See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/44570466

The Influence of a Sports Drink on the

Postexercise Metabolism of Elite Athletes as
Investigated by NMR-Based Metabolomics
Article in Journal of the American College of Nutrition October 2009
DOI: 10.1080/07315724.2009.10719787 Source: PubMed

CITATIONS

READS

28

98

10 authors, including:
Federico Marini

Giorgio Capuani

Sapienza University of Rome

Sapienza University of Rome

115 PUBLICATIONS 1,385 CITATIONS

35 PUBLICATIONS 530 CITATIONS

SEE PROFILE

SEE PROFILE

Maria Enrica Di Cocco

Maurizio Paci

Sapienza University of Rome

University of Rome Tor Vergata

43 PUBLICATIONS 719 CITATIONS

205 PUBLICATIONS 3,104 CITATIONS

SEE PROFILE

All in-text references underlined in blue are linked to publications on ResearchGate,

letting you access and read them immediately.

SEE PROFILE

Available from: Federico Marini

Retrieved on: 29 July 2016

Original Research

The Influence of a Sports Drink on the Postexercise

Metabolism of Elite Athletes as Investigated by
NMR-Based Metabolomics
Alfredo Miccheli, MD, Federico Marini, PhD, Giorgio Capuani, PhD, Alberta Tomassini Miccheli, PhD, Maurizio Delfini,
PhD, Maria Enrica Di Cocco, PhD, Caterina Puccetti, PhD, Maurizio Paci, PhD, Marta Rizzo, MD, Antonio Spataro, MD
Department of Chemistry, Sapienza University of Rome (A.M., F.M., G.C., A.T.M., M.D., M.E.D.C., C.P.), Department of Chemical
Science and Technology, University of Rome Tor Vergata (M.P.), Medical Committee of the Italian Rowing Federation
(M.R., A.S.), Rome, ITALY
Key words: metabolomics, sports drink, physical exercise, NMR spectroscopy, multivariate analysis
Objective: The aim of this study is to evaluate the systemic effects of an isotonic sports drink on the
metabolic status of athletes of the Italian Olympic rowing team during recovery after strenuous and prolonged
physical exercise by means of nuclear magnetic resonance (NMR)-based metabolomics analysis on plasma and
urine.
Methods: Forty-four male athletes of the Italian Olympic rowing team were enrolled in a double-blind
crossover study. All subjects underwent 2 evaluations at 1-week intervals. The evaluation was performed on a
rowing ergometer after strenuous physical exercise to produce a state of dehydration. Afterward, the athletes
were rehydrated either with a green teabased carbohydrate-hydroelectrolyte drink or with oligomineral water.
Three blood samples were drawn for each subject: at rest, after the exercise, and following rehydratation, while
2 urine samples were collected: at rest and after the rehydratation period. Biofluid samples were analyzed by
high-resolution 1H NMR metabolic profiling combined with multilevel simultaneous data-analysis (MSCA) and
partial-least squares-discriminant analysis (PLS-DA).
Results: The between-subject variations, as evaluated by MSCA, reflected the variations of lactate levels
induced by the physical exercise. Analysis of the within-individual variance using multilevel PLS-DA models of
plasma and urine metabolic profiles showed an effect of the green teabased sports drink on glucose, citrate, and
lactate levels in plasma and on acetone, 3-OH-butyrate, and lactate levels in urine. The increase of caffeine and
hippuric acid levels in urine indicated the absorption of green tea extract components.
Conclusions: NMR-based metabolomics allowed the complex effects of a green tea extractbased
carbohydrate/hydroelectrolyte beverage on the energy metabolism of athletes during recovery by postexercise
rehydration to be evaluated.

INTRODUCTION

resulting from the loss of water and electrolytes in sweat [1,2].

For this reason, the commonly formulated sports drinks contain
added sugars and electrolytes to supply substrates, prevent
dehydration, replace lost electrolytes, and rapidly rehydrate
during the postexercise period [3].
An important aspect of fluid homeostasis during prolonged
exercise is the ability to absorb ingested fluids [4]. The benefit
of ingested carbohydrate is related to its role not only in
supplementing endogenous stores in the later stages of exercise
[5] but also in promoting water absorption in the small

There is a growing interest in functional foods such as

sports drinks designed to improve performance in athletes. The
athlete ingests drinks before, during, or after training or
competition to improve performance by minimizing the impact
of the factors that cause fatigue and impair the performance of
skilled tasks. The factors that have been considered to
contribute most to the onset of fatigue in exercise are the
depletion of glycogen stores and the onset of dehydration

Address reprint requests to: Alfredo Miccheli, Dipartimento di Chimica, Sapienza Universita` di Roma, P.le Aldo Moro 5, 00185 Rome, ITALY. E-mail:
alfredo.miccheli@uniroma1.it

Journal of the American College of Nutrition, Vol. 28, No. 5, 553564 (2009)
Published by the American College of Nutrition
553

Metabolomics and Sports Drink Effects

intestine [6]. Intestinal glucose absorption is an active, energyconsuming process linked to the transport of sodium. Although
water movement in the proximal segment of the small intestine
is a passive process driven by the local osmotic gradient, it is
closely linked to the active transport of solute. On this basis,
the osmolality of ingested fluids is considered to play a key
role in the flux of water across the upper part of the small
intestine to achieve an optimal rehydration [7,8]. Despite the
recommendations by the American College of Sports Medicine
[9] that state that cool water is the best fluid to be ingested
during endurance exercise, a considerable body of evidence
supports the consumption of carbohydrate-electrolyte sports
drinks for high-intensity exercise lasting about 60 minutes
[10]. Therefore, carbohydrate-electrolyte beverages, considered a functional food, are becoming one of the most
extensively investigated sports nutrition topics, but in most
cases, the benefits claimed by advertising have not been
substantiated by scientific evidence of actual effects on the
whole metabolism of athletes during or after exercise. The
concept of functional foods requires the understanding of the
mechanism of prevention and protection, the identification of
the biologically active molecules, and the demonstrated
efficacy of these molecules.
Modern analytical platforms, based on such technology as
nuclear magnetic resonance (NMR) spectroscopy or mass
spectrometry associated with multivariate data analysis, allow
the simultaneous measurement of multiple metabolites from
biofluids, thus providing new insights into our understanding
of the functionalities of whole organisms [11].
Metabonomics, or metabolomics, has formally been
defined as the quantitative measurement of the dynamic
multiparametric metabolic response of living systems to
pathophysiological stimuli or genetic modification [12]. The
potential of metabolomics in pharmacology and in toxicology
has been clearly recognized [13], and its role in human
nutrition is an emerging field of investigation [14].
One of the key aims of nutritional metabolomics is to
identify those small molecules responsible for the difference
between the effects of different diets and, in so doing, deepen
our knowledge of human health and the interacting and
regulatory roles of nutrition [15,16].
The aim of this study was to evaluate the systemic effects
of an isotonic sports drink on the metabolic status of athletes of
the Italian Olympic rowing team during recovery after
strenuous and prolonged physical exercise by means of
NMR-based metabolomics analysis of plasma and urine.
Since the effects of nutritional treatment tend to be smaller
than the biological variation found between individuals, we
designed a crossover experimental protocol allowing comparison at the individual level and applied the combination of
multilevel simultaneous data analysis (MSCA) and partialleast squares-discriminant analysis (PLS-DA) to evaluate the

554

impact of a sports drink on athletes metabolism, as

recommended by van Velzen et al. [17].

MATERIALS AND METHODS

Study Protocol
Forty-four male athletes of the Italian Olympic rowing team
were enrolled in a double-blind crossover designed study. The
average age was 25.7 6 3.7 years (range, 1834 years), the
average height was 187.5 6 6.6 cm (range, 174198 cm), and
the average body mass was 84.7 6 9.4 kg (range, 7094 kg).
Each athlete was previously informed about the experimental
research protocol and signed an informed consent form
agreeing to the study. All of the subjects, divided into 2
subgroups (A and B), were subjected to 2 evaluations at an
interval of 1 week. In both sessions, each athlete was evaluated
in the morning, at rest, fasting (without ingesting any fluids for
at least 8 hours), and at least 24 hours after the latest training
session. The evaluation was performed on a rowing ergometer
(Concept 2, Morrisville, VT) and consisted of a 20-minute
warm up, 1000 m at maximal exercise, and submaximal
exercise for about 50 minutes to produce a state of
dehydration. Subsequently, the 2 groups were rehydrated,
either with a green tea extractbased carbohydrate-hydroelectrolyte drink (Isote`, Coop, Italy; group A) or with oligomineral
water (group B). The rehydration protocol called for the
administration of 25 ml of liquid every 510 minutes, to
restore the body weight at rest, with a minimum ingestion of
500 ml of fluid. The environmental temperature and humidity
were monitored during the tests. One week later, the protocol
was repeated, swapping the 2 groups; that is, group A was
rehydrated by administration of oligomineral water and group
B by administration of the hydroelectrolyte drink. The total
average volume of oligomineral water or sports beverage
ingested by athletes within 120 minutes after performing
exercise was 750 6 250 ml and 700 6 250 ml, respectively.
Three blood samples were drawn for each subject: at rest,
immediately after exercise, and 120 minutes after the end of
exercise, while 2 urine samples were collected: at rest and
120 minutes after exercise (rehydration period).

Sports Beverage Chemical Composition

The green teabased carbohydrate-hydroelectrolyte drink
(Isote`) is a natural beverage without chemical additives or
preservatives. Its resulting chemical composition is water,
fructose, lemon juice, acerole juice, mineral salts (sodium citrate,
sodium chloride, monopotassium phosphate, magnesium carbonate), natural flavors, and green tea extract rich in polyphenols.
The measured average quantities of the most important
compounds are reported in Table 1.

VOL. 28, NO. 5

Metabolomics and Sports Drink Effects

1

Table 1. Average Quantities for 100 ml of Product

Total sugars
Fructose*
Citrate*
Polyphenols
Sodium
Potassium
Chloride
Magnesium

4.2
3.2
254
57
46
14
25
2.6

g
g
mg
mg
mg
mg
mg
mg

* Measured by nuclear magnetic resonance spectroscopy on the final product.

Sample Pretreatment
Plasma. Blood samples were drawn into Li/Heparine
vacuum containers (BD, Franklin Lakes, NJ). Plasma was
separated by centrifugation and immediately frozen at 280uC.
Afterward, 700 ml of each sample was deproteinized by
ultrafiltration with 10-kD filters (Amicon Ultra-4; Millipore,
Billerica, MA) according to a previously reported procedure
[18]. A total of 500 ml of the resulting solution was added to
100 ml of a PBS-buffered D2O solution containing the sodium
salt of trimethylsilyl-2,2,3,3-tetradeuteropropionic acid (TSP;
2 mM final concentration) as a quantitative and chemical shift
reference; the resulting pH was 7.2. The final number of
subjects included in the study was 24 due to the drop out of
some of the athletes or to accidental causes, such as hemolysis.
This led to 144 plasma samples.
Urine. Urine samples were collected, centrifuged, sterilized
by 0.22-mm filtration (Millipore), and frozen at 280uC.
Samples for NMR analysis were prepared by taking a 700-ml
aliquot and adding 70 ml of a solution of 20.2-mM TSP in D2O
(2 mM final concentration). The pH was adjusted to 2.5 with
HCl 2N. A total of 600 ml of the resulting solution was
analyzed. The total number of subjects was 31, which yielded
124 urine samples.
1

H NMR Spectra Acquisition

Monodimensional spectra of plasma samples were obtained

by collecting 64 scans for each spectrum on a Bruker Avance
400 spectrometer (Bruker Spectrospin, Rheinstetten, Germany)
operating at 9.4 T at 298 K using a 90u detection pulse and
acquiring the FIDs in 64K points; spectral width was 6000 Hz.
Presaturation was used to suppress the solvent signal, and the
relaxation delay was set at 15 seconds to avoid relaxation
effects. TOCSY and HSQC bidimensional spectra were
performed on a few samples to allow signal assignment. Urine
spectra were performed on a Bruker Avance 700 spectrometer
(16.45 T) under the same conditions as those used for plasma
spectra; TOCSY and HSQC bidimensional spectra were
performed as well.

JOURNAL OF THE AMERICAN COLLEGE OF NUTRITION

H NMR Data Analysis

1

H NMR spectra were analyzed using 1H-Manager version

11.0 software (Advanced Chemistry Development, Inc.,
Toronto, Canada). An exponential window function with a
line-broadening factor of 0.18 Hz was applied to the FIDs
before being Fourier transformed. Each of the spectra thus
obtained was then manually phase and baseline corrected. The
reference signal from the TSP methyl resonance was adjusted
to 0.00 ppm. Bidimensional spectra were analyzed using
Bruker Topspin version 1.3 software.
Areas from the most outstanding NMR signals were
measured either by direct integration or by deconvolution
with Lorentzian line shapes (tolerance 1027).
Data from urine spectra were normalized to the methyl
peak integral of creatinine at 3.05 ppm to correct for
concentration effects. To minimize biological variability, for
each subject and for each measured variable, the plasma data
set was generated by subtracting the after-exercise values from
the after-rehydration values; likewise, the urine data set was
generated by subtracting the values at rest from the afterrehydration values.

Data Analysis Protocol

To analyze the metabolic variations induced in the athletes
by the consumption of the sports drink with reference to the
supply of oligomineral water, an approach analogous to that of
van Velzen et al. [17] was followed. In particular, for each of
the 2 NMR data sets (plasma and urine samples), the 2 main
contributions to the overall variability observed in the data set
(intersubject and intrasubject variation) were identified by
means of a multilevel analysis, and further data processing was
carried out separately on each of the 2 subsets.
Structure of the NMR Data Set. In both cases, the data set
consisted of I subjects on which J NMR-integrated intensities
were measured under K different occasions. While the
numbers of variables and subjects differed in the 2 cases (40
variables measured on 24 individuals for plasma and 51
variables measured on 31 individuals for urine), the number of
intervention occasions was the same in both studies: 2 (supply
of oligomineral water and of the sports drink). Accordingly,
the total number of NMR spectra in the data set is given by M
(I 3 K). Statistical analysis was performed in a Matlab R2008a
environment (The MathWorks, Inc., Natick, MA); in particular, multilevel data analysis was carried out using the routines
written by the BDA group of the University of Amsterdam
(freely downloadable at http://www.bdagroup.nl/index.php/
downloads/software), while all other routines were written in
house.
Biological versus Nutritional Variation. The variation
observed in the data is due to at least 2 main factors: the

555

Metabolomics and Sports Drink Effects

biological variation between the subjects involved in the study
and the nutritional variation induced by supplying the athletes
with either oligomineral water or the investigated sports drink.
Therefore, when performing an overall analysis of this data set,
it is quite likely that, since both sources of variation are
contributing factors, the resulting model is hard to interpret. To
overcome this problem, in this study we followed the scheme
proposed by van Velzen et al. [17], which takes a multilevel
splitting approach to partition the overall variance into the
contributions of the different terms involved. In particular, the
NMR data matrix X is first split into the sum of an offset
contribution (Xoff) and of a between-subject term (biological
variation, Xb) and a within-subject term (nutritional variation,
Xw), according to the equation
X~Xoff zXb zXw :

The decomposition of the matrix into these 3 terms is carried

out in 2 consecutive centering steps: the first centering step is
applied to the entire data set X and results in a mean-centered
data block (Xc) and an offset term (Xoff 5 1MxmT, where 1M is
a column vector of ones having dimension M and xm is the
overall mean vector).
Subsequently, a second centering step is performed on the
mean-centered data (Xc): in this case, mean centering is
applied per subject i over K intervention occasions. For each
subject, this results in splitting the corresponding block of the
centered matrix into a centered mean (xbi) and a net variation
around this Xwi:
Xci ~1K xTbi zXwi :

Here, Xci (K 3 J) contains the mean-centered data for subject i

and is part of matrix Xc; 1K (K 3 1) that contains ones; xbiT (1
3 J) contains the mean values of each column in Xci, and Xwi
(K 3 J) contains the mean-centered data per subject i. These
data are then concatenated for each individual into the
corresponding matrices Xb and Xw, as described by equation 1.
One of the main advantages of this approach is that the 3
sources of variation identified in equation 1 are orthogonal to
each other, so it is possible to calculate the magnitude of the
different sources of variation based on the hypotheses of the
analysis of variance (ANOVA):
kXk2 ~kXoff k2 zkXb k2 zkXw k2 :

main goal was to study the effects of the consumption of a

sports drink compared with water, a classification analysis
based on PLS-DA on the within-individual variability was
used for this purpose.
In a multivariate regression method such as PLS-DA, the
response variable y (class labels) is used to guide the
projections in meaningful directions and to provide informa w and y according to y 5
tion on the relationship between X
Xwb + e, where b (J 3 1) is the regression coefficient vector, e
(M 3 1) contains the y model residuals, and y (M 3 1) contains
the class labels, which is equal for all subjects at intervention
occasion k. The y values of the subjects who were given water
only (k1) are assigned to class 0, while the sports drink group
(k2) is assigned to class +1.

RESULTS
1

H NMR Spectra
1

H NMR monodimensional spectra of human biofluids such

as plasma and urine are composed of hundreds of signals
relative to all the hydrogen atoms belonging to any metabolite
in solution. Therefore, the spectral regions relative to common
chemical groups are characterized by a high degree of
superimposition of signals relative to specific groups from
different molecules having similar chemical structure. Disambiguation of such superimposed regions is hard, and only a
small portion of the overall signals can be safely assigned
either by correlations from 2D-NMR spectra or comparison
with literature and NMR spectra databases [1921], or by
adding the candidate pure compound to the solution. In some
cases, although the signal assignment is definite, results have
to be presented as the sum of different metabolites because the
same signals, which are clearly separated in some spectra,
collapsed in some others, and the relative contributions cannot
be separated even by deconvolution. The relative standard
deviation expressed as interday CV% of the NMR signal
integral was less than 3% for signals relative to metabolites at
1023 M concentration and less than 5% for signals relative to
metabolites at 1025 M concentration.
Table 2 reports the signals that were taken into account in
the present study for plasma and urine, respectively.

Plasma Samples
Multivariate Data Analysis on the Different Subsets.
After splitting the data set into these blocks, each block was
analyzed separately to explore the metabolic sources of the
systematic variations observed. In particular, principal component analysis on the between-subjects data block was used to
identify the metabolites that are mainly responsible for the
difference among individuals. On the other hand, since our

556

According to equation 1, the total variation present in the

plasma data set can be split into 3 contributions: a general
offset term, which corresponds to the overall mean (Xm), and a
between- (Xb) and a within-individual (Xw) variation term. On
this basis, it is possible to estimate the relative contribution of
each of these terms by an ANOVA-like analysis: by inspecting
the sum of squares of the elements of these matrices, it is

VOL. 28, NO. 5

Metabolomics and Sports Drink Effects

observed that the offset term accounts for 35.51% of the total
variation, the remaining 64.49% being due to interindividual
and intraindividual sources. In particular, 28.00% of the total
variation can be ascribed to differences among individuals,
while the remaining 36.49% can be ascribed to the differences
within each individual induced by the ingestion of either water
or beverage. This means that more than 40% of the relevant
variation is due to differences among individuals: this
consideration supported our choice to use a multilevel
approach to facilitate interpretation of the results.
Analysis of Within-Individual Variation. Once the
original data matrix was partitioned as described in the
previous subsection, the 2 biologically relevant matrices Xb
and Xw were analyzed separately.
Analysis of the within-subject data block Xw represents the
core part of our statistical analysis, since this block accounts
for metabolic variations due to the supply of either water or the
sports drink. To investigate the effect of the different
treatment, a PLS-DA classification model was applied to this
matrix after suitable scaling. In particular, as suggested in the
literature, Pareto scaling, which uses the square root of the
standard deviation as the scaling factor, was chosen [17]: this
kind of scaling allows the difference in the signal strengths to
be compensated while partially retaining the original spectral
structure.
A 6-fold cross-validation procedure was used to determine
the optimal complexity of the PLS-DA classification model
and to estimate its predictive ability. The best model was
chosen as the one including 2 latent variables (74.53% Xvariance, 60.76% Y-variance), which resulted in the same 2
observations per class being misclassified both in the modeling
and the cross-validation stages. These misclassified blood
samples belong to subjects 5 and 36 and are those that fall very
close to the discrimination boundary, as shown in Fig. 1.
To determine which metabolites were mainly responsible
for the separation between water and drink, and to gain some
insight into the metabolic effect of the drink itself, variable
importance for prediction (VIP) scores and the PLS regression
coefficients were analyzed.
In greater detail, the VIP of a predictor is a value that
expresses the contribution of the individual variable in the
definition of the F-latent vector model. In particular, it is
defined by the formula
v
u
2
PF  2 T  
u
c t tk wjk kwk k
,
4
VIPj ~tNvars k~1 PkF k  2 T 
k~1 ck tk tk
where ck is the coefficient of the PLS inner relation
corresponding to the kth LV, tk is the kth score vector, Nvars
is the number of experimental variables, and wk is the weight
vector for the kth PLS latent variable, whose elements are wjk.

JOURNAL OF THE AMERICAN COLLEGE OF NUTRITION

Since the average of squared VIP scores equals 1, the

greater than 1 rule is generally used as a criterion to identify
the most significant variables.
As far as the VIP scores are concerned (Fig. 2), of the 40
variables used as predictors in the analysis, only 7 contribute
significantly to the classification model (namely, the 4 peaks of
glucose, 1 of the 2 peaks of lactate and citrate, and the peak
resulting from the sum of 2-oxoglutarate, lysine, and creatine).
While the VIP score is always a positive value, inspection of
the regression coefficients on the other hand provides us with
an estimate of the direction of the difference in the metabolite
concentration changes when the athletes are given water or the
sports drink: a positive coefficient means that the level change
of that metabolite is higher when the athletes drink the sports
beverage than when they drink water, while a negative
coefficient will imply the opposite (Table 3). Athletes
ingesting the sports beverage showed positive delta (recovery-exercise values) of plasma glucose and citrate levels
120 minutes after the end of exercise (+0.18 6 0.38 mM and
+0.03 6 0.05 mM, respectively), whereas athletes ingesting only
water displayed negative delta values of these metabolite levels
(20.16 6 0.30 mM and 20.03 6 0.04 mM, respectively).
Conversely, 120 minutes after exercise, negative delta values of
plasma lactate were observed in both treatment groups, indicating
a recovery trend toward basal lactate levels after the increase
occurring during exercise (20.92 6 1.41 mM and 20.76 6
0.78 mM in athletes ingesting sports beverage or oligomineral
water, respectively). This means that consumption of the sports
drink results in increased glucose, citrate, and 2-oxoglutarate+
lysine+creatine plasma levels and in a more pronounced lowering
of the lactate levels compared with water ingestion.
Analysis of Between-Individual Variation. The contribution of the difference among individuals to the overall
observed variation was further studied by computing a PCA
model of the Xb matrix from the plasma data set, after Pareto
scaling. More than 60% of the explained variance was
accounted for by the first principal component, which, as can
easily be seen from the loadings plot in Fig. 3A, is almost
exclusively correlated with the lactate variation.
Comparison between Rest and Recovery Plasma Data.
The comparison between rest and recovery plasma data
obtained from athletes who had ingested sports drink or
oligomineral water showed significant increases in glucose (p
, 0.05) and citrate levels (p , 0.001) as evaluated by paired
Students t test. However, the increase in citrate levels in sports
drinkrehydrated athletes reflects the contribution of the citrate
content in the ingested sports beverage (about 0.25 g/dL).

Urine Samples
A similar data analysis protocol was followed in the case of
the urine data set: again, the total variation was split into the

557

Metabolomics and Sports Drink Effects

Table 2. Assignment of Signals Detected by 1H NMR Spectroscopy of Plasma (pH 7) and Urine (pH 2.5)
Abbreviation

d (ppm)

Multiplicity

0.83
0.88
1.00
1.03
1.07
1.13
1.16
1.21
1.27
1.34
1.46
1.50
1.92
2.13
2.23
2.28
2.38
2.372.44
2.522.56
2.993.06
3.05
3.20
3.23
3.24
3.27
3.28
3.38
3.42
3.55
3.97
4.06
4.11
5.24
6.85
7.01
7.16
7.277.45
7.69
7.72
8.46

d
t
d
d
d
d
d
d
d
d
d
d
s
s
s
s
s
m
dd
m
s
s
s
m
s
s
s
m
m
s
s
q
d
d
s
d
m
s
s
s

CH3 2-Hydroxyvalerate
CH3 2-Hydroxybutyrate
cCH3 Isoleucine
CH3 Valine
CH3 3-Hydroxyisobutyrate
CH3 Isobutyrate

0.54
0.94
1.03
1.16
1.26
1.33
1.34
1.41
1.44
1.52
1.57
2.08
2.22
2.33
2.34
2.38
2.72
2.92

m
m
dd
d
d
s
d
d
s
d
s
s
s
s
t
s
t
dd

CH3 Biliaric acids

cCH3 Isoleucine + CH3 Leucine
CH3 Valine
CH3 2-Methyl,3-hydroxybutyrate
CH3 3-Hydroxybutyrate
CH3 3-Hydroxyisovalerate
CH3 Threonine
CH3 Lactate
CH3 2-Hydroxyisobutyrate
CH3 Alanine

Metabolite

Plasma
2-OH-Vale
2-OH-But
Ileu
Val
(R)3-OH-IsoBut
IsoBut
U1
(R)3-OH-But
Thr
Lac1
Ala
U2
Ac
Met
Acetone
AcAc
Pyr
Gln+Suc
Cit
2-OxoGlut+Lys+Cr
Crn1
Cho
U3
Glc1
TMAO
U4
U5
Glc2
Glc3
U6
Crn2
Lac2
Glc4
Tyr1
His1
Tyr2
Phe
1-MetHis
His2
For
Urine
Biliaric acids
Ile+Leu
Val
2Met3OHBut
3OHBut
3OHIsovale
Thr
Lac
2OHIsoBut
Ala
U1
Ac
Acetone
AcAc
N-AcGln
Pyr
DMA
Cit

558

CH3
CH3
CH3
CH3

3-Hydroxybutyrate
Threonine
Lactate
Alanine

CH3 Acetate
CH3 Methionine
CH3 Acetone
CH3 Acetoacetate
CH3 Pyruvate
cCH2 Glutamine + CH2 Succinate
CH2 Citrate
aCH2 2-Oxoglutarate, eCH2 Lysine, CH3 Creatine
CH3 Creatinine
CH3 Choline
H2 Glucose
CH3 Trimethylamine oxide

H5,7 Glucose
H3 Glucose
CH2 Creatinine
CH Lactate
H11 Glucose
H2,6 Tyrosine
H5 Hystidine
H3,5 Tyrosine
H2,3,4,5,6 Phenylalanine
H4 1-Metylhistidine
H2 Histidine
H Formate

CH3
CH3
CH3
CH3
CH3
CH3
CH2

Acetic acid
Acetone
Acetoacetate
N-acetylglutamine
Pyruvate
Dimethylamine
Citrate

VOL. 28, NO. 5

Metabolomics and Sports Drink Effects

Table 2. Continued
Abbreviation
Cr
Crn1
U2
U3
Caff1
Ins
Methanol
Caff2
TMAO
1-MetHis1+U4
3-MetHis1+U5
Hipp1
U6
Crn2
Trig1
Uracil
U7
Gua
U8
U9
Phe+His
Hipp2
Hipp3
Xant
Hypoxant1
Oxypur
U10
Hypoxant2
3-MetHis2
His
1-MetHis2
U11
Trig2
Trig3
U12

d (ppm)

Multiplicity

3.11
3.13
3.19
3.23
3.32
3.34
3.35
3.50
3.54
3.86
3.88
4.18
4.2
4.29
4.44
5.79
5.84
5.92
6.36
7.037.18
7.297.44
7.55
7.64
8.00
8.24
8.25
8.32
8.49
8.63
8.68
8.69
8.82
8.88
9.20
9.27

s
s
s
s
s
s
s
s
s
s
s
d
s
s
s
d
d
m
s
m
m
pseudo-t
pseudo-t
s
s
s
d
s
s
s
s
m
t
s
m

Metabolite
CH3 Creatine
CH3 Creatinine

CH3
CH2
CH3
CH3
CH3
CH3
CH3
CH2

(C12)caffeine
scyllo-inositol
Methanol
(C14)caffeine
Trimethylamine oxide
1-Metylhistidine, U4
3-Metylhistidine, U5
Hippurate

CH2 Creatinine
CH3 Trigonelline
H5 Uracil
H2 Guanosine

H2,3,4,5,6 Phenilalanine, H5 Histidine

H3,5 Hippurate
H4 Hippurate
H2 Xanthine
H2 Hypoxanthine
H9 Oxypurinol
H7
H2
H2
H4

Hypoxanthine
3-Methylhistidine
Histidine
1-Methylhistidine

H5,3 Trigonelline
H1 Trigonelline

U 5 unassigned. Multiplicity abbreviations: s 5 singlet, d 5 doublet, t 5 triplet, dd 5 double doublet, q 5 quadruplet, m 5 multiplet.

contributions of the offset, the between- and the withinindividual terms. The contribution of each of the terms to the
total variation is comparable to that observed in the plasma
data set. Indeed, the offset term contributes to 28.25% of the
total variation, with the remaining 71.75% being due to
intraindividual and interindividual factors that contribute
almost to the same extent (36.98% for the between-individual
and 34.77% for the within-individual variability).

Fig. 1. Cross-validated predictions from PLS-DA analysis of the

within-individual plasma subset. The classification threshold is
indicated by a full line. Subjects 5 and 36 are represented by semifilled
top and semifilled bottom symbols, respectively.

JOURNAL OF THE AMERICAN COLLEGE OF NUTRITION

Analysis of Within-Individual Variation. Also in the case

of urine, our attention was mainly focused on the intraindividual source of variation, which was used to ascertain which
changes in the levels of urine metabolites were induced by the
consumption of the sports drink with respect to oligomineral
water supply. To investigate the effect of the different
treatment, a PLS-DA classification model was applied to this
matrix after Pareto scaling.
A 7-fold cross-validation procedure was used to determine
the optimal complexity of the PLS-DA classification model

559

Metabolomics and Sports Drink Effects

Fig. 2. Variable importance for prediction (VIP) score of variables from the PLS-DA analysis of plasma within-individual data subset.

and to estimate its predictive ability. The best model was

ascertained as the one including 10 latent variables (95.93% Xvariance, 91.03%Y-variance), which resulted in no observation
being misclassified in the modeling and/or in the crossvalidation stages. This situation is evidenced in Fig. 4, where
the cross-validated predictions from the model are shown. To
investigate which metabolites were mainly responsible for the
separation between water and drink, and to gain insight into the

Table 3. Signs of the Regression Coefficients for the Most

Relevant Variables as Evaluated by Variable Importance for
Prediction Score
Plasma
Lactate
Citrate*

Q
q

2-OxoGlut+Lys+Cr*
Glucose*

q
q

Q
q
q
q
q
Q

Caffeine (Thein)*
1-MetHis+U4
Hipp1
U6
U9*
Phe+His

q
q
q
Q
Q
q

Urine
Valine
3-Hydroxybutyrate
Lactate
Acetate
Acetone*
U3

* Metabolites whose delta values show a significant difference also when

compared by paired Students t test between the sports beverage and the
oligomineral water groups.

560

metabolic effect of the drink itself, VIP scores and the PLS
regression coefficients were also analyzed in this case. With
respect to plasma, a larger number of variables show a
significant value of the VIP score, with the most outstanding
being U9, lactate, caffeine, acetone, 1-methylhistidine, 3hydroxybutyrate, acetate, U3, hippuric acid, U6, and
phenylalanine+histidine (Fig. 5). Inspection of the directions
of changes indicated by the regression coefficients and
reported in Table 3 further facilitates the interpretation of the
results of PLS-DA analysis by indicating the metabolites
whose levels increase and those that decrease after ingestion
of the sports drink. With reference to the differences between
postexercise recovery and the rest condition, caffeine (theine)
and hippurate in urine showed positive delta values (+0.5 6
5.6 and +10.5 6 9.6 mmol/mmol creatinine, respectively) in
athletes who ingested the sports beverage and negative delta
values in athletes ingesting only water (24.7 6 9.9 and
213.8 6 7.7 mmol/mmol creatinine, respectively), although
only caffeine delta values were significantly different (p 5
0.024 by paired Students t test). A significant variation in
delta values was also displayed by acetone, which increased
in athletes who ingested the sports beverage (2.6 6 2.2 and
1.3 6 1.6 mmol/mmol creatinine in the sports beverage and
water group, respectively, p 5 0.013, by paired Students t
test).

VOL. 28, NO. 5

Metabolomics and Sports Drink Effects

Fig. 4. Cross-validated predictions from PLS-DA analysis of the

within-individual urine subset.

Effect of the Training Set Size on Model Accuracy

Fig. 3. Loadings of the first principal component of the betweenindividuals subsets from plasma (A) and urine (B).

PLS-DA analysis included also delta values of 3-hydroxybutyrate, acetate, valine, and phenylalanine+histidine as
variables discriminating between the sports beverage and the
water group, although they did not show significant differences
in the paired Students t test.
Analysis of Between-Individual Variation. As in the
analysis of plasma data, also in the case of the urine data set
the sources of variation in metabolite concentration due to
differences among individuals was taken into account by
performing a PCA analysis on the between-subjects block Xb,
after suitable scaling. Pareto scaling was also used in this case
for the reasons already given. As observed in the case of the
plasma samples, a large proportion of the between-individuals
variability is explained by the first principal component. A
careful inspection of the loading plot in Fig. 3B confirms what
already emerges from an analysis of the plasma data set, that
is, that lactate (the only variable having a significant loading
on PC1) is the metabolite whose concentration varies the most
among the subjects.

JOURNAL OF THE AMERICAN COLLEGE OF NUTRITION

Lastly, the effect of training set size on the accuracy of the

prediction was assessed by repeating the statistical analyses on
randomly selected subsets of progressively decreasing dimensionality. In particular, for each total number of athletes, the
analysis was repeated 20 times to account for subset to subset
variability, and a minimum of 8 athletes were used to build the
smallest training set. The accuracy of the models thus
constructed was comparable with the one built on the whole
set of results up to a numerosity of 14 athletes, after which
performance started to decrease significantly. Therefore, even
though each data set should be considered separately, this
would suggest that the minimum requirement in terms of
training set dimension to obtain accurate results is about 14
individuals per treatment.

DISCUSSION
In the past few years, nutritional metabolomics has been
shown to be a successful approach to the investigation of the
difference between the effects of different diets on human
health, deepening our knowledge of the interacting and
regulatory roles of nutrition [15,16]. More recently, metabolomics has been also applied in sports medicine studies to
investigate the physiological alterations of the metabolic
profile induced by strenuous physical exercise [22] and longterm training in athletes [23].
In this study, NMR-based metabolomics analysis has been
applied to the investigation of the nutritional impact of a green
teabased isotonic sports beverage on the metabolism of
Olympic team rowers during recovery after strenuous physical
exercise. To compare the effect of the sports drink at the
individual level with respect to oligomineral water as a placebo
control, we performed a double-blind crossover-designed study
also entailing the use of a specific multivariate analysis,

561

Metabolomics and Sports Drink Effects

Fig. 5. VIP score of variables from the PLS-DA analysis of urine within-individual data subset.

MSCA/PLS-DA, to split the between-subject variation from

the within-subject variation. In crossover-designed metabolomics experiments, the use of multilevel data analysis is to be
preferred over classical megavariate methods, such as PCA or
PLS-DA, since in the latter methods, the interindividual
variance acts as a source of variation that can overwhelm the
intraindividual variation [17]. The between-subject variations,
as evaluated by MSCA, agreed with metabolic variations
reflecting the individual response to physical exercise and were
mainly represented by lactate level changes. A reduction of the
within-subject variation was further enhanced by expressing
the values as delta between recovery and exercise data for
plasma and between recovery and rest data for urine. The
MSCA analysis of within-subject variations allowed us to
highlight the metabolic changes induced by water or beverage
ingestion during postexercise at individual level.
The multilevel PLS-DA models of plasma and urine
metabolic profiles supported an effect of the green teabased
sports drink on energy metabolism and on glucose homeostasis
after exercise. Indeed, the most important energy metabolism
linked variables in the discrimination were represented by
signals from glucose, citrate, lactate, and 2-oxoglutarate+lysine+creatine in plasma and from acetone, 3-hydroxybutyrate,
acetate, and lactate in urine. According to the NMR results
obtained, the postrehydration levels of blood glucose and
insulin were significantly higher in the athletes drinking the

562

green teabased sports beverage with respect to oligomineral

water, as independently measured by routine tests on the same
samples. In particular, the glucose plasma levels were 88.8 6
10.6 and 81.9 6 8.0 mg/dL (p , 0.05), and the insulin levels
were 10.4 6 4.7 and 6.5 6 1.8 mU/ml ( p , 0.001) after intake
of beverage and of water, respectively.
The subtle physiological systemic changes observed in
energy metabolism, both in plasma and urine, can be
accounted for by the biochemical composition of the green
teabased carbohydrate/hydroelectrolyte drink.
In athletes rehydrated by drinking the green teabased
carbohydrate/hydroelectrolyte beverage, positive delta values
of plasma glucose were observed. Conversely, in rowers who
drank water, decreased glucose plasma levels after recovery
with respect to exercise were evident (negative recoveryexercise delta values). The significant increase in glucose
plasma levels observed after sports drink rehydration probably
can be explained in terms of the fructose content of the sports
beverage as well as in terms of fructose transformation to
glucose in the liver. The absence of NMR-detectable fructose
signals in the sports drinkrehydrated group supported a liver
metabolism for this substrate during the 120-minute postexercise period. Our results seem to be in agreement with the
argument favoring a poststrenuous exercise ingestion of
fructose with respect to glucose, considering fructose as a
readily available energy source that does not stimulate insulin

VOL. 28, NO. 5

Metabolomics and Sports Drink Effects

release and consequently does not inhibit fatty acid mobilization and beta-oxidation [3]. Indeed, plasma insulin levels in
sports beveragerehydrated athletes were below the lipolysis
inhibiting concentration (13 mU/ml), as reported by Borer [24].
The urine acetone, 3-hydroxybutyrate, and acetate levels were
higher than in water-rehydrated athletes, suggesting a ketone
body production maintained by fat oxidation when the sports
drink was ingested.
Ketone formation takes place mostly in the liver and, to a
lesser extent, in the kidneys, using plasma free fatty acids as
the substrate. Most liver-produced ketone bodies are released
into the circulation to serve as energy substrate for other
tissues, mainly muscle and brain. The observed significant
increase in acetone alone and not in acetoacetate levels could
be explained by a slow nonenzymatic spontaneous decarboxylation of acetoacetate occurring in urine. The increased
acetate levels may also be related to an increased liver fatty
acid oxidation, when lipogenesis is inhibited, in agreement
with Folch et al., who reported a sustained high fat oxidation
and a totally suppressed de novo lipogenesis after exercise
[25].
On this basis, the discriminant metabolic variables, as
identified by PLS-DA in both plasma and urine, may be
regarded as evidence of a complex effect that could be
interpreted on the basis of a stimulation of liver glucose
production by fructose, of the ketone body liver production by
fat oxidation depending on the green teabased beverage
ingestion, and of the peculiar postexercise metabolic state.
In our poststrenuous exercise experimental conditions, the
energy metabolism appeared to be still dependent on fatty acid
oxidation, as shown by the increase in urine ketone body levels
and by the decrease in plasma lactate concentration, and this
was also in agreement with an effect mediated by green tea
extracts as previously observed in humans [2629]. It has been
reported that green tea extract might stimulate fat oxidation
through the inhibition of the noradrenaline degrading enzyme
catechol O-methyltransferase [30]. A reduction in noradrenaline degradation could potentially prolong adrenergic drive and
increase lipolysis, even if the mechanism of the green tea
extractdependent stimulation of fatty acid oxidation may be
more complex and not yet completely clarified.
Interestingly, a decrease in valine levels and an increase in
aromatic amino acids, namely, phenylalanine, histidine, and
methyl-histidine, were associated with an increase in acetone
and 3-hydroxybutyrate levels in the urine of sports drink
rehydrated rowers. Although these variations had very low
intensity, the covariance of these metabolites suggests a more
sustained branched amino acid muscle oxidative metabolism in
the athletes during recovery from exercise.
The actual absorption of green tea components on
rehydration with the sports drink was demonstrated by a slight
increase in NMR signals relative to caffeine and hippuric acid

JOURNAL OF THE AMERICAN COLLEGE OF NUTRITION

in the urine of sports drinkrehydrated rowers, in agreement

with previous results [29,31]. Hippuric acid is a product of
dietary flavonoids. The majority of the polyphenols are not
absorbed in the small intestine but metabolized by the gut
microflora and absorbed by the colon before being excreted
into the urine [32].
Consequently, the metabolic effects observed in our
experimental set can be mainly ascribed to the green tea
extracts contained in the sports drink, even if the molecular
mechanism by which tea polyphenols affect physiological
changes is not yet completely understood.
Numerous metabolomics studies have demonstrated the
efficacy of single nutritional supplements, but the synergistic
effects at a systemic level due to the administration of different
formulations of micronutrients and macronutrients have been
poorly investigated. During the revision of our article, an
approach similar to ours was applied to evaluate the metabolic
response to the ingestion of different beverage complex
formulations during the recovery phase after strenuous
physical exercise by using gas-chromatography/time-of-flight
mass spectrometry and multivariate analysis of data from
human blood serum [33].

CONCLUSION
This study represents an original demonstration that NMRbased metabolomics allows the complex metabolic effects of a
green tea extractbased carbohydrate/hydroelectrolyte beverage to be evaluated in athletes during the recovery phase after
physical exercise. In particular, the metabolomic approach in
which MSCA/PLS-DA is used in a double-blind cross-over
designed study allowed us to investigate the effect of a
nutritional intervention on energy metabolism at the individual
level, although the observed changes were smaller than the
biological variations between individuals.

ACKNOWLEDGMENTS
We are grateful to the Italian Rowing Federation and COOP
ITALIA for their financial support, as well as to the Institute of
Sports Medicine and Science, Italian National Olympic
Committee (Rome, Italy), for logistic support and subsidiary
laboratory analyses.

REFERENCES
1. Maughan RJ: Fluid and electrolyte loss and replacement in
exercise. In Harries M, Williams C, Stanish WD, Micheli LL
(eds): Oxford Textbook of Sports Medicine. Oxford, UK:
Oxford University Press, pp 8293, 1994.

563

Metabolomics and Sports Drink Effects

2. Kay D, Marino FE: Fluid ingestion and exercise hyperthermia:
implications for performance, thermoregulation, metabolism and
the development of fatigue. J Sports Sci 18:7182, 2000.
3. Maughan RJ: The sports drink as a functional food: formulations
for successful performance. Proc Nutr Society 57:1523, 1998.
4. Gisolfi CV, Duchman SM: Guidelines for optimal replacement
beverages for different athletic events. Med Sci Sports Exerc
24:679687, 1992.
5. Coyle EF: Fuels for sport performance. In Lamb DR, Murray R
(eds): Perspectives in Exercise Science and Sports Medicine
Optimising Sport Performance, vol. 10. Carmel, IN: Benchmark
Press, pp 95138, 1997.
6. Below P, Mora-Rodriguez R, Gonzalez-Alonso J, Coyle EF: Fluid
and carbohydrate ingestion independently improve performance
during 1h of intense cycling. Med Sci Sports Exerc 27:200210, 1995.
7. Lambert GP, Chang RR, Xia T, Summers RW, Gisolfi CV:
Absorption from different intestinal segments during exercise. J
Appl Physiol 83:204212, 1997.
8. Hill RJ, Bluck LJ, Davies PS: The hydration ability of three
commercially available sports drinks and water. J Sci Med Sport
11:116123, 2008.
9. American College of Sports Medicine: Position stand on
prevention of thermal injuries during distance running. Med Sci
Sports Exercise 16:ixxiv, 1984.
10. Ryan M: Sports drinks: research asks for reevaluation of current
recommendations. J Am Diet Assoc (suppl. 2):S197S198, 1997.
11. Bollard ME, Stanley EG, Lindon JC, Nicholson JK, Holmes E:
NMR-based metabonomic approaches for evaluating physiological
influences on biofluid composition. NMR Biomed 18:143162,
2005.
12. 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 29:11811189, 1999.
13. Nicholson JK, Connelly J, Lindon JC, Holmes E: Metabonomics: a
platform for studying drug toxicity and gene function. Nat Rev
Drud Discover 1:153161, 2002.
14. Lenz EM, Bright J, Wilson ID, Hughes A, Morrisson J, Lindberg
H, Lockton A: Metabonomics, dietary influences and cultural
differences: a 1H NMR-based study of urine samples obtained
from healthy British and Swedish subjects. J Pharm Biomed Anal
36:841849, 2004.
15. Gibney MJ, Walsh M, Brennan L, Roche HM, German B, van
Ommen B: Metabolomics in human nutrition: opportunities and
challenges. Am J Clin Nutr 82:497503, 2005.
16. Rezzi S, Ramadan Z, Fay LB, Kochhar S: Nutritional metabonomics:
applications and perspectives. J Proteome Res 6:513525, 2007.
17. van Velzen EJ, Westerhuis JA, van Duynhoven JP, van Dorsten
FA, Hoefsloot HC, Jacobs DM, Smit S, Draijer R, Kroner CI,
Smilde AK: Multilevel data analysis of a crossover designed
human nutritional intervention study. J Proteome Res 7:4483
4491, 2008.
18. Tiziani S, Emwas AH, Lodi A, Ludwig C, Bunce CM, Viant MR,
Gunther UL: Optimized metabolite extraction from blood serum
for 1H nuclear magnetic resonance spectroscopy. Anal Biochem
377:1623, 2008.
19. Lindon JC, Nicholson JK, Everett JR: NMR spectroscopy of
biofluids. Ann Rep Nucl Magn Reson Spectrosc 38:188, 1999.

564

20. Engelke U, Moolenaar S, Hoenderop S, Morava E, van der Graaf

M, Heerschap A, Wevers R: Handbook of 1H-NMR Spectroscopy in Inborn Errors of Metabolism: Body Fluid NMR Spectrum
and in Vivo MR Spectroscopy. Heilbronn, Germany: SPS
Verlagsgesellschaft mbH, 2006.
21. Human Metabolome Database: Accessed at: www.hmdb.ca
22. Pohjanen E, Thysell E, Jonsson P, Eklund C, Silfver A, Carlsson
IB, Lundgren K, Moritz T, Svensson MB, Antti H: A multivariate
screening strategy for investigating metabolic effects of strenous
physical exercise in human serum. J Proteome Res 6:21132120,
2007.
23. Yan B, Jiye A, Wang G, Lu H, Huang X, Liu Y, Zha W, Hao H,
Zhang Y, Liu L, Gu S, Huang Q, Zhang Y, Sun J: Metabolomic
investigation into variation of endogenous metabolites in professional athletes subject to strength-endurance training. J Appl
Physiol 106:531538, 2008.
24. Borer KT (ed): Hormonal regulation of fuel use in exercise. In
Exercise Endocrinology.Champaign, IL: Human Kinetics, pp.
97120, 2003.
25. Folch N, Peronnet F, Massicotte D, Duclos M, Lavoie C, HillaireMarcel C: Metabolic response to small and large 13C-labelled pasta
meals following rest or exercise in man. Br J Nutr 85:671680, 2001.
26. Venables MC, Hulston CJ, Cox HR, Jeukendrup AE: Green tea
extract ingestion, fat oxidation, and glucose tolerance in healthy
humans. Am J Clin Nutr 87:778784, 2008.
27. Dulloo AG, Duret C, Rohrer D, Girardier L, Mensi N, Fathi M,
Chantre P, Vandermander J: Efficacy of a green tea extract rich in
catechin polyphenols and caffeine in increasing 24-h energy
expenditure and fat oxidation in humans. Am J Clin Nutr 70:1040
1045, 1999.
28. Ota N, Soga S, Shimotoyodome A, Haramizu S, Inaba M, Murase
T, Tokimitsu I: Effects of combination of regular exercise and tea
catechins intake on energy expenditure in humans. J Health Sci
51:233236, 2005.
29. Van Dorsten FA, Daykin CA, Mulder TPJ, Van Duynhoven JPM:
Metabonomics approach to determine metabolic differences
between green tea and black tea consumption. J Agr Food Chem
54:69296938, 2006.
30. Chen D, Wang CY, Lambert JD, Ai N, Welsh WJ, Yang CS:
Inhibition of human liver catechol-O-methyltransferase by tea
catechins and their metabolites: structure-activity relationship and
molecular-modeling studies. Biochem Pharmacol 69:15231531,
2005.
31. Daykin CA, Van Duynhoven JP, Groenewegen A, Dachtler M, Van
Amelsvoort JM, Mulder TP: Nuclear magnetic resonance spectroscopic based studies of the metabolism of black tea polyphenols in
humans. J Agr Food Chem 53:14281434, 2005.
32. Halliwell B, Rafter J, Jenner A: Health promotion by flavonoids,
tocopherols, tocotrienols, and other phenols: direct or indirect
effects? Antioxidant or not? Am J Clin Nutr 81:268S276S, 2005.
33. Chorell E, Moritz T, Branth S, Antti H, Svensson M: A predictive
metabolomics evaluation of nutrition-modulated metabolic stress
responses in human blood serum during the early recovery phase
of strenous physical exercise. J Proteome Res 8:29662977, 2009.

Received January 13, 2009; revision accepted June 29, 2009.

VOL. 28, NO. 5