STATE-OF-THE-ART PAPER
Application of Metabolomics
to Cardiovascular Biomarker and Pathway Discovery
Gregory D. Lewis, MD,* Aarti Asnani, BS,* Robert E. Gerszten, MD*
Charlestown, Boston, and Cambridge, Massachusetts
Emerging technologies based on mass spectrometry and nuclear magnetic resonance enable the monitoring of
hundreds of metabolites from tissues or body fluids, that is, "metabolomics." Because metabolites change rapidly
in response to physiologic perturbations, they represent proximal reporters of disease phenotypes. The profiling
of low molecular weight biochemicals, including lipids, sugars, nucleotides, organic acids, and amino acids,
that serve as substrates and products in metabolic pathways is particularly relevant to cardiovascular diseases.
In addition to serving as disease biomarkers, circulating metabolites may participate in previously unanticipated
roles as regulatory signals with hormone-like functions. Cellular metabolic pathways are highly conserved among
species, facilitating complementary functional studies in model organisms to provide insight into metabolic
changes identified in humans. Although metabolic profiling technologies and methods of pattern recognition and
data reduction remain under development, the coupling of metabolomics with other functional genomic approaches
promises to extend our ability to elucidate biological pathways and discover biomarkers of human
disease. (J Am Coll Cardiol 2008;52:117­23)  2008 by the American College of Cardiology Foundation
Metabolism refers to the body's conversion of food stores into
energy currencies that can be used to perform work. Although
decades of research in biochemistry, nutrition, and physiology
have revealed specific metabolic pathways, systematic surveys
of pathways altered in human disease states, such as diabetes,
obesity, and cardiovascular disease, have yet to be performed.
An emerging set of tools, based on mass spectrometry, nuclear
magnetic resonance, and other technologies, enables the monitoring
of dozens to hundreds of metabolites from biological
samples. Although these technologies are still under development,
they complement other functional genomic approaches,
such as high-throughput genome sequencing, ribonucleic acid
expression analysis, and proteomics, and promise to transform
our ability to profile samples with the goal of illuminating
biology and discovering valuable clinical biomarkers.
The Birth of Metabolomics
Small biochemicals are the end result of all regulatory
complexity present in the cell, tissue, or organism, including
transcriptional regulation, translational regulation, and
post-translational modification (Fig. 1). Metabolic changes
are thus the most proximal reporters of alterations in the
body in response to a disease process or drug therapy. In
1971, Arthur Robinson and Linus Pauling conceived the
core idea that information-rich data reflecting the functional
status of a complex biological system resides in the
quantitative and qualitative pattern of metabolites in body
fluids (1). In the same year, Horning and Horning (2) first
used the term metabolic profiling to describe the gas
chromatography (GC) output from a patient sample. This
emerging approach to the quantitative metabolic profiling of
large numbers of small molecules in biofluids was ultimately
termed "metabonomics" by Nicholson et al. (3), and
"metabolomics" by others. Recently, more focused analyses
of specific metabolite families or subsets have even given rise
to new terms such as "lipidomics." Although the majority of
biomarkers have emerged as extensions of "targeted" physiological
studies, it has become evident that a metabolite
profile derived in an unbiased manner may be informative
even if the constituents or their relationships to the disease
are initially unknown (Fig. 2).
To date, the majority of metabolomics studies have been
performed in model organisms. Studies have elucidated the
genetic control of metabolites in plants, such as arabidopsis,
and have determined "metabolic footprints" of genetically
altered yeast (S. cerevisiae) (4,5). In the latter report, metabolic
profiling of conditioned media was used to "diagnose" otherwise
silent mutant phenotypes. Tandem mass spectrometry
(MS) has also been used to profile 36 acylcarnitine species
in mice overexpressing hepatic malonyl-coenzyme A
From the *Center for Immunology and Inflammatory Diseases and Cardiology
Division, Massachusetts General Hospital, Charlestown; Center for Immunology
and Inflammatory Diseases and Cardiology Division, Harvard Medical School,
Boston; and the Broad Institute of MIT and Harvard, Cambridge, Massachusetts.
Supported by the National Institutes of Health (R01 HL072872 and
U01HL083141), the Donald W. Reynolds Foundation and the Fondation Leducq (to
Dr. Gerszten), the Heart Failure Society of America (to Dr. Lewis), the Harvard/
MIT Clinical Investigator Training Program and the American Heart Association
Fellow-to-Faculty Award (to Dr. Lewis), and a pre-doctoral award from the Sarnoff
Cardiovascular Research Foundation (to Mr. Asnani). Cardiovascular Genomic
Medicine Series is edited by Geoffrey S. Ginsburg, MD, PhD.
Manuscript received May 30, 2007; revised manuscript received February 22, 2008,
accepted March 24, 2008.
Journal of the American College of Cardiology Vol. 52, No. 2, 2008
 2008 by the American College of Cardiology Foundation ISSN 0735-1097/08/$34.00
Published by Elsevier Inc. doi:10.1016/j.jacc.2008.03.043
decarboxylase, yielding novel information
regarding muscle betahydroxybutyrate
levels and insulin
sensitivity (6).
The vision for metabolic profiling
to diagnose human disease,
however, extends from seminal
studies of inborn errors of metabolism
in infants. Millington and
colleagues pioneered the use of
tandem MS-based methods for
monitoring fatty acid oxidation, as well as organic and selected
amino acids (7). Their work has culminated in universal
neonatal screening for metabolism disorders in the state of
North Carolina (8). It is anticipated that a global metabolomic
analysis of more common diseases might identify new biomarkers
or spotlight pathways for dietary or drug modulation. The
application of metabolomics to complex cardiovascular diseases,
however, is likely to be more difficult than its application
to inherited inborn errors of metabolism.
Technologies to Define the Human Metabolome
The global collection of metabolites in a cell or organism is
often called the metabolome; this refers to all small molecules
that exclude nucleic acids and proteins (Fig. 1, Table 1).
Present estimates suggest that the human metabolome consists
of approximately 3,000 endogenous metabolites (Human
Metabolome Project [9,10], Kyoto Encyclopedia of Genes and
Genomes [11]). As with the human genome, the exact size of
the human metabolome remains under debate. Estimates of
the metabolome will likely be revised as technologies to detect
metabolites become more sensitive and comprehensive. Moreover,
some argue that nutritional compounds, xenobiotics
modified by human enzymes, as well as microbial metabolites
present in the gut must also be taken into consideration when
defining the human metabolome.
The metabolome spans a variety of chemical compound
classes, including those that are anionic versus cationic and
lipophilic versus hydrophilic (Table 1). Metabolites in tissue
or body fluids are present across a broad range of concentrations.
Therefore, no single analytical method is capable of
analyzing all metabolites. However, capturing a subset of
"sentinel" metabolites in critical pathways may prove to be a
more tractable problem than proteomics. Estimates suggest
that post-translational modifications may bring the total
number of protein species to 106
, and perhaps 108
to 109
if immunoglobulins are included. Thus, the metabolome
may be less complex than the human proteome. Cellular
metabolic pathways are highly conserved across species.
Therefore, once metabolic changes are identified in humans,
complementary functional studies in model organisms may
rapidly provide insight into homeostatic and disease pathways.
Metabolites can be measured by several available analytical
methods (for reviews of metabolomics technologies, see
articles by Dunn et al. [12,13] and Lindon et al. [14]).
Chromatographic procedures such as GC, high-performance
liquid chromatography (LC), and capillary electrophoresis have
been used to identify and quantify specific metabolite subsets
(e.g., amino acids [15,16] or purine metabolites [16,17]) but
are best used for initial compound separation in combination
with other detection techniques. Recently, 2 high-throughput
technologies have garnered the most use for profiling a large
number of metabolites simultaneously: nuclear magnetic resonance
(NMR) spectroscopy and MS. Mass spectrometry distinguishes
metabolites on the basis of mass/charge ratio (m/z)
PHENOTYPE
Genome Proteome
DNA RNA Protein
Modified
Proteins Metabolites
N~2 x 104
N~10 6
N ~10 7
Transcriptome
Enzymes
E N V I R O N M E N T
N~3x103
Metabolome
Figure 1 Integration of Metabolomics With Other "Omics" Approaches and Relationship to Phenotype
Abbreviations
and Acronyms
GC gas chromatography
LC liquid
chromatography
MS mass spectrometry
NMR nuclear magnetic
resonance
TCA tricarboxylic acid
118 Lewis et al. JACC Vol. 52, No. 2, 2008
Metabolomics in Biomarker Discovery July 8, 2008:117­23
and requires a separation of the metabolite components using
either GC after chemical derivitization or LC, with a new
method of ultraperformance LC being increasingly used. Mass
spectrometry also permits absolute quantification of metabolite
levels via the standard addition method, which entails using
spiked-in internal standards across a range of concentrations.
When available, isotope-labeled standards can be easily differentiated
from the endogenous metabolite by the appropriate
number of mass units.
Nuclear magnetic resonance spectroscopy uses magnetic
properties of nuclei to determine the number and type of
chemical entities in a molecule. Proton (1
H) NMR spectroscopy
can detect soluble proton-containing molecules
with a molecular weight of 20 kD or less. The NMR
spectra serve as the raw material for pattern recognition
analyses, which simplifies the complex multivariate data into
2 or 3 dimensions that can be readily understood and
evaluated. Both NMR and LC-MS systems can be applied
to in vivo tissues or to biological fluids, such as serum,
plasma, urine, and so on, obtained from humans (18). The
advantages of NMR are that it requires relatively little
sample preparation, is nondestructive, and can provide
information about the precise structure of metabolites (4).
However, NMR sensitivity is related to magnet strength,
and presently available instrumentation can unambiguously
detect only the most abundant metabolites in plasma.
However, more sensitive systems are rapidly evolving.
In contrast, the most important advantage of MS coupled
with up-front chromatography is far greater sensitivity than
NMR. Mass spectrometry-based systems have been used to
resolve compounds in the nanomole/liter to picomole/liter
and even femtomole/liter range, whereas identification of
compounds by 1
H-NMR requires concentrations of 1
nmol/l or higher (19). In human plasma, limits of detection
between 0.1 and 1 mol/l for a series of compounds
analyzed by GC-MS have been described. Normal plasma
concentrations for these metabolites are in the micromole/
liter range, well above the limits of detection established for
most MS technologies.
Targeted Versus Pattern Recognition Analyses
Perturbations of the metabolome that arise either as a cause
or consequence of disease manifest as particular patterns of
0
100
200
300
Year
NumberofPublications
Figure 2
Rapid Incorporation of
Metabolomics in Biomedical Research
Criteria for inclusion of publications included use of "metabolome,"
"metabolomics," or "metabonomics" as a keyword or in the title.
Endogenous Compounds ProfiledWith Current Metabolomics Technologies
Table 1
Endogenous Compounds Profiled
With Current Metabolomics Technologies
Metabolite Class
Acetylcarnitines
Acylglycines
Alcohols
Aldehydes
Alkanes/alkenes
Amides/amines
Amino acids
Aromatics
Bile acids
Branched-chain hydrocarbons
Carbohydrates
Carnitines
Catecholamines
Cholesterols
Coenzyme A derivatives
Esters
Ethers
Fatty acids/alcohols/aldehydes
Glycerols/glycerophospholipids
Glycols/glycolipids
Heterocyclics
Histidine metabolites
Hydroxyacids
Imidazoles
Isoprenoids
Ketones
Lactams
Lactones
Lipids
Minerals
Nitrogenous compounds
Nucleosides/nucleotides
Organic acids
Peptides
Phenols
Phosphates
Porphyrins
Prostaglandins
Pterins
Purines
Pyrimidines
Quinones
Sphingolipids
Steroids/sterols
Sugars
Sulfates
Thiols
Vitamins
Data from Wishart et al. (9).
119JACC Vol. 52, No. 2, 2008 Lewis et al.
July 8, 2008:117­23 Metabolomics in Biomarker Discovery
metabolites in a tissue or body fluid. This patterning
concept has been the basis for recent efforts to discover
proteomic or metabolomic "signatures" in tissue or serum.
Mass spectrometry and NMR techniques can rapidly generate
well-defined sets of peaks from a sample across a broad
range of mass/charge. A growing controversy is whether
such "metabolite signatures" can be used to accurately
distinguish disease states from normal. A significant time
advantage of direct profile comparisons derives from skipping
the far more laborious task of unambiguously identifying
the entities that underlie the peaks. Thus, rapid
screening of patient samples is possible.
Using a pattern of peaks to diagnose disease without
knowing the represented metabolites, however, raises some
concerns. One issue is that of reproducibility. Because most
MS or NMR instruments were not designed as clinical
tools, it is hard to generate consistent results from machine
to machine or from operator to operator. Some contend that
the patterns are mostly "noise" and do not discriminate
biologically meaningful information. Without unequivocal
identifications, one cannot independently confirm findings
with complementary technologies. Others contend that the
peaks profiled by the methodologies used to date only
represent the most abundant plasma or tissue constituents.
The most important consequence of not unequivocally
identifying spectral peaks, however, is that little insight is
gained into the biology, either to understand disease pathways
through basic cellular mechanisms or as a check on the
biological consistency and reasonableness of the data. Overfitting
of data is also a common problem when algorithms
are generated from hundreds or thousands of peaks. Blinded
prospective studies must ultimately be organized to better
address the controversy.
Human metabolomics studies are also complicated by
potentially confounding clinical variables such as diet or
drug effects, particularly if NMR- or MS-based profiling
techniques are used in which metabolite peaks are not
unambiguously identified. Because of the various limitations
inherent to pattern discovery, many have championed
metabolomics applications in targeted approaches. The user
targets a predefined set of metabolites to be quantified by
monitoring specific chromatographic retention times, as
well as parent and daughter mass-to-charge ratios of analytes.
With the targeted approach being more focused, relying
on a predefined set of entities, researchers have more confidence
in the end results, because they know what is giving rise
to the signals. Although this approach has many advantages, it
is blind to changes in metabolites whose retention times and
MS characteristics have not been incorporated into the analysis
method. As efforts to define the human metabolome grow
(20), we anticipate increasingly comprehensive targeted platforms
for biomarkers and pathway discovery. Improvements in
MS and databases to enable identification of unknown peaks
will also be critical.
Statistical Approaches to Metabolomic
Data Reduction and Pathway Analysis
Although a high-throughput metabolomics approach to
biomarker discovery brings many advantages, it also brings
a danger of generating false-positive associations due to
multiple testing and overfitting of data, as noted in the
preceding text. Application of traditional statistical approaches
(e.g., Bonferroni correction) in this setting tends to
levy an insurmountable statistical penalty that can obscure
biologically relevant associations. Even newer statistical
techniques, such as advanced resampling methods or control
of the false discovery rate (21), do not adequately address
the fundamental problem of how to detect subtle but
important changes in multiple variables identified in an
"omics" approach.
For metabolites participating in known biological pathways,
a bioinformatics approach using pathways analysis can
harness the vast information gathered in genomics or
metabolomics experiments and turn it into a strength.
Specifically, although measurement error in the marker
discovery phase often prevents high confidence in any one
particular metabolite's correlation, the observation that multiple
metabolites in a particular biological pathway are
moving in tandem brings confidence that a particular
pathway, and therefore any biomarkers in that pathway,
truly is correlated with the perturbation. By using a more
principled selection process for candidate marker triage, this
approach increases the likelihood that candidate biomarkers
will be validated in subsequent prospective validation studies.
This approach also enhances one's ability to use the
metabolomic data collected in the biomarker discovery
phase to gain insight into disease biology.
Systematic analysis of functional trends has become
widespread and important in the analysis of deoxyribonucleic
acid microarray data from model organisms (22). The
value of this approach in human studies was illustrated in a
recent analysis of high-throughput differential messenger
ribonucleic acid (mRNA) expression (23). Expression of
mRNA was assessed on over 22,000 genes comparing
patients with type 2 diabetes mellitus and unaffected control
subjects (patients with normal glucose tolerance). A group
of genes with depressed expression in diabetes versus control
subjects was identified and tested for association with a
collection of other gene characteristics. It was found that
this gene set was enriched for genes involved in oxidative
phosphorylation. Although individual oxidative phosphorylation
genes were not dramatically reduced in expression, as
a group the trend was highly significant. Furthermore, the
effect was attributable to a subset of oxidative phosphorylation
genes regulated by peroxisome proliferator-activated
receptor coactivator 1, a regulator of mitochondrial biogenesis.
Thus, the analysis of trends among differentially
expressed genes led directly to insight into altered metabolism
in diabetes patients and hinted at therapeutic hypoth-
120 Lewis et al. JACC Vol. 52, No. 2, 2008
Metabolomics in Biomarker Discovery July 8, 2008:117­23
eses involving the modulation of oxidative phosphorylation
pathways.
There are several statistical issues complicating functional
trends analysis of high-throughput data that have been
rigorously addressed in software under development, including
"FuncAssociate," recently described by Berriz et al. (24).
Although the analysis software was developed for use with
high-throughput mRNA expression data, the general approach
may be used in conjunction with essentially any
high-throughput experimental approach for identifying or
ranking "interesting genes," FuncAssociate has generally
been used in conjunction with controlled-vocabulary functional
annotation (e.g., Gene Ontology annotation), but can
be used in conjunction with many different sources of
gene/protein/metabolite annotation (e.g., expression pattern
in other studies, phenotype, protein complex membership,
disease association, or phylogenetic profile).
Several data reduction strategies using supervised learning
multivariate analysis can be used to construct multivariate
metabolite biomarker profiles (25). In supervised learning,
an algorithm is used to transform the multivariate data from
metabolite profiles into a lower dimensionality with biological
interest (e.g., health vs. disease). Metabolite data (inputs)
and disease status (outputs or targets) form pairs that
are used in the calibration of the model, with the goal of the
model being to correctly associate the inputs with the
targets. Discriminant analysis is a cluster analysis-based
algorithm for categorical variables (26), partial least squares
is a popular linear regression-based method, and artificial
neural networks offer the advantage of a machine-based
method that can learn nonlinear mappings (27).
Principal components analysis applies raw data to a
normalized matrix, which is then projected onto a specific
scoring schema (4,28). This schema is based upon a series of
orthogonal principal components, the first of which describes
the maximum variance in the original dataset. The
Principal components analysis can be used for dimensionality
reduction in a data set by retaining those characteristics
of the dataset that contribute most to its variance, by
keeping lower-order principal components and ignoring
higher-order ones. Similarly, Fisher discriminant analysis
creates normalized data matrices that decrease sample variability
within a specific condition (e.g., cells grown aerobically)
while maximizing sample variability between different
conditions (e.g., aerobic vs. anaerobic growth) (29). These
relatively unbiased methods of reducing large datasets have
the potential to define previously unknown relationships
between metabolites in a given physiologic state.
A limitation of current analytical approaches in metabolomics
is that they rely on the relative or absolute concentrations
of a metabolite in a given tissue or plasma sample
and do not take into account varying enzymatic activities
controlling metabolic flux through biological pathways.
Studies examining metabolic flux incorporate isotopelabeled
metabolites to map the fate of enzymatic substrates.
Studies of tricarboxylic acid (TCA)-cycle anaplerosis in
energy metabolism (30) and fractional synthetic rate of fatty
acids and cholesterol (31) have generally focused on small,
well characterized pathways in specific tissues such as
muscle or liver. The integration of current analytical methods
with assessment of enzymatic activities contributing to
metabolic flux will help to elucidate biologically relevant
metabolic changes.
Novel Roles for
Metabolites in Human Physiology
Secreted small ligands such as catecholamines play central
roles in cardiovascular physiology. A growing body of
literature suggests previously unanticipated roles for metabolites
that have been traditionally thought to function
exclusively as intracellular signals. He et al. (32) recently
discovered that the "orphan" G-protein­coupled receptors
GPR91 and GPR99, which are highly expressed in the
kidney, bind the TCA-cycle intermediates succinate and
-ketoglutarate, respectively. The working hypothesis is
that a local mismatch of energy supply and demand, altered
metabolism of TCA-cycle intermediates, or injury leads to
mitochondrial dysfunction and the release of succinate and
-ketoglutarate from tissues. Once released into the circulation,
the metabolites function in a hormone-like manner,
binding their receptors in the renal cortex and triggering the
release of renin and activation of the renin-angiotensin
system. In the case of tissue ischemia from volume loss, this
process might be adaptive to match metabolic demands. In
other conditions associated with high succinate production,
such as congestive heart failure, resultant increases in blood
pressure might prove maladaptive. This recent work highlights
roles for new types of circulating metabolites functioning
as hormones in the body.
Application of Metabolomics to Unique
Human Cardiovascular Disease Models
Novel metabolomics techniques still suffer from signal-tonoise
issues, however, and applications to humans may be
limited by interindividual variability. Although recent studies
have evaluated the diurnal and even seasonal variation of
hemostatic and inflammatory proteins (e.g., fibrinogen,
D-dimer, and C-reactive protein), systematic studies have
yet to be performed for metabolites in humans. Studies to
identify novel disease-related pathways are also restricted by
the inherent unpredictability of the onset of pathological
states. As noted previously, human metabolomics studies are
also at high risk for potential clinical confounders, such as
diet or drug effects, as well as age, gender, and comorbidities.
It has been advocated that the analysis of samples from
large patient cohorts, stratified by known risk factors or
exposures, may minimize the impact of clinical confounding
variables (33). However, the throughput of most current
metabolomics technologies, particularly those that are MS
based, precludes the analysis of large patient cohorts.
121JACC Vol. 52, No. 2, 2008 Lewis et al.
July 8, 2008:117­23 Metabolomics in Biomarker Discovery
To help circumvent these problems, investigators have
begun to apply these emerging technologies to unique
clinical scenarios where serial sampling can be performed in
patients both before and after a controlled perturbation,
thereby allowing each patient to serve as his or her own
biological control. Clinical cardiology is uniquely suited for
such investigation. As proof of principle, a targeted MSbased
metabolomics platform was applied to patients undergoing
exercise stress testing with myocardial perfusion
imaging (34). Eighteen patients had no evidence of ischemia
(control group), whereas 18 patients had inducible
ischemia (case group). Plasma was fractionated by highperformance
LC and metabolites analyzed using a triple
quadrupole mass spectrometer to monitor hundreds of ion
pairs by targeted MS. The majority of metabolites displayed
concordant changes in cases and controls (i.e., increased in
both or decreased in both). For example, lactic acid, as well
as hypoxanthine and inosine, end products of adenosine
monophosphate catabolism, increased in both case and
control groups. However, 6 metabolites yielded highly
significant discordant changes in case and control groups.
Using a metabolic risk score derived from metabolites with
discordant changes in the 2 groups, case subjects could be
distinguished from control subjects with a high degree of
accuracy (p 0.0001; c-statistic 0.95).
Strategies emphasizing the in-depth analysis of small,
extremely well phenotyped patient cohorts are ideal in light
of current technological limitations. However, such an
approach has potential limitations that should be considered.
First, although serial sampling in patients serving as
their own biological controls helps diminish interindividual
variability and signal-to-noise issues, populations studied to
date are nevertheless small. Further testing in larger cohorts
may be powered to detect more subtle metabolic changes
and will provide sufficient precision in the estimates of the
utility of each marker to allow for appropriate relative
weighting of each component.
Second, although metabolite profiling of serum or plasma
offers the advantage of simple sample collection, and may
reflect the sum of metabolic changes occurring throughout
the body, sampling specific tissues that serve as proximal
sources of metabolites (35,36) enables localization of metabolic
changes and may help to gauge the sensitivity and
specificity of signature metabolic profiles in plasma. Alternatively,
for metabolites that are rapidly cleared from the
circulation, it may be more appropriate to perform metabolic
profiling on urine samples.
Integration of Metabolomics With
Other "Omics" Technologies in a Systems
Biology Approach to Cardiovascular Disease
The identification of new pathways and biomarkers in
cardiovascular disease will depend on the complementary
power of genetics, transcriptional profiling, proteomics, and
metabolomics. For example, Mayr et al. (36) recently used
metabolomics and proteomics to characterize metabolic
profiles of atrial tissue that predispose patients to the
development of atrial fibrillation. Genome-wide association
studies that provide an unbiased scan of genomic sequence
variants will also catalyze integrative "omics" approaches.
For example, 3 groups recently identified several loci,
including chromosome 9p21, associated with early-onset
myocardial infarction (37­39). The chromosomal regions
identified to date do not contain genes recognizably associated
with established coronary heart disease risk factors such
as plasma lipoproteins. However, the integration of metabolic
and proteomic data from these same patients may
provide clues as to how the variants modulate the atherosclerotic
process.
Conclusions
An emerging set of analytical and bioinformatics tools have
made it possible to profile hundreds of metabolites in
complex mixtures such as plasma. Although these technologies
are still under development, when coupled with other
functional genomic approaches, metabolomics promises to
transform our ability to profile samples with the goal of
elucidating biological pathways and discovering valuable
clinical biomarkers.
Reprint requests and correspondence: Dr. Robert E. Gerszten,
Cardiology Division and Center for Immunology and
Inflammatory Diseases, Massachusetts General Hospital East-
8307, 149 13th Street, Charlestown, Massachusetts 02129. Email:
rgerszten@partners.org.
REFERENCES
1. Pauling L, Robinson AB, Teranishi R, Cary P. Quantitative analysis
of urine vapor and breath by gas-liquid partition chromatography. Proc
Natl Acad Sci U S A 1971;68:2374­6.
2. Horning EC, Horning MG. Metabolic profiles: gas-phase methods
for analysis of metabolites. Clin Chem 1971;17:802­9.
3. Nicholson JK, Lindon JC, Holmes E. "Metabonomics": understanding
the metabolic responses of living systems to pathophysiological stimuli
via multivariate statistical analysis of biological NMR spectroscopic
data. Xenobiotica 1999;29:1181­9.
4. Raamsdonk LM, Teusink B, Broadhurst D, et al. A functional
genomics strategy that uses metabolome data to reveal the phenotype
of silent mutations. Nat Biotechnol 2001;19:45­50.
5. Allen J, Davey HM, Broadhurst D, et al. High-throughput classification
of yeast mutants for functional genomics using metabolic
footprinting. Nat Biotechnol 2003;21:692­6.
6. An J, Muoio DM, Shiota M, et al. Hepatic expression of malonylCoA
decarboxylase reverses muscle, liver and whole-animal insulin
resistance. Nat Med 2004;10:268­74.
7. Roe CR, Millington DS, Maltby DA. Identification of 3methylglutarylcarnitine.
A new diagnostic metabolite of 3-hydroxy-
3-methylglutaryl-coenzyme A lyase deficiency. J Clin Invest 1986;77:
1391­4.
8. Frazier DM, Millington DS, McCandless SE, et al. The tandem mass
spectrometry newborn screening experience in North Carolina: 1997­
2005. J Inherit Metab Dis 2006;29:76­85.
9. Wishart DS, Tzur D, Knox C, et al. HMDB: the Human Metabolome
Database. Nucleic Acids Res 2007;35:D521­6.
10. Human Metabolome Project. The Human Metabolome Database.
Available at: http://www.hmdb.ca. Accessed March 2008.
122 Lewis et al. JACC Vol. 52, No. 2, 2008
Metabolomics in Biomarker Discovery July 8, 2008:117­23
11. KEGG: Kyoto Encyclopedia of Genes and Genomes. Available at:
http://www.genome.jp/kegg/. Accessed March 2008.
12. Dunn WB, Bailey NJ, Johnson HE. Measuring the metabolome:
current analytical technologies. Analyst 2005;130:606­25.
13. Dunn WB, Ellis D. Metabolomics: current analytical platforms and
methodologies. Trends Anal Chem 2005;4:285­94.
14. Lindon JC, Holmes E, Nicholson JK. Metabonomics techniques and
applications to pharmaceutical research and development. Pharm Res
2006;23:1075­88.
15. Backstrom T, Goiny M, Lockowandt U, Liska J, Franco-Cereceda A.
Cardiac outflow of amino acids and purines during myocardial ischemia
and reperfusion. J Appl Physiol 2003;94:1122­8.
16. Mei DA, Gross GJ, Nithipatikom K. Simultaneous determination of
adenosine, inosine, hypoxanthine, xanthine, and uric acid in microdialysis
samples using microbore column high-performance liquid chromatography
with a diode array detector. Anal Biochem 1996;238:
34­9.
17. Zemgulis V, Ronquist G, Bjerner T, et al. Energy-related metabolites
during and after induced myocardial infarction with special emphasis
on the reperfusion injury after extracorporeal circulation. Acta Physiol
Scand 2001;171:129­43.
18. Cheng LL, Chang IW, Louis DN, Gonzalez RG. Correlation of
high-resolution magic angle spinning proton magnetic resonance
spectroscopy with histopathology of intact human brain tumor specimens.
Cancer Res 1998;58:1825­32.
19. Zhang X, Wei D, Yap Y, Li L, Guo S, Chen F. Mass spectrometrybased
"omics" technologies in cancer diagnostics. Mass Spectrom Rev
2007;26:403­31.
20. Human Metabolome Project. What is metabolomics? Available at:
http://www.metabolomics.ca/About/overview.htm. Accessed March
2008.
21. Storey JD, Tibshirani R. Statistical significance for genomewide
studies. Proc Natl Acad Sci U S A 2003;100:9440­5.
22. Tavazoie S, Hughes JD, Campbell MJ, Cho RJ, Church GM.
Systematic determination of genetic network architecture. Nat Genet
1999;22:281­5.
23. Mootha VK, Lindgren CM, Eriksson KF, et al. PGC-1alpharesponsive
genes involved in oxidative phosphorylation are coordinately
downregulated in human diabetes. Nat Genet 2003;34:267­73.
24. Berriz GF, King OD, Bryant B, Sander C, Roth FP. Characterizing
gene sets with FuncAssociate. Bioinformatics 2003;19:2502­4.
25. van der Greef J, Stroobant P, van der Heijden R. The role of analytical
sciences in medical systems biology. Curr Opin Chem Biol 2004;8:
559­65.
26. Manly B. Multivariable Statistical Methods: A Primer. London:
Chapman & Hall, 1994.
27. Bishop C. Neural Networks for Pattern Recognition. Oxford: Clarendon
Press, 1995.
28. Baxter IR, Borevitz JO. Mapping a plant's chemical vocabulary. Nat
Genet 2006;38:737­8.
29. Villas-Boas SG, Moxley JF, Akesson M, Stephanopoulos G, Nielsen
J. High-throughput metabolic state analysis: the missing link in
integrated functional genomics of yeasts. Biochem J 2005;388:669­77.
30. Gibala MJ, Young ME, Taegtmeyer H. Anaplerosis of the citric acid
cycle: role in energy metabolism of heart and skeletal muscle. Acta
Physiol Scand 2000;168:657­65.
31. Bederman IR, Reszko AE, Kasumov T, et al. Zonation of labeling of
lipogenic acetyl-CoA across the liver: implications for studies of
lipogenesis by mass isotopomer analysis. J Biol Chem 2004;279:
43207­16.
32. He W, Miao FJ, Lin DC, et al. Citric acid cycle intermediates as
ligands for orphan G-protein-coupled receptors. Nature 2004;429:
188­93.
33. Kirschenlohr HL, Griffin JL, Clarke SC, et al. Proton NMR analysis
of plasma is a weak predictor of coronary artery disease. Nat Med
2006;12:705­10.
34. Sabatine MS, Liu E, Morrow DA, et al. Metabolomic identification
of novel biomarkers of myocardial ischemia. Circulation
2005;112:3868­75.
35. Howarth KR, LeBlanc PJ, Heigenhauser GJ, Gibala MJ. Effect of
endurance training on muscle TCA cycle metabolism during exercise
in humans. J Appl Physiol 2004;97:579­84.
36. Mayr M, Yusuf S, Weir G, et al. Combined metabolomic and
proteomic analysis of human atrial fibrillation. J Am Coll Cardiol
2008;51:585­94.
37. Samani NJ, Erdmann J, Hall AS, et al. Genomewide association
analysis of coronary artery disease. N Engl J Med 2007;357:443­53.
38. McPherson R, Pertsemlidis A, Kavaslar N, et al. A common allele on
chromosome 9 associated with coronary heart disease. Science 2007;
316:1488­91.
39. Helgadottir A, Thorleifsson G, Manolescu A, et al. A common variant
on chromosome 9p21 affects the risk of myocardial infarction. Science
2007;316:1491­3.
Key Words: metabolomics y biomarkers y prognosis.
123JACC Vol. 52, No. 2, 2008 Lewis et al.
July 8, 2008:117­23 Metabolomics in Biomarker Discovery