Biotechnol. Appl. Biochem.

(2007) 47, 7184 (Printed in Great Britain)

doi:10.1042/BA20060221

71

REVIEW
Metabolomics as a complementary tool in cell culture
Soo Hean Gary Khoo* and Mohamed Al-Rubeai1
*Department of Chemical Engineering, University of Birmingham, Edgbaston, Birmingham B15 2TT, U.K., and School of
Chemical and Bioprocess Engineering, University College Dublin, Belfield, Dublin 4, Republic of Ireland
Metabolomics, the global study of metabolite changes
in a biological system, has drawn a significant amount
of interest over the last few years. It can be said to
be an amalgam of traditional areas such as metabolite
analysis, bioanalytical development and chemometrics.
Thus, piecing these areas together into the cohesive science of metabolome analysis has proved to be difficult.
Most work to date has been focused on plant, microbial, as well as tissue and biofluid samples. However,
the diverse potential of metabolomics in many fields,
including cell engineering, has made it a universal tool
for industrial, medical and research purposes. It is also
a vital component of a systems biology approach, as it
is believed to be a good reflection of the phenotype of
any cell or tissue. At the heart of metabolomics growth
is the issue of method development, including sample
preparation, instrument analysis, data processing and
bioinformatics. Here, we look at the cell-culture applications of metabolomics and the issues that can
transform metabolomics into a mature omics
science.

Introduction
With the systematic genomic sequencing of various organisms, an unprecedented amount of information has
now been revealed. Deciphering such a blueprint via
the understanding of functions and interactions within a
complex biological system has been the focus of the postgenomic era. That has fuelled the growth of other omic
sciences, such as transciptomics and proteomics, which,
together with the rapid development of bioinformatics and
statistical tools, can now be used in many research and medical applications. One of those relatively new omic sciences
is the field of metabolomics. The metabolome was first
described by Oliver et al. [1] as being the set of all
the low-molecular-mass compounds synthesized by an
organism. Metabolomics is therefore the analysis of small
molecules that constitute metabolism, and it offers the
closest direct measurement of a cells physiological activity

[2]. Hence it follows on that metabolome analysis can

be considered as the measurement of the change in the
relative concentrations of metabolites as the result of
the deletion or overexpression of a gene . . . [and thus]
should allow the target of a novel gene product to be
located on the metabolic map. Another definition of the
metabolome states that it consists of only of those native
small molecules (definable non-polymeric compounds) that
are participants in general metabolic reactions and that are
required for the maintenance, growth and normal function
of a cell [3]. This would exclude peptides, and even many
larger lipids, as metabolites. Realistically, metabolites can
be considered a class of naturally occurring compounds,
diverse in their chemical structure, that are less than 1 kDa
in molecular mass. These compounds function as carriers,
substrates or products in biochemical pathways. Despite
these attempts to define the metabolome, there remains
some vagueness which will slowly be resolved as the field
develops. Some view metabolomics as the vital piece of the
Rossetta stone [4] needed for deciphering the puzzle of
complex systems as seen in Figure 1. The field of metabolomics could be said to fuse metabolite analysis,
bioanalytical science and chemometrics. In its present state,
the global analysis of all metabolites seems to be a long
way off and thus, in the pure sense, metabolomics now only
consists of fragments of biochemical and metabolite analysis.
Despite the ambiguity and lack of comprehensiveness, the
field of metabolomics grows towards the critical objective
of extracting useful knowledge from metabolite pools.
Therefore, for the purpose of its utility, metabolomics is

Key words: cell culture, metabolism, metabolite analysis, metabolomics,

systems biology.
Abbreviations used: CE, capillary electrophoresis; ESI, electrospray ionization;
EST, expressed sequence tag; FT-ICR, Fourier-transform ion cyclotron
resonance; FTIR, Fourier-transform infrared; HCA, hierarchical cluster
analysis; ICA, independent component analysis; LC, liquid chromatography;
MST, mass-spectral tag; NIR, near infrared; NLM, non-linear mapping; PAD,
photodiode array detection; PCA, principal component analysis; PLS-DA,
partial least squares-discriminant analysis; SBML, systems biology markup
language; SOM, self-organizing map; TOF, time-of-flight; UPLC,
ultra-performance liquid chromatography.
1
To whom correspondence should be addressed (email
m.al-rubeai@ucd.ie).


C

2007 Portland Press Ltd

72

S. H. G. Khoo and M. Al-Rubeai

Figure 1 Overview of the interactions between different omics within a cell

best described as an area of science, rather than an analytical

approach, that characterizes a metabolic phenotype under
a specific set of conditions which links these phenotypes to
their correspondent genotypes [5].
A number of other diverse applications of metabolomic
analysis exist. These include the commercial applications in
agriculture, industrial biotechnology and xenobiochemistry
[6], medical applications like biomarker discovery and
nutritional health, as well as environmental applications,
such as environmental toxicology, developmental growth of
organisms and pathogenhost interactions. The application
of metabolomics in the area of mammalian cell culture is
relatively undeveloped and thus the aim of the present
review is to provide an insight into the issues pertaining
to metabolome analysis as well as to explore its possible
applications in cell culture.

Understanding the complexity of

metabolome analysis
Metabolomics requires the unbiased identification and
quantification of all of the metabolites present in a specific
biological sample (from an organism or in vitro) [7]. Such a
requirement is difficult to meet with the present analytical
technologies. In addition, the exact number of metabolites


C

2007 Portland Press Ltd

in a system is not known. It is believed that the number

of metabolites for a particular cell type should be lower
than the number of genes and proteins in a cell [8], which
would give a number less than 10 000 [9]. However, if
an analysis platform is to be used universally for various
organisms, this number increases very significantly. It is
speculated that there are an estimated 200 000 different
metabolites in the Plant Kingdom [10], with the numbers
in the mammalian systems being lower. To put this into
perspective, present microarray technologies for transcripts
have an upper limit of about 15 00020 000 ESTs (expressed
sequence tags) per array, whereas 2-DE [two-dimensional
(polyacrylamide) electrophoresis] can readily differentiate a
few thousand proteins, with 10 000 proteins as an upper limit
[11]. Present metabolomic analyses can resolve anything
from a low of 70 metabolites [12] to over 4000 metabolites
[13]. In addition, their diverse chemical properties make
complete metabolite analysis difficult. Genes are composed
of a linear four-letter code, whereas proteins have a 20-letter
code of primary amino acids as a foundation. Metabolites
do not have any fixed codes, and thus a universal method of
characterization is difficult. Present methods use the specific
chemical properties of these entities to separate, identify
and decipher their structures. Combinatorial approaches
allow for a greater coverage. It is therefore important to
define criteria for such analyses. Hence an ideal metabolomic
analysis should provide the following:

Metabolomics and cell culture

r
r
r
r

give an instantaneous snapshot of all metabolites in any

given system
use analytical methods that have high recovery, experimental robustness, reproducibility, high resolving power
and high sensitivity [14] whilst being able to be applied
universally
provide the unambiguous quantification and identification
of metabolites
allow distinguishing factors to be highlighted while
easily being incorporated into biochemical network
models

To overcome present analytical failings, the metabolomic community tends to confine their use of specific analytical approaches (categories) to help answer specific types of
questions. These categories include metabolite or metabolic
profiling, metabolic fingerprinting and metabonomics. Metabolite fingerprinting aims to look at the evidence of major
metabolic effects as a result of perturbations. Rapid classification of samples according to their origin or their biological relevance allows the maintenance of high-throughput
analysis. In such cases it might not be necessary to determine
the levels of all metabolites individually, as patterns of spectra
may be sufficient for classification.
On the other hand, metabolic profiling is used to elucidate the function of a whole pathway or intersecting pathways
and does not require the characterization of the entire
metabolome. Thus such analysis focuses on a chosen class
of compounds (such as amino acids or carbohydrates).
A similar definition to metabolite profiling is termed
metabonomics. The term metabonomics was first coined
by Nicholson and his colleagues [15] in 1996 to describe the
studies of metabolite profiles in biofluids, such as plasma or
urine, from whole organisms. As further elaborated, metabonomics is the quantitative measurement of the dynamic
mutliparametric metabolic responses of living systems to
pathophysiological stimuli or genetic modification and the
key feature in this type of analysis is pattern recognition
[16]. The metabolon thus refers to co-ordinated channelling
of substrates through tightly connected enzyme complexes
[17]. As the field of metabolomics broadens, the lines that
separate these classes become blurred, owing to the need
for greater comprehensiveness in data extraction.

Methodology
Sample preparation and extraction
The first step to ensuring the simultaneous detection of a
large number of metabolites is to have adequate methods
for sample preparation and extraction. Erroneous sample

preparation can lead to misleading or inaccurate data, even

with the most sensitive instruments. Sampling techniques,
extraction, storage and pre-analysis preparation are just
some of the necessary steps taken before instrument
analysis. As metabolic processes may be rapid, varying from
milliseconds to minutes [18,19], the first necessary step
is to rapidly stop any inherent enzymatic activity or any
changes in the metabolite levels. This is sometimes termed
quenching. In addition, sampling methods should not be
biased towards any group of molecules, but this challenge is
presently unresolved. The time and method of sampling are
important issues to be considered to ensure reproducibility
in the analytical sample, especially since a large number of
biological replicates is commonly used.
Traditional quenching methods include freeze-clamping
(with lower-temperature receptacles), immediate freezing
in liquid nitrogen or by acidic (e.g. perchloric acid or nitric
acid) treatments [20]. Freezing in liquid nitrogen is generally
considered to be the easiest way of stopping enzyme activity
provided that cells or tissues are not allowed to partially
thaw before extracting metabolites. In order to prevent this
from happening, enzyme activity is inhibited by freeze-drying
or by immediate addition of organic solvents while applying
heat. Freeze-clamping is a faster process of freezing cells that
avoids the potential artefacts caused by wound response
[10]. Acidic treatments can severely decrease the number
of metabolites detected as a result of degradation due to
the low pH. Acidic treatments also pose severe problems
for many analytical methods that follow, so the acids have to
be removed. Cold organic solvents may be directly added
to tissue samples and kept below temperatures of 20 C
during the entire sample preparation.
Cells are subsequently disrupted, releasing the metabolites. Frozen samples may be ground down by sonication,
homogenization by mechanical means (for example mortars
or ball mills) in pre-chilled holders [21] or directly in
an extraction solvent [22]. Most frequently, polar organic
solvents such as alcohols are added directly to frozen
samples to extract polar compounds, whereas non-polar
solvents such as chloroform or dichloromethane allow the
extraction of lipids and other hydrophobic compounds.
Sometimes, adding a mixture of polar and non-polar solvents
allows for extraction of both classes of metabolites. Hot
alcoholic extractions are also performed routinely. A procedure for the extraction and separation of metabolites,
proteins and mRNA from a single sample has also been
reported [23]. The mixture of cell debris, protein and the
desired metabolites need to be separated. This can be done
by centrifugation or filtration.
Dynamic sampling methods allow for the determination
of kinetic rates of substrate change. In a Bioreactor equipped
with specialized equipment, this can be done by spraying
samples rapidly into a moving belt of sample tubes containing


C

2007 Portland Press Ltd

73

74

S. H. G. Khoo and M. Al-Rubeai

Table 1

Summary of analytical platforms and their general operational ranges

Analytical platform

Analysis time/throughput

Metabolites measured

Sensitivity/limits of detection

Comments

NMR

Up to 20 min per sample (at

least 500 l); autosampler
available with spectrometers

Typically 2050 metabolites

identified

Micromolar concentrations;
sensitivity depends on
peak obsevation in spectra

Non-invasive method
with a proven trackrecord in the medical field;
provides structural
information

TOF/FT-ICR MS

A 1 min analysis time; thousands

of samples a day; sample volume
<1 l, typically in the nanolitre
range; array formats available

About 200010 000 mass

ions measured

About 5 p.p.m. for standard mass

spectrometers (<1 p.p.m. for
FT-ICR); limits of detection
depend on ionization of molecule

Can be used with

different ionization
modes; ion suppression
an important factor

GC-MS

Ranges from 10 min (run

time) to 1 h (including
derivatization time); about
1000 samples/month; 1
10 l sample volatilized;
autosampler typically available

About 2001000
metabolites

Picomolar concentrations;
sensitivity dependent on MS

Cheap to run and

commonly used due to
long history of usage;
not all samples can
be analysed by GC-MS

LC-MS

Ranges from 10 min to 2 h;

injection range can be from
0.2 l (capillary columns) to
50 l (HPLC columns);
autosampler typically available

Up to 3000 putative
metabolites measured.

Femtomolar concentrations;
sensitivity dependent on MS

Suffers from substantial matrix

effects as well as ion
suppression

quenching liquid [24]. This allows for the automated collections of a large number of samples (up to 4.5 samples/s),
which is ideal in cell culture or bioreactor situations. Many
of the traditional methods mentioned above are slow and
laborious, sometimes leading to the degradation of metabolites. Hence integrated procedures that allow for sampling,
quenching and extraction to be simplified into a single
step have been devised [25]. Samples can be taken rapidly
(less than 1 s) into a sample tube which quenches as well
as disrupts the cells. This is ideal for use with bioreactors
and allows for sampling to be completed in seconds. Intracellular metabolite concentrations are subsequently determined by subtracting the metabolite content of the cell-free
extracellular medium from the resultant mixture.

Analytical instrument platforms

Once metabolites have been extracted, analytical analysis
takes the form of classical separation and identification.
For this purpose, there exist a whole variety of established
instruments, each with their own pros and cons. To understand the reasons for the use of various instrument
platforms, one must understand the desired characteristics
for metabolite analysis. First, the instruments need to have
excellent sensitivity that is, to be able to analyse multiple
metabolite classes without loss of resolution. Peaks should
therefore not represent the merger of several components.
This also means that instruments should be able to handle
a large range of concentrations, ranging from picomolar
to millimolar concentrations. It should also allow easy


C

2007 Portland Press Ltd

identification and quantification of metabolites and allow for

the comparison of relative changes in metabolite abundances
in comparative experiments. Lastly, it should have a rather
short analysis time to increase the number of samples
analysed. High throughput can be understood in two ways:
rapid analysis (short analysis time per sample) and/or having
a wide coverage of components. One way of increasing
resolution is by reducing the number of metabolites that are
simultaneously analysed by the instrument, thus allowing
for the reduction of time for the separation process.
Fractionation can thus be applied to solve this issue [10]. To
determine unknown structures of metabolites, information
from the intact molecule needs to be obtained, such as
functional groups, size (mass) and elemental composition.
Therefore, hyphenated instruments have been shown to be
extremely useful as they allow for an extra dimension of
analysis. Dimensionally they do two things: they increase
the separation of compounds by different characteristics
(such as retention time or different physical properties
such as mass) and provide additional structural information
for identification. It is also possible to combine data from
separate instruments to form a comprehensive study of
metabolism [2628]. Table 1 gives a summary of the different
instruments used.

NMR spectroscopy
NMR is a non-invasive, highly discriminatory, highthroughput method that can analyse rather crude samples.
A single metabolite typically gives several signals in the

Metabolomics and cell culture

inter-laboratory reproducibility, as seen in the COMET

(Consortium for Metabonomic Toxicology) project [33]. In
a ground-breaking use of NMR, metabolite concentrations
in a yeast mutant were monitored and correlated with the
function of genes. This was termed FANCY (Functional
ANalysis by Co-responses in Yeast) [34]. Peak identification
software is also commercially available from Chenomx
(Edmonton, AB, Canada).

Figure 2 Overlaid spectra of replicate extractions of metabolites extracted

from mouse myeloma NS0 cells
The horizontal axis represents p.p.m., while the vertical axis represent the
intensity in arbitrary units.

NMR spectra, causing a problem in the resolution of

individual metabolites. This is disadvantageous, as it makes
identification of metabolites difficult in convoluted complex
spectra. Typical NMR analysis normally allows for the identification of 2050 metabolites. Acquisition time for NMR can
be between 10 and 15 min per sample and is favourable with
transcriptomic and proteomic approaches. However, NMR
falls short of resolution and sensitivity compared with MS
methods. The sensitivity depends on the natural abundance
of the atoms studied (1 H, 31 P, 13 C etc.) or the artificial
introduction of the isotopes into the sample. Significant
amounts of culture (at least 3 million cells) are required
for metabolite extraction. By increasing the strength of
magnetic fields, increased specificity with greater resolution
and separation of signature chemical shifts is possible. In
addition, longer analysis times or the use of cryogenic probes
can help [29]. Another disadvantage of NMR-based analysis
is the fact that the sample tubes are generally large in volume
(at least 500 l), and hence the sample volumes required are
comparatively large (less than 10 l for MS applications).
Details of NMR metabolomic techniques have also been
reviewed [30,31].
By far the most popular means of metabolomic measurement is one-dimensional proton NMR (see Figure 2).
The resulting spectra are generally complex, with overlapping peaks, and thus require significant data processing.
However, to remove ambiguity to the assignment of metabolites, an additional dimension may be added. These
include the J-resolved, homonuclear-shift COSY (correlated
spectroscopy), TOCSY (total correlation spectroscopy) and
NOESY (nuclear overhauser effect spectroscopy) [32].
NMR has also been shown to have reasonably good

MS
MS instruments are by far the most widely used in the field
of metabolomics (including hyphenated technologies). DIMS
(direct injection MS) is the direct injection of samples into
low-resolution ESI (electrospray ionization) MS instruments,
resulting in a quick (less than 1 min/sample) and useful way
of getting high throughput (more than 100 samples per
day) with sufficient information. By controlling the factors
such as ion fragmentation and sample matrix, Dunn et al.
[13] were able to analyse up to 250 plant samples/day
using an ESI-TOF (time-of-flight)-MS. High-throughput MS
can also be achieved by adapting a hybrid FT-ICR (Fouriertransform ion cyclotron resonance) MS with a Nanomate
chip-based nanoelectrospray assembly (Advion BioSciences,
Hethersett, Norwich, U.K.). Typical injection volumes are
in the nanolitre range, with a separation of between 3000
and 10 000 molecular species without chromatographic
separation [9]. Ion suppression is, however, a major problem
in MS. Recent technological advancements have made TOFMS acquisition times quicker and mass determination very
accurate. TOF instruments can provide mass resolutions
of greater than 4000 peaks at mass 200, which allows the
resolution and the detection of metabolites of the same
nominal mass but at different monoisotopic masses [8].
MALDI (matrix-assisted laser-desorptionionization) MS
methods are advantageous, as it is a TOF instrument
that gives direct mass-to-charge ratios, but has substantial
interference from the matrix used. However, to circumvent
the problem, a matrix-free system was developed called the
DIOS (desorption ionization on silicon) method [35],
which, when coupled with a TOF instrument, becomes
a powerful tool for metabolite quantification [36,37].
Unfortunately the major problem with MS methods is that
they cannot differentiate chemical isomers with identical
mass-to-charge ratios, such as those of common hexoses.
Furthermore, owing to the disruption of chemical bonds
during ionization, structural identification from intact masses
of the molecule is lost.
The mass accuracy of these instruments is typically
only 5 p.p.m., and overlapping peaks could result in mass
differences lower than that of the threshold. FT-ICR MS can
overcome this problem, as it has lower limit of detection
(< 1 p.p.m.). The disadvantages of FT-ICR are that it has


C

2007 Portland Press Ltd

75

76

S. H. G. Khoo and M. Al-Rubeai

smaller dynamic ranges and still fails when structural isomers

having the same monoisotopic mass are employed. This new
instrument will become increasingly popular in metabolomic
analyses, and, when coupled with software that can exploit
the information in isotope patterns, it can produce the
empirical formula for metabolites directly [38].
Chromatographic and other column separations
The most common forms of chromatography are GC and
LC (liquid chromatography). GC runs are relatively long,
at about 60 min or more [39]; however, deconvolution
software allows for the decrease in run times [40]. In
LC there is a shift from standard HPLC to UPLC (ultraperformance liquid chromatography), which can significantly
increase resolution sensitivity and peak capacity [41] due to
the reduced particle size, while decreasing sample volumes
(2 l compared with 2050 l in HPLC) and mobile phases
(around 500 l/min compared with 1 ml/min) [42]. UPLC
systems operate at high operating pressures and use
sub-2-m porous packing. The move towards smaller-bore
capillary size columns is also advantageous, as complex
samples require high sensitivities. Unlike pressured systems
such as LC, CE (capillary electrophoresis) makes use of an
electric field to move molecules towards the detector, much
like gel electrophoresis. CE coupled with UV or LIF (laserinduced fluorescence) detectors are highly sensitive, but lack
selectivity [43]. With different detection modes available, it
is possible to add two detectors to a single chromatography
step, as in the example of LC instruments (LC-UVPAD, where PAD is an electrochemical method termed
photodiode array detection) [44]. However, LC-UV-PAD
methods require compounds to contain chromophores.
Electrochemical detectors together with LC instruments
are used commonly as an alternative detection step.
There are several modes of electrochemical detection, such
as amperometric and coulometric [45].
Other vibrational spectroscopies: FTIR
(Fourier-transform infrared), NIR (near-infrared)
and Raman
Vibrational spectroscopies are relatively insensitive, but FTIR
allows for high-throughput screening of biological samples in
an unbiased fashion. Samples require little or no preparation,
with as little as 0.520 l of sample required [8]. However,
IR has some drawbacks. Similar to NMR, water signals
pose a problem and must be subtracted electronically or
attenuated total reflectance may be used. Compared with
the other methods it is one of the least sensitive, but
its unbiasness to compounds and ability to analyse large
numbers of samples in a day (1000 spectra or more) makes
it a plausible method for screening purposes. FTIR has been
used in the preliminary analysis of the yeast metabolome


C

2007 Portland Press Ltd

Figure 3 First-derivative spectra of NIR spectra obtained from

supernatants of yeast fermentation
Illustration provided courtesy of Dr G. McLeod, Department of Chemical
Engineering, University of Birmingham, Birmingham, U.K.

[1] (Figure 3 gives an example of a NIR spectra from yeast

fermentations) as well as providing a non-invasive method
with which to study the overproduction of metabolites
from Escherichia coli and Staphylococcus aureus in vivo [46].
Metabolite fingerprinting techniques for the diagnosis of
disease by the analysis of tissues and biofluids using FTIR
are also possible [4749].
Hyphenated instruments
LC-MS and GC-MS methods are the most common hyphenated technologies. Furthermore, GC-MS is considered the
gold standard in metabolite detection and quantification
[17]. They offer good sensitivity (limits of detection being
picomolar or nanomolar) and selectivity, but have relatively
longer analysis times, owing to the GC separation times [43].
LC-MS methods typically have a somewhat lower chromatographic resolution than GC-MS methods, but, as GC-MS
methods require samples to be rendered volatile, LCMS methods can analyse higher mass ranges and compounds
not easily rendered volatile. GC-MS is a relatively low-cost
method that provides high separation efficiencies (about 300
metabolites/run] [50] that can resolve complex biological
mixtures. To search for physical properties unique to
each metabolite while distinguishing it from neighbouring
peaks, a deconvoluting software is often used on the mass
spectra [40,51] (AMDIS: http://chemdata.nist.gov/mass-spc/
amdis). Potentially, the number of metabolites can increase
to 1000 [52,53]. MSFACTS [54] is another program that can
elucidate a list of metabolites from a database such as KEGG
(Kyoto Encyclopedia of Genes and Genomes) based on the
MS profiles and specific GC retention times. Recently, a

Metabolomics and cell culture

database for GC-MS data on metabolites was proposed [55].

This database would make use of MSRI (mass spectral and
retention time index), which would also contain MST (massspectral tag) profiles from various sources. A detailed review
of GC-MS-based metabolite-profiling technologies has been
given by Kopka [56]. LC-MS make use of IR-UV detectors
that can detect double-bonded or aromatic substances,
while different separation modes can be chosen. Liquidphase methods suffer from significant matrix effects, a major
one being the presence of non-volatile compounds, which
may reduce the evaporation of volatile ions during the electrospray process [57]. This effect is termed ion suppression
and can only be circumvented by reducing the size of liquid
droplets [58]. Typical run times for LC-MS methods range
from short runs of 10 min [41] to 2 h [59]. CE-MS is another
method that can be used to separate a variety of cationic and
anionic compounds, nucleotides and coenzyme metabolites
while identifying and quantifying them by MS [60].
Since LC-MS and NMR analyses are complementary
[41], on-line LC-NMR and LC-NMR-MS approaches have
been developed [61,62]. LC-NMR techniques were first
developed together with other online separation techniques
[63,64], as they were able to overcome the shortcomings of
NMR spectroscopy. 13 C isotopes have been used together
with LC-ESI-MS/MS instruments to follow the metabolism of
yeast cells, thereby eliminating the drawbacks such as nonlinear responses or matrix effects. This isotopic experiment
was termed MIRACLE (mass isotopomer ratio analysis of
U-13 C-labelled extracts) [65]. In addition, the use of multicolumn separations seems logical, as this adds an additional
level of separation in an automated-online fashion [41].
GC GC-TOF-MS is another innovative multidimensional
system that has been developed [66].

Data processing and databases

Making sense of the data collected on multiple metabolites in
the spectra is just as important as the data collection. There
can be several objectives. First, one can aim to identify as
many metabolites as possible to obtain a quantitative or
qualitative measure of changes in the metabolic network
of the cell. These metabolites are assigned to various
functions in the metabolism of the cells and used to
reconstruct the metabolic networks according to the design
of the comparative experiment. Secondly, one can focus on
identifying changes in the metabolic pathways by looking
for emerging/disappearing metabolites as well as statistically
significant metabolite changes without identifying other
non-discriminating metabolites. The identification of these
metabolites can be associated with certain physiological
characteristics such as high production in cell lines, cellular
differentiation or robust growth in cells. In addition, once

Table 2 Common methods for data preprocessing, reduction and

processing
Data preprocessing
Normalization of datadata transforms
Normalization of data using internal standard(s)
Baseline correction, peak shifting and noise removal
Missing value correction
Deconvolution of peaks
Data reduction
Limiting data analysis to specified representative region of data
Excluding variables or samples that do not have
consistent replicates or lie outside the analytical limits
Exclude sample outliers
Data processing methods
Univariate and multivariate statistics
Coefficient of variation
ANOVA or MANOVA
Correlation or regression
Unsupervised methods
PCA, ICA and subtypes
Clustering, HCA, k-means
SOMs
Supervised methods
Fisher discriminant analysis
Partial least squares
Neural networks (artificial and polynomial)
Genetic programming and algorithms

a specific profile has been identified, screening or rapid

identification can be carried out by pattern recognition.
One application of the latter could be in cell line screening
where clones with a known metabolic pattern for higher
productivities can be screened. Data processing is therefore
an important process for making sense of the soup [67],
and the steps taken depend on what ones objectives are.
Goodacre et al. [68] differentiates the interpretation of
data into hypothetico-deductive-driven or inductive-driven
approaches. Most work to date has largely taken the
hypothetico-deductive approach, as very little is known
about the metabolome within a cell. While it is agreed that
statistical analysis must be performed to ensure analytical
rigour [43], how this should done remains a matter of
debate, since different methods could give varying results.
Certainly, for useful knowledge to be derived, chemometric,
comparative and visualization tools will help. Table 2 shows
some of the common methods used for preprocessing,
reduction and processing of data and demonstrate that
pattern recognition and multivariate discrimination analysis
are important steps. As with all omic sciences, large
datasets will be required to be stored, retrieved and analysed
in an open consistent format. In addition, with the use of
multiple instruments, data standardization is necessary for
integrated data analysis. Likewise, with transcriptomic data,
universal data standardization is pertinent if data sharing and
databases are to be established [69]. The Standard Metabolic


C

2007 Portland Press Ltd

77

78

S. H. G. Khoo and M. Al-Rubeai

Reporting Structures working group has outlined a format,

which includes the experimental and data processing stages,
as to how metabolic data should be reported.
To start, the data have to be converted into a format
that can be processed by statistical packages. For NMR
data, this involves separating spectra or plots into discrete
intensity sections. Some of the methods commonly used
include binning and area calculation. This can now be
transformed into a matrix for processing. MS data can also
be converted into similar matrices [70]. Complete determination of the metabolite concentrations can be done at
this stage by cross-referencing with known standards (peak
positions or MSTs). However, this is not always accomplished, mostly because of the lack of a comprehensive database of metabolite concentrations run under the similar
matrix conditions. Instead, it is often the aim to study the
significant changes in the metabolism of the cell and to
identify these specific features. The data matrix is therefore
used in combination with machine-learning methods to
achieve this. Pattern recognition and multivariate analysis
are not distinct, as there is a great deal of similarities
between the two. Both aim to reduce the dimensionality
of the measurement vector. Pattern recognition methods
(inductive machine learning) include non-supervised
methods like NLM (non-linear mapping), HCA (hierarchical
cluster analysis), PCA (principal component analysis) [16],
k-means clustering and SOMs (self-organizing maps) [71].
Unsupervised methods allow the data to make classification
without a training dataset (not a priori output) and can be
used to follow the physiology of a cell in the course of
a culture. For example, the expansion of undifferentiated
stem cells in culture can be studied, as the outcome
of metabolic changes is not known, or when developing
chemically defined media for industrial applications. In both
cases the changes in the physiological response of the
cells in culture can be studied and understood in greater
detail. PCA has become the norm in visualization of data
(Figure 4). It makes use of a statistical method for reducing
multidimensional data (such as multiple spectra) down to a
few dimensions that can be readily comprehended. Another
more recently introduced method is ICA (independent
component analysis; [72]), a linear representation of components that are statistically independent [73] that is ideal
for non-Gaussian datasets from a large number of samples
and a small number of variables. With ICA we are able
to separate the original source of the components by
observing the signal mixtures of the components, hence
it is useful in the separation of different sample classes.
Furthermore, it can also extract features from the recorded
spectra, which is useful for the identification of the specific
metabolites (biomarkers) from the samples. MCA (multilevel
component analysis) has been recently developed [74]
to deal with datasets that have variations in different


C

2007 Portland Press Ltd

Figure 4

PCA plot of mouse myeloma NS0 samples

C represents control cultures, while I represents the induced cultures which

have been activated to overexpress the p21 cytostatic gene.

levels, like variations between organisms and variations in

time. Supervised methods used in pattern recognition also
exist, but require a training dataset in addition to a priori
knowledge for it to function well. This can be used if the
origins of the samples (classes) are known, for example,
when studying the different between high producer clones
and their parental cell lines, hence allowing screening
of clones based on rules obtained from the metabolome.
Examples of these include SIMCA (soft independent modelling of class analogy) and neural networks, partial least
squares [or PLS-DA (partial least squares-discriminant
analysis)], support vector machines, genetic programming
and genetic algorithms. McGovern et al. [75] used genetic
programming and genetic algorithms to decipher spectral
data from different spectroscopic sources.
Once discriminate analysis has been done, it has been
suggested that statistical significance can be determined by
applying classical statistics such as Students t test, ANOVA
or MANOVA (multiple ANOVA) [10]. Given that large
statistical variation exists in biological samples, this may
not be fruitful. Another way of dealing with data after the
preprocessing step is to try to detect significant correlations
of components within a data matrix [73]. For a larger
number of samples with sets of identified and quantified
metabolites, the correlations between them can be found
using Persons correlation coefficient [7], where the distance
of biological connectivity of all the measured metabolites to
be found and enables such distance maps to be visualized.
One recent introduction is the use of correlation matrices
as a step taken after supervised or unsupervised data mining
[73]. Such correlations between metabolites allow for the
understanding of regulatory characteristics in the metabolic

Metabolomics and cell culture

Table 3 Online databases of biochemical references and metabolite profile

references
Abbreviations: BRENDA, BRaunschweig ENzyme Database; EMP, Enzymes and
Metabolic Pathways database; IUBMB, International Union of Biochemistry
and Molecular Biology; DOME, Database of OMEs; AMDIS, Automated Mass
Spectral Deconvolution and Identification System.
Name

URL

Biochemical references
KEGG
BRENDA
The EMP project
IUBMB Enzyme Nomenclature
EcoCyc
MetaCyc

http://www.genome.ad.jp/kegg
http://www.brenda.uni-koeln.de
http://www.empproject.com
http://www.chem.qmul.ac.uk/iubmb/enzyme
http://ecocyc.org
http://metacyc.org

Metabolite profile references

ArMet
DOME
AMDIS
MetAlign
MetaGeneAlyse
MeT-RO

Metabolic tools
AraCyc
MapMan
MetNet
Biosilico

(Magnetic Resonance Metabolomics Database of Linkoping,

University of Linkoping,
Linkoping,
Sweden), ESMRMB
(European Society for Magnetic Resonance in Medicine and
Biology, Basel, Switzerland; mdl.imv.liu.se) and NMRShiftDB
(Cologne University Bioinformatics Centre, Cologne,
Germany; nmrshiftdb.cubic.uni-koeln.de). Although still in
its infancy, metabolomic communities around the world
agree that certain reporting standards are required. ArMET
(architecture for metabolomics), is one such framework
for reporting metabolomic data and experiments for data
storage [78].

Applications in cell culture

http://www.Armet.org
http://medicago.vbi.vt.edu/dome.html
http://chemdata.nist.gov/mass-spc/amdis
http://www.metalign.nl
http://metagenealyse.mpimp-golm.mpg.de
http://www.metabolomics.bbsrc.ac.uk/
MeT-RO.htm

http://www.Arabidopsis.org/biocyc/index.jsp
http://gabi.rzpd.de/projects/MapMan
http://metnet.vrac.iastate.edu/
http://biosilico.kaist.ac.kr

network and the system has been applied to plant samples

[76].
For the proper assignment of metabolites, whether
in terms of identity or models of networks, bioinfomatic
tools and databases are required. Databases allow for the
referencing of metabolites and the elucidation of chemical
structures. However, for databases to be widely used
by different communities, the curation system should be
user-friendly and biology-orientated, thus avoiding different
computational standards which will limit access to data. One
way of introducing standards is the use of SBML (Systems
Biology Markup Language) [77], which will allow interoperability between different models. Two main types of
databases are used in metabolomics: the reference
biochemical databases and the metabolite profile databases.
Although there are many online reference databases of
both types currently being developed (shown in Table 3),
comprehensive databases for intracellular metabolites are
greatly lacking. To date, few public databases exist. Those
that do exist are collections of metabolites from literature
or standards and may not bear any resemblance to the
actual sample composition of cellular mixtures. Examples
of NMR-based collections of metabolites include the BRMB
(Biological Magnetic Resonance Data Bank, Department of
Biochemistry, University of Wisconsin-Madison, Madison,
WI, U.S.A.; www.bmrb.wisc.edu/metabolomics/), the MDL

The rapidly changing face of metabolome analysis, instrumentation development and data processing are fundamentally driven by specific application needs. Metabolomics
diverse importance in the medical, nutrition, health and
environmental fields is its critical factor for growth.
As already mentioned, the potential of metabolomics in
traditional and emerging areas in cell culture has yet to
be realized. One such area that has created a great interest
is in the area of pharmacokinetics and drug testing. With
the initiative to implement the 3R (refinement, reduction
and replacement) principles [79] in animal testing, tissue cell
culture is set to develop into in vitro models as an alternative
means of drug testing. Metabolomics, together with the
other omics, as well as the bioreactor culture of human
tissues, can therefore contribute to the 3R principles
[80]. Metabolomics can be used to characterize tissue
growth in these bioreactors, thereby validating their use as
replacements for drug testing. Furthermore, metabolomics
is the single best window into the cellular state [81] and
therefore is ideal for drug development and testing. It has
been noted that present testing of drugs on animals (namely
dogs, mice and rats) is insufficient in clinical testing and
that these human cell cultures may be an alternative for
understanding the specific pharmacokinetic metabolism of
drug candidates [8183]. A metabolomic approach to drug
discovery using in vitro cultures of cell lines and tissues
is particularly considered to be an appropriate way by
which SPARs (structurepathwayactivity relationships) can
be determined [84]. These tissues and pathogen-specific cell
lines allow dosage and metabolic effects to be characterized.
Such validation and characterization of drug candidates is
critical in obtaining full clinical approval. This has lead to the
USA FDA (Food and Drug Administration) to acknowledge
the need for such methods to be used together with
manufacturing scale-up development and clinical trials of
biologics [85]. NMR metabolic profiling has been used to
characterize human hepatoma cells under specific culture
conditions [86], thus paving the way for understanding the


C

2007 Portland Press Ltd

79

80

S. H. G. Khoo and M. Al-Rubeai

Figure 5 PCA plot of metabolic profile of a NS0 myeloma batch culture

The circles represent samples taken from cultures across different stages of growth; early (red)-, mid (blue)- and late (green)-exponential phases of growth. The
arrow shows the trajectory of the metabolic profile changes as the culture grows.

metabolism in these cancer cells. Some examples of drug and

toxin metabolism studies include work in endometrial cells
[87] and the response of plant [Silene cucubalus (bladder
campion)] cell cultures to environmental toxins such as
cadmium [88].
Other applications include plant and mammalian cell
metabolic engineering. While plant metabolomics has been
around for the last decade or so, mammalian cell metabolomics for industrial cell lines is greatly lacking. The use of
metabolomics in plant cell cultures allows for the determination of important secondary metabolites [89] such as
isoflavone and taxol, which have been proved to be effective
pharmaceutical ingredients [90,91]. In the case of flavonal
metabolic engineering, the transformation of ubiquitous
naringenin to the isoflavone genistein as a means of introduction of isoflaviods into legumes [92,93] has been carried out.
Taxol, a form of taxoid compound, is an anticancer drug
commercially available on the market. Metabolomic studies
on Taxus (yew) cell cultures have allowed for the identification new taxoids [94], which could lead to the further
understanding of the induction of taxoid production in Taxus
cells [95] and the production of synthetic versions in other
plant cells [96]. Furthermore, the need to understand the
elicitor induction of secondary metabolite production in
plant cells also opens up new doors for metabolomic analysis
[97] and also emphasizes how metabolomics can be a functional genomic tool in studying transgenic plants [43]. One


C

2007 Portland Press Ltd

side of the spectrum, the production of biopharmaceuticals

from animal cells, has not embraced metabolomics as a
tool. This is mostly because metabolites are now not the
primary focus, and the relationship between metabolism
and protein production is not fully understood. Yet it is
precisely for this reason that metabolomics can bridge the
gap of understanding as to the dynamics of metabolism,
cell growth and protein production. One demonstration of
this relationship is shown in Figure 5, where the metabolite
profile of cell cultures varies across the age of a batch
culture is depicted. In addition, it may be used to optimize
conditions of bioreactors or the development of chemically
defined media. It can been seen that the rapid growth of cells
gives rise to a metabolic profile that can be distinguished
from that of slow-growing cells (S. H. G. Khoo and M.
Al-Rubeai, unpublished work). Therefore the effect of
specific chemicals on metabolism can be screened rapidly
using metabolomics methods, thus linking chemical entity
to high or robust growth and even productivity. Miniature
bioreactors that allow for a variety of conditions to be
monitored can be used together with metabolomic studies
with rapid sampling [98]. Metabolomics has been used to
monitor cell transfections using similar techniques [99]
as well as being used as a determinant of apoptosis in
cell culture [100]. In the latter example, cells sensitive to
apoptosis induction were distinguished from resistant cells
and hence showed the potential of cell line selection in

Metabolomics and cell culture

industrial applications. Characterizing cell lines and selection

of cell lines are a vital step in the process development of
biologics. Once the clone is selected for production, it is
unlikely to be changed for the entire lifespan of production.
Hence stable and efficient protein production lies on the vital
stage of clonal selection [101]. In addition, cell line genetic
drifts and clonal variations, which are important validating
steps in cGMP (current Good Manufacturing Practice)
production, can be monitored by using high-throughput
metabolomics methods.
Monitoring mutagenesis and gene deletions in cell
cultures can also be carried out using metabolomics.
Gene knockouts and genetic strategies using RNAi (RNA
interference) can be used with phenotypic screens such as
metabolomics to determine the physiological functions of
enzymes. It has been used to map silent mutations [34,102]
as well as to observe pleiotropic effects of single genetic
alterations [102], but it has mostly been applied to microbial
systems. Hence it is also a vital research tool and an essential
part of any cellular modelling process. As the metabolic
network in a living cell is viewed as being a complex
network of reactions that are tightly connected [103], any
perturbations in the transcriptome and proteome can filter
down to the concentration of metabolites. This makes the
metabolome data a form of integrated data, which gives
it strength as an omic science. Metabolite concentration
and fluxes have been shown to not be regulated by gene
expression alone (e.g. glycolysis in trypanosomes; [104]).
This may again disprove the idea of a simple hierarchical
regulation of function by gene expression, as is the case
with proteins and mRNA. Metabolic flux analysis can
also integrate metabolite concentration data, making flux
data more accurate than hitherto. Principally, using stable
isotopes, flux can be determined by feeding strategies as well
as rapid sampling methods [24,105]. Flux analysis with intracellular metabolite concentrations can be combined with
traditional isotopomer flux analysis and constraints-based
modelling to give a better picture of cell phenotypes [106].

omics is not always simple, owing to mRNA splicing and

translation control issues [108]. In addition, a single gene
may code for isoenzymes reacting with multiple metabolite
substrates [109]. The difficulty in determining the timing of
different events, that is, transcription and protein activity
[110], also contribute to the difficulty in integrating data.
Hence, in order for metabolomics to be used in systems biology, novel strategies will need to be created. One step forward in such an integration process is the functional assignments between protein/gene and metabolite within a
system of interest [111,112]. This can be done by creating decomposable models [113] where basic biochemical
pathways are first modelled using static data. Subsequently,
time-dependent concentrations of other types of components (transcriptomic and/or proteomic) will then be
incorporated. Databases and standardization also play key
roles in this process, as data sharing will be the cornerstone
of any major cellular reconstruction model.
Of particular interest is the coining of the terms
synthetic biology and systems biotechnology, that is, the
engineering of cells with novel abilities [114,115]. In these
proposed processes, systems biology is considered to be
the complete characterization of a cell of interest. This
characterization involves the transmission of data by using
matrices to an in silico process. Modelling and simulation of
the cellular components follows, and what results will lead
to the redesigning of the cellular system. Since this is an
iterative process, the feedback loops allow for the improved
learning of cellular dynamics. As with most developments,
this is still many years away and the likelihood of its
implementation in cell culture systems rest upon the initial
developments in microbial organisms such as E. coli. Clearly,
the exciting prospect of this happening is limited by the
present state of technological development and the lack
of data-handling approaches in dealing with the volume of
information. Yet, in due time, the widespread use of metabolomics in understanding biological processes will see
boundless benefits.

Towards systems biology

References

Data integration is not limited to flux data. The concept

of systems biology is similar to functional genomics in its
approach, but is slightly different in its objectives. Systems
biology encompasses a holistic approach to the study
of biology and the objective is to simultaneously monitor
all biological processes operating as an integrated system.
It involves the iterative interplay between linked activities
which allow for the study of signal-processing elements
that lie downstream of the signal initiator. These help
one understand how cross-talk may be occurring between
pathways [107]. However, as mentioned above, linking

1 Oliver, S. G., Winson, M. K., Kell, D. B. and Bagnanz, F. (1998)

Trends Biotechnol. 16, 373377
2 Beecher, C. W. W. (2002) Innov. Pharm. Technol., June, 5764
3 Beecher, C. W. W. (2003) In Metabolic Profiling: Its Role in
Biomarker Discovery and Gene Function Analysis (Harrigan,
G. G. and Goodacre, R., eds), pp. 311319, Kluwer Academic
Publishers, Amsterdam
4 Trethewey, R. N., Krotzky, A. J. and Willmitzer, L. (1999)
Curr. Opin. Plant Biol. 2, 8385

5 Villas-Boas,
S. G., Rasmussen, S. and Lane, G. A. (2005)
Trends Biotechnol. 23, 385386


C

2007 Portland Press Ltd

81

82

S. H. G. Khoo and M. Al-Rubeai

6 Bugrim, A., Nikolskaya, T. and Nikolsky, Y. (2004) Drug Discov.

Today 9, 127135
7 Weckwerth, W. (2003) Annu. Rev. Plant Biol. 54, 669689
8 Dunn, W. B. and Ellis, D. (2005) Trends Anal. Chem. 24,
285294
9 Brown, S. C., Kruppa, G. and Dasseux, J.-L. (2005)
Mass Spectrom. Rev. 24, 223231
10 Fiehn, O. (2002) Plant Mol. Biol. 48, 155171
11 Klose, J. and Kobalz, U. (1995) Electrophoresis 16, 10341059
12 Tweeddale, H., Notley-McRobb, L. and Ferenci, T. (1998)
J. Bacteriol. 180, 51095116
13 Dunn, W. B., Overy, S. and Quick, W. P. (2005) Metabolomics
1, 137148
14 Fiehn, O. (2001) Comp. Funct. Genom. 2, 155168
15 Nicholson, J. K., Lindon, J. C. and Holmes, E. (1999)
Xenobiotica 29, 11811189
16 Mitchell, S., Holmes, E. and Carmichael, P. (2002) Biologist 49,
217221
17 Fiehn, O. and Spranger, J. (2003) In Metabolite Profiling: Its Role
in Biomarker Discovery and Gene Function Analysis (Harrigan,
G. G. and Goodacre, R., eds.), pp. 199216, Kluwer Academic
Publishers, Amsterdam
18 Gerdtzen, Z. P., Daoutidis, P. and Hu, W.-S. (2004) Metab. Eng.
6, 140154
19 Taoka, S. and Banerjee, R. (2002) J. Biol. Chem. 277,
2242122425
20 ap Rees, T. and Hill, S. A. (1994) Plant Cell Environ. 17,
587599
21 Fiehn, O., Kopka, J., Trethewey, R. N. and Willmitzer, L. (2000)
Anal. Chem. 72, 35733580
22 Orth, H. C. J., Rentel, C. and Schmidt, P. C. (1999) J. Pharm.
Pharmacol. 51, 193200
23 Rossi, D. T. and Sinz, M. W. (2002) Mass Spectrometry in Drug
Discovery, Marcel Dekker, New York
24 Buchholz, A., Hurlebaus, J., Wandrey, C. and Takors, R. (2002)
Biomol. Eng. 19, 515
25 Schaub, J., Schiesling, C., Reuss, M. and Dauner, M. (2006)
Biotechnol. Prog. 22, 14341442
26 Kaderbhai, N. N., Broadhurst, D. I., Ellis, D. I., Goodacre, R. and
Kell, D. B. (2003) Comp. Funct. Genom. 4, 376391
27 Abel, C. B. L., Lindon, J. C., Noble, D., Rudd, B. A. M.,
Sidebottom, P. J. and Nicholson, J. K. (1999) Anal. Biochem.
270, 220230
28 Lenz, E. M., Bright, J., Knight, R., Wilson, I. D. and Major, H.
(2004) J. Pharm. Biomed. Anal. 35, 599608
29 Keun, H. C., Beckonert, O., Griffin, J. L., Richter, C.,
Moskau, D., Lindon, J. C. and Nicholson, J. K. (2002)
Anal. Chem. 74, 45884593
30 Griffin, J. L. (2003) Curr. Opin. Chem. Biol. 7, 648654
31 Defernez, M. and Colquhoun, I. J. (2003) Phytochemistry 62,
10091017
32 Fan, T. W.-M. (1996) Prog. Nucl. Magn. Reson. Spectrosc. 28,
161219


C

2007 Portland Press Ltd

33 Lindon, J. C., Nicholson, J. K., Holmes, E., Antti, H., Bollard,

M. E., Keun, H., Beckonert, O., Ebbels, T. M., Reily, M. D.,
Robertson, D. et al. (2003) Toxicol. Appl. Pharmacol. 187,
137146
34 Raamsdonk, L. M., Teusink, B., Broadhurst, D., Zhang, N.,
Hayes, A., Walsh, M. C., Berden, J. A., Brindle, K. M., Kell,
D. B., Rowland, J. J. et al. (2001) Nat. Biotechnol. 19, 4550
35 Wei, J., Burlak, J. M. and Siuzdak, G. (1999) Nature 399,
243246
36 Go, E. P., Prenni, J. E., Wei, J., Jones, A., Hall, S. C., Witkowska,
H. E., Shen, Z. and Siuzdak, G. (2003) Anal. Chem. 75,
25042506
37 Go, E. P., Shen, Z., Harris, K. and Siuzdak, G. (2003)
Anal. Chem. 75, 54755479
38 Aharoni, A., Ric de Vos, C. H., Verhoeven, H. A., Maliepaard,
C. A., Kruppa, G., Bino, R. and Goodenowe, D. (2002) OMICS:
J. Integ. Biol. 6, 217234
39 Roessner, U., Wagner, C., Kopka, J., Trethewey, R. N. and
Willmitzer, L. (2000) Plant J. 23, 131142
40 Halket, J. M., Przyborowska, A., Stein, S. E., Mallard, W. G.,
Down, S. and Chalmers, R. A. (1999) Rapid Commun. Mass
Spectrom. 13, 279284
41 Wilson, I. D., Plumb, R., Granger, J., Major, H., Williams, R. and
Lenz, E. M. (2005) J. Chromatogr. B 817, 6776
42 Novakova, L., Matysova, L. and Solich, P. (2006) Talanta 68,
908918
43 Sumner, L. W., Mendes, P. and Dixon, R. A. (2003)
Phytochemistry 62, 817836
44 Fraser, P. D., Pinto, M. E., Holloway, D. E. and Bramley, P. M.
(2000) Plant J. 24, 551558
45 Gonzalez de la Huebra, M.-J., Bordin, G. and Rodriguez, A. R.
(2003) Anal. Bioanal. Chem. 375, 10311037
46 Winson, M. K., Goodacre, R., Timmins, E. M., Jones, A., Alsberg,
B. K., Woodward, A. M., Rowland, J. J. and Kell, D. B. (1997)
Anal. Chim. Acta 348, 273282
47 Heise, H. M., Voigt, G., Lampen, P., Kupper, L., Rudloff, S. and
Werner, G. (2001) Appl. Spectrosc. 55, 434443
48 Chiriboga, L., Xie, P., Yee, H., Vigorita, V., Zarou, D., Zakim, D.
and Diem, M. (1998) Biospectroscopy 4, 4753
49 Schmitt, J., Beekes, M, Brauer, A., Udelhoven, T., Lasch, P. and
Naumann, D. (2002) Anal. Chem. 74, 38653868
50 Fiehn, O., Kopka, J., Dormann, P., Altmann, T., Trethewey, R. N.
and Willmitzer, L. (2000) Nat. Biotechnol. 18, 11571161
51 Stein, S. E. (1999) J. Am. Soc. Mass Spectrom. 10, 109110
52 Lu, X., Cai, J. L., Kong, H. W., Wu, M., Hua, R. X., Zhao, M. Y.,
Liu, J. F. and Xu, G. W. (2003) Anal. Chem. 75, 44414451
53 Blumberg, L. M. (2003) J. Chromatogr. A 985, 2938
54 Duran, A. L., Yang, J., Wang, L.-J. and Sumner, L. W. (2003)
Bioinformatics 19, 22832293
55 Schuaer, N., Steinhauser, D., Strelkov, S., Schomburg, D., Allison,
G., Moritz, T., Lundgren, K., Roessner-Tunali, U., Forbes, M. G.,
Willmitzer, L. et al. (2005) FEBS Lett. 579, 13321337
56 Kopka, J. (2006) J. Biotechnol. 124, 312322

Metabolomics and cell culture

57 Niessen, W. M. (2003) J. Chromatogr. A 1000, 413436

58 Bahr, U., Pfenninger, A., Kara, M. and Stahl, B. (1997)
Anal. Chem. 69, 45304535
59 Rochfort, S. (2005) J. Nat. Products 68, 18131820
60 Soga, T., Ohashi, Y., Ueno, Y., Naraoka, H., Tomita, M. and
Nishioka, T. (2003) J. Proteome Res. 2, 488494
61 Lindon, J. C., Nicholson, J. K. and Wilson, I. D. (2000)
J. Chromatogr. B 748, 233258
62 Corcoran, O. and Spraul, M. (2003) Drug Discov. Today 8,
624631
63 Albert, K. (1995) J. Chromatogr. A 703, 123147
64 Lindon, J. C., Nicholson, J. K. and Wilson, I. D. (1996) Prog.
Nucl. Magn. Reson. Spectrosc. 29, 149
65 Mashego, M. R., Wu, L., Van Dam, J. C., Vinke, J. L., Van Winden,
W. A., Van Gulik, W. M. and Heijnen, J. J. (2004) Biotechnol.
Bioeng. 85, 620628
66 Shellie, R. A., Welthagen, W., Zrostlikova, J., Spranger, J.,
Ristow, M., Fiehn, O. and Zimmermann, R. (2005)
J. Chromatogr. A 1086, 8390
67 Kell, D. B. (2004) Curr. Opin. Microbiol. 7, 296307
68 Goodacre, R., Vaidyanathan, S., Dunn, W. B., Harrigan, G. G.
and Kell, D. B. (2004) Trends Biotechnol. 22, 245252
69 Lindon, J. C., Nicholson, J. K., Holmes, E., Keun, H. C.,
Craig, A., Pearce, J. T. M., Bruce, S. J., Hardy, N., Sansone, S.-A.,
Antti, H. et al. (2005) Nat. Biotechnol. 23, 833838
70 Broeckling, C. D., Reddy, I. R., Duran, A. L., Zhao, X. and
Sumner, L. W. (2006) Anal. Chem. 78, 43344341
71 Kohonen, T. (1995) Self-Organizing Maps, Springer-Verlag,
New York
72 Scholz, M., Gatzek, S., Sterling, A., Fiehn, O. and Selbig, J. (2004)
Bioinformatics 20, 22472454
73 Weckwerth, W. and Morgenthal, K. (2005) Drug Discov.
Today: Targets 10, 15511558
74 Jansen, J. J., Hoefsloot, H. C. J., van der Greef, J., Timmerman,
M. E. and Smilde, A. K. (2005) Anal. Chim. Acta 530, 173183
75 McGovern, A. C., Broadhurst, D., Taylor, J., Kaderbhai, N.,
Winson, M. K., Small, D. A., Rowland, J. J., Kell, D. B. and
Goodacre, R. (2002) Biotechnol. Bioeng. 78, 527538
76 Morgenthal, K., Weckwerth, W. and Steuer, R. (2006)
Biosystems 83, 108117
77 Hucka, M., Finney, A., Sauro, H. M., Bolouri, H., Doyle,
J. C., Kitano, H., Arkin, A. P., Bornstein, B. J., Bray, D.,
Cornish-Bowden, A. et al. (2003) Bioinformatics 19, 524531
78 Jenkins, H., Hardy, N., Beckmann, M., Draper, J., Smith, A. R.,
Taylor, J., Fiehn, O., Goodacre, R., Bino, R., Hall, R. et al. (2004)
Nat. Biotechnol. 22, 16011606
79 Balls, M. and Fentem, J. H. (1999) Toxicol. In Vitro 13, 837846
80 Kroeger, M. (2006) Trends Biotechnol. 24, 343346
81 Beecher, C. W. W. (2005) Innov. Pharm. Technol. 2005, 1925
82 Li, A. P. (2004) Drug Discov. Today 9, 687693
83 Li, A. P. (2004) Chem.Biol. Interact. 150, 2733
84 Cho, C. R., Labow, M., Reinhardt, M., van Oostrum, J. and
Peitsch, M. C. (2006) Curr. Opin. Chem. Biol. 10, 294302

85 Kozlowski, S. and Swann, P. (2006) Adv. Drug Deliv. Rev. 58,

707722
86 Miccheli, A. T., Miccheli, A., Clemente, R. D., Valerio, M.,
Coluccia, P., Bizzarri, M. and Conti, F. (2006) Biochim. Biophys.
Acta 1760, 17231731
87 Griffin, J. L., Pole, J. C. M., Nicholson, J. K. and Carmichael, P. L.
(2003) Biochim. Biophys. Acta 1619, 151158
88 Bailey, N. J. C., Oven, M., Holmes, E., Nicholson, J. K. and Zenk,
M. H. (2003) Phytochemistry 62, 851858
89 Oksman-Caldentey, K.-M. and Inze, D. (2004) Trends Plant Sci.
9, 432440
90 Liu, C.-J., He, X. G., Lindenmaiser, M., Yang, J., Cleary, M.,
Qiu, S. X. and Cordell, G. A. (2000) J. Agric. Food Chem.
48, 354365
91 Muir, S. R., Collins, G. J., Robinson, S., Hughes, S., Bovy, A.,
De Vos, C. H. R., van Tunen, A. J. and Verhoeyen, M. E. (2001)
Nat. Biotechnol. 19, 470474
92 Jung, W., Yu, O., Lau, S.-M. C., OKeefe, D. P., Odell, J.,
Fader, G. and McGonigle, B. (2000) Nat. Biotechnol. 18,
208212
93 Liu, C.-J., Blount, J. W., Steele, C. L. and Dixon, R. A. (2002)
Proc. Natl. Acad. Sci. U.S.A. 99, 1457814583
94 Ketchum, R. E. B., Rithner, C. D., Qiu, D., Kim, Y. S., Williams,
R. M. and Croteau, R. B. (2003) Phytochemistry 62, 901909
95 Wang, Y.-D., Yuan, Y.-J. and Wu, J.-C. (2004) Biochem. Eng. J.
19, 259265
96 Besumbes, O., Sauret-Gueto, S., Phillips, M. A., Imperial, S.,
Rodiguez-Concepcion, M. and Boronat, A. (2004) Biotechnol.
Bioeng. 88, 168175
97 Zhao, J., Davis, L. C. and Verpoorte, R. (2005) Biotechnol. Adv.
23, 283333
98 Eklund, S. E., Snider, R. M., Wikswo, J., Baudenbacher, F.,
Prokop, A. and Cliffel, D. E. (2006) J. Electroanal. Chem. 587,
333339
99 Griffin, J. L., Mann, C. J., Scott, J., Shoulders, C. C. and
Nicholson, J. K. (2001) FEBS Lett. 509, 263266
100 Lutz, N. W. (2006) C. R. Chimie 9, 445451
101 Barnes, L. M. and Dickson, A. J. (2006) Curr. Opin. Biotechnol.
17, 381386
102 Weckwerth, W., Loureiro, M. E., Wenzel, K. and Fiehn, O.
(2004) Proc. Natl. Acad. Sci. U.S.A. 101, 78097814
103 Nielsen, J. (2003) J. Bacteriol. 185, 70317035
104 ter Kuile, B. H. and Westerhoff, H. V. (2001) FEBS Lett. 500,
169171
105 Harada, K., Fukusaki, E., Bamba, T., Sato, F. and Kobayashi, A.
(2006) Biotechnol. Prog. 22, 10031011
106 Lee, W. N. P. (2006) Metabolomics 2, 3139
107 Kell, D. B. (2005) Biochem. Soc. Trans. 33, 520524
108 Hanash, S. M. and Beretta, L. M. (2002) Brief. Funct. Genom.
Proteomics 1, 1022
109 Schwab, W. (2003) Phytochemistry 62, 837849
110 Nicholson, J. K., Holmes, E., Lindon, J. C. and Wilson, I. D.
(2004) Nat. Biotechnol. 22, 12681274


C

2007 Portland Press Ltd

83

84

S. H. G. Khoo and M. Al-Rubeai

111 Bino, R. J., Hall, R. D., Fiehn, O., Kopka, J., Saito, K.,
Draper, J., Nikolau, B. J., Mendes, P., Roessner-Tunali, U., Beale,
M. H., Trethewey, R. N., Lange, B. M., Wurtele, E. S. and
Sumner, L. W. (2004) Trends Plant Sci. 9, 418425
112 Saghatelian, A. and Cravatt, B. F. (2005) Curr. Opin. Plant Biol.
9, 6268
113 Arita, M., Robert, M. and Tomita, M. (2005) Curr. Opin.
Biotechnol. 16, 344349

C

2007 Portland Press Ltd

114 Barrett, C. L., Kim, T. Y., Kim, H. U., Palsson, B. O. and Lee, S. Y.
(2006) Curr. Opin. Biotechnol. 17, 492
115 Lee, S. Y., Lee, D.-Y. and Kim, T. Y. (2005) Trends Biotechnol.
23, 349358
Received 3 November 2006/8 February 2007; accepted 20 March 2007
Published on the Internet 18 May 2007, doi:10.1042/BA20060221