ToxSci Advance Access originally published online on April 9, 2007
Toxicological Sciences 2007 98(1):271-285; doi:10.1093/toxsci/kfm076
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Sensitivity of 1H NMR Analysis of Rat Urine in Relation to Toxicometabonomics. Part I: Dose-Dependent Toxic Effects of Bromobenzene and Paracetamol
* Department of Pharmacology, N.V. Organon, Molenstraat 110, 5340 BH Oss, The Netherlands
Department of Medicinal Chemistry, N.V. Organon, Molenstraat 110, 5340 BH Oss, The Netherlands
Department of Toxicology and Drug Disposition, N.V. Organon, Molenstraat 110, 5340 BH Oss, The Netherlands
Department of Analytical Research, TNO Quality of Life, Utrechtseweg 48, PO Box 360, 3700 AJ Zeist, The Netherlands
1 To whom correspondence should be addressed. Fax: +31-412-663532. E-mail: willem.schoonen{at}organon.com.
Received October 23, 2006; accepted March 22, 2007
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
1H nuclear magnetic resonance (NMR) spectroscopy of rat urine in combination with pattern recognition analysis was evaluated for early noninvasive detection of toxicity of investigational chemical entities. Bromobenzene (B) and paracetamol (P) were administered at five single oral dosages between 2 and 500 mg/kg and between 6 and 1800 mg/kg, respectively. The sensitivity of the proposed method to detect changes in the NMR spectra 24 and 48 h after single dosing was compared with histopathology and biochemical parameters in plasma and urine. Both B and P applied at the highest dosages induced liver necrosis and markedly increased aspartate aminotransferase (AST) and alanine aminotransferase (ALT) plasma levels. At dosages of 125 mg/kg B and 450 mg/kg P, liver necrosis and changes in AST and ALT were less pronounced, while at lower dose levels these effects could not be detected. Changes in kidney pathology or standard urine biochemistry were not observed at any of these dosages. Evaluation of the total NMR dataset showed 80 signals to be sensitive for B and P dosing. Principal component analysis on the reduced dataset revealed that NMR spectra were significantly different at dosages above 8 mg/kg (B) and 110 mg/kg (P) at both sampling times. This implies a 4- to 16-fold increased sensitivity of NMR versus histopathology and clinical chemistry in recognizing early events of liver toxicity.
Key Words: metabonomics; urinalysis; hepatotoxicity; bromobenzene; paracetamol; necrosis; biomarkers.
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
Despite the developments of toxicological and pharmacological test systems for over 40 years, attrition rates in the development phase of pharmaceutical drug candidates due to adverse effects remain high (Caldwell et al., 2001
; Cuatrecasas, 2006; Nicholson et al., 2002; Schuster et al., 2005; Smith and Schmid, 2006). To address the issue in early stages of compound discovery, assessment of hepatotoxicity of series of compounds is of prime importance. Therefore, medium throughput tox screening methods are warranted in the early research phase. Noninvasive techniques are preferred for minimization of animal sacrifices and to enable longitudinal studies.
In the present study, 1H nuclear magnetic resonance (NMR) spectroscopy of urine from rats in combination with principal component discriminant analysis (PCDA) (Hoogerbrugge et al., 1983; Kowalski, 1975; Kowalski and Bender, 1974; Massart et al., 1988; Tiedeman et al., 1967; Vogels et al., 1993) is evaluated for early toxicity studies. Urinalysis by (proton) NMR will produce a fingerprint of the endogenous metabolism of the animal. Realize that the spectrum will generally contain information on organic molecules in the molecular weight range of roughly below 1000 Da and at concentrations in the low millimolar range. This may be seen as a limitation but should also be considered a not yet fully understood wealth of information. Interestingly, we do not need to be able to fully interpret all spectral signals to use changes in the signals as biomarkers for toxicity and/or pharmacology.
Previous studies have demonstrated relations between induced toxicology and changes in the proton NMR spectra of urine from animal model systems (Beckwith-Hall et al., 1998, 2002; Bollard et al., 2005; Gartland et al., 1991; Griffin et al., 2001; Holmes et al., 2000; Nicholson et al., 1999; Robertson et al., 2000; Van der Greef, 2003; Waters et al., 2001). Changes in the urine NMR spectra demonstrated disturbances in physiological metabolites upon administration of several hepatotoxins, like 
-naphthyl isothiocyanate, carbon tetrachloride, D-(+)-galactosamine, and butylated hydroxytoluene, as well as several nephrotoxins, like 2-bromoethylamine, 4-aminophenol, and paraquat (Bairaktari et al., 1998; Beckwith-Hall et al., 1998; Robertson et al., 2000). Moreover, initiatives are taken to extend the use of NMR urinalysis to pharmacometabonomic phenotyping and personalized drug treatment (Clayton et al., 2006; Doorn et al., 2007). These studies support the applicability of urine NMR spectroscopy in general but did not always measure the sensitivity in relation to classical toxicology assessment.
The goal of our study is to correlate the effects of a large dose range of bromobenzene (B) and paracetamol (P) on NMR urinalysis versus histopathological and biochemical parameters to identify the most sensitive technique for the detection of early toxicity. Both B and P are well-known hepatotoxic and nephrotoxic compounds (Casaret and Doul's Toxicology, 1986; Dixon et al., 1975; Vries, 1996). Many investigations on the toxicology of these two compounds were reported earlier (Hetu et al., 1983; Jollow et al., 1974; Koen et al., 2000; Miller et al., 1990; Niesink et al., 1996; Siegers et al., 1978; Tredger et al., 1985; Vandenberghe, 1996; Williams and Davis, 1977; Wong et al., 2000, Zampaglione et al., 1973). If each of these compounds is given at relatively high dosages, hepatotoxicity is observed by liver necrosis and by an increase in the plasma levels of aspartate aminotransferase (AST) and alanine aminotransferase (ALT). Glutathione depletion in the liver plays a predominant role in this process. With exceeding dosages also nephrotoxicity can be identified with both B and P, since at these concentrations reactive metabolites like 3,4-bromobenzene oxide and N-acetyl-p-benzoquinonemine are no longer detoxified in the liver. In an earlier pilot experiment (Schoonem, Zandberg and Horbach, unpublished data), it was established that 1000 and 2000 mg/kg B as well as 1800 mg/kg P induced slight changes in urinary parameters such as protein content and glucose and ß-N-acetyl-glucosamidase levels in some of the treated animals. These clinical chemistry data are indicative of defects in the glomerular and proximal tubuli of the kidney. These small changes in the urinary parameters, however, were not yet accompanied by overt kidney pathology.
In order to establish the NMR sensitivity, a dose range of B and P was investigated from 2, 8, 32, 125, and 500 mg/kg for B and 6, 25, 110, 450, and 1800 mg/kg for P. The animals were studied for either 1 or 2 days after the exposure to a single dose of toxicant at the start of the experiment.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
Experimental setup.
The studies described below are in compliance with the Dutch Act on Animal Experimentation. The studies were approved by the Animal Experimentation Committee.
In this study, two toxic compounds, i.e., B and P (both Sigma-Aldrich, St Louis, MO), were used and given at five different dosages 2, 8, 32, 125, and 500 mg/kg for B and 6, 25, 110, 450 and 1800 mg/kg for P (Supplementary Table 1). Each different dose group as well as the vehicle group consisted of five animals. For each group, the experiment was carried out with both 1 and 2 days of treatment (Supplementary Table 1). The experimental protocol was as follows.
Male Wistar rats (Harlan) of 6 weeks old (140–170 g) were placed in metabolism cages, and 24 h later urine samples were collected (day 0). At the start of the treatment, either vehicle or a fixed dose of compound was administered orally in 10% gelatine/mannitol (at 2 ml/kg the Organon standard). For P, an exception was made at the highest dosage of 1800 mg/kg. At this dosage, two subsequent injections with 900 mg of compound were given directly after each other due to solubility problems at this high dose level. Urine samples were collected after both 1 day (0–24 h) and 2 days (24–48 h) of treatment. Urine samples were stored at – 20°C for clinical chemistry and at 4°C on ice for NMR. Both at 1 and 2 days after compound administration, blood samples were collected of five rats, whereafter these rats were rapidly decapitated. Blood samples were stored at 4°C on ice and later at – 20°C before analysis. Plasma and urine samples were analyzed on biochemical parameters, while livers and kidneys were dissected for histopathology.
Pathology and clinical chemistry.
Pathology and clinical chemistry were performed according to standard procedures. Samples of the livers and kidneys were preserved in 10% buffered formalin. Subsequently, tissue samples were dehydrated and embedded in paraffin wax. The paraffin-embedded liver and kidney samples were processed at 5-µm thick sections. Sections were stained with hematoxylin and eosin (sections) and used for histological examination by light microscopy. Plasma and urine samples were analyzed for biochemical parameters.
NMR sample preparation.
After collecting the urine samples, these samples were stored on ice, filtrated on cotton, and centrifuged to remove solids during collection. Per sample, a volume of 1 ml urine was lyophilized overnight and dissolved in 1 ml 0.1M phosphate buffer (pH 7.0) using D2O as aqueous solvent to obtain a urine preparation in a stable buffer solution. Between collection of urine, NMR sample preparation and analysis only 24 up to 28 h passed.
NMR pattern recognition.
All spectra were recorded on a Bruker Avance DRX 400 MHz spectrometer operating at a temperature of 300 K. The proton 1D spectra were recorded using a 45° detection pulse, a 50 ms CarrPurcellMeiboomGill pulse train with a spin echo delay of 1 ms to suppress protein signals, and with presaturation during the relaxation delay of 2 s to suppress residual water. The spectra were recorded with the following acquisition parameters: spectral width 8000 Hz, 64 k complex data points, and 256 transients. The spectra were processed using the XWINNMR software (Bruker). The phase and baseline of the spectra were corrected manually using an exponential window function and fifth order polynomial fit. The chemical shifts of the signals were referenced against 3.047 ppm for peak synchronization. Thereafter, the peaks in each NMR spectrum were normalized in intensity relative to the 3.047-ppm signal in that specific NMR spectrum. To convert the NMR data to a format that can be used by multivariate analysis, the position and intensity/abundance of all the lines in the NMR spectra above the noise level were collected in line listing which were then converted into a single data matrix. All NMR peaks were synchronized manually on their peak pattern to eliminate the small disturbances (
0.001 ppm) in the position between lines from identical compounds within different spectra by a manual correction device developed at TNO (Zeist, The Netherlands) to optimize the peak listing. Main advantage of using this unique method, developed by TNO (Vogels et al., 1996) is that it, contrary to the methods used by Nicholson et al. (1999), retains the high resolution of the original NMR spectra. Similar analysis methods described by Cloarec et al. (2005) indicate that these improvements led to better data interpretations and better NMR spectral analysis.
The presence of the drugs and/or their derivatives in urine could mask the toxicologically relevant changes. Therefore, these particular drug-related peaks were indirectly discarded from the spectra, and only those peaks were taken into account that appeared in both treated and untreated samples. After elimination of all peaks that do not occur in all categories of the spectra, a specific set of unique peak positions remained. These remaining peaks were further analyzed by PCDA.
We have to make a remark on the signal we used for peak synchronization (3.047 ppm). This peak is identified as a combined peak from creatinine and creatine together with a very small contribution of one of the triplet peaks of -ketoglutarate. Creatinine and creatine peaks are also often used in other NMR metabonomics applications (Clayton et al., 2003, 2004; ppm's at 4.05 and 3.93, respectively). Creatinine is a very commonly used reference compound for the calibration/normalization of urine clinical parameters (Chung et al., 2003; Dyson et al., 1992; Narayaran and Appleton, 1980). It is also a reliable indicator of a good glomerular filtration rate as is the creatinine/creatine ratio (Chung et al., 2003). PCDA analysis on the chemical shift corrected datasets from the control samples and samples from the lower concentrations of the administered drugs showed that the peak at 3.047 ppm was the most constant peak in the analysis. This peak is well recognized and not influenced in intensity/abundance by the treatment with the compounds B and P. The least influenced peak of 3.047 ppm was immediately followed by the other peaks belonging to creatine (3.970 ppm) and creatinine (4.050 ppm) in our PCDA analysis. This shows that the creatinine-creatine ratio is constant for all our samples, which is confirmed by clinical analysis of creatinine. Therefore, this 3.047 peak was chosen as the reference peak for all samples. Since normally the ratio of creatinine and creatine is used for balancing the other urine metabolite quantitities, there is a high confidence in using this combined creatinine and creatine peak.
Principal component (discriminant) analysis, (differential) factor spectra, and the procrustus method.
After summarizing all signals of the NMR spectra of all samples into a single matrix, the influence of the different types of metabolites in the samples were analyzed. The first step was to perform principal component analysis (PCA). In PCA, the principal components (PCs) of the NMR spectra were determined. PCs are linear combinations of highly correlated NMR signals. PCA has been thoroughly described by Massart et al. (1988). In order to identify the influence of different types of metabolites, it was decided to use a correlation matrix in all calculations. The PCs can be used to look at the differences and similarities of samples. A useful method is to project the samples into a plot spanned by the first few PCs. This will result in a low-dimensional representation of the total data matrix with only a minimal amount of information getting lost in the process. In such a plot, generally referred to as a "score plot", similar samples will tend to cluster together, while spectra from more dissimilar samples will be found at larger mutual distances.
The next step is to perform discriminant analysis by PCDA on these PCs. PCDA is a multivariate analysis method pioneered by Hoogerbrugge et al. (1983). The choice of how many PCs to use is not very defined, but n should preferably not exceed the number of patterns divided by four (Tiedeman et al., 1967). The number of peaks included in the PCDA model was reduced using a Procrustus rotation method as described by Héberger and Andrade (2004). In the Procrustus algorithm, a PCDA is performed on the full data, and the resulting PCDA score plot is stored. The first step is to remove one of the original variables and repeat the PCDA analysis. The deviation between the original and the new score plot is calculated by a classical Procrustus analysis according to Gower and Dijksterhuis (2004). While running this procedure for all peaks, the peak that has the least influence on the result was eliminated. The extraction of peaks was stopped at 80 remaining peaks, i.e., at the point where the slope of the plot of the error versus the remaining number of peaks showed a sharp increase in the deviation between the original data and the reduced dataset. In this paper, this method was adapted to work with PCDA score plots instead of PCA plots. The minimal set of 80 peaks was then used to perform all subsequent multivariate discriminant analysis as described by Hoogerbrugge et al. (1983).
To interpret the result of PCDA calculations, we used so-called score plots. In the score plot, the position of the original variables is given so that the projection length of the variable vector parallel to a certain discriminant axis is proportional to the importance (loading) of that variable to that axes (Windig et al., 1983). Similar to PCs, the linear discriminant axi are also linear combinations of the original NMR peaks, which can be visualized in differential factor spectra. Factor spectra were used to correlate the position of clusters in the score plots to the original features in the spectra by a graphical rotation of the loading vectors as described by Windig et al. (1983). In such a differential factor spectrum, the contribution of each of the NMR peaks in the original spectra is plotted as a bar with a length between –1 and +1 at a position corresponding to the position of that peak in the original NMR spectrum. A high positive or negative bar for a given peak corresponds to a relative high (above average) or low concentration of that metabolite in the samples, respectively. In cases where a single metabolite is responsible for the formation of a single discriminant axis, i.e., if a small group of samples contains an inordinate amount of this metabolite, then the factor spectrum in the direction of this group of samples will generally have a few positive peaks indicating the single metabolite and many negative peaks indicating all metabolites that are lower than average in these few spectra.
A more detailed description of the PCA and PCDA is given in Supplementary Data.
Identification of metabolites.
The preliminary identity of the peaks used in PCDA was determined with an internal specialized database of 400 compounds for compound identification in rat urine, an external database (http://www.liu.se/hu/mdl/main/) (Agar et al., 1991; Fan, 1996a,b; Govindaraju et al., 2000) and several literature comparisons (Bollard et al., 2005; Clayton et al., 2006; Robertson et al., 2000). These assignments are not definitive and should be handled with care because small chemical shifts due to pH differences between the spectra in the internal database and the spectra in our study can be present. Additionally, due to the vast number of components in rat urine overlap can occur. Final identification should be done with spiking the urines and 2D-NMR measurements. However, the main scope of our study was to show the NMR profiling differences.
Overview analysis.
The different steps in the analysis are as follows:
- 1) Acquire NMR spectra.
- 2) Normalization.
- 3) Collect all peaks above the signal-to-noise-ratio in the spectra.
- 4) Combine all spectra into a single data matrix.
- 5) The spectra have their chemical shift calibrated to the creatinine and creatine peak at 3.047 ppm and or normalized on height/intensity to that peak as well.
- 6) Synchronized peaks belong to the same compound in different spectra by manual correction (differences
0.001 ppm), leading to high resolution of the original NMR spectra.
- 7) Removal of all compound-related peaks. The remaining peaks are present in urines of both treated and untreated animals. In this way, signals from the administered drugs and their metabolites are excluded from our analysis.
- 8) Discriminant analysis on full (compound/metabolite independent) large biomarker set.
- 9) Selection with Procrustus rotation method of the most relevant biomarkers.
- 10) PCDA on most relevant biomarker set as determined in step 9.
- 11) Identify directions of the highest discrimination i.e., difference between treatment and control.
- 12) Use differential factor spectra analysis to select the most discriminating peaks.
- 13) Check discrimination by analyzing the level of the most discriminating peaks in the original data. This is expressed in % and indicated in brackets in all figures showing the distribution of the data in PCDA analysis;
- 14) Comparing the peaks of interest with the databases for compound identification for rough assignment.
- 2) Normalization.
Steps 10–13 of PCDA analysis were repeated on different selections of treated rats (including the control set) to distinguish between the effects on dose, time after dosing, and the response of the animal to the kind of drug administered:
- A) B at 1 day after drug administration.
- B) B at 2 days after drug administration.
- C) P at 1 day after drug administration.
- D) P at 2 days after drug administration.
- E) Combination of all data of A to D.
- F) Combination of all data of E excluding the highest dose of P for both 1 and 2 days treatment.
- B) B at 2 days after drug administration.
All calculations and data preprocessing were performed on a Siemens Personal Computer using the Windows-based TNO-Winlin NMR pattern recognition program.
Statistical procedure.
The observed differences for ALT and AST levels in the blood plasma as well as the differences in intensities of individual NMR peaks were tested for statistical significance by means of an one-way ANOVA test, followed by a paired Student's t-test (p < 0.05).
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
Pathology
The highest dosage for both compounds clearly induced liver damage after 1 and 2 days of treatment. With B at 500 mg/kg, slight (four animals) or marked to severe (respectively, four or one animal) centrilobular necrosis was observed in the liver. One animal only showed minor changes by mild vacuolation of hepatocytes. Treatment with 1800 mg/kg P induced marked centrilobular liver necrosis in nine animals, while vacuolation of hepatocytes was observed in one animal. In the kidneys treated with either B or P, no pathological abnormalities were observed. At the second highest dosage, less severe effects were found for both drugs. Treatment with B at 125 mg/kg, only resulted in one animal into minimal centrilobular necrosis after 1 day of treatment and five animals with increased mild basophilicity after 2 days of treatment. Treatment for 1 day with P at 450 mg/kg led to either moderate (two animals) or marked (two animals) liver centrilobular necrosis, while pathology was absent after 2 days of treatment. At lower dosages, no histopathological abnormalities were observed in either liver or kidney.
Clinical Chemistry
Plasma concentrations of AST and ALT were markedly increased after 1 and 2 days of treatment with the highest dosage of B (500 mg/kg) and P (1800 mg/kg) (Supplementary Figs. 1 and 2). The animals showing vacuolation of hepatocytes had normal AST and ALT levels. At the second highest dosage, an increase of AST and ALT levels was only found after 1 day of treatment for both drugs. With B, two animals with the highest enzyme levels also showed signs of histopathology. With P, two animals showed a twofold increase in AST and ALT levels, while for two other animals, a 50- to 100-fold increase was observed. Only in the two animals with the highest AST and ALT levels, a significant centrilobular necrosis was observed. In the animal without histopathological findings of this group, AST levels were similar to the control level, whereas ALT levels were twofold increased. After 2 days of treatment at the second highest dosage with B and P, the AST and ALT levels were comparable to the controls. At the lower dosages, no change in AST and ALT levels was observed.
Other parameters in plasma, including bilirubin, alkaline phosphatase and gamma glutamyl transferase, albumin, and total protein remained unchanged even at the highest dosages. For urine, the parameters including creatinine, osmolality, glucose, ureum, calcium, phosphate, ß-N-acetyl-glucosamidase, total protein, sodium, potassium, and chloride were also not significantly different from the control values (p < 0.05, one-way ANOVA, paired Student's t-test) (not shown).
Description of the NMR Spectra
In Figures 1A–C, the NMR spectrum of a representative control urine sample is shown together with spectra of urine samples at high-dose treatment of B and P. Partial assignments are given in Table 1 and Figure 1D. Spectral data reduction (e.g., elimination of exogenous compound and metabolite peaks) resulted in a pattern of 174 peaks (Fig. 2A). The Procrustus rotation method further reduced the number of relevant signals present in all urine samples (treated and untreated animals) to 80 biomarker peaks (Figs. 1D and 2B and Table 1). These 80 biomarker signals are expected to be related to drug-induced changes in the liver, as toxicity in the kidney was not observed by clinical chemistry and histopathology.
|
|
|
Most changes in the NMR spectra are related to endogenous metabolites involved in carbon metabolism such as the citric acid cycle intermediates, e.g., citrate and
-ketoglutarate, and the glycolysis end product L-lactate. Changes in the levels of formate and allantoin, which are indicative for stress responses, are also observed. Moreover, changes in dimethylglycine and betaine may indicate changes in the choline shunt or folate metabolism. Besides these changes, also hippurate, dimethylamine, and trimethylamine N-oxide are indicative for liver toxicity.
Pattern recognition—B.
In Figures 3A and 3B, the fingerprint of 80 biomarker peaks is compared by PCDA after 1 and 2 days of treatment. Clear differences are found for samples of the B groups treated with 500, 125, and 32 mg/kg with respect to the control group. The group treated with 8 mg/kg B, however, partly overlapped the control group. The 2-mg/kg B–treated group, in contrast, differed again significantly from the control group. We have no explanation for the latter observation.
|
Comparison of the differential factor spectra for 1 and 2 days shows that excretion into the urine of many endogenous metabolites is largely decreased upon dosing (Figs. 3C and 3D). This is the case for
-ketoglutarate (2.446, 2.458, 2.476, 3.010, 3.025), citrate (2.540 and 2.694), betaine (3.312), taurine (3.274, 3.312, 3.43, 3.441), tyrosine (7.220), and hippurate (7.536 and 7.830), as we can see from the positive lines in the factor spectra. On the other hand, the production of lactate (1.355), "metabolite 11" (2.072), dimethylglycine (2.934), "metabolites 35, 39, and 40" (3.688, 3.940, and 3.954), allantoin (5.393), uridine (5.795), "metabolites 44, 45, 46, 48, 50, and 53" (6.679, 6.879, 6.978, 7.348, and 7.516), and formate (8.476) appeared to be increased after 1 day of treatment, which is indicated by the negative lines. The profile differed after 2 days of treatment; only "metabolites 11, 24, and 39" (2.072, 3.398, and 3.940) are increased, while the levels in many other compounds are still decreased (Fig. 3D).
Pattern recognition—P.
As can be seen in Figure 4 (A, B, E, and F), a clear clustering is found with PCDA after 1 and 2 days of treatment. But the spectra from the 1800-mg/kg–treated animals seem to completely differ from trends seen in the 450, 110, 25, and 6 mg/kg–treated groups (Figs. 4A and 4B). While the 450- and 110-mg/kg–treated groups show significant clustering away from the control group, the 25- and 6-mg/kg–treated groups hardly differ from control (Figs. 4A and 4B).
|
When the 1800-mg/kg group data is excluded, all groups from 450, 110, 25, and 8 mg/kg become separated from control (Figs. 4E and 4F). Separation was most evident for the 450-mg/kg group, followed by the mid-dose group of 110 mg/kg. The next lowest and lowest dose groups could still be discriminated from the control group at both sampling times, showing the sensitivity of this NMR PCDA method.
Comparison of the differential factor spectra for 1 and 2 days (Figs. 4C and 4D) shows that excretion of many endogenous metabolites is increased for the first 24 h, in contrast to what was seen for B. An initial decrease in concentration is seen after 24 h only for the following metabolites: acetylcholine (2.178), "metabolites 16 and 30" (2.803 and 3.526), creatinine (4.054), and "metabolite 54" (7.674). Most of the other peaks were largely enhanced by P treatment. After 48 h, the overall production level changed, and many metabolites were now decreased. An increase in concentration could still be seen for the main peaks of citrate (2.540), "metabolite 17" (2.887), trimethylamine N-oxide (3.283), and formate (8.476) in the P-treated group. For all the other peaks, normal control levels were reestablished. But the above analysis might be disturbed by the exceptional 1800-mg/kg dosing data.
After elimination of the 1800-mg/kg data, the PCDA analysis and differential factor spectra analysis revealed other changes in the profiles (See Figs. 4G and 4H). After 1 day of treatment, a decrease in concentration of "metabolites 3, 5, 6, and 7" (1.589, 1.698, 1.721, and 1.775), citrate (2.540 and 2.694), dimethylglycine (2.934), "metabolite 38" (3.915), tyrosine (7.220), hippurate (7.536 and 7.830), and "metabolite 54" (7.674) is observed. Most of the other peaks such as of -ketoglutarate (2.446 and 3.025) and metabolites 8, 12, 25, 40, 44, 45, 48, 49, 50, 51, and 52 (1.884, 2.151, 3.330, 3.954, 6.679, 6.879, 7.231, 7.307, 7.348, 7.366, and 7.380) were largely enhanced by P treatment. After 2 days, the overall production level changed, and many metabolites were now decreased in the treated groups in relation to the control group (Fig. 4H). An increase in concentration could only be seen for the main peaks of metabolites 5, 8, 9, 12, 14, 17, 19, 23, 35 (1.698, 1.884, 1.968, 2.151, 2.287, 2.887, 2.989, 3.193, and 3.688), trimethylamine N-oxide (3.282), and formate (8.476).
Pattern recognition—B and P combined.
Data of all the individual dose groups with B and P were combined in one analysis. Again, a clear separation was found between the data from the highest dose groups (B 500 and P 1800 mg/kg) and controls (Fig. 5A). Due to the doubtful contribution of the P 1800 mg/kg group to the overall analysis, this high dose group was removed again, as shown in Figure 5B. The next highest dosages, i.e., B 125 mg/kg and P 450 mg/kg, still show clear separation from control at 1 day after administration. After 48 h, this effect was still seen for B but not for P. The spectral data from P at the second lowest (P 25 mg/kg) and lowest dose (P 6 mg/kg) did not differ from control (Fig. 5B), whereas spectra from the B-treated groups at the second lowest and lowest dose (B 8 mg/kg and B 2 mg/kg) still show marked differences from control. The mid-dose (B 32 mg/kg and P 110 mg/kg) data produced intermediate results. At least for B a difference was present at both days, while for P this difference was only found 1 day after treatment.
|
In Figures 5C and 5D, the differential factor spectra are shown. Differences in biomarker concentrations are notable in the area of citrate and
-ketoglutarate between 2.446 and 3.025 ppm but also in the area of the sugars between ppm 3.18 and 3.69.
B administration results in changes of "metabolite 11" (2.072), dimethylglycine (2.934), "metabolites 39 and 40" (3.940 and 3.954), allantoin (5.393), "metabolites 44 and 46" (6.679 and 6.978), and hippurate (7.536 and 7.830). P administration results in changes of acetylcholine (2.178), "metabolite 16" (2.803), citrate (2.540 and 2.694), -ketoglutarate (2.446, 2.548, 2.476, and 3.02), creatinine (4.054), "metabolite 54" (7.674), and hippurate (7.536 and 7.830).
Dose-dependent effects on individual peaks.
In Figures 6 and 7, eight of the 80 biomarker peaks with the most prominent changes were selected, and their averaged spectral intensities (normalized against the 3.047 peak of creatinine and creatine) were plotted against the dose. The dose dependence deserves individual discussion for each of these eight biomarkers.
- "Metabolite 11" at 2.072 ppm. After 1 day of B treatment, a dose-dependent increase in the "metabolite 11" level was seen. After 2 days of B administration, the increase in "metabolite 11" levels was only seen in the highest dose group. In the case of the P treatment, an increase in "metabolite 11" levels was only observed in the highest dose group both after 1 and 2 days.
- Acetylcholine at 2.178 ppm. A completely different profile was observed with B and P for this metabolite. B did not reveal any changes in the levels of acetylcholine, except for a small nonsignificant reduction at the lowest dose after 1 day of treatment and at the highest dose after 2 days of treatment. On the other hand, P showed after 1 day of treatment, a nice proportional dose dependence except for a sharp decrease in the amount at the highest dose. After 2 days of P treatment only in the highest dose group, an increase in the acetylcholine levels was observed.
- -Ketoglutarate at 2.446 ppm. With B, only small changes in the levels of -ketoglutarate could be observed. These changes are again proprotional to the dose except for the highest dosages. With P, a large increase in the -ketoglutarate level at the highest dose level was observed after 1 day of treatment, which was even more excessive after 2 days of treatment. At all the other dosages, the levels were more or less comparable to control levels.
- Trimethylamine N-oxide at 3.282 ppm. With B, an inconsistent profile pattern was observed. The dose-response relationship seemed reciprocal for the 1-day treatment and proportional for 2 days of treatment, but the changes were statistically not significant in both cases. With P, a large increase in the trimethylamine N-oxide level was observed only at the highest dose level both after 1 and 2 days of treatment. At the other dose levels, there was not a clear difference from the control group.
- "Metabolite 39" at 3.940 ppm. With both B and P, a clear enhancement in these levels was observed at only the highest dose levels. All the other dosages showed comparable levels with respect to the control samples, with exception of the lowest dose of B after 1 day of treatment. Here, a decrease in the levels was observed. With B, a similar change in both peaks is seen, as expected.
- "Metabolite 40" at 3.954 ppm. The pattern is similar to the previous "metabolite 39" at 3.940 ppm.
- "Metabolite 24" at 3.249 ppm. With B, no changes were observed in "metabolite 24" levels both after 1 and 2 days of treatment. With P, a clear increase in "metabolite 24" levels was observed at the highest dose level after 1 and 2 days of treatment. All the other dosages did not induce differences with respect to the control levels.
- "Metabolite 50" at 7.307 ppm. After treatment with B, no differences in "metabolite 50" levels were observed except for a decrease at the highest dose after 2 days of treatment. With P, a dose-dependent increase in the "metabolite 50" level was identified after 1 day of treatment at the two highest dose levels. After two days of treatment, this effect on "metabolite 50" levels was only observed at the highest dose level.
|
|
The individual peaks that were discussed in detail (Figs. 6 and 7) show different types of dose-dependent responses. In the majority of the cases, it was shown that the dose response is flat except for an increase at the highest dose. Another group of metabolites show a proportional relationship, i.e., metabolite 11 after 1 day of treatment. For some of these observations, the highest concentration does not follow the same trend. In a third group, a dose-response relationship was not found.
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
The usefulness of 1H NMR spectroscopy of urine from rats in combination with PCDA and the Procrustus rotation method is evaluated for early noninvasive detection of toxicity. The method appears to be very valuable as changes in endogenous metabolism become apparent at much lower dosages when applying test compounds B and P than observed with histopathology. The NMR data processing protocol was improved by using peak listing instead of bucketing (Cloarec et al., 2005
; Vogels et al., 1996), by alignment of peaks using manual correction with home-built (TNO) software and by focusing on the most relevant peaks selected by the Procrustus rotation method. In this way, we focus on more relevant NMR biomarkers resulting in a better discrimination of toxicity profiles. Using traditional toxicity screening techniques, both hepatotoxic compounds showed overt changes in plasma levels of AST and ALT at both 1 and 2 days after treatment at the highest dosages. This coincides with the severity of slight or marked centrilobular necrosis, which is in line with earlier data for these compounds. Histopathology and clinical chemistry could only observe significant changes at 125 mg/kg for B and 450 mg/kg for P. However, the current NMR pattern analysis was able to demonstrate differences at the mid-dose levels for B (32 mg/kg) and P (110 mg/kg), which indicates the potential of this new technique for the rapid toxicological analysis of investigational compounds. In case of P, it was even shown that both the second lowest and lowest dose levels discriminated themselves from control. This means that NMR has a detection limit which is at least fourfold lower for both B and P, and in some cases even 16-fold or more, compared with standard approaches. Thus, NMR followed by PCDA analysis seems very sensitive as an early toxicity identifier by using 80 relevant biomarker signals.
Comparison of the urine NMR data for B and P with those obtained by other investigators is difficult, as only a few studies have been reported (Heijne et al., 2005). Observations upon treatment with P with NMR have been described for mouse plasma after 4 h of treatment (Coen et al., 2003). Here a dramatic increase is seen in lactate, acetate, pyruvate, and glucose as well as 3-hydroxybutyrate. This indicates that glycolysis is increased and oxidative phosphorylation from fatty acids is hampered by P. The decrease of citrate and -ketoglutarate levels with B in our study also pinpoints to a decrease in oxidative phosphorylation. Moreover, in case of P, citrate was accumulating, while -ketoglutarate was decreased. Additionally, a role of the following metabolites dimethylglycine, trimethylamine N-oxide, taurine, betaine, and hippurate after B and P treatment is confirmed by others (Bollard et al., 2005; Clayton et al., 2006; Gartland et al., 1989; Heijne et al., 2005). Dimethylglycine and betaine are metabolites in the choline shunt. Accumulation of betaine and dimethylglycine can occur due to a hampered dimethylglycine oxidase and sarcosine oxidase activity in the mitochondria (Binzak et al., 2001; Wittwer and Wagner 1980, 1981). Both enzymes are also known as dehydrogenases, but inadequate oxidative phosphorylation reduces the activity of these two enzymes.
It is seen in the current study that NMR urinalysis is a sensitive approach to detect changes in the normal endogenous metabolism of test animals. These changes might reflect toxicological effects as well as pharmacological effects. Surrogate markers of treatment efficacy and toxicity can be utilized to optimize the monitoring of preclinical studies, thus providing wide applicability of the technique. NMR urine fingerprints can potentially help to identify side effects of drug candidates that might not otherwise be identified until after relatively expensive and lengthy preclinical studies. Thus far, this type of technology has not yet been subjected to a rigorous evaluation of specificity, robustness, and cost-effectiveness as a general modality.
The procedure of NMR is fast and noninvasive, and rats may be reused for testing other compounds after a certain washout period. The presented method of NMR analysis and pattern recognition can be handled within two to three working days. Immediately after urine collection, the urine is freeze dried, and on the next day dissolved in buffer for NMR analysis in between 24 and 28 h after collection of the urine samples. On the next morning, the recordings can be processed for the water phase correction and peak alignment. If all samples are processed, the PCDA analysis can be carried out within an hour, implying that this process can be done semiautomatically within 48 h after collection of the urine samples. In comparison, the histopathology usually will take at least 1 week before all data will become available.
Currently, the studies are highly dependent on histopathology for validation but it is conceivable that eventually this screening tool may be applied in a stand-alone manner, assuming the role of clinically relevant biomarkers being fully validated. Using toxic model compounds, a database can be built which classifies toxicity on the basis of specific biomarker patterns. In order for this type of screening to be successful, it is essential that the biochemical mechanisms leading to NMR changes are related to mechanisms of cell injury. The further validation for hepatotoxicity should proceed with the testing of other toxic liver compounds and identification of the NMR biomarkers. Moreover, nontoxic liver compounds should be tested as a negative control and to classify other types of toxicity.
|
SUPPLEMENTARY DATA |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
Supplementary data are available online at http://toxsci.oxfordjournals.org/.
|
ACKNOWLEDGMENTS |
|---|
We would like to thank Dr S. H. Moolenaar for her contribution on improvement of the manuscript and Mrs N. Bisseling and Mr J. van Orsouw for the preparation of all figures.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
Agar NS, Rae CD, Chapman BE, Kuchel PW. 1H NMR spectroscopic survey of plasma and erythrocytes from selected marsupials and domestic animals of Australia. Comp. Biochem. Physiol. (1991) 99:575–597.[CrossRef][Medline]
Bairaktari E, Katopodis K, Siampoulos KC, Tsolas O. Paraquat-induced renal injury studied by 1H nuclear magnetic resonance spectroscopy of urine. Clin. Chem. (1998) 44:1256–1261.
Beckwith-Hall BM, Nicholson JK, Nicholss AW, Foxall PJD, Lindon JC, Connor SC, Abdi M, Connelly J, Holmes E. Nuclear magnetic resonance spectroscopic and principal components analysis investigation into biochemical effects of three model hepatoxins. Chem. Res. Toxicol. (1998) 11:260–272.[CrossRef][Web of Science][Medline]
Beckwith-Hall BM, Holmes E, Lindon JC, Gounarides J, Vivkers A, Shpiro M, Nicholson JK. NMR based metabonomics studies on the biochemical effects of commonly used drug carrier vehicles in the rat. Chem. Res. Toxicol. (2002) 15:1136–1141.[CrossRef][Web of Science][Medline]
Binzak BA, Wevers RA, Moolenaar SH, Lee Y-M, Hwu W-L, Poggi-Bach J, Engelke UFD, Hoard HM, Vockley JG, Vockley J. Cloning of dimethylglycine dehydrogenase and a new inborn error of metabolism, dimethylglycine dehydrogenase deficiency. Am. J. Hum. Genet. (2001) 68:839–847.[CrossRef][Web of Science][Medline]
Bollard ME, Stanley EG, Lindon JC, Nicholson JK, Holmes E. NMR-based metabonomic approaches for evaluating physiological influences on biofluid composition. NMR Biomed. (2005) 18:143–162.[CrossRef][Web of Science][Medline]
Caldwell GW, Ritchie DM, Masucci JA, Hageman W, Yan Z. The new pre-clinical paradigm: Compound optimization in early and late phase drug discovery. Curr. Top. Med. Chem. (2001) 1:353–366.[CrossRef][Medline]
Casarett and Doull's Toxicology. The basic Sciences of poisons. Klaassen CD, Amdur MO, Macmillan JD, eds. (1986) 3rd ed. Publishing Company, New York, Collier Macmillan Canada, Inc. Toronto, Collier Macmillan Publishers London.
Chung Y-L, Wassif WS, Bell JD, Hurley M, Scott DL. Urinary levels of creatine and other metabolites in the assessment of polymyositis and dermatomyositis. Rheumatology (2003) 42:298–303.
Clayton TA, Lindon JC, Everett JR, Charuel C, Hanton G, Le Net J-L, Provost J-P, Nicholson JK. An hypothesis for a mechanism underlying hepatotoxin-induced hypercreatinuria. Arch. Toxicol. (2003) 77:208–217.[Web of Science][Medline]
Clayton TA, Lindon JC, Everett JR, Charuel C, Hanton G, Le Net J-L, Provost J-P, Nicholson JK. Hepatotoxin-induced hypercreatinaemia and hypercreatinuria: Their relationship to one another, to liver damage and to weakened nutritional status. Arch. Toxicol. (2004) 78:86–96.[CrossRef][Web of Science][Medline]
Clayton TA, Lindon JC, Cloarec O, Antti H, Charuel C, Hanton G, Provost J-P, Le Net J-L, Baker D, Walley RJ, Everett JR, Nicholson JK. Pharmaco-metabonomic phenotyping and personalized drug treatment. Nature (2006) 440:1073–1077.[CrossRef][Medline]
Cloarec O, Dumas ME, Trygg J, Craig A, Barton RH, Lindon JC, Nicholson JK, Holmes E. Evaluation of the orthogonal projection on latent structure model limitations caused by chemical shift variability and improved visualisation of biomarker changes in 1NMR spectrocopic metabonomic studies. Anal. Chem. (2005) 77:517–526.[Medline]
Coen M, Lenz EM, Nicholson JK, Wilson ID, Pognan F, Lindon JC. An integrated metabonomics investigation of acetaminophen toxicity in the mouse using NMR spectroscopy. Chem. Res. Toxicol. (2003) 16:295–303.[CrossRef][Web of Science][Medline]
Cuatrecasas P. Drug discovery in jeopardy. J. Clin. Invest. (2006) 116:2837–2842.[CrossRef][Web of Science][Medline]
Dixon MF, Fulker MJ, Walker BE, Kelleher J, Losowsky MS. Serum transaminase levels after experimental paracetamol-induced hepatic necrosis. Gut (1975) 16:800–807.
Doorn M, van Vogels J, Tas A, Hoogdalem EJ, van Burggraaf J, Cohen A, van der Greef J. Evaluation of metabolite profiles as biomarkers for the pharmacological effects of thiazolidinediones in Type 2 diabetes mellitus patients and healthy volunteers. Br. J. Clin. Pharmacol (2007) 63:562–574.[CrossRef][Web of Science][Medline]
Dyson EH, Will EJ, Davison AM, O'Malley AH, Shepherd HT, Jones RG. Use of the urinary protein creatinine index to assess proteinuria in renal transplant patients. Nephrol. Dial. Transplant. (1992) 7:450–452.
Fan T. Metabolite profiling by one- and two-dimensional NMR analysis of complex mixtures. Progr. NMR Spec. (1996a) 28:161–219.
Fan T. Recent advances in profiling plant metabolites by multinuclear and multidimensional NMR. In: Nuclear Magnetic Resonance in Plant Biology—Shachar-Hill Y, Pfeffer PE, eds. (1996b) Rockville, MD: American Society of Plant Physiologists. 181–254.
Gartland KPR, Bonner FW, Trimbel JA, Nicholson JK. Biochemical characterisation of para-aminophenol-induced nephrotoxic lesions in the F344 rat. Arch. Toxicol. (1989) 63:97–106.[CrossRef][Web of Science][Medline]
Gartland KPR, Beddell CR, Lindon JC, Nicholson JK. Application of pattern recognition methods to the analysis and classification of toxicological data derived from proton nuclear magnetic resonance spectroscopy of urine. Mol. Pharmacol. (1991) 39:629–642.[Abstract]
Govindaraju V, Young K, Maudsley AA. Proton NMR chemical shifts and coupling constants for brain metabolites. NMR Biomed. (2000) 13:129–153.[CrossRef][Web of Science][Medline]
Gower JC, Dijksterhuis GB. Procrustus Problems (2004) New York: Oxford statistical science series, 30, Oxford University Press.
Griffin JL, Walker LA, Shore RF, Nicholson JK. Metabolic profiling of chronic cadmium exposure in the rat. Chem. Res. Toxicol. (2001) 14:1428–1434.[CrossRef][Web of Science][Medline]
Héberger K, Andrade JM. Procrustus and pair-wise correlation: A parametric and a non-parametric method for variable selection. Croatica Chem. Acta (2004) 77:117–125.
Heijne WH, Lamers RJAN, van Bladeren PJ, Groten JP, van Nesselrooij JH, van Ommen B. Profiles of metabolites and gene expression in rats with chemically induced hepatic necrosis. Toxicol. Pathol. (2005) 33:425–433.[Web of Science][Medline]
Hetu C, Dumont A, Joly JG. Effect of chronic ethanol administration on bromobenzene liver toxicity in the rat. Toxicol. Appl. Pharmacol. (1983) 67:166–177.[CrossRef][Web of Science][Medline]
Holmes E, Nicholls AW, Lindon JC, Connor SC, Connelly JC, Haselden JN, Damment SJ, Spraul M, Neidig P, Nicholson JK. Chemometric models for toxicity classification based on NMR spectra of biofluids. Chem. Res. Toxicol. (2000) 13:471–478.[CrossRef][Web of Science][Medline]
Hoogerbrugge R, Willig SJ, Kistemaker PG. Discriminant analysis by double stage principal component analysis. Anal. Chem. (1983) 55:1710–1712.
Jollow D, Mitchell JR, Zampaglione N, Gillette JR. Bromobenzene induced liver necrosis. Protective role of glutathione and evidence for 3,4-bromobenzen oxide as the hepatotoxic metabolite. Pharmacology (1974) 11:151–169.[CrossRef][Web of Science][Medline]
Koen YM, Williams TD, Hanzlik RP. Identification of three protein targets for reactive metabolites of bromobenzene in rat liver cytosol. Chem. Res. Toxicol. (2000) 13:1326–1335.[CrossRef][Web of Science][Medline]
Kowalski BR. Measurement analysis by pattern recognition. Anal. Chem. (1975) 47:1152A.
Kowalski BR, Bender FC. The application of pattern recognition to screening prospective anticancer drugs. Adenocarcinoma biological activity test. J. Am. Chem. Soc. (1974) 96:916–918.[CrossRef][Web of Science][Medline]
Massart DL, Vandeginste BMG, Deming SN, Michotte Y, Kaufman L. Chemometrics a Textbook (1988) Amsterdam: Elsevier.
Miller NE, Thomas D, Billings RE. Bromobenzene metabolism in-vivo and in-vitro. The mechanism of 4-bromocatechol formation. Drug Metab. Dispos. (1990) 18:304–308.[Abstract]
Narayanan S, Appleton HD. Creatinine: A review. Clin. Chem. (1980) 26:1119–1126.
Nicholson JK, Lindon JC, Holmes E. Metabonomics: Understanding the metabolic response of living systems to pathophysiological stimuli via multivariate statistical analysis of biological NMR spectroscopic data. Xenobiotica (1999) 29:1181–1189.[CrossRef][Web of Science][Medline]
Nicholson JK, Connelly J, Lindon JC, Holmes E. Metabonomics: A platform for studying drug toxicity and gene function. Nat. Rev. (2002) 1:153–161.
Niesink RJM, de Vries J, Hollinger MA. Toxicology: Principles and Applications (1996) Boca Raton, FL: CRC Press, Inc.
Robertson DG, Reily MD, Sigler RE, Wells DF, Paterson DA, Braden TK. Metabonomics: Evaluation of nuclear magnetic resonance (NMR) and pattern recognition technology for rapid in vivo screening of liver and kidney toxicants. Toxicol. Sci. (2000) 57:326–337.
Schoonen WGEJ, Kloks CPAM, Ploemen J-PHTM, Horbach GJ, Smit MJ, Zandberg P, Mellema J-R, Thijssen-van Zuylen C, Tas AC, van Nesselrooij JHJ, Vogels JTWE. Uniform procedure of 1H NMR analysis of rat urine and toxicometabonomics. Part II: Comparison of NMR profiles for classification of hepatotoxicity. Toxicol. Sci. (2007) doi: 10.1093/toxsci/kfm077.
Schuster D, Laggner C, Langer T. Why drugs fail—a study on side effects in new chemical entities. Curr. Pharm. Des. (2005) 11:3545–3559.[CrossRef][Web of Science][Medline]
Siegers CP, Strubelt O, Schutt A. Relations between hepatotoxicity and pharmacokinetics of paracetamol in rats and mice. Pharmacology (1978) 16:273–278.[CrossRef][Web of Science][Medline]
Smith DA, Schmid EF. Drug withdrawals and the lessons within. Curr. Opin. Drug Discov. Devel. (2006) 9:38–46.[Web of Science][Medline]
Tiedeman DV, Tatsuoka MM, Langmuir CR. Multivariate Statistics for Personal Classification (1967) New York: John Wiley and Sons.
Tredger JM, Smith HM, Read RB, Portmann B, Williams R. Effects of ethanol ingestion on the hepatotoxicity and metabolism of paracetamol in mice. Toxicology (1985) 36:341–352.[CrossRef][Web of Science][Medline]
Vandenberghe J. Hepatotoxicology: Mechanisms of liver toxicity and methodological aspects. In: Toxicology, Principles and Applications—Niesink RJM, de Vries J, Hollinger MA, eds. (1996) chapter 23. Boca Raton, FL: CRC Press, Inc. 702–723.
Van der Greef J, Davidov E, Verheij ER, Vogel J, Heijden van der R, Adourian AS, Oresic M, Marple EW, Naylor S. The role of metabolomics in systems biology. In: Metabolic Profiling: Its Role in Biomarker Discovery and Gene Function Analysis—Harrigan GG, Goodacre R, eds. (2003) Boston, MA: Kluwer Academic Publishing. 170–198.
Vogels JTWE, Tas AC, van den Berg F, van der Greef J. A new method for classification of wines based on proton and carbon-13 NMR spectroscopy in combination with pattern recognition techniques. Chemometr. Intell. Lab. Syst. Lab. Inf. Manag. (1993) 21:249–258.
Vogels JTWE, Tas AC, Venekamp J, van der Greef J. Partial linear fit: A new NMR spectroscopy preprocessing tool for pattern recognition applications. J. Chemometrics (1996) 10:425–438.[CrossRef]
Vries de J. Cytotoxicity: Molecular mechanisms of cell death. In: Toxicology, Principles and Applications—Niesink RJM, Vries de J, Hollinger MA, eds. (1996) chapter 10. Boca Raton, FL: CRC Press, Inc. 702–723.
Waters NJ, Holmes E, Williams A, Waterfield CJ, Farrant RD, Nicholson JK. NMR and pattern recognition studies on the time-related metabolic effects of -naphtylisothiocyanate on liver, urine and plasma in the rat: An integrative metabonomic approach. Chem. Res. Toxicol. (2001) 14:1401–1412.[CrossRef][Web of Science][Medline]
Williams R, Davis M. Pathology and biochemistry of paracetamol induced liver damage. In: Clinical Toxicology—Duncan WA, Leonard BJ, eds. (1977) Amsterdam: Excerpta Medica.
Windig W, Haverkamp J, Kistemaker PG. Interpretation of sets of pyrolysis mass spectra by discriminant analysis and geographical rotation. Anal. Chem. (1983) 55:81–88.
Wittwer AJ, Wagner C. Identification of folate-binding proteins of mitochondria as dimethylglycine dehydrogenase. Proc. Natl. Acad. Sci. USA (1980) 77:4484–4488.
Wittwer AJ, Wagner C. Identification of folate binding protein of rat liver mitochondria as dimethylglycine dehydrogenase and sarcosine dehydrogenase: Flavoprotein nature and enzymatic properties of the purified protein. J. Bio. Chem. (1981) 256:4109–4115.
Wong SGW, Jeffrey W, Card W, Racz WJ. The role of mitochondral injury in bromobenzene and furosemide induced hepatoxicity. Toxicol. Lett. (2000) 116:171–181.[CrossRef][Web of Science][Medline]
Zampaglione N, Jollow DJ, Mitchell JR, Stripp B, Hamrick M, Gillette JR. Role of detoxifying enzymes in bromobenzene induced liver necrosis. J. Pharmacol. Exp. Ther. (1973) 187:218–227.
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||