the rotor and then tap water was added, resulting in a homogenous mixture of beads and H2 O. Because of the large magnetic susceptibility field from the glass beads, this model system represents an upper limit that could be obtained in a biological sample, i.e., tissues close to an air cavity or a tissue–bone interface. Unless otherwise specified, the beads + H2 O system will be the model phantom system for experiments throughout this chapter. Further experimental details can be found in refs. (16, 22) for the 1 H 2D PASS experiments and the 1 H PHORMAT, respectively. The only difference between the published experiments and those described in this chapter is that no water suppression is applied in the current experiment because H2 O is the signal to be observed. The 1 H T1 of the H2 O in the mixture of glass beads + H2 O, as measured by the conventional inversion recovery method at spinning rates ranging from 30 to 500 Hz, is 1.59 ± 0.03 s. This observation confirms the prediction by Eq. [34b] that T1 is independent of molecular diffusion. 1 H T2 measured by a CPMG pulse sequence π − (τ − π − τ −)n − acq 2

352

Hu

Table 20.1 1 H T of H O in a mixture of glass beads + H O at different 2 2 2 spinning rate Spin Rate (Hz)

T2f (ms)

Percentage of T2f

T2 s (ms)

Percentage of T2s

T2 (one exp fit)

100

–

–

27

100

27

200

14

76

70

24

31

400

11

69

76

31

33

500

13

65

86

35

39

where τ is the rotor period of the spinning, is generally multi exponential and can fit well with two components. The resultant values obtained at spinning rates from 100 to 500 Hz are summarized in Table 20.1. Recall that in our measurement, n started from 1 instead of 0, and the component with a very short T2 value is significantly reduced at a low spinning rate compared to that of the component with a longer T2 value. Despite this fact, Table 20.1 still indicates that the measured T2 value is decreased at a lower spinning rate due to molecular diffusion. This is because the π pulses in the CPMG pulse sequence used are less efficient in suppressing the effect of molecular diffusion at a low spinning rate than that at a higher spinning rate. 3.2. 1 H PHORMAT Experiments

Figure 20.6 shows the spectra along the isotropic dimension of the PHORMAT experiment at various sample spinning rates ranging from 1 to 50 Hz. The projection spectra along the acquisition dimension at these spinning rates were found essentially to be the same with a linewidth of 3745 Hz, defined as the full width at the half-height positions of the resonant line. It is known from Fig. 20.6 that line narrowing along the isotropic dimension of the PHORMAT experiment increases at increasing spinning rate. For example, a line narrowing factor of 22 was obtained at 50 Hz, while only a factor of 4 was achieved at 1 Hz. A plot of the isotropic linewidth subtracted by the natural linewidth obtained at a sample spinning rate of 1 kHz, ν1/2 , versus the spinning rate f is provided in Fig. 20.7, from which the following relationship is obtained with a correlation coefficient of 0.9991: ν1/2 = 911.31 × f −0.4718

[48]

The results given by Eq. [48] can be compared with the the−t1 ) oretical prediction using Eq. [44], since the term exp(−2 T3T 1 in Eq.[44] causes only a fixed attenuation at a low spinning rate because t1 << T. The linewidth along the isotropic dimension (t1 )

Slow Magic Angle Sample Spinning

353

Fig. 20.6. The linewidths along the isotropic dimension of the 1 H PHORMAT experiment of H2 O in a mixture of H2 O + glass beads at various sample spinning rates.

Fig. 20.7. The line broadening along the isotropic dimension of the PHORMAT experiment as a function of the sample spinning rate.

predictedby Eq. [44] is determined by the term related only to 1 2 2 2 diffusion exp(− 27 D0 γ G0 t1 T ) , which corresponds to a Gaussian decay relative to t1 . It follows from ref. (40) that the halflinewidth associated with this term is given by

354

Hu 1/2

ν1/2 = 0.6409 × γ G0 × D0

× f −0.5

[49]

Equation [49] indicates that given a constant gradient field G0 , the theory predicts that the line broadening caused by molecular diffusion in a PHORMAT experiment is proportional to f −0.5 , where f is the sample spinning rate. The experimental result given in Eq. [48] is in good agreement with this prediction, and the small discrepancy is within the experimental error range. Comparing Eqs. [48] and [49], we have 0.6409 × γ G0 × 1/2 D0 = 911.31, where the units of D0 and γ G0 are cm2 /s and Hz/cm, respectively. Assuming that D0 = 2 × 10−5 cm2 /s, the diffusion constant of free water at room temperature, we obtain γ G0 = 317.9 kHz/cm. This result is reasoned in the following by examining the packing of the glass beads and the associated susceptibility broadening. The diameters of the used glass beads ranged from 210 to 250 μm. We will take an average, i.e., 230 μm, as the diameter of the beads used. Assuming a dense packing as pictured in Fig. 20.8. The static linewidth of H2 O in the glass beads is 3745 Hz, and the natural linewidth is 24 Hz. Thus, the susceptibility broadening is 3721 Hz. To obtain the analytical form of the gradient inside the space between the beads, we would need to obtain the derivative of Eq. [5] and sum the contribution from all the beads in the sample tube, which requires a computer simulation. Instead the assumption of the constant gradient is used. It follows that γ G0 l = 3721 Hz

[50]

r b c

b = 0.57735 r = 66.4μm c = 0.15467 r = 17.8μm

Fig. 20.8. Glass beads packing and associated geometries. The centers of three beads form a triangle with equal sides. c is the distance between the center of the triangle and the nearest point of the surface of beads and b is the distance between the center of the triangle and the contact point of the two beads. r is the radian of the beads, i.e., r = 115 μm. There is one bead above and one below the triangle to form a compact packing. Water molecules reside in the spaces between the beads.

Slow Magic Angle Sample Spinning

355

where l is the length. Using the dimensions b and c in Fig. 20.8, we obtain (γ G0 )min = 560 kHz/cm and (γ G0 )max = 2090 kHz/cm, respectively. It is very interesting to note that the result, γ G0 = 317.9 kHz/cm, which was obtained from the PHORMAT experiment by using Eqs. [48] and [49], is, qualitatively, in agreement with (γ G0 )min . Besides, a better agreement between theory and experimental results would be obtained if a precise knowledge of the diffusion constant D0 in the glass beads is available since D0 in beads can only be reduced compared with that of the free water. A reduced D0 would give an increase in theoretical γ G0 . This result indicates that a constant field gradient approximation in glass beads is in fact a reasonable assumption for a sample spinning rate ranging from 1 to 50 Hz. Since the maximum time available for molecular diffusion is two-thirds of the rotor period (2T/3) along the isotropic dimension of the PHORMAT experiment, the maximum lengths " that water molecules diffuse through space are calculated by D0 2T /3, which produces 36.5 and 18.25 μm for 1 and 4 Hz spinning rates, respectively. This means that at spinning rates higher than 4 Hz, the assumption of free water diffusion is reasonably valid. 3.3. Justification of the Constant Gradient Field Approach in a PHORMAT Experiment Using the Glass Beads + H2 O Mixture

The 1 H PHORMAT experiment was performed on the above beads + H2 O system but without sample spinning and at a magnetic field of 2.0 T using the in vivo slow MAS probe described in ref. (23). In the PHORMAT sequence of Fig. 20.3, the value of 1/T is set between 0.5 and 25 Hz to simulate the diffusioninduced line broadening in a real spinning PHORMAT experiment. Since the sample is static during the experiment, the gradient field remains constant during the experiment. Figure 20.9 gives the diffusion-induced line broadening along the evolution dimension of the PHORMAT as a function of the equivalent sample spinning rate. Clearly, at a constant gradient field G0 , the line broadening caused by molecular diffusion in a PHORMAT

Static 1H PHORMAT 350 300 250 Line broadening 200 along the evolution dimension ( Δυ1/2 Hz) 150 100 50 0

y = 227.04 f – 0.4968 R2= 0.9923

0

5

10

15

20

25

30

Equivalent sample spinning rate (Hz):f = 1/T

Fig. 20.9. The line broadening along the evolution dimension of a static 1 H PHORMAT experiment as a function of the equivalent sample spinning rate.

356

Hu

experiment is indeed proportional to f −0.5 , where f is the equivalent sample spinning rate. This not only justifies the theory given in Eq. [49] but also justifies that the constant gradient field approach used in the above sections is valid in the H2 O + glass beads system. The coefficients 911 at 7.05T and 227 at 2T, a factor of four times difference, are approximately equal to the magnetic field difference, i.e., 3.5. This is because the magnetic susceptibility field due to either the spherical or the cylindrical geometry is proportional to the external magnetic field strength (see Eqs. [4]–[7]); hence, its corresponding gradient field obtained by taking its first spatial derivative is also proportional to the main magnetic field. According to Eq. [49], the diffusion-induced line broadening would also be proportional to the magnetic field strength. 3.4. 1 H 2D PASS Experiments

Using Eq. [46c], for the case of θ = 0, Eq. [47] becomes

FID(0, t2 ) = att(T ) ×

∞

a(k) exp i(ωiso + kωr )t2

k=−∞

[51a] # 1 exp − D0 γ 2 G02 t23 exp(−t2 /T2 ) 3 1 att(T ) = M0 exp(−(T /T2 )) × exp − D0 γ 2 G02 T 3 [51b] 108

The summation in Eq. [51] represents a conventional FID with isotropic resonance located at ωiso and consists of a family of spinning sidebands spaced by the rotor frequency ωr . The sideband spectrum is broadened by both the spin–spin relaxation time T2 and molecular diffusion. The term outside the summation causes a net attenuation of the FID from both T2 and molecular diffusion due to the use of the constant one rotor period. It is known from Table 20.1 that T2 measured by the rotorsynchronized CPMG pulse sequence is double exponential at a spinning rate as high as 500 Hz. Assuming D0 is a single value, the generalized form for the attenuation factor in Eq. [51] becomes att(T ) = A × exp(−T /T2f ) + B × exp(−T /T2s ) 1 × exp − D0 γ 2 G02 T 3 108

[52]

Using T2f and T2s , as well as their respective ratios measured at a spinning rate of 400 Hz (see Table 20.1) to approximate the T2 values, it follows from Eq. [52]

Slow Magic Angle Sample Spinning

357

att(T ) =M0 × 0.69 × exp(−T /11) + 0.31 × exp(−T /76) × exp(−k × T 3 ) [53a] k=

1 D0 γ 2 G02 108 × 109

[53b]

where T is in units of ms, γ G0 in units of Hz/cm, and D0 in units of cm2 /s. Figure 20.10 shows the decay of the integral signal obtained from the first increment, i.e., θ = 0, of the 2D PASS experiment as a function of the rotation rotor period T. With Eq. [53a], the calculated data using k = 1.2 × 10–3 reasonably follow the experimental data. Assuming D0 = 2 ×10–5 cm2 /s, it follows from Eq. [53b] that a value of γ G0 = 2545 kHz/cm is obtained, which is close to (γ G0 )max = 2090 kHz/cm found from the geometry of the glass beads above, but differs by a factor of 8 from the data derived from the PHORMAT experiment, where γ G0 = 317.9 kHz/cm was obtained. The discrepancy between the PHORMAT and the 2D PASS data is mainly due to the difference in sample spinning associated with these two different experiments. PHORMAT is good for spinning rates less than 50 Hz, while PASS is good for spinning rates larger than 50 Hz in the glass beads + H2 O system. At higher spinning rates, the diffusion of the molecules could be enhanced due to the enhanced flow of the H2 O as a result of sample spinning. If D0 is enhanced at a higher spinning rate, then γ G0 will be reduced according to Eq. [53b] and a better agreement between PHORMAT and PASS would be obtained. It is known from Eq. [47] and Fig. 20.5 that f (θ) is a function of θ along the evolution dimension of the 2D PASS experiment. This means that the attenuation of the FID due to molecular diffusion, i.e.,exp(−D0 γ 2 G02 T 3 f (θ)), differs when θ changes. The effect of f (θ) to the 2D PASS spectrum is examined in Fig. 20.11 by multiplying the FID corresponding to each θ value 250 200 Integral 150 Intensity 100

50 0 0

5

1 Rotor Period T

1

2

Fig. 20.10. The decay of the integral signal obtained from the first increment, i.e., θ = 0, of the 2D PASS experiment as a function of the rotation rotor period T. The symbol •denotes experimental data, while ♦ represents the data calculated using M0 = 2367 and k = 1.2 × 10–3 .

358

Hu

Fig. 20.11. The 2D PASS center band spectra acquired at 250 Hz on H2 O + glass beads with diameter of 210–250 μm. The intensity of FID corresponding to each θ increment is multiplied by a factor equal to exp(kf (θ)) before Fourier transformation.

according to exp(kf (θ)), where k is a constant. The impact of k on the residual spinning sidebands due to diffusion is illustrated in Fig. 20.11. The spectra given in Fig. 20.11a is the center band spectra (isotropic spectra) from a conventional 2D PASS experiment acquired at a sample spinning rate of 250 Hz. Small residual sideband is observed (see the 16-fold expansion spectrum). Figure 20.11b shows the center band from the manipulated 2D PASS with k = –4 applied. It is apparent that the residual sideband intensity is enhanced. This means that the effect of θ-dependent molecular diffusion attenuation is to cause residual spinning sidebands in the 2D PASS spectrum. It is thus expected that if the condition k = D0 γ 2 G02 T 3 is satisfied, the residual sidebands will be greatly suppressed. Also when k > D0 γ 2 G02 T 3 , the sidebands should be inverted. These trends are indeed observed in Fig. 20.11c, d, respectively. In Fig. 20.11c, the sideband intensity is suppressed by nearly a factor of 2 when k = 4 is applied to manipulate the FID. The resultant γ G0 is 1768 kHz/cm, which is in reasonable agreement with the value found from the data in Fig. 20.10, where γ G0 = 2545 kHz/cm was obtained. The center band spectra of the 1 H 2D PASS experiment acquired on the mixture of H2 O + glass beads at several selected sample spinning rates are plotted in Fig. 20.12. The linewidth

Slow Magic Angle Sample Spinning

359

Fig. 20.12. The center bands of 1 H 2D PASS at several selected spinning rates acquired on a mixture of H2 O + glass beads. The sidebands indicated by arrows are due to the θ-dependent attenuation arising from molecular diffusion.

of the isotropic peak versus spinning rate is summarized in Table 20.2. The linewidth of the isotropic peak from both singlepulse and PHORMAT experiments is also included in the table for comparison. Figure 20.12 indicates that the residual sidebands due to the θ-dependent attenuation arising from molecular diffusion increase at decreased spinning rate, which by is apparently caused the increased attenuation factor, i.e.,exp −D0 γ 2 G02 T 3 f (θ) , as a result of the increased rotor period T. In contrast to the PHORMAT experiment, the isotropic linewidth of the center band from the 2D PASS spectrum only slightly increases at decreased spinning rate (see the data in Table 20.2), which can be reasoned by the following two competitive facts. (A) The fixed attenuation from T2 due to the use of one full rotor period as the constant evolution time suppresses the components with short T2 values. This would result in reduced linewidth at reduced spinning rate. It is known from Table 20.2 that the isotropic spectral linewidth, obtained by left shifting one full rotor period of data at the beginning of the FID from a single-pulse experiment, is very similar to that observed from the 2D PASS experiment at the same spinning rate. But the sacrifice is a dramatic reduction in signal intensity. For example, at a spinning rate of 65 Hz, only about 1% of the

360

Hu

Table 20.2 The isotropic linewidths at various sample spinning rates obtained from 2D PASS, single–pulse, and PHORMAT experiments on the sample of a mixture of H2 O + glass beads ν 1/2 (Hz) 2D PASS

Single pulsea

25

–

–

223

30

50 (∼0.1%)

–

–

40

42

–

179

Spin rate (Hz)

Single pulse

PHORMAT

50

39.3 (1.8%)

–(0.2%)

156

65

36.0 (4.7%)

45 (1%)

–

70

34.3 (5.5%)

43 (1.4%)

–

75

32.0 (6.8%)

38 (1.9%)

–

100

30.5 (14.6%)

36 (4.3%)

82

–

150

29.6 (31.2%)

29 (12.4%)

78

–

200

26.5 (48.2%)

26 (20.5%)

64

–

250

25.6 (60.1%)

24 (32.7%)

40

–

300

22.0 (66.1%)

21 (53.7%)

32

–

350

23.7 (73.5%)

–

–

–

400

23.9 (77.5%)

–

–

–

500

25.4 (∼80%)

26 (85.5%)

28

–

20.4 —

24 (93.1%)

23

–

1000

a Left-shifted one full rotor period linewidth. The data included in the parentheses denote the percentage of the

observed integral signal intensity relative to what is obtained from a single π /2 pulse experiment under the same experimental setup.

total signal is left in the shifted spectrum, and a slightly larger signal that is equal to approximately 4.7% of the total magnetization is observed from the 2D PASS due to the refocusing of the π pulses. (B) The sidebands are broadened along the acquisition dimension by molecular diffusion. Since in an ideal 2D PASS experiment the sidebands are separated according to the sideband order, this would result in increased linewidth for the isotropic peak at decreased spinning rate. This would be the dominant mechanism accounting for the slightly increased linewidth at decreased spinning rate. It is also known from Table 20.2 that the isotropic linewidth obtained from the PHORMAT experiment at a sample spinning rate of 50 Hz is 156 Hz and is substantially larger than that of 2D PASS (i.e., 39.3 Hz). This is because PHORMAT does not differentiate the components with different T2 values and the effect of the molecular diffusion to the line broadening is also different from that of PASS (see Section 3.2 for details).

Slow Magic Angle Sample Spinning

3.5. Conclusions

361

In this chapter, the detailed theory of slow MAS techniques for eliminating the magnetic susceptibility-induced line broadening is presented. Both 1 H PHORMAT and 1 H PASS are capable of producing a high-resolution 1 H-NMR metabolite spectrum in the presence of the magnetic susceptibility gradient field, regardless of the strength of the gradient field if translational molecular diffusion does not exist or if the combined impact of the magnetic susceptibility field gradient and molecular diffusion is small (valid for most biological applications). 1 H PHORMAT is applicable at very slow or ultra-slow sample spinning rates of 1 Hz or less as long as the spin-lattice relaxation time T1 is longer than or comparable to the period of sample rotation. In contrast, 1 H PASS is applicable at a sample spinning rate of higher than 20 Hz or above because PASS is a constant time evolution experiment. Because of these differences, 1 H PHORMAT and the like are the methods of choice for non-destructive applications of live subjects with sample sizes as large as a small animal, i.e., a laboratory mouse. 1 H PASS is mainly usable for small-sized objects, such as excised tissues, cells attached to solid surfaces, biofilms, food, and oil seeds, due to the restriction of the sample spinning rate used. The other difference between 1 H PHORMAT and 1 H PASS is sensitivity. Because of the use of π pulses, 1 H PASS is nearly four times more sensitive than PHORMAT, which uses two projection pulses. Therefore, 1 H PASS is the method of choice if the sample size is small, i.e., a few millimeter or less. However, in the extreme cases where both the magnetic susceptibility field gradient is large and the translational molecular diffusion is fast, the diffusion of metabolites through the magnetic susceptibility field gradient induces line broadening that cannot be averaged out by either the 1 H PHORMAT or the 1 H PASS method. For the PHORMAT experiment, the diffusion-induced 1/2 line broadening is jointly proportional to γ G0 × D0 × f −0.5 , where γ G0 is the strength of the gradient, D0 is the diffusion constant, and f is the sample spinning rate. Therefore, at ultraslow sample spinning rates, the diffusion-induced line broadening can be severe if both the gradient field is large and the molecular diffusion is fast. Nevertheless, even in the extreme case of glass beads with a susceptibility line broadening of ∼3700 Hz, a line narrowing factor of about 4 is still achieved by the 1 H PHORMAT experiment. For the PASS method, molecular diffusion will modulate the amplitudes of the FIDs along the evolution dimension. As a result, unwanted residual spinning sidebands will be seen in the isotropic spectrum of the PASS experiment. In conclusion, the molecular diffusion-induced line broadening needs to be considered when PHORMAT and PASS are applied to extreme cases such as tissues close to air cavities and tissues on the surface of bones, where both the magnetic susceptibility field gradient is unusually large and molecular diffusion is still fast.

362

Hu

4. Notes 1. The detailed pulse sequence and the steps in setting up the high-resolution 1 H PHORMAT experiment on excised tissues using a sample spinning rate as low as 1.0 Hz and with H2 O suppression can be found in ref. (22). 2. The extension of the high-resolution 1 H PHORMAT method to study live whole small laboratory animals can be found in ref. (23). 3. The principles and the experimental details in setting up the localized 1 H PHORMAT/LOCMAT technique, where localized high-resolution 1 H-NMR spectrum of various tissues and organs in a live small animal, can be found in ref. (24). 4. The experimental details of performing high-resolution 1 H PASS with water suppression can be found in ref. (16). 5. A comprehensive review (with many examples) of the applications of the high-resolution slow or ultra-slow MAS techniques can be found in ref. (26). 6. 1 H PHORMAT is applicable for all kinds of sample geometries regardless of sample shape, i.e., samples with or without a preferred orientation. In case of an oriented sample, the beginning of the pulse sequence is synchronized to a fixed sample point. Unlike PHORMAT, 1 H PASS requires a powder average. Thus, the kind of biological samples suitable for PASS analysis should be as homogeneous as possible, including soft tissues, a pack of dense cells, or a pack of cells attached to spherical solid surfaces/beads, food seeds. Here the random orientation of the cells, beads and seeds, etc. in the sample creates an orientation average analogous to a powder average.

Acknowledgments This work was supported by a NIH/National Center for Research Resources (NCRR) under grant 1R21RR025785-01. The research was performed in the Environmental Molecular Sciences Laboratory (a national scientific user facility sponsored by the DOE Biological and Environmental Research) located at the Pacific Northwest National Laboratory and operated for DOE by Battelle. Dr. Robert A. Wind is acknowledged for his enthusiastic

Slow Magic Angle Sample Spinning

363

support during the early stage development of the various slow MAS techniques for biological applications.

References 1. Mitchell, D. G., Cohen, M. S. (2004) MRI principles, 2nd edn, SAUNDERS (An Imprint of Elsevier) Philadelphia, Pennsylvania. 2. Callaghan, P. T. (1993) Principles of Nuclear Magnetic Resonance Microscopy, Oxford University Press, New York, NY. 3. Petroff, O. A. C., Prichard, J. W., Ogino, T., Shulman, R. G. (1998) Proton magneticresonance spectroscopic studies of agonal carbohydrate-metabolism in rabbit brain. Neurology 38, 1569–1574. 4. Cheng, L. L., Ma, M. J., Becerra, L., Ptak, T., Tracey, I., Lackner, A., Gonzalez, R. G. (1997) Quantitative neuropathology by high resolution magic angle spinning proton magnetic resonance spectroscopy. Proc Natl Acad Sci USA 94, 6408–6413. 5. Weybright, P., Millis, K., Campbell, N., Cory, D. G., Singer, S. (1998) Gradient, highresolution, magic angle spinning 1 H nuclear magnetic resonance spectroscopy of intact cells. Magn Reson Med 39, 337–345. 6. Bollard, M. E., Murray, A. J., Clarke, K., Nicholson, J. K., Griffin, J. L. (2003) A study of metabolic compartmentation in the rat heart and cardiac mitochondria using highresolution magic angle spinning 1 H NMR spectroscopy. FEBS Lett 553, 73–78. 7. Tate, A. R., Foxall, P. J., Holmes, E., Moka, D., Spraul, M., Nicholson, J. K., Lindon, J. C. (2000) Distinction between normal and renal cell carcinoma kidney cortical biopsy samples using pattern recognition of 1 H magic angle spinning (MAS) NMR spectra. NMR Biomed 13, 64–71. 8. Righi, V., Mucci, A., Schenetti, L., Tosi, M. R., Grigioni, W. F., Corti, B., Bertaccini, A., Franceschelli, A., Sanguedolce, F., Schiavina, R., Martorana, G., Tugnoli, V. (2007) Ex vivo HR-MAS magnetic resonance spectroscopy of normal and malignant human renal tissues. Anticancer Res 27, 3195–3204. 9. Chen, J. H., Wu, Y. V., Decarolis, P., O’Connor, R., Somberg, C. J., Singer, S. (2008) Resolution of creatine and phosphocreatine 1 H signals in isolated human skeletal muscle using HR-MAS 1 H NMR. Magn Reson Med 59, 1221–1224. 10. Bollard, M. E., Garrod, S., Holmes, E., Lindon, J. C., Humpfer, E., Spraul, M., Nichol-

son, J. K. (2000) High-resolution 1 H and 1 H–13 C magic angle spinning NMR spec-

11.

12.

13. 14.

15.

16.

17.

18.

19.

troscopy of rat liver. Magn Reson Med 44, 201–207. Garrod, S., Humpfer, E., Spraul, M., Connor, S. C., Polley, S., Connelly, J., Lindon, J. C., Nicholson, J. K., Holmes, E. (1999) High-resolution magic angle spinning 1 H NMR spectroscopic studies on intact rat renal cortex and medulla. Magn Reson Med 41, 1108–1118. Garrod, S., Humpher, E., Connor, S. C., Connelly, J. C., Spraul, M., Nicholson, J. K., Holmes, E. (2001) High-resolution 1 H NMR and magic angle spinning NMR spectroscopic investigation of the biochemical effects of 2-bromoethanamine in intact renal and hepatic tissue. Magn Reson Med 45, 781–790. Lindon, J. C., Nicholson, J. K., Holmes, E. (eds.) (2007) The Handbook of Metabonomics and Metabolomics, Elsevier, New York, NY. Antzutkin, O. N., Shekar, S. C., Levitt, M. H. (1995) Two-dimensional sideband separation in magic-angle-spinning NMR. J Magn Reson A 115, 7–19. Hu, J. Z., Wang, W., Liu, F., Solum, M. S., Alderman, D. W., Pugmire, R. J., Grant, D. M. (1995) Magic-angle-turning experiments for measuring chemical–shift-tensor principal values in powdered solids. J Magn Reson A 113, 210–222. Wind, R. A., Hu, J. Z., Rommereim, D. N. (2001) High resolution 1 H NMR spectroscopy in organs and tissues using slow magic angle spinning. Magn Reson Med 46, 213–218. Bertram, H. C., Hu, J. Z., Rommereim, D. N., Wind, R. A., Andersen, H. J. (2004) Dynamic high-resolution 1 H and 31 P NMR spectroscopy and 1 H T2 measurements in postmortem rabbit muscles using slow magic angle spinning. J Agric Food Chem 52, 2681–2688. Hu, J. Z., Wind, R. A., Rommereim, D. N. (2006) 1 H relaxation times of metabolites in biological samples obtained with non-destructive ex vivo slow-MAS NMR. Magn Reson Chem 44, 269–275. Hu, J. Z., Rommereim, D. N., Minard, K. R., Woodstock, A., Harrer, B. J.,

364

20.

21.

22.

23.

24.

25.

26.

27.

28. 29.

Hu Wind, R. A., Phillips, R. P., Sime, P. J. (2008) Metabolomics in lung inflammation: a high-resolution 1 H NMR study of mice exposed to silica dust. Toxicol Mech Meth 18, 385–398. Hu, J. Z., Wind, R. A., Mclean, J., Gorby, Y. A., Resch, C. T., Fredrickson, J. K. (2004) High-resolution 1 H NMR spectroscopy of metabolically active microorganisms using non-destructive magic angle spinning. Spectroscopy 19, 98–102. Minard, K. R., Guo, X., Wind, R. A. (1998) Quantitative 1 H MRI and MRS microscopy of individual V79 lung tumor spheroids. J Magn Reson 133, 368–373. Hu, J. Z., Rommerein, D. N., Wind, R. A. (2002) High resolution 1 H NMR spectroscopy in rat liver using magic angle turning at a 1 Hz spinning rate. Magn Reson Med 47, 829–836. Wind, R. A., Hu, J. Z., Rommereim, D. N. (2003) High-resolution 1 H NMR spectroscopy in a live mouse subjected to 1.5 Hz magic angle spinning. Magn Reson Med 50, 1113–1119. Wind, R. A., Hu, J. Z., Majors, P. D. (2006) Localized in vivo isotropic–anisotropic correlation 1 H NMR spectroscopy using ultraslow magic angle spinning. Magn Reson Med 55, 41–49. Wind, R. A., Hu, J. Z., Majors, P. D. (2005) Slow-MAS NMR: a new technology for in vivo metabolomic studies. Drug Discov Today Technol 2, 291–294. Wind, R. A., Hu, J. Z. (2006) In vivo and ex vivo high-resolution 1 H NMR in biological systems using low-speed magic angle spinning. Prog Nucl Magn Reson Spectrosc 49, 207–259. Lüdeke, K. M., Röschmann, P., Tischler, R. (1985) Susceptibility artifacts in NMR imaging. Magn Reson Imaging 3, 329–343. Jackson, J. D. (1962) Classical Electrodynamics, 2nd edn, Wiley, New York. Boxerman, J. L., Hamberg, L. M., Rosen, B. R., Weisskoff, R. M. (1995) MR contrast due to intravascular magnetic

30. 31.

32.

33.

34.

35.

36.

37. 38. 39. 40.

susceptibility perturbations. Magn Reson Med 34, 555–566. Kennan, R. P., Zhong, J., Gore, J. C. (1994) Intravascular susceptibility contrast mechanisms in tissues. Magn Reson Med 31, 9–21. Edmond, D. T., Wormald, M. R. (1988) Theory of resonance in magnetically inhomogeneous specimens and some useful calculations. J Magn Reson 77, 223–232. Chu, S. C., Xu, Y., Balschi, J. A., Springer, C. S. Jr. (1990) Bulk magnetic susceptibility shifts in NMR studies of compartmentalized samples: Use of paramagnetic reagents. Magn Reson Med 13, 239–262. Yablonskiy, D. A., Haacke, E. M. (1994) Theory of NMR signal behavior in magnetically inhomogeneous tissues: The static dephasing regime. Magn Reson Med 32, 749–763. Bihan, D. L., Basser, P. J. (1995) Molecular diffusion and nuclear magnetic resonance, in (Bihan, D. L., ed.), Diffusion and Perfusion Magnetic Resonance Imaging, pp. 5–17. Raven Press, New York, NY. Leu, G., Tang, X., Peled, S., Maas, W. E., Singer, S., Cory, D. G., Sen, P. N. (2000) Amplitude modulation and relaxation due to diffusion in NMR experiments with a rotating sample. Chem Phys Lett 332, 344–350. Cotts, R. M., Hoch, M. J. R., Sun, T., Markert, J. T. (1989) Pulsed field gradient stimulated echo methods for improved NMR diffusion measurements in heterogeneous systems. J Magn Reson 83, 252–266. Torrey, H. C. (1956) Bloch equation with diffusion terms. Phys Rev 104, 563–566. Callaghan, P. T. (1993) Principles of Nuclear Magnetic Resonance Microscopy, pp. 160–161. Clarendon Press, Oxford. Abragam, A. (1961) The Principles of Nuclear Magnetism, p. 61. Oxford University Press, Oxford. Ernst, R. R., Bodenhausen, G., Wokaun, A. (1997) Principles of Nuclear Magnetic Resonance in One and Two Dimensions, p. 332, Clarendon Press, Oxford Science Publication.

Chapter 21 Processing and Modeling of Nuclear Magnetic Resonance (NMR) Metabolic Profiles Timothy M.D. Ebbels, John C. Lindon, and Muireann Coen Abstract Modern nuclear magnetic resonance (NMR) spectroscopy generates complex and information-rich metabolic profiles. These require robust, accurate, and often sophisticated statistical techniques to yield the maximum meaningful knowledge. In this chapter, we describe methods typically used to analyze such data. We begin by describing seven goals of metabolic profile analysis, ranging from production of a data table to multi-omic integration for systems biology. Methods for preprocessing and pretreatment are then presented, including issues such as instrument-level spectral processing, data reduction and deconvolution, normalization, scaling, and transformations of the data. We then discuss methods for exploratory modeling and exemplify three techniques: principal components analysis, hierarchical clustering, and self-organizing maps. Moving to predictive modeling, we focus our discussion on partial least squares regression, orthogonal partial least squares regression, and genetic algorithm approaches. A typical set of in vitro metabolic profiles is used where possible to compare and contrast the methods. The importance of validating statistical models is highlighted, and standard techniques for doing so, such as training/test set and cross-validation are described. Finally, we discuss the contributions of statistical techniques such as statistical total correlation spectroscopy, and other correlation-based methods have made to the process of structural characterization for unknown metabolites. Key words: Preprocessing, normalization, scaling, PCA, hierarchical clustering, self-organizing maps, PLS, genetic algorithms, STOCSY, CLASSY.

1. Introduction Modern nuclear magnetic resonance (NMR) technology allows the acquisition of highly reproducible and highly resolved spectra on complex biological mixtures such as biofluids or tissue extracts. The spectra contain thousands of resonances, which T.O. Metz (ed.), Metabolic Profiling, Methods in Molecular Biology 708, DOI 10.1007/978-1-61737-985-7_21, © Springer Science+Business Media, LLC 2011

365

366

Ebbels, Lindon, and Coen

can be overlapped and noisy, reporting on perhaps hundreds of metabolites. Careful data processing and statistical analysis are therefore required to derive useful information from these profiles. In contrast to some other molecular profiling techniques (e.g., DNA microarrays), a large proportion of the signals observed cannot easily be associated with a chemical structure. For unknowns, structural characterization is a labor-intensive and time-consuming task, while even for widely observed metabolites, it is not easy to automatically quantify their levels in crowded spectra obtained in high-throughput experiments. The goals of data analysis in NMR metabolic profiling can be summarized as follows: 1. Production of a data array amenable to statistical analysis, usually a table of N rows (samples) by K columns (metabolic variables, typically peak areas) 2. Exploration and overview of the data, including quality control and identification of outliers 3. Observation of relationships between samples and variables, including clusters and trends 4. Determination of which metabolic variables are responsible for the patterns observed 5. Structural characterization of unidentified metabolite signals 6. Interpretation of metabolic effects at the level of biological pathways 7. Systems biology integration of metabolic information with that from other omics techniques This chapter describes the steps which are taken to address some of these goals. We limit ourselves to 1-D 1 H NMR spectra which are currently the method of choice for high-throughput untargeted metabolic profiling.

2. Materials Two data sets are used to exemplify the methods described in this chapter. The first was obtained from the Consortium for Metabonomic Toxicology (1) and is a sample of 1050 1-D 1 H NMR spectra of urine from control (untreated) Sprague Dawley rats from a large number of preclinical toxicology studies. Each study lasted 7 days, and these data correspond to 24 h urine collected on day 2 (i.e., 24–48 h excretion) after the intervention, e.g. dosing with a toxin, and profiled at a 1 H observation frequency of 600 MHz as detailed in (2). The second data set corresponds to 24 1-D 1 H NMR metabolic profiles of aqueous extracts from rat primary

Processing and Modeling of NMR Metabolic Profiles

367

hepatocytes treated with trichostatin-A (TSA). Cells were sampled at 24 and 48 h post-dose, extracted with chloroform/methanol, and profiled at 600 MHz as detailed in (3). All methods demonstrated in this chapter were performed using in-house routines written in MATLAB (R2009a; the MathWorks, Natick, MA) including the SOM toolbox (http://www.cis.hut.fi/projects/somtoolbox/) and computed on a dual-processor quad core 3.2 GHz Intel Xeon Apple Mac Pro. Nonetheless, commercial and open-source software is available for most techniques, and all methods can be run straightforwardly on standard desktop machines.

3. Methods 3.1. Preprocessing and Pretreatment 3.1.1. Instrument-Level Processing

NMR observations are recorded by the instrument as free induction decays (FIDs) and require instrument-level processing to transform them to a form that is interpretable to the spectroscopist. In the ideal case, each FID consists of a linear superposition of exponentially decaying sinusoids. In the first processing step, known as apodization, the FID is multiplied by a window function, typically a decaying exponential. This suppresses later parts of the FID which mainly contain noise, improving the signal to noise ratio of the resulting spectrum. Exponential apodization also broadens peaks, reducing the resolution of the spectrum, and thus selection of optimal apodization parameters is a trade-off between resolution and signal to noise. Conversely, it is also possible to manipulate the FID using more complex window or weighting functions to improve resolution with only a modest trade-off in signal to noise ratio. Fourier transformation maps the data from the time domain to the frequency domain such that each exponentially decaying sinusoid in the FID corresponds to a single Lorentzian peak in the spectrum. The peak shape will only be Lorentzian (and the time decay exponential) if the magnetic field is homogeneous. Note that the FID is usually acquired in quadrature mode resulting in two components (real and imaginary), and thus a complex Fourier transform must be used, resulting in both real and imaginary components to the final spectrum. At this stage, spectral peaks usually contain phase errors because of, for example, the finite time delay between the NMR excitation pulse and the activation of the receiver. Usually a correction of the phase, which is achieved using a linear combination of the real and imaginary parts, is sufficient to return all peaks to an “absorption-mode” Lorentzian shape, which offers

368

Ebbels, Lindon, and Coen

the best resolution and interpretability. The final instrument-level processing stages (applied to the absorption mode spectrum only) are that of correcting baseline offsets, for example, by subtracting a low-order polynomial fit and calibrating the origin of the chemical shift scale to the position of an internal or other standard. All of these stages have been comprehensively detailed in the literature (4–6). While manual processing has been standard for many years, the automated processing of large numbers of spectra from metabolic profiling experiments has only become possible in the last decade or so. This is largely due to the complexity of the spectra. Phasing is particularly hard to automate, since many algorithms depend on selection of suitable peaks whose phase can be optimized or of regions of baseline devoid of peaks, which may change from sample to sample. Even with the new algorithms, phasing must be performed manually in some cases. 3.1.2. Data Reduction and Deconvolution

Once properly phased, baseline-corrected, and chemical shiftcalibrated spectra are available, the data must be transformed into a format amenable to statistical analysis. Typically one aims for a data table where rows correspond to biological samples and columns to metabolic variables. At this stage, one has a choice between two fundamentally different types of representation: (a) identified and quantified metabolite resonances (7, 8) or (b) raw (9) or binned (10, 11) spectra. There is a strategic difference between the two representations. The peak fitting approach allows quick and straightforward interpretation of statistical models in terms of the metabolites responsible for clustering and class separation. However, automated identification and estimation of metabolite levels from 1-D NMR spectra are hard, due to the presence of peak overlap, variations in peak positions, noise, high dynamic range, differential relaxation effects, and the inherent difficulty of identifying resonances by chemical shift and multiplicity alone. A particular disadvantage here is that new metabolites cannot be recognized as only known peaks are modeled. The alternative approach of modeling the spectra directly allows the analyst to bypass this complex preprocessing step and proceed directly to statistical modeling. This allows for the expensive and timeconsuming process of metabolite identification and quantification to be focused on a minority of relevant peaks in the spectra highlighted by the statistical models. In this representation, the effects of peak shifts may be mitigated somewhat by binning the spectra to a lower resolution or applying spectral alignment algorithms. When analyzing raw or binned spectra, one is faced with the problem of artifacts and unwanted signals present in most biological NMR spectra. These include residual water or other solvent signals (even when solvent suppression pulse sequences are used), internal standards, and resonances resulting from applied

Processing and Modeling of NMR Metabolic Profiles

369

chemical treatments and their metabolites. The simplest and most common method is to delete chemical shift variables corresponding to these peaks, though more sophisticated methods are available in some cases (12, 13). One of the most frustrating problems with NMR metabolic profiles from the statistical point of view is the presence of small but significant shifts in the positions of individual peaks between different samples. Peak fitting methods can take account of such shifts for known peaks, although their presence is a severe hurdle to most algorithms. Another approach is to align peaks so that sample-to-sample shifts are corrected. Many algorithms have been proposed but none are perfect. In particular, few are able to cope when the degree of peak overlap varies significantly between samples, especially when the order of peaks on the chemical shift axis can swap (14). Most algorithms work by segmenting the spectra, matching peaks, and shifting their positions either linearly or with a nonlinear stretch. This latter difference will clearly affect the linear relationship between peak areas and metabolite concentrations, making interpretation of statistical models potentially more difficult. A further issue is the choice of metric used to assess the quality of alignment. Many approaches employ the Pearson correlation, but this is known to be biased toward large peaks (15). 3.1.3. Normalization

Normalization is one of the most significant steps in the processing of NMR spectra. We take all procedures which are intended to remove unwanted variation between profiles to be types of normalization. Normalization is often aimed at removing variation in the total amount of material between samples. This is particularly important for urine studies in which overall sample concentration can vary by orders of magnitude (5), but is also essential for in vitro and tissue extract studies in which the total mass of cells or tissue is difficult to control (for example, due to differential solvent extraction efficiencies). Other intersample inconsistencies can also be accounted for by normalization, such as differential relaxation or variations in RF pulse calibration. Typically, each row of the data table is multiplied by a constant (16), which can be computed in many different ways. The best method depends on the problem at hand. For example, are relative or absolute concentrations required? Is the analysis global or targeted? Use of an internal standard of known concentration (e.g. the chemical shift standard), addition of a known amount of an endogenous substance, an endogenous metabolite quantified by an independent method, or sample dry weight can achieve absolute normalization. Statistical approaches can be very useful when the analysis is untargeted. One of the longest standing of such techniques is that of normalizing to constant total integrated intensity (TII) across the whole profile (11). Despite many advantages, TII normalization has the disadvantage

370

Ebbels, Lindon, and Coen

that it can induce spurious inter-metabolite correlations (16), since, for example, an increase in one metabolite must be compensated by a decrease in the remaining intensity to maintain a constant TII. Alternative methods require selection of a “reference” spectrum against which to normalize the remaining “test” spectra. The “probabilistic quotient” method (17) computes the normalization factor which makes the median fold change between the test and the reference variables a constant across all spectra. “Histogram” normalization (18) minimizes the fit error between the intensity histograms of the test and the reference spectra. 3.1.4. Scaling and Transformations

In any given sample, metabolites can range in concentration over many orders of magnitude. Furthermore, variation in metabolite levels is often linked to concentration, such that higher concentration metabolites have higher variation. Figure 21.1 illustrates this effect via the relationship between the mean and the standard deviation of spectral intensity for a large sample of urines from normal laboratory rats. The curve is seen to divide into two parts. At low mean values, the standard deviation is roughly constant. At higher intensities, the variance rises approximately linearly with the mean. This general form is typical of data from many analytical procedures across a wide range of sciences. The high dynamic range and mean–variance relationship can strongly influence the subsequent statistical analysis. For example, techniques such as principal components analysis (PCA) and partial least squares (PLS) regression model the variance or covariance in the data. Therefore, the most abundant metabolites that exhibit the highest variation will be the strongest influences on these

Fig. 21.1. Standard deviation versus mean intensity for a set of 1050 normalized 1 H NMR spectra of urine from normal laboratory rats (Data courtesy of the COMET project (1)).

Processing and Modeling of NMR Metabolic Profiles

371

models. A practical consequence is that a small number of the highest concentration species (sometimes termed the “usual suspects”) were often found to be the most significant metabolites in early studies. To avoid this problem the variables can be rescaled to adjust their influence in such models. Common scaling methods are given below. 1. Unit variance scaling: Given the mean–variance relation, a natural way to give each variable equal weighting is to divide through by its standard deviation, resulting in a variance of unity for the scaled variable. This works well for peak integrated data, but is not advisable for raw spectra or binned data since the influence of variables containing only noise is greatly amplified. 2. Pareto scaling: Each variable is divided through by the square root of the standard deviation, providing an intermediate approach between no scaling and unit variance scaling. The influence of small, low variance, peaks is increased without greatly amplifying noise variables. 3. Range scaling: Each variable is divided through by its range (maximum – minimum values). This has similar advantages and drawbacks as unit variance scaling, but is less robust to the effects of outliers. 4. Level scaling: Each variable is divided through by its mean. This often has a similar effect to unit variance scaling (due to the mean–variance relation) and therefore has similar advantages and drawbacks. 5. Log scaling: Each variable is log transformed. The objective is to render the noise additive and/or the variable distribution more normal. There are obvious problems when any data values are zero or negative – a common occurrence with both raw spectral/binned and peak integrated data. 6. Generalized log scaling (19): Each variable is transformed by a function that approximates a logarithm at high values but does not diverge at low values. A parameter is required to set the scale of the function, which usually requires a number of experimental replicates to determine the signal variance. One should note that scaling is not always needed, and selection of the optimal method will depend on the modeling approach used. 3.2. Unsupervised and Exploratory Modeling

The output of preprocessing and pretreatment is a data table of N samples (rows) by K variables (columns). Each row will typically correspond to one NMR spectrum from one biological sample, while each column corresponds to a single NMR metabolic variable (e.g., integrated intensity of a given peak). The next stage

372

Ebbels, Lindon, and Coen

is to explore the data to discover its overall structure, including any trends or groupings that may be apparent. At this exploratory stage, unsupervised methods should be used – those that do not assume any prior knowledge, allowing as far as possible an unbiased view of the data set. One important product of this stage will be the identification of outliers. The quality control (QC) aspect of this stage of analysis is particularly important, since some trends or groupings may result from confounding factors in the experiment. An iterative process can be followed in which outliers are excluded and the data remodeled, revealing further, more subtle outliers which are then also removed, until no outlying samples remain. It is important to note that there has to be a valid analytical (e.g., poor signal to noise ratio or inadequate water peak suppression) or biological reason (e.g., from other data the sample is known to be anomalous) to exclude spectra, and it is not allowable to remove them because they do not fit any preconceived hypothesis. Throughout this process, the overall structure and the major sources of variation in the data set become clear. In fortunate situations, the expected phenotypic groupings or trends are then clearly observed. If not, and the hoped-for effects are subtle, we may wish to progress to supervised methods to draw out the relevant information. We highlight below some of the most commonly used methods and for each, give the aim and a basic description of the algorithm and discuss example results. 3.2.1. Principal Components Analysis (PCA)

PCA is probably the most commonly used multivariate statistical technique in metabolic profiling. It is primarily used to reduce the dimensionality of the data, enabling relationships between samples and variables to be visualized in low (two or three)dimensional plots. The basic premise of PCA (as with all latent variable models) is that the data exist in a space of much lower dimension than the dimension of the space in which they are recorded. For example, a 1-D NMR spectrum with 32k data points may be considered as a point in a space of 32k dimensions. However, because of strong correlations between variables, the majority of the variance of a typical metabolic profile data set of a few tens of NMR spectra can often be summarized by less than 10 principal components (PCs), meaning that the data can be well approximated as lying in a subspace of less than 10 dimensions. At the outset, PCA identifies the direction of greatest variance in the data. This direction is designated the first principal component (PC1), and the data are projected (orthogonally) to this new “axis.” The projected coordinates are termed the “scores,” while the cosines of the angles made by the PC with the original coordinate axes are termed the “loadings.” Subsequent components can

Processing and Modeling of NMR Metabolic Profiles

373

then be calculated with the proviso that each must (a) be orthogonal to the previous ones and (b) summarize the greatest variance remaining at that point. So, for example, PC2 will select the direction of greatest variance from those orthogonal to PC1. There are many algorithms to calculate PCA, though one of the most popular in chemometrics (and thus metabolic profile analysis) is nonlinear iterative partial least squares (NIPALS (20)), which is implemented in most common software packages for multivariate statistics. An example of a typical PCA using 1 H NMR metabolic profiles of aqueous extracts from rat primary hepatocytes treated with trichostatin-A (TSA) and controls is shown in Fig. 21.2. The plot of the first two score vectors shows some clustering along the first principal component, with the control group split by samples from the TSA-dosed group. Clear clustering according to time point is also evident with 24- and 72-h control groups clustering at the right and left of the plot, respectively. One 72-h TSA sample appears on the far side of the 72-h control group – a possible outlier, the reasons for which could be investigated further. The corresponding loadings plot indicates the variables linked to the structure seen in the scores plot. In this model, the variables correspond to adjacent regions of integrated NMR intensity (0.01 ppm width) and have been Pareto scaled to reduce the influence of large peaks. In general, directions on the scores and loadings plots can be linked. For example, variables on the right of the loadings plot must be higher in samples on the right of the scores plot than those on the left. In the figure, some variables corresponding to resonances from glucose and glycogen are indicated. Since these are on the same side of the origin as the 24-h control samples, one can infer that glycogen and glucose decrease over time and are lower in the 72-h control samples, which are on the opposite side of the scores plot. This is one reason why

Fig. 21.2. PCA of NMR spectra from aqueous extracts of rat hepatocytes. a Scores plot for the first two components showing samples from controls (open circles) and TSA-treated hepatocytes (filled squares). Labels give the time in hours post-dose. b Loadings for the first two components. Labels refer to the central 1 H chemical shift of each 0.01 ppm width bin. Some variables corresponding to resonances from glucose and glycogen are ringed.

374

Ebbels, Lindon, and Coen

PCA and other projection methods are so useful in metabolic profiling; interpretation of biological changes and differences is fast and straightforward. One must bear in mind, however, that a PC model only explains a proportion of the total variation in the data set (69.6% for the first two components in this case) and that one should examine further PCs to be sure that the correct interpretation is made. This raises a key question in PC modeling, which is the choice of the number of components to use. There are many methods of selecting the optimal number of components, and the optimal one will depend on the intended use of the PC model. For example, is one interested in data exploration (as above) or are we using PCA to preprocess the data for another technique (when perhaps a larger number of components would be desirable)? In most cases we recommend the use of cross-validation to assess how well left-out data fits the model, as described in Section 3.4. 3.2.2. Hierarchical Clustering

Hierarchical cluster analysis (HCA, see, e.g. (21)) is widely used in modeling “omics” data and has been extensively applied to metabolic profiles from several platforms. Its ability to group profiles according to their similarity without any prior knowledge makes it very useful in exploratory data analysis. The most common “agglomerative” form requires the choice of two input functions. The first is the metric to be used to measure similarity between profiles. Common choices for this metric in metabolic profiling include Euclidean and Pearson correlation distances. The second is the so-called linkage function, which defines how similarities between clusters of profiles are calculated. For example, centroidal linkage defines the similarity between two clusters as the similarity between their centroids, while in single (sometimes called nearest neighbor) linkage the similarity of clusters is defined as the similarity of their two most similar members. A wide array of alternative linkage functions exist, such as complete, average, and Ward’s (21). The choice of metric and linkage as it will have a significant impact on the clustering structure observed. The algorithm begins by designating each individual profile as a cluster at the lowest level. The algorithm then iterates the following two steps: (1) find the most similar pair of clusters and (2) designate these as a new cluster and return to step (1). In this way, small clusters are grouped into larger clusters in a hierarchical fashion, hence the name of the technique. The final result is a list of clusters and their memberships at each level of the hierarchy. The hierarchy is usually visualized as a tree diagram or dendrogram that gives an intuitive illustration of the hierarchical structure.

Processing and Modeling of NMR Metabolic Profiles

375

Fig. 21.3. HCA dendrograms of rat hepatocyte data. a Centroidal linkage, b Single linkage. Labels indicate dose group (1 – control, 2 – TSA) and time point (24 or 72 h).

Examples of HCA dendrograms for the rat hepatocyte data set using different linkages are shown in Fig. 21.3. It is immediately clear that the two different linkages give different clustering results. No clear set of clusters is visible in either dendrogram. However, on closer inspection of the centroidal plot, one observes a strong effect of time point with all but two of the 24-h samples clustering into two groups at the top and bottom of the plot. The clustering structure of the 72-h samples is more heterogeneous. One 72-h class 2 (TSA-treated) sample appears to be an outlier in both plots (lowest branch of dendrogram in Fig. 21.3b). In the centroidal plot, some links of the tree appear to have reversed direction such that two clusters merge at a higher level of similarity than their corresponding subclusters. This is a characteristic of centroidal linkage, and although visually awkward, this does not usually impede interpretation of cluster structure. Such reverse links do not appear in the single linkage dendrogram. In comparison with PCA, an advantage of HCA is that similarities in the full multidimensional space can be visualized in a single plot. Conversely, drawbacks are that relationships between objects are 1-D (i.e., similarity is a univariate parameter) and trees become hard to read when there are many objects. A common pitfall when interpreting dendrograms is to make inferences about similarity from the 1-D ordering of objects on the leaves of the tree. This is erroneous since each sub-tree in the hierarchy can be rotated, i.e., its ordering reversed without changing the representation of hierarchical structure. The correct way to interpret the dendrogram is from the root downward (right to left in Fig. 21.3), noting which objects or subclusters are grouped together. For example, in Fig. 21.3a, it is tempting to suggest that all the 72-h samples cluster together. In fact, the 72h samples are part of a large cluster also including the eight 24-h samples at the top of the plot. A common approach is to look for the longest continuous region on the similarity axis where the

376

Ebbels, Lindon, and Coen

number of clusters is maintained. However, this may not be a great deal longer than the next longest region, and therefore it is often hard to clearly define the optimal number of clusters. 3.2.3. Self-Organizing Maps

The self-organizing map (SOM) (22) is another unsupervised clustering and visualization tool which has been widely used in analysis of omic data. The technique maps vectors in a high K-dimensional input space to a regular array of nodes in a low (usually two)-dimensional space – the map. Each node corresponds to a K-dimensional reference or “codebook” vector, which defines the prototype for that position on the map. In the classic SOM training algorithm, reference vectors are initialized either randomly or via a PCA of the training set, and each object is presented to the map sequentially. For each object, the node whose reference vector is closest (the “winning” node) is noted, and its reference vector is updated to make it more similar to the presented object. Reference vectors in the neighborhood of the winning node are also updated, although to a lesser extent than the winning node, using a “neighborhood function” (usually a 2-D Gaussian or similar decaying function of distance). The algorithm then returns to the first object with the new reference vectors and proceeds until the reference vectors do not change position significantly between iterations. In this way, a smooth map is built in which adjacent nodes have more similar reference vectors than distant nodes; the SOM nodes model the density of data in the K-dimensional input space. The original algorithm described above suffered from the problem that different maps could be built with the same data, depending on the order in which the training examples were presented. However, a batch SOM algorithm is now commonly used in which the ordering of the data has no effect (23). Figure 21.4 illustrates the SOM approach applied to the rat hepatocyte data using a 3 × 8 hexagonal lattice. The SOM can be visualized in many ways. One of the most useful is the “hit histogram” in which the number of samples mapping to each cell is displayed. As with the PCA, the controls split into two groups, locating at opposite corners of the map. The TSA-treated samples map in between the two control clusters. The other widely used visualization is the “component plane,” which displays the contribution to the reference vectors of a single variable. In this case, we examine the variable at a 1 H NMR chemical shift of 3.235 ppm corresponding to glycerophosphocholine (GPC). The color scale indicates that GPC is elevated in both control groups compared to TSA-treated group. The relationship between the SOM and the PCA is illustrated in Fig. 21.4b where the reference vectors and the original data are projected into the space of the first two PCs. One can see how the map covers different parts of the space with different density resulting in a nonlinear transformation of the input coordinates.

Processing and Modeling of NMR Metabolic Profiles

377

Fig. 21.4. Self-organizing map analysis of the rat hepatocyte data. a Hit histogram where number of samples mapping to each unit is represented by the size of the hexagon (black – control, gray – TSA treated), b Component plane visualization for the variable 3.235 ppm (GPC), c Visualization of the map and input data in the space of the first two principal components. Solid lines connect nodes on the map. Open circles – controls, filled squares – TSA treated.

An important difference between the SOM and other mapping techniques such as PCA is that the map is discrete. Each input vector is mapped to a single winning node. New data can easily be classified in this way, and one can visualize the areas of the map to which test data of different classes are mapped. A drawback of this granularity, however, is that, when there are few training data, the map must be small to prevent over-fitting. In this case, it is hard to visualize subtle differences between objects since a relatively large change is required to map to a different winning node. Another difficulty with SOMs is that of local minima. The procedure can result in a “twisted” map where similar objects are not mapped to adjacent nodes. These problems can be avoided by using resampling techniques and checking for consistency of the map structure. 3.3. Supervised Modeling

Supervised methods are those which use a priori known structure to learn patterns and rules with which to predict new data. Regression is an example of a supervised approach in which a relationship is learnt between a matrix of predictors (often denoted X) and a matrix or vector of responses (usually denoted Y). In regression, the responses are usually taken to be continuous parameters, e.g., age, blood pressure, enzyme activity. The other major type of supervised method is classification in which one searches for a rule for classifying examples into one of several (possibly overlapping) classes. Regression and classification are closely linked, and most regression methods can be used in classification mode by using a suitable response vector which encodes the class membership. Supervised methods are powerful, and it is often possible to develop a model which explains the data very well (e.g., 100%

378

Ebbels, Lindon, and Coen

successful classification), even when there is no true relationship in the data. This is largely a by-product of the high dimensionality of omics data sets; there are just too many degrees of freedom with which to predict the response. Thus, validation of the model is extremely important as discussed below in Section 3.4. As with unsupervised methods, suitable supervised methods for omics analysis must be able to handle highly collinear noisy variables, usually with many more variables than samples. Classical multiple linear regression (MLR) fails under these conditions since covariance matrices are poorly conditioned, leading to extremely unstable estimation of regression coefficients. Thus, one must move to more specialized techniques. A wide variety of methods have been used; here we exemplify two widely used supervised approaches: partial least squares and genetic algorithms. 3.3.1. Partial Least Squares Regression and Discriminant Analysis

Partial least squares (PLS) regression (see, e.g., (24) and references therein) is often considered the natural extension of PCA to supervised modeling. Like PCA, PLS is a latent variable approach that assumes the data can be well approximated by a low-dimensional subspace within the high-dimensional input space. Latent variables (the PLS components) are assumed to be linear combinations of the input variables which span this subspace. This means that PLS also involves linear projection of the data to the latent variable space and that it is also well suited to dealing with many highly correlated variables. In PLS, as with PCA, components explaining small amounts of variance are not retained in the model, and thus noise can be rejected. The NIPALS algorithm for PLS is also able to deal with a degree of missing data. PLS is also able to construct a model in which more than one response is related to the predictors, so-called twoblock regression. All the above properties make PLS ideally suited to modeling of metabolic profile data. A PLS model is composed of two linked submodels of the X and Y spaces. Each space is modeled by a set of latent variables summarized by scores and loadings. The objective of PLS is to maximize the covariance between the scores in the X and Y spaces (25). Note that it is covariance, not correlation, which is maximized, and thus the scale of individual variables will affect the model. As for PCA, components are orthogonal (uncorrelated) to each other and ordered by decreasing magnitude of covariance. In PLS, a set of weights are additionally calculated which link each component in the X space to its partner in the Y space. Interpretation of the model usually relies on investigation of these weights rather than the conventional loadings. In metabolic profiling, the approach is most commonly used to classify samples from two or more groups and is termed PLS discriminant analysis (PLS-DA). In this case, each Y variable con-

Processing and Modeling of NMR Metabolic Profiles

379

Fig. 21.5. PLS-DA of rat hepatocyte data. a Scores plot for the first two PLS-DA components. b Loadings plot for the PLSDA model. Variables corresponding to phosphocholine (PC, 3.225 ppm) and glycerophosphocholine (GPC, 3.235 ppm) are ringed.

sists of a “dummy” vector, one for each class, set to one if the corresponding sample belongs to the class and zero otherwise. This “one-of-K” encoding must be used in preference to a single column of values 1. . .K when there are more than two classes since the classes cannot usually be assumed to be ordered. Figure 21.5 illustrates PLS-DA applied to the rat hepatocyte data and should be compared to the PCA result (Fig. 21.2). As expected, use of the supervised method has resulted in a clearer separation between the control and the TSA-dosed classes. However, some subclustering according to time is still apparent in both classes. The corresponding loadings plot clearly highlights two variables (ringed) which differentiate the classes at 72 h. The interpretation of the PLS weights is identical to that of PCA loadings. One can infer that glycerophosphocholine (GPC, 3.235 ppm) is reduced in TSA-dosed samples compared to controls at 72 h. Conversely, phosphocholine (PC, 3.225 ppm) is elevated at this time point. Orthogonal PLS (O-PLS) (26) is a more recent development of PLS in which the regression model is split into two parts. The first part (the “predictive” components) models variation in X which is linearly related to the response Y, while the second part (the orthogonal components) comprises variation that is linearly uncorrelated to the response. While the model has identical predictive power to standard PLS (for the same number of components), its advantage lies in improved interpretability since only the predictive components need be investigated to find variables associated with the response. In this way, variation not associated with the response can be filtered out. This so-called structured noise is often induced by uncontrolled confounding factors such as population heterogeneity due to differences in subject age or diet, or temporal effects.

380

Ebbels, Lindon, and Coen

Fig. 21.6. O-PLS-DA analysis of the rat hepatocyte data using a (1+1) model. a Scores plot of the predictive versus orthogonal components. b Correlation – covariance plot of the O-PLS regression coefficients with 1 H chemical shift indicated by shading. The two variables corresponding to PC (3.225 ppm) and GPC (3.235 ppm) are ringed.

Figure 21.6 shows O-PLS-DA applied to the rat hepatocyte data. The data have been modeled using one predictive and one orthogonal component (a so-called 1 + 1 component model), and it is instructive to compare the scores plot to that of the PLS-DA model in Fig. 21.5. In the PLS-DA case, separation of the classes required two components, and thus the weights of both components had to be examined to interpret the model. With O-PLSDA, the model is rotated so that class discrimination is apparent on the first component only; there is no variation correlated to the class differences in the orthogonal component. Thus only the first component weights need to be examined to ascertain variables important in separating the groups. One helpful method for investigating the model is to plot the weights versus their “backscaled” version (in which each weight is scaled by the unit variance scale factor for the corresponding variable (9)). This plot, similar to the “S-plot” of ref. (27), summarizes the correlation and covariance of each variable with the response. Influential and reliable variables are those at the top right/bottom left of the plot. Again, for this data set we can observe that GPC levels have an important role in separating the classes, while PC is also influential in the opposite direction, although to a lesser extent. 3.3.2. Genetic Algorithms and Programming

Genetic algorithms (GAs, see, e.g., (28)) are a class of global optimization algorithms widely used in machine learning for a variety of tasks. They are especially useful in cases where the search landscape is multimodal, i.e., there are many solutions which are locally optimal, but perhaps not close to the global optimum. In metabolic profiling, GAs have often been applied to the problem of variable selection, for example, refining the very large number of measured metabolic variables down to a small set which are highly predictive of the desired biological property.

Processing and Modeling of NMR Metabolic Profiles

381

GAs are inspired by the process of biological evolution. Each solution (or individual) is represented by a string of numbers known as a chromosome. Often the representation is binary, as in the variable selection problem, where all chromosome elements (the “genes”) are zero except those representing selected variables, which are set to one. The GA simulates a population of solutions, some of which are “fitter” (more optimal) than others. The algorithm ensures that the genes from the fitter individuals survive while genes in less-fit individuals die out. The general algorithm runs as follows. Begin 1. Initialize the population chromosomes to random binary strings. 2. Repeat i. Evaluate the fitness of all solutions in the population. ii. Repeat i. Select parents to reproduce according their fitness, with fitter individuals more likely to be selected. ii. Produce children using genes from pairs of parents using genetic crossover. iii. Introduce small random variations in the population by using mutation. Until new population is full. Until the maximum number of generations has been reached. End The process of crossover typically picks a random point inside each parent and creates the children by swapping the genes after the crossover point between the parents. Mutation allows the search process to explore new territory by randomly changing a small number of genes from zero to one or vice versa. It must be remembered that the GA is just a search tool and, in order to help model metabolic profiles, it must be coupled to a classification or regression function so that fitness can be calculated. A typical fitness function would be the success rate of classification of the training set via cross-validation. Another point is that a GA is a stochastic algorithm and must be run multiple times and the solutions aggregated in order to achieve robust conclusions. In the variable selection problem, one can examine the frequency with which the solutions choose each variable. In metabolic profiling, this is easily visualized as a histogram of selection frequency superimposed on the profile itself or a table of the most frequently selected variables. If a classifier based on the optimal variable set is desired, it is best to train a new classifier on this set rather than selecting solutions from the GA run, as each is subject to stochastic fluc-

382

Ebbels, Lindon, and Coen

Fig. 21.7. Metabolic profile classification by a simultaneous sample and variable selection genetic algorithm. a PCA scores plot for one solution showing samples from liver (triangles) and kidney (circles) toxins. Filled points are those selected by the GA, and the PCA corresponds to the space of the selected variables. The solid line shows the 1-nearest neighbor decision surface. b Kernel density visualization of sample selection. Dark regions correspond to parts of the input space that are over selected compared to a null model. Reproduced from Cavill et al. (29) by permission of Oxford University Press.

tuations and not all optimal variables may be present in each individual. One example of the GA approach applied to metabolic profiles is illustrated in Fig. 21.7. In this case (29), the task was to classify NMR spectra of urine from rats treated with model toxins as to whether the target organ was the liver or kidney. The GA was asked to select both variables and samples from the training set to find the optimal nearest neighbor classifier. The figure shows two visualizations. In the left panel, one sees a PCA scores plot of the space of the selected variables for one particular solution. Only a small proportion of the samples have been selected by the GA, and the nearest neighbor decision surface shows that few samples are incorrectly classified. The right panel is a plot developed to visualize samples which are over selected by the GA, i.e., more often than would be expected at random. The plot is a PCA in the full variable space with a weighted kernel density map placed on top of the data points. The kernels are weighted by the number of times a given sample was selected by the GA, and the gray scale compares this to the 95th percentile of a null model. The dark regions indicate that just a few samples are selected more often than would be expected at random and are thus prototypical of their class. The approach of selecting both samples and variables simultaneously not only resulted in a better classification error but also considerably improved the efficiency and speed of the whole approach.

Processing and Modeling of NMR Metabolic Profiles

3.4. Model Validation

383

The process of model validation is one of the most important parts of the modeling process. As mentioned above, in highly multivariate, noisy data sets, it is usually very easy to produce a model that appears to explain the data in hand (e.g., with good classification accuracy), but how do we know that we can rely on its predictions and interpretation? Various methods exist for testing model validity, including examination of the model fit and determination of uncertainties in model coefficients. However, perhaps the most important principle is that a valid model should be predictive, i.e., it should be able to model unseen data almost as well as the data used in its construction. The most intuitive way to assess the predictivity of a model is using training/test set validation. In this scenario, the available data are randomly split into two groups. The model is built with one part (the training set) and then the other set (the test set) is predicted using this model. The fit or prediction error of the test data gives an idea whether the model is valid in the predictive sense or not. Typically a third to a half of the initial data are reserved for the test set, which is sampled in such a way as to make it representative of the whole data set (e.g., ensuring equal proportions from each class). One of the drawbacks of this approach is that not all the data are used to build the model from which the final interpretation is drawn. This is especially difficult when the data are few. An approach which circumvents this problem is that of cross-validation. In this method, the data are divided into m non-overlapping test sets such that each sample is a member of only one set. Each set is left out one at a time, the model built with the remaining data, and the test set predicted. At the end of the m “folds” of cross-validation, the prediction errors are combined to assess overall performance in the same way as for simple training/test set validation. Typically, only 10% or fewer samples are included in any one of the m test sets. In this way, all samples are predicted exactly once and a much higher proportion of the data are used to train the model, leading to more accurate parameter estimation. Other resampling validation schemes exist (e.g., Monte Carlo cross-validation, bootstrap, jacknife), a discussion of which is beyond the scope of this chapter. Models that are able to fit the training data well, but exhibit large errors on the test set, are often said to be over-fitted. This implies that the algorithm is not modeling reliable structure in the data, but is fitting noise. The avoidance of over-fitting is an important concern when optimizing parameters which control model complexity, such as the number of components in a PC model. As the model is made more complex, the fit to the training data will improve. However, while the test set prediction error initially

384

Ebbels, Lindon, and Coen

decreases, there comes a level of complexity beyond which prediction error begins to increase. After this point, the model is said to be over-fitted, and the optimal complexity is often chosen to be that which minimizes prediction error. 3.5. Statistical Methods for Structural Characterization of Metabolites

Five years ago the contribution of statistical modeling to metabolic profiling experiments would typically have ended with the interpretation of the predictive model in terms of metabolic features associated with the phenotype. If the predictive feature was an unidentified peak, the full range of chemical analytical technologies could be employed in elucidating the metabolite responsible. However, in recent years, statistical approaches have appeared which can aid this process, allowing expensive and timeconsuming experiments to be more effectively targeted and/or suggesting possible candidates for further investigation. Of these, the most prominent have been those based on the idea of statistical spectroscopy in which statistical correlations between spectral variables are examined to reveal signals originating from the same molecule. While the idea was originally developed for near-infrared spectroscopy (30), its first application in the metabolic profiling arena was as statistical total correlation spectroscopy (STOCSY) (31). In STOCSY, the pair-wise Pearson correlations between 1-D

Fig. 21.8. Two-dimensional STOCSY plot of the 8.85–9.3 ppm resonances of N-methylnicotinamide (NMND) from 1 H NMR spectra of a sample of 1050 urines from normal laboratory rats. Pearson correlations above 0.7 are shown in gray scale. Mean 1-D spectra are shown along each axis. Data courtesy of the COMET project (1).

Processing and Modeling of NMR Metabolic Profiles

385

NMR spectral variables across a set of samples are computed. The correlation values can then be visualized as an image (2-D STOCSY) or mapped onto a 1-D spectrum using a color code. Figure 21.8 shows an example derived from a large sample of 1-D 1 H NMR spectra of urine from normal laboratory rats in the COMET project (1). The mean spectra are plotted along the axes of the figure, and statistical correlations between two doublets and a singlet from the metabolite N-methylnicotinamide (NMND) can be clearly identified away from the diagonal of the plot. The features in the plot appear distorted due to the effects of small changes in peak position between different 1-D spectra, perhaps due to minor variations in pH or ionic strength, but this anomaly can be mitigated if a peak alignment algorithm is applied. For most data sets, it is possible to select a threshold above which almost all correlations reflect structural relationships (32), indicating that the tool can be rigorously used in structural assignment tasks. The STOCSY approach has been applied in a wide variety of NMR metabolic profiling contexts. For structural assignment, the technique has been applied to LC-NMR data (33) and diffusionedited NMR (34), where correlations derive from LC elution profiles and diffusion attenuation, respectively. Combining data from several different observed nuclei in heteronuclear NMR experiments results in HET-STOCSY (35–37), which has enabled cross-assignment of resonances between 1 H, 31 P, and 19 F spectra, as well as editing of the homonuclear STOCSY according to heteronuclear correlations. A further extension allows crossplatform correlation, such as between NMR spectroscopy and mass spectrometry data for improved biomarker characterization (38). An interesting recent development in this area is cluster analysis statistical spectroscopy (CLASSY) (39) shown in Fig. 21.9. The approach starts by detecting peaks in the mean spectrum and computing Pearson correlations between them (as in STOCSY). The correlations are then iteratively thresholded, and the resulting connectivity matrix is clustered to find cliques of fully connected peaks. These cliques will often represent the individual resonances derived from a single metabolite. The algorithm proceeds at the next level by applying hierarchical clustering between these identified cliques and visualizing the data in a heat map where colors correspond to fold change of the metabolite with respect to a control level. The approach has a number of advantages over basic STOCSY. For example, by applying the recursive local clustering, correlations between peaks from the same molecule are identified with higher accuracy than by using a simple thresholding approach. In this way, variables can be more intuitively ordered (i.e., all variables from the same metabolite together) than using a chemical shift ordering (as in

386

Ebbels, Lindon, and Coen

Fig. 21.9. Cluster analysis statistical spectroscopy (CLASSY). a Peaks detected in a set of control rat urine spectra are identified. b The locally clustered correlation network. c Heat map of Pearson correlations, ordered by hierarchically clustering the local clusters. Reprinted with permission from (39). Copyright 2009 American Chemical Society.

STOCSY). The following hierarchical fold–change visualization is thus greatly enhanced since the response of individual and related metabolites can easily be seen and interpreted as a group. It will be interesting to see further applications of this technique in the near future. 3.6. Conclusions

In this chapter, we have summarized the main steps involved in processing and modeling of NMR metabolic profiles. We have described and demonstrated some of the most commonly used techniques and illustrated some of their strengths and weaknesses. Overall, it must be remembered that the optimal data analysis method will vary depending on the data acquired and goals of the experiment. Yet, the basic stages described above will usually be present in every analysis, and the corresponding issues should be considered when designing the analysis strategy. The analysis of metabolic profile data is a field that is still evolving rapidly, and there are many challenges still to be addressed. For NMR data, some of the most important are the following: 1. Automated assignment and quantification of metabolites from high-throughput global profiling. 2. Analysis of metabolic profile data at the pathway level, including automated mapping of profile data to pathways.

Processing and Modeling of NMR Metabolic Profiles

387

3. Cross-platform integration of NMR metabolic profiles with those from LC- and GC-MS. 4. Systems biological integration of metabolic profiles with, for example, genomic, transcriptomic, and proteomic profiles. Many researchers are active in these and other areas, and new methods and tools are appearing every week. We look forward to a multitude of exciting new developments in the years to come.

Acknowledgments The members of the Consortium on Metabonomic Toxicology and of the EU Framework Programme 6 integrated project CARCINOGENOMICS (contract no. PL037712) are acknowledged for the data used to demonstrate methods in this chapter. References 1. Lindon, J. C., et al. (2005) The consortium for metabonomic toxicology (COMET): aims, activities and achievements. Pharmacogenomics 6, 691–699. 2. Ebbels, T. M. D. et al. (2007) Prediction and classification of drug toxicity using probabilistic modeling of temporal metabolic data: the consortium on metabonomic toxicology screening approach. J Proteome Res 6, 4407–4422. 3. Ellis, J. K. et al (2009) Effect of the histone deacetylase inhibitor trichostatin a on the metabolome of cultured primary hepatocytes. J Proteome Res 9(1), 413–419. 4. Claridge, T. D. W. (2009) High-resolution NMR techniques in organic chemistry, in (Baldwin, J., Williams, R. M., Backvall, J.-E., eds.), Tetrahedron Organic Chemistry, 2nd ed., Vol. 27, Elsevier Science, Amsterdam, p. 398. 5. Nicholson, J. K., et al. (1989) Highresolution proton magnetic-resonance spectroscopy of biological-fluids. Prog Nucl Magn Reson Spectrosc 21, 449–501. 6. Lindon, J. C., et al. (2001) Pattern recognition methods and applications in biomedical magnetic resonance. Prog Nucl Magn Reson Spectrosc 39, 1. 7. Weljie, A. M., et al. (2006) Targeted profiling: quantitative analysis of 1 H NMR metabolomics data. Anal Chem 78, 4430– 4442. 8. Lewis, I. A., et al. (2007) Method for determining molar concentrations of metabolites in complex solutions from two-dimensional

9.

10.

11.

12. 13. 14.

15.

16.

H-1-C-13 NMR spectra. Anal Chem 79, 9385–9390. Cloarec, O., et al. (2005) Evaluation of the orthogonal projection on latent structure model limitations caused by chemical shift variability and improved visualization of biomarker changes in 1 h NMR spectroscopic metabonomic studies. Anal Chem 77, 517. Holmes, E., et al. (1994) Automatic data reduction and pattern-recognition methods for analysis of 1 H nuclear-magneticresonance spectra of human urine from normal and pathological states. Anal Biochem 220, 284. Spraul, M., et al. (1994) Automatic reduction of NMR spectroscopic data for statistical and pattern recognition classification of samples. J Pharm Biomed Anal 12, 1215–1225. Ebbels, T. M. D. et al. (2004) Methods for Spectral Analysis and Their Applications: Spectral Replacement, United States. Antoine, J. -P., et al. (2000) Water peak suppression: time-frequency vs time-scale approach. J Magn Res 144, 189–194. Csenki, L., et al. (2007) Proof of principle of a generalized fuzzy hough transform approach to peak alignment of onedimensional 1 h NMR data. Anal Bioanal Chem 389, 875–885. Veselkov, K., et al. (2009) Recursive segment-wise peak alignment of biological 1 h NMR spectra for improved metabolic biomarker recovery. Anal Chem 81, 56–66. Craig, A., et al. (2006) Scaling and normalization effects in NMR spectroscopic

388

17.

18. 19.

20.

21. 22. 23.

24. 25. 26. 27.

28. 29.

30.

Ebbels, Lindon, and Coen metabonomic data sets. Anal Chem 78, 2262–2267. Dieterle, F., et al. (2006) Probabilistic quotient normalization as robust method to account for dilution of complex biological mixtures. Application in H-1 NMR metabonomics. Anal Chem 78, 4281–4290. Torgrip, R. J. O., et al. (2008) A note on normalization of biofluid 1d H-1-NMR data. Metabolomics 4, 114–121. Durbin, B. P., et al. (2002) A variancestabilizing transformation for geneexpression microarray data. Bioinformatics 18 (Suppl 1), S105–S110. Wold, H. (1966) Estimation of principal components and related models by iterative least squares, in (Krishnaiaah, P. R., Ed.), Multivariate Analysis, Academic, New York, NY, pp. 391–420. Duda, R. O., et al. (2000) Pattern Classification, 2nd ed, Wiley, New York, NY, p. 680. Kohonen, T. (1990) The self-organizing map. Proc IEEE 78, 1464. Kohonen, T. (2001) Self-organizing maps, in (Huang, T. S., Kohonen, T., Schroeder, M. R., eds), Springer Series in Information Sciences, 3 ed, Springer, New York, NY, p. 501. Wold, S., et al. (2001) PLS-regression: a basic tool of chemometrics. Chemom Intell Lab Syst 58, 109–130. Höskuldsson, A. (1988) PLS regression methods. J Chemom 2, 211–228. Trygg, J., et al. (2002) Orthogonal projections to latent structures (O-PLS). J Chemom 16, 119–128. Wiklund, S. et al. (2007) Visualization of GC/TOF-MS-based metabolomics data for identification of biochemically interesting compounds using OPLS class models. Anal Chem 80, 115–122. Mitchell, M. (1996) An Introduction to Genetic Algorithms, MIT Press, Cambridge, MA, p. 205. Cavill, R., et al. (2009) Genetic algorithms for simultaneous variable and sample selection in metabonomics. Bioinformatics 25, 112–118. Noda, I. (1990) 2-Dimensional infrared (2d ir) spectroscopy – theory and applications. Appl Spectrosc 44, 550–561.

31. Cloarec, O., et al. (2005) Statistical total correlation spectroscopy: an exploratory approach for latent biomarker identification from metabolic 1 H NMR data sets. Anal Chem 77, 1282. 32. Couto Alves, A., et al. (2009) Analytic properties of statistical total correlation spectroscopy (STOCSY) based information recovery in 1 h NMR metabolic data sets. Anal Chem 81, 2075–2084. 33. Cloarec, O., et al. (2007) Virtual chromatographic resolution enhancement in cryoflow LC-NMR experiments via statistical total correlation spectroscopy. Anal Chem 79, 3304–3311. 34. Smith, L. M., et al. (2007) Statistical correlation and projection methods for improved information recovery from diffusion-edited NMR spectra of biological samples. Anal Chem 79, 5682–5689. 35. Coen, M., et al. (2007) Heteronuclear 1 h–31p statistical total correlation NMR spectroscopy of intact liver for metabolic biomarker assignment: application to galactosamine-induced hepatotoxicity. Anal Chem 79, 8956–8966. 36. Keun, H. C., et al. (2008) Heteronuclear 19f-1 h statistical total correlation spectroscopy as a tool in drug metabolism: study of flucloxacillin biotransformation. Anal Chem 80, 1073–1079. 37. Wang, Y., et al. (2008) Magic angle spinning NMR and 1 h–31p heteronuclear statistical total correlation spectroscopy of intact human gut biopsies. Anal Chem 80, 1058–1066. 38. Crockford, D. J., et al. (2006) Statistical heterospectroscopy, an approach to the integrated analysis of NMR and UPLCMS data sets: application in metabonomic toxicology studies. Anal Chem 78, 363–371. 39. Robinette, S. L., et al. (2009) Cluster analysis statistical spectroscopy using nuclear magnetic resonance generated metabolic data sets from perturbed biological systems. Anal Chem 81, 6581–6589.

SUBJECT INDEX

A Acylcarnitines . . . . . . . . . 55–71, 78, 231, 233, 237, 243–244 Aging. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7, 11 Alignment . . . . . . . . 198, 203, 209, 230, 234, 240, 242, 244, 250–251, 254, 256, 278–280, 283–284, 291, 295, 368, 369, 385 Amino acids . . . 25–52, 55–57, 68, 73–74, 77–78, 101, 141, 147, 192, 214, 230–231, 237, 244, 301

166–167, 192, 195–197, 201–203, 207–208, 210, 235, 240, 259, 281 Detector, UV . . . . . . . . . . . . . . . . . . . . 178, 180, 308, 314, 316 Diagnostic medicine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 Drug discovery . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173–189 metabolism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73, 96 Dynamic range . . . . . . . . . . . 48, 52, 135, 253, 303, 322, 326, 368, 370

B

E

Bile acids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119–128, 301 Body fluids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214, 247–256 Breath . . . . . . . . . . . . . . . . . . . . . . . . . 6–8, 14, 16, 21, 149, 210

Electron ionization . . . . . . . . . . . . . . . . . . . 151, 153, 197, 281 Electrospray ionization . . . . . . . . . . 119–128, 134, 144, 168, 229–245, 259–274, 281 Enzyme . . . . . . . . . . . . . . . . . . . . . 55, 119–120, 147, 195, 377 Extraction liquid-liquid . . . . . . . . . 78, 302, 305–306, 309, 317, 332 liquid-solid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177–178 solid-phase . . . . . . . . . . . . . . . . . . . . . . . . 78, 302, 307–308

C Capillary electrophoresis . . . . . . . . . . . . . . . . . . . 100, 229–245 Cell . . . . . . . 8, 23, 39, 56–57, 105, 120, 134–135, 148, 161, 164–166, 168–169, 178, 188, 213–214, 231, 235–236, 239, 248–249, 251–252, 255, 264–265, 306, 308–309, 314–316, 325–327, 330–333, 336, 376 Chromatography gas . . . . . . . . 4, 6, 78, 81, 84, 86–92, 100, 120, 131–145, 147–156, 160, 191–203, 205–210, 259, 277 two-dimensional . . . . . . . . . . . . . . . . . . . . . . . . 205–210 liquid anion exchange. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161 high performance . . . . . . . . . . . . . . 100, 173–189, 308 hydrophilic interaction . . . . . . . . . . . . . . 214–216, 231 reversed-phase . . . . . . . . . . . . . . . . . . . . . . . . 26, 39, 49 ultra performance . . . . . . . . . . . . . 119–128, 178–179, 249–250, 283 Citric acid cycle . . . . . . . . . . . . . . . see Tricarboxylic acid cycle Clinical-biochemical diagnosis . . . . . . . . . . . . . . . . . . . 99–115 Culture . . . . . . . . . 12, 86, 149–150, 164, 224, 227, 330–333

F Fatty acid oxidation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56, 68 Fragmentation . . . . . . . 6, 8, 49, 92, 153, 175–176, 181, 186, 191–192, 255, 259, 264, 273, 278, 280

G Genetic algorithm . . . . . . . . . . . . . . . . . . . . . . . . 378, 380, 382 Glycolysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 131, 147 Glycolytic intermediates. . . . . . . . . . . . . . . . . . . . . . . . 131–145 Group specific internal standard technology (GSIST) . . . . . . . . . . . . . . . . . . . . . . . 161, 163–164

H Head injury . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 Hierarchical clustering . . . . . . . . . . . . . . . . . . . . 286, 374, 385

D Data analysis . . . . . . . . . . . . . . 14, 16, 166, 168, 197–201, 225, 234, 240, 250–251, 260, 277–279, 286, 366, 374, 386 modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286, 374, 378 processing . . . . . . . . . . . 49, 192, 225, 233, 250, 254, 256, 272, 278–279, 282, 366 Database . . . . . . . . . . . . . . . 19, 148, 192, 199, 201, 206, 234, 245, 251, 278, 294–295 Deconvolution . . . . . . . . . . . . . . . . . . . 280, 282–283, 368–369 Derivatization . . . . . . . . . . . . 26, 57, 66, 70, 78, 86, 114, 135, 141–142, 145, 148, 151–152, 155, 161–163,

I Infusion . . . . . . . . . . . . . . . . . . . . . . . . . 67, 174, 248, 260–261, 267, 269–273 Inherited metabolic disease . . . . . . . . . . . . . . . . . . . . 25, 57, 68 Ion suppressor . . . . . . . . . . . . . . . . . . . . . . . 133, 137–138, 144 Ischemia . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100, 328 Isotope coding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161 pattern . . . . . . . . . . . . . . . . . . . . . . 181, 183, 245, 282–283 stable . . . . . . . . . 2, 31, 37, 39, 47–48, 147–148, 160, 322 Isotopologue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134–135, 322 Isotopomer . . . . . . . . . . . . . . . . . . . . . . . . . . 134, 140, 148, 322

T.O. Metz (ed.), Metabolic Profiling, Methods in Molecular Biology 708, c Springer Science+Business Media, LLC 2011 DOI 10.1007/978-1-61737-985-7,

389

METABOLIC PROFILING

390 Subject Index K

O

Krebs cycle . . . . . . . . . . . . . . . . . . . see Tricarboxylic acid cycle

Organic acids . . . . . . . . . . . . . . . . . 73–97, 192, 226, 240, 301 Organic acidurias . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28, 74 Orthomolecular medicine . . . . . . . . . . . . . . . . . . . . . . . . . . . 3, 8 Oximation . . . . . . . . . . . . . . . 78, 85, 141, 195–197, 202, 355

L Lipidomics . . . . . . . . . . . . . . . . . . . . . . . . . . 247–256, 259–274 shotgun. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248, 259–274 Lipids . . . . . . . . . 73, 143, 150, 191, 226, 247–249, 251–255, 259, 265–266, 268, 271, 273, 301

M Magic angle spinning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336 Mass spectrometer ion-trap . . . . . . . . . . . . . . . . . . . . . . . . . . 194, 231, 245, 261 orbitrap . . . . . . . . . . . . . . . . . . . . . 175, 250–251, 254–255 quadrupole single . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84, 87, 133 triple . . . . . . . 27, 49, 67, 84, 92, 133, 194, 216, 227, 264–265, 268–269, 273 time-of-flight . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194, 250 Mass spectrometry isotope dilution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134 tandem . . . . . . . 26, 40–41, 55–71, 78, 95, 254, 259–274 Mass spectrum . . 56, 71, 174–176, 183, 191, 254, 264, 272 Matrix . . . . . . . . . . . . 59, 66, 77, 83, 134, 145, 160, 174, 177, 198–199, 203, 214, 245, 283, 285, 287, 292, 299, 301–302, 315, 317, 322, 339, 342, 377, 385 Media . . . . . . . . 134, 148, 164, 180, 202, 205, 213, 330–333 Metabolic fingerprinting . . . . . . . . . . . . . . . . . . . . . . . . . 196, 213–214 flux . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322 Metabolism . . . . . 1–2, 11, 25–52, 73–97, 99–115, 120, 131, 147–148, 175–177, 181, 188–189, 206, 234, 237, 241, 312–313, 321, 330 inborn errors . . . . . . . . . . . 2, 25–52, 73–97, 99–115, 237 Metabolite anionic . . . . . . . . . . . . . . . . . . . . . . 233, 236, 238–240, 244 cationic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236–240 identification . . . . . . . . . . . . . . . . . . . . . 177, 183, 199–201, 206, 209, 230, 241, 245, 279, 282, 294–295, 300–301, 304, 368 parent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217–223 Metabolomics . . . . . 2–3, 10, 14, 20, 57, 148, 191–203, 205, 213–228, 229–245, 277–296, 321–333, 336–337 Methoxyamine-HCl . . . . . . . . . . . . . . . . . . . . . . 132, 141–142 Multiple reaction monitoring . . . . . . . . 71, 92, 132, 170, 214 Multiple sclerosis . . . . . . . . . . . . . . . . . . . . . . . . . . 5, 7, 13, 100 Multivariate analysis194, 198, 206, 240, 278–279, 285–294

N Neurodegeneration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 N-methyl-N-trimethylsilyltrifluoroacetamide (MSTFA), 81, 86, 133, 141–142, 207–208, 210 Normalization . . . . . . . . . . . . .12, 15, 19, 199, 209–210, 230, 234, 240, 250–251, 254, 256, 260, 279–280, 284–285, 369–370 Nuclear magnetic resonance spectroscopy carbon-13 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302, 312, 322 fluorine-19 . . . . . . . . . . . . . . . . . . . . . . . 303–309, 313–317 nitrogen-15 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322 proton . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205, 299 Nuclei . . . . . . . . . . . . . . . . . . . . . . . . 21, 99, 321–322, 345, 385

P Partial least squares discriminant analysis . . . . . . . . 278, 286 Pattern recognition . . . . . . . . . . . . . . . . . . 2, 12, 209, 214, 240 Peak . . . . . . . . . . . . . . . . . . . . 4, 15, 42, 46, 50–52, 56, 61, 71, 90–92, 96, 107, 110, 113–114, 125, 135, 137, 140–141, 144–145, 151, 153–154, 166, 168, 175, 181, 183–184, 186, 188, 196, 199, 201, 203, 209, 225, 227, 230, 234, 237, 241, 244–245, 250–251, 254, 256, 280–284, 295, 302, 311–316, 332, 359–360, 366–369, 371–373, 385–386 Pentose phosphate pathway . . . . . . . . . . . . . . . . . . . . . 159–170 Plasma . . 26, 31, 38–40, 51, 55–71, 77, 100–101, 106, 122, 125, 148, 152, 157, 177–180, 192, 194–197, 199–201, 203, 231, 235–236, 249, 251–252, 266, 268, 270, 272, 300–301, 306, 310, 317, 323 Precipitation . . 26, 39, 47, 61, 115, 177–178, 180, 192, 196, 242, 302, 306–307, 310–311, 318, 332 Preventive medicine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11–12 Principal components analysis . . . . . 199, 278, 286, 370, 372 Protein . . 1–3, 19–21, 27–28, 31, 38–40, 48–49, 61, 65, 75, 77, 84, 106, 110–113, 143, 177–178, 180, 188, 192, 195–197, 202, 233, 235–236, 242, 248, 250–252, 255, 270, 273, 301–302, 306, 310, 318, 323, 326–328, 330, 332 Purines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99–115 Pyrimidines. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .77, 99–115

Q Quantification . . . . . . . . . . . . . . . . . 31–39, 41, 46–49, 51–52, 113–114, 135, 148, 151, 159–170, 192, 230, 234, 237, 240–241, 244, 259–261, 272, 279–280, 305, 368

R Radiolabel . . . . 174–175, 177–180, 182, 184–186, 303, 308 Resolution . . . . . . . . 4, 7–8, 15–16, 40, 56, 61, 67, 143, 174, 183–185, 189, 229–231, 237–238, 240, 242–244, 248, 254–256, 265, 280, 294–295, 302, 317, 335–363, 367–368 Retention time . . . . 32, 39, 41–42, 46, 49–51, 87, 113, 141, 153–154, 163, 166, 168, 188, 191, 199, 201, 209, 224–226, 228, 251, 254–256, 278, 280–284, 290–291, 306, 315

S Sample processing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215 Scan neutral loss . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260, 266–268 precursor ion . . . . . . . . . . . 56, 61, 67, 260, 263, 265–266 product ion . . . . . . . . . . . . . . . . . . 260, 263, 265, 268, 272 Selected ion monitoring . . . . . . . . . . . . . . . . . . . . 87, 143, 170 Self-organizing map . . . . . . . . . . . . . . . . . . . . . . . . . . . 376–377 Silylation . . . . . . . . . . . . . . . . . . . . 78, 141–142, 193, 195–197 Szent-gy¨orgyi-krebs cycle . . . . . . see Tricarboxylic acid cycle

METABOLIC PROFILING Subject Index 391 T Tissue . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3, 14, 20–21, 28, 120, 148, 150–152, 155, 163, 177, 179, 207, 247–256, 260, 271, 286, 322, 324, 328–330, 335–337, 339, 351, 361–362, 365, 369 Tricarboxylic acid cycle . . . . . . . . . . . . . . . . . . . . . . . . . 147–156 Trimethylsilyl . . . . . . . . . . . . 81, 86, 141, 148, 193, 200, 202, 324, 332

U Urine . . . . . . . . . . . . . . . . . . . 2, 4, 6–8, 13–16, 20–21, 25–26, 28–29, 31, 33, 38–40, 57, 68–69, 73–74, 77–78,

81–87, 91–92, 94–96, 100–103, 106, 110–113, 115, 177, 179, 182–183, 187, 192, 194, 196–197, 203, 231, 236, 239–240, 242, 244, 300–302, 305–307, 310, 315, 317, 326–329, 331, 366, 369–370, 382, 385–386

V Volatile . . . 7, 16, 21, 39, 48, 84, 86, 96, 120, 132, 140, 151, 156, 187, 192, 236

X Xenobiotic . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 174, 176–177