代写 Pharmaco-metabonomic phenotyping and personalized drug
100%原创包过,高质代写&免费提供Turnitin报告--24小时客服QQ&微信：120591129
代写 Pharmaco-metabonomic phenotyping and personalized drug treatment
Pharmaco-metabonomic phenotyping and
personalized drug treatment
T. Andrew Clayton1, John C. Lindon1, Olivier Cloarec1, Henrik Antti2, Claude Charuel3, Gilles Hanton3,
Jean-Pierre Provost3, Jean-Loı¨c Le Net3, David Baker4, Rosalind J. Walley5, Jeremy R. Everett5
& Jeremy K. Nicholson1
There is a clear case for drug treatments to be selected according to
the characteristics of an individual patient, in order to improve
efficacy and reduce the number and severity of adverse drug
reactions1,2. However, such personalization of drug treatments
requires the ability to predict how different individuals will
respond to a particular drug/dose combination. After initial
optimism, there is increasing recognition of the limitations of
the pharmacogenomic approach, which does not take account of
important environmental influences on drug absorption, distribution,
metabolism and excretion3–5. For instance, a major factor
underlying inter-individual variation in drug effects is variation
in metabolic phenotype, which is influenced not only by genotype
but also by environmental factors such as nutritional status, the
gut microbiota, age, disease and the co- or pre-administration of
other drugs6,7. Thus, although genetic variation is clearly important,
it seems unlikely that personalized drug therapy will be
enabled for a wide range of major diseases using genomic knowledge
alone. Here we describe an alternative and conceptually new
‘pharmaco-metabonomic’ approach to personalizing drug
treatment, which uses a combination of pre-dose metabolite
profiling and chemometrics to model and predict the responses
of individual subjects.We provide proof-of-principle for this new
approach, which is sensitive to both genetic and environmental
influences, with a study of paracetamol (acetaminophen) administered
to rats. We show pre-dose prediction of an aspect of the
urinary drug metabolite profile and an association between predose
urinary composition and the extent of liver damage sustained
after paracetamol administration.
1H nuclear magnetic resonance (NMR) spectroscopy has been
widely applied as a metabolite profiling tool for metabonomic
studies, as it enables many endogenous metabolites to be quantified
rapidly and reproducibly without derivatization or separation8–11. In
one of many potential applications, NMR-based metabonomic
analysis of post-dose rodent biofluids has been developed as a
rapid and non-invasive means of assessing the toxicity of potential
drug compounds, and this has involved studying the effects of a
variety of model toxins12. In one such study, we administered
galactosamine hydrochloride (800 mg kg21) to a group of ten rats
and found the extent of the induced liver effects to be so variable that
the rats could be classified as either ‘responders’ or ‘non-responders’.
Searching for the cause of this variation, we performed principal
component analysis (PCA) on the NMR spectra of the relevant
pre-dose urine samples and observed some discrimination between
responder and non-responder groups in terms of their pre-dose
metabolite profiles (Fig. 1a). Although this result was not sufficient
to say that galactosamine responder/non-responder behaviour is
predictable, it suggested that information on individual responses
to xenobiotics might be contained in the metabolite patterns of
pre-dose biofluids. We thus conceived the possibility of ‘pharmacometabonomics’,
which we define as ‘the prediction of the outcome
(for example, efficacy or toxicity) of a drug or xenobiotic intervention
in an individual based on a mathematical model of pre-intervention
metabolite signatures’.
We also conducted a larger study on the severity of liver damage
induced in rats by allyl alcohol (50 mg kg21), and found a weak but
statistically significant association between the extent of the induced
damage and the pre-dose urinary data (Fig. 1b), although in this case
the discriminating factor was the total pre-dose excretion of organic
compounds as estimated by 1H NMR spectroscopy, rather than the
代写 Pharmaco-metabonomic phenotyping and personalized drug treatmentdual rat, with colour-coding
according to post-dose behaviour. b, A scores plot from PCA of urinary data
obtained before dosing male Sprague–Dawley rats with allyl alcohol
(50 mg kg21). Each point represents an individual rat, with colour-coding
according to the extent of post-dose liver damage. Red signifies greatest
damange, green signifies least damage, mid-damage group excluded.
c, Schematic of our pharmaco-metabonomic hypothesis.
1Biological Chemistry, Biomedical Sciences Division, Faculty of Natural Sciences, Sir Alexander Fleming Building, Imperial College London, South Kensington, London SW7 2AZ,
UK. 2Department of Chemistry, Umea° University, 901 87 Umea°, Sweden. 3Pfizer Global Research & Development, Centre de Recherche, 37401 Amboise Cedex, France. 4Pfizer
Inc., 2800 Plymouth Road, Ann Arbor, Michigan 48105, USA. 5Pfizer Global Research & Development, Ramsgate Road, Sandwich, Kent CT13 9NJ, UK.
Vol 440|20 April 2006|doi:10.1038/nature04648
1073
© 2006 Nature Publishing Group
diet, ethnicity and disease13–17, and recognizing that many of the
factors that influence the metabolism of drugs would normally
operate on endogenous and dietary substances, we decided to test
the hypothesis that the pre-dose metabolite profile of an individual
animal contains sufficient information to allow the prediction of
aspects of drug metabolism and toxicity in that animal without any
prior knowledge of the animal’s genomic profile (Fig. 1c). We chose
the commonly used analgesic paracetamol (acetaminophen) for this
investigation.
We collected pre- and post-dose urine samples from 65 rats given a
single toxic-threshold dose of paracetamol (600 mg kg21, a treatment
that resulted in no mortality or clinical signs), and analysed all of
these samples by 1H NMR spectroscopy (Fig. 2). The post-dose
spectra showed characteristic signals of both endogenous and paracetamol-
related metabolites, with the major paracetamol-related
metabolites being identified as paracetamol sulphate (S), paracetamol
glucuronide (G), the mercapturic acid (MA) derived from
paracetamol, and paracetamol (P) itself. Using these spectra in
conjunction with the known urinary volumes, we determined the
amounts and relative proportions of paracetamol-related metabolites
excreted by each animal, and modelled the variation in these data
in relation to the multivariate endogenous metabolite profiles
obtained from the pre-dose samples. Liver damage of variable
severity was identified and quantified by clinical chemistry and
histopathology in samples taken at approximately 24 h post-dose,
with the extent of the microscopically visible damage being scored in
each of five liver lobes and a mean histology score (MHS) derived for
each rat (see Supplementary Fig. 1 and Supplementary Tables). The
inter-animal variation in MHS was also modelled in relation to the
multivariate endogenous metabolite profiles obtained from the
pre-dose samples, in an attempt to integrate classical end-point
histopathology with a pre-dose metabonomic data set. We then
tested the predictive ability of the various models.
The mole ratio of paracetamol glucuronide to paracetamol (G/P)
was found to be the most convincingly predicted of the various postdose
metabolite quantities. Thus, we built and validated a PLS model
(projection to latent structure; see Methods) for predicting, from
their pre-dose metabolite profiles, the G/P values obtained post-dose
for individual animals (Fig. 3). The most important factors underlying
this model were a positive correlation (r ¼ 0.48) between G/P
and the integral of the d5.06–5.14 region of the pre-dose NMR
spectra, and negative correlations (r ¼ 20.56 and 20.54) between
G/P and the integrals of the d8.98–9.10 and d0.50–0.86 regions of the
pre-dose spectra (Fig. 3b). The integrals for the latter two regions
were correlated to one another (r ¼ 0.90), but showed no clear
relationship to the integral for the d5.06–5.14 region. Inspection of
the pre-dose spectra revealed some small but distinct signals in
the d5.06–5.14 region, with the d8.98–9.10 region and much of the
d0.50–0.86 region being relatively featureless but positive relative to
the zero line.
Our validation (Fig. 3c, d) confirms that the predictive model for
G/P is robust, and the positive correlation between G/P and the
d5.06–5.14 region of the pre-dose spectra is consistentwith metabolic
control. The glucuronide part of paracetamol glucuronide produces
a doublet in this region post-dose, and the signals in the d5.06–5.14
region of the pre-dose spectra may arise, correspondingly, from
endogenous ether glucuronides. It is logical that G/P would be
positively correlated to the pre-existing tendency of each individual
rat to form and excrete such ether glucuronides, and the integral for
the d5.06–5.14 region could be a good reflection of that tendency,
being relatively free of other signals. The significance of the negative
correlations between G/P and the specified regions of the pre-dose
spectra is not understood, but might reflect some dependency on
urinary protein.
Although we did not obtain a fully validated model for predicting
post-dose histology, our analysis did show a statistically significant
association between the nature of the pre-dose urinary biochemical
profile and the post-dose histological outcome. Having assigned each
animal to one of three histology classes (described in Table 1), PCA
was carried out on the pre-dose spectral data (62 animals), with
partial separation between histology classes 1 and 3 being found on
principal component 2 (PC2) (Fig. 4a). Furthermore, a weak but
statistically significant correlation (r ¼ 20.34; P ¼ 0.007) was found
between the PC2 scores and MHS (Fig. 4b). PCAwas also carried out
Figure 2 | Representative 1H NMR spectra. a, b, 1H NMR spectra of
pre-dose (a, 248 h to 224 h) and post-dose (b, 0 to þ24 h) urine samples
from a rat dosed with paracetamol (600 mg kg21). The inset in a is an
expansion of the d1.9–1.0 region, indicating the complexity of the
endogenous profile and the richness of the embedded information. 2-OG,
2-oxoglutarate; G, paracetamol glucuronide.
Figure 3 | Pre-dose prediction of the urinary mole ratio of paracetamol
glucuronide to paracetamol (G/P) obtained in paracetamol-dosed rats.
a, Observed versus predicted G/P values for a two-component PLS model in
which all predictions relate to model-building data. b, The twelve regions of
the pre-dose 1H NMR spectra most important to the above model. Each
of the 0.04 p.p.m.-wide spectral segments is identified by the chemical shift
at its mid-point. VIP, variable influence on projection. c, Internal
validation of the above model, showing clear decreases in performance as the
G/P data are permuted relative to the pre-dose data. R2 describes how
well the derived model fits the data, and is the proportion of the sum of
squares explained by the model. Q2 describes the predictive ability of the
derived model and is the cross-validated R2. d, The result of a seven-round
cross-validation exercise in which every point represents test data not used
in the model-building (regression line is represented by the equation:
observed value ¼ (0.89 £ predicted value) þ 0.29; root mean square error
of predictions 0.32).
LETTERS NATURE|Vol 440|20 April 2006
1074
© 2006 Nature Publishing Group
on the pre-dose data for classes 1 and 3 only (32 animals), with partial
separation of those classes again being observed on PC2 (Fig. 4c) and
the statistical significance of that separation being confirmed by a
Mann–Whitney U-test (P ¼ 0.002).
The main pre-dose factors underlying the histology class discrimination
are the levels of taurine, trimethylamine-N-oxide (TMAO)
and betaine (Fig. 4d). A higher pre-dose level of taurine is associated
more with class 1 than with class 3, whereas a higher combined
pre-dose level of TMAO and betaine is associated more with class 3
than with class 1. The beneficial relationship found between pre-dose
urinary taurine and the extent of paracetamol-induced liver damage
is consistent with earlier findings for the degree of liver damage
induced by a variety of other liver toxins, and with the protective
effect observed when taurine is administered to rats before or soon
after a hepatotoxic dose of paracetamol18,19. The significance of the
pre-dose level of taurine might lie in its known defensive properties20,
but urinary taurine might also reflect the availability of inorganic
sulphate (to which taurine is metabolically related), which is a
precursor of the paracetamol-sulphating agent phosphoadenosine
phosphosulfate (PAPS)21,22. The latter interpretation is consistent
with our observation that most of the animals with a higher degree of
liver necrosis (MHS . 2.5) showed a low proportion of paracetamol
sulphate in their post-dose urine. In contrast, a higher pre-dose level
of TMAO is associated with more paracetamol-induced liver
damage, suggesting that the gut bacteria might have a role in
determining the extent of such damage23,24—and the contribution
of the gut bacteria to a bayesian probabilistic (‘Pachinko’) model of
metabolism has already been postulated11. The occasional pre-dose
presence of a significant quantity of betaine, which produces anNMR
signal that overlaps that of TMAO, may be a confounding factor and
certainly contributed to the abnormal position of one of the class 1
animals on the scores plot shown in Fig. 4c.
Whatever the basis of the observed pre-dose discrimination, our
paracetamol study findings identify statistically significant relationships
between variation in the pre-dose data and the post-dose
variation in histopathology and in the urinary level of a drug
metabolite relative to its parent. Thus, the two apparently independent
models provide the first demonstration of the concept of
pharmaco-metabonomics, in which drug-induced responses in individuals
are potentially predictable from their pre-dose metabolite
profiles, which serve as biochemical signatures that report simultaneously
on multiple factors of relevance to drug metabolism and
drug effects.
Looking forward, we propose that this pharmaco-metabonomic
approach, which amounts to response-targeted pre-dose phenotyping,
might provide the basis of a future population-screening
tool for selecting individuals according to their suitability for treatment
with particular drugs, drug classes or drug doses. Furthermore,
this approach should have certain advantages over methods for drug
and dose selection that are dependent on the use of test compounds25
to characterize particular aspects of an individual’s metabolic
phenotype. In particular, the pharmaco-metabonomic
approach has the potential to provide automatic identification and
weighting of multiple response-determining factors. A further
inherent benefit of the pharmaco-metabonomic approach is the
identification of new biomarkers that are predictive of individual
responses.
The main potential application we envisage for pharmaco-metabonomics
is with respect to personalized human healthcare, and
there is clearly a need for further development and to investigate how
well the approach can be transferred from single-strain laboratory
animals to human subjects, for whom much greater genetic and
environmental variation would be expected. Recent developments in
data analysis should assist pharmaco-metabonomic modelling and
biomarker identification26,27. Other analytical techniques such as
LC–MS and GC–MS might also be used, with the likelihood of
detecting a large number of low-concentration metabolites not
normally accessible by NMR. Metabolite profiling of fluids other
than urine, such as blood and faecal extracts, should also provide
additional information. In practice, the success or otherwise of the
pharmaco-metabonomic approach would be expected to vary from
drug to drug, and would depend on the nature of the challenge posed
by each drug, on the response of interest and on the extent to which
the relevant response-controlling factors are reported in the pre-dose
data. However, in principle, by using this methodology, adverse drug
reactions could potentially be avoided and drugs and dose levels
could be targeted more effectively according to the metabolic and
other characteristics of each individual. We envisage that optimal
personalization of drug treatments may eventually involve a variety
of response-prediction approaches, including both pharmacometabonomics
and pharmacogenomics. However, pharmacometabonomics
has an important theoretical advantage over
pharmacogenomics in that it can potentially take account of both
genomic and environmental factors affecting drug-induced
responses. Furthermore, although pharmaco-metabonomics would
normally relate to predicting drug- or xenobiotic-induced responses,
we envisage that similar methodology could also be applied
Figure 4 | Pre-dose discrimination of the degree of liver damage obtained in
paracetamol-dosed rats. a, A scores plot from PCA of the pre-dose NMR
data. Each point represents a single rat and is colour-coded by its histology
class (with increasing severity of damage, class 1 is green, class 2 is blue, class
3 is red; see Table 1). b, Plot of mean histology score (MHS) versus the PC2
score obtained from the above PCA, with colour-coding as before. c, Ascores
plot from PCAof the pre-doseNMRdata for rats in histology classes 1 and 3.
Each point represents a single rat,with colour-coding as before. d, A loadings
plot corresponding to c, showing the variables making the largest
contributions to PC2, and the direction of each contribution. Individual
0.04 p.p.m.-wide spectral segments are identified by the chemical shifts at
their midpoints, and variables corresponding to particular compounds are
identified by name. Tau, taurine; Citr, citrate; Oxog, 2-oxoglutarate; TMAO,
trimethylamine-N-oxide; Bet, betaine. ‘2Tau’ indicates doubling of the Tau
values.
Table 1 | Liver histopathology in paracetamol-dosed rats at about 24
* It was not possible to obtain mean histology scores of 1.5 or 2.5 because of the
methodology used, in which individual scores across five liver lobes were averaged. See
Supplementary Information for details.
NATURE|Vol 440|20 April 2006 LETTERS
1075
© 2006 Nature Publishing Group
to predicting individual responses to broader medical, dietary,
microbiological or physiological challenges.
METHODS
This section relates to the paracetamol study only. Further details are provided in
the Supplementary Methods.
Animal treatment and sampling. The paracetamol study was performed in
accordance with relevant national legislation using 75 male Sprague-Dawley rats
(Crl:CD (SD)IGS BR) obtained from Charles River. The rats (approximately
seven weeks old) were placed in individual cages in a controlled environment,
with free access to water and a commercial feed. Sixty-five animals received, by
oesophageal intubation, a single oral dose of paracetamol (600 mg kg21) as an
aqueous suspension containing methylcellulose (0.5% w/v) and Tween 80
(0.1% w/v). A further ten rats were used as a control group and were orally
dosed with vehicle only. Individual pre-dose (248 to 224 h) and post-dose
(0–24 h) urine samples were collected into ice-cooled vessels containing 0.5 ml
of an aqueous 100 mg ml21 solution of sodium azide as a preservative. After
post-dose urine collection, individual blood samples were collected and
centrifuged to recover plasma for clinical chemistry. The rats were then killed
using CO2 and the liver of each animal was examined, weighed and sampled
for histopathology.
NMR sample preparation, spectral acquisition, processing and construction
of data sets. Urine samples were prepared as described in the Supplementary
Information. 1H NMR spectra were acquired (7200 Hz spectral width, 65,536
time domain points, 8 dummy scans, 64 real scans) at 600 MHz, at a nominal
303 K, on a Bruker DRX 600 NMR spectrometer operated by XWINNMR
software (both Bruker Biospin) with the ‘noesypresat’ pulse sequence28 used to
suppress the water signal during a 3-s relaxation delay and during the 0.1-s
mixing time. After Fourier transformation with 0.3-Hz line broadening and a
single zero-filling, pre-dose spectra were manually phased and baselinecorrected,
and the chemical shift scale set by assigning the value of d0 to the
signal fromthe added TSP (sodium3-trimethylsilyl-[2,2,3,3,-2H4]-1-propionate).
Each of these processed spectra was then ‘data-reduced’ using AMIX software
(Bruker Biospin), where the spectral regions d . 9.5, d6.1–5.5, d5.0–4.5 and
d , 0.5 were discarded before dividing the remainder of each spectrum into
sequential segments (‘bins’) of 0.04 p.p.m. width and obtaining an integral for
each segment. These integrals were then normalized to give the same total
integral for each data-reduced spectrum. Each bin was initially identified by the
chemical shift at its midpoint, but quantities relating to certain compounds
were derived by making appropriate bin combinations as described in the
Supplementary Information.
Post-dose spectra were processed as above and subsequently with resolution
enhancement, in order to determine the amounts and the relative proportions of
the paracetamol-related compounds29,30 excreted by each animal, by reference to
the cluster of N-acetyl signals in the range d2.11–2.22, the TSP signal and the
post-dose urinary volume. The amounts excreted were corrected to unit body
mass.
Clinical chemistry. Blood plasma samples were analysed at 30 8C on an AU600
clinical analyser (Olympus) for a variety of parameters, and univariate statistical
analyses were performed. Details and results are provided in the Supplementary
Information.
Histopathology. For each animal, ten representative samples of the liver (two
each from the left, right, left middle, right middle and caudate lobes) were
examined, the changes in each lobe scored, a mean histology score (MHS)
calculated and a histology class assigned as described in Table 1.
Multivariate modelling. Pirouette (v.2.7 and 3.1, Infometrix) and SIMCA Pþ
(v.10.0 and 10.5, Umetrics) software packages were used. After initial investigations,
three animals were excluded from all subsequent modelling because of
abnormal pre- or post-dose behaviour.
Predictive multivariatemodelling was carried out in SIMCA, with a supervised
pattern-recognition method (projection to latent structure, PLS) being used to
model the co-variation betweenNMR-derived pre-dose data (X) and selected postdose
response variables (Y). Model validation was performed as described in the
Supplementary Information.VIP (variable influence on projection) values characterize
the relative overall importance of the individualXvariables to themodel, and
the weights for each component of the model reveal the nature of the
relationship between each X variable and the predicted quantity, Y.
For predictive modelling of the G/P mole ratio, total area-normalized predose
spectral data were used with unit variance variable scaling. An unsupervised
pattern recognition method (principal component analysis, PCA) was also used.
Histology-coded PCA was performed, using Pirouette, on the pre-dose spectral
data normalized to constant total spectral area, with mean-centred variable
scaling and with the histology class of each animal assigned as described in
Table 1. The relevant loadings describe how the original variables contribute to
each principal component.
Univariate statistical analysis and tests of correlation. All significance tests
were two-sided. See Supplementary Information for details.
Received 17 October 2005; accepted 14 February 2006.
1. Spear, B. B., Heath-Chiozzi, M. & Huff, J. Clinical application of
pharmacogenetics. Trends Mol. Med. 7, 201–-204 (2001).
2. Pagliarulo, V., Datar, R. H. & Cole, R. J. Role of genetic and expression profiling
in pharmacogenomics: the changing face of patient management. Curr. Issues
Mol. Biol. 4, 101–-110 (2002).
3. Ingelman-Sundberg, M. Genetic variability in susceptibility and response to
toxicants. Toxicol. Lett. 120, 259–-268 (2001).
4. Nebert, D. W., Jorge-Nebert, L. & Vesell, E. S. Pharmacogenomics and
“individualized drug therapy”: high expectations and disappointing
achievements. Am. J. Pharmacogenomics 3, 361–-370 (2003).
5. van Aken, J., Schmedders, M., Feuerstein, G. & Kollek, R. Prospects and limits
of pharmacogenetics: the thiopurine methyl transferase (TPMT) experience.
Am. J. Pharmacogenomics 3, 149–-155 (2003).
6. Rang, H. P., Dale, M. M. & Ritter, J. M. Pharmacology (Churchill Livingstone,
Edinburgh, 1995).
7. Tannock, G. W. Normal Microflora: An Introduction to Microbes Inhabiting the
Human Body (Chapman and Hall, London, 1995).
8. Nicholson, J. K., Lindon, J. C. & Holmes, E. ‘Metabonomics’: understanding the
metabolic responses of living systems to pathophysiological stimuli via
multivariate statistical analysis of biological NMR data. Xenobiotica 29,
1181–-1189 (1999).
9. Lindon, J. C., Nicholson, J. K., Holmes, E. & Everett, J. R. Metabonomics:
metabolic processes studied by NMR spectroscopy of biofluids combined with
pattern recognition. Concepts Magn. Reson. 12, 289–-320 (2000).
10. Nicholson, J. K., Connelly, J., Lindon, J. C. & Holmes, E. Metabonomics: a
platform for studying drug toxicity and gene function. Nature Rev. Drug Discov.
1, 153–-161 (2002).
11. Nicholson, J. K. & Wilson, I. D. Understanding ‘global’ systems biology:
metabonomics and the continuum of metabolism. Nature Rev. Drug Discov. 2,
668–-676 (2003).
12. Lindon, J. C. et al. The Consortium for Metabonomic Toxicology
(COMET): aims, activities and achievements. Pharmacogenomics 6, 691–-699
(2005).
13. Shockcor, J. P. & Holmes, E. Metabonomic applications in toxicity screening
and disease diagnosis. Curr. Top. Med. Chem. 2, 35–-51 (2002).
14. Rawson, E. S., Clarkson, P. M., Price, T. B. & Miles, M. P. Differential response
of muscle phosphocreatine to creatine supplementation in young and old
subjects. Acta Physiol. Scand. 174, 57–-65 (2002).
15. Zuppi, C. et al. 1H NMR spectra of normal urines: reference ranges of the major
metabolites. Clin. Chim. Acta 265, 85–-97 (1997).
16. Zoratti, R. A review on ethnic differences in plasma triglycerides and highdensity-
lipoprotein cholesterol: is the lipid pattern the key factor for the low
代写 Pharmaco-metabonomic phenotyping and personalized drug treatment
