Metabonomics: Evaluation of Nuclear Magnetic Resonance (NMR) and Pattern Recognition Technology for Rapid in Vivo Screening of Liver and Kidney Toxicants

Donald G. Robertson*,1, Michael D. Reily{dagger}, Robert E. Sigler*, Dale F. Wells*, David A. Paterson* and Timothy K. Braden*

* Departments of Worldwide Preclinical Safety and {dagger} Discovery Technologies, Parke-Davis Pharmaceutical Research, Division of Warner Lambert Company, Ann Arbor, Michigan

Received March 13, 2000; accepted May 24, 2000


    ABSTRACT
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
The purpose of this study was to evaluate the feasibility of metabonomics technology for developing a rapid-throughput toxicity screen using 2 known hepatotoxicants: carbon tetrachloride (CCl4) and {alpha}-naphthylisothiocyanate (ANIT) and 2 known nephrotoxicants: 2-bromoethylamine (BEA) and 4-aminophenol (PAP). In addition, the diuretic furosemide (FURO) was also studied. Single doses of CCl4 (0.1 and 0.5 ml/kg), ANIT (10 and 100 mg/kg), BEA (15 and 150 mg/kg), PAP (15 and 150 mg/kg) and FURO (1 and 5 mg) were administered as single IP or oral doses to groups of 4 male Wistar rats/dose. Twenty-four-h urine samples were collected pretest, daily through Day 4, and on Day 10 (high dose CCl4 and BEA only). Blood samples were taken on Days 1, 2, and 4 or 1, 4, and 10 for clinical chemistry assessment, and the appropriate target organ was examined microscopically. NMR spectra of urine were acquired and the data processed and subjected to principal component analyses (PCA). The results demonstrated that the metabonomic approach could readily distinguish the onset and reversal of toxicity with good agreement between clinical chemistry and PCA data. In at least 2 instances (ANIT and BEA), PCA analysis suggested effects at low doses, which were not as evident by clinical chemistry or microscopic analysis. Furosemide, which had no effect at the doses employed, did not produce any changes in PCA patterns. These data support the contention that the metabonomic approach represents a promising new technology for the development of a rapid throughput in vivo toxicity screen.

Key Words: NMR; metabonomics; pattern recognition; toxicity screening; carbon tetrachloride; {alpha}-naphthylisothiocyanate; 2-bromoethylamine; 4-aminophenol; nephrotoxicity; hepatic toxicity.


    INTRODUCTION
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
An emerging approach to enable high-throughput in vivo toxicology is metabonomics, a technology pioneered by Jeremy Nicholson, Elaine Holmes, and John Lindon at Imperial College in London (Holmes and Shockcor, 2000Go; Nicholson et al., 1999Go). Metabonomics combines the techniques of high resolution NMR with pattern recognition technology to rapidly evaluate the metabolic status of an animal, such that the onset, duration, severity, and target organ localization can all be determined from a peripheral sample such as urine. Toxicants, by definition, disrupt the normal composition and flux of endogenous biochemicals in, or through, key intermediary cellular metabolic pathways. These alterations may be reflected, either directly or indirectly, in the blood that percolates through the target tissues, which in turn may produce characteristic biomolecular traces in urine after filtration by the kidney. The diagnostic utility of any one trace biomolecule is limited, due to the number of variables affecting its concentration anywhere along its route to the urine, and by the commonality of biochemical processes disrupted by toxicants. However, if a significant number of trace molecules can be monitored, the overall pattern produced may be more consistent and predictive than any one marker. Obtaining such comprehensive biochemical information is made possible using high-field NMR equipment and, which is an essential component of metabonomics technology. Proton (1H) NMR spectroscopy will detect all soluble proton-containing molecules with a molecular weight of approximately 20 kD or less at concentrations > 100 µM, with signals providing both quantitative and qualitative (structural) information. The NMR spectra serve as the raw material for pattern recognition analyses, which simplifies the complex multivariate data into 2 or 3 dimensions that can be readily understood and evaluated. Acquiring this wealth of information non-invasively enables a potential for developing relatively high throughput, in vivo toxicity-screening technologies. The principles of the technique are described in detail elsewhere (Anthony et al., 1994aGo; Beckwith-Hall et al., 1998Go; Holmes et al., 1992aGo, 1998aGo; Holmes and Shockcor, 2000Go; Nicholson et al., 1999Go) and NMR analysis of biofluids has recently been reviewed in detail (Lindon et al., 1999Go). An extensive literature exists on the use of metabonomics to evaluate nephrotoxicants (Anthony et al., 1994bGo; Gartland et al., 1989aGo, bGo; Holmes et al., 1992bGo, 1998bGo), hepatic toxicants (Beckwith-Hall et al., 1998Go; Holmes et al., 1998aGo), and to a lesser extent, reproductive toxicants (Anthony et al., 1995Go; Gray et al., 1986Go). The purpose of this work was to evaluate metabonomics technology with the eventual goal of developing a rapid-throughput in vivo toxicity screen in rats, using urine samples. A byproduct of metabonomics studies is the capacity to identify potential biomarkers, which may be considered of even greater significance than the use of the technology for screening applications. However, biomarker identification was not the main objective of this study and will only be dealt with peripherally. Two hepatotoxins, carbon tetrachloride (CCl4 ) and {alpha}-naphthylisothiocyanate (ANIT) and 2 nephrotoxins, 2-bromoethylamine (BEA) and 4-aminophenol (PAP) were chosen for investigation. The toxicity of all 4 compounds is well established (Holmes et al., 1992bGo; Jean and Roth, 1995Go; Kossor et al., 1995Go; Pirmohamed and Park, 1996Go; Rao et al., 1997Go; Shao and Tarloff, 1996Go) and NMR evaluations of BEA, ANIT, and PAP in rats were previously published. (Anthony et al., 1994aGo, 1995Go; Beckwith-Hall, 1998; Holmes et al., 1992aGo, 1998aGo). Evaluation of ANIT, BEA, and PAP allowed comparison of results with published data to assess the reproducibility and robustness of the approach while evaluation of CCl4 added to the existing knowledge base with regard to the use of metabonomic technology to assess the effects of this classic hepatotoxicant. Unlike previous studies, an additional aspect of these investigations was to assess the dose-response relationship with regard to NMR assessment, in order to determine the sensitivity of the metabonomic approach. Two doses of each compound were assessed. The high dose was chosen to provide an overtly toxic dose as determined by classical clinical pathology and histopathology. The low dose chosen was anticipated to produce minimal clinical pathology effects or morphologic changes. In addition to these 4 compounds, an attempt was made to evaluate the effects of a high and low therapeutic dose of the diuretic furosemide (FURO) to determine what, if any, effects diuresis in the absence of toxicity would produce in the NMR spectra of urine.


    MATERIALS AND METHODS
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
Chemicals.
All chemicals were obtained from Sigma Chemical Co. (St. Louis, MO).

Animals.
Random bred, male albino Wistar (Crl:WI)BR were purchased from Charles River Laboratory (Raleigh, NC). Before being placed on study, animals were acclimated for 5 to 7 days in individual stainless steel wire-mesh cages in temperature (70–78°F) and humidity (30–70% RH) controlled rooms with a 12-h light cycle. Animals were fed (Purina Certified Rodent Chow®-5002) and watered ad libitum.

Study design.
To facilitate logistical considerations (primarily metabolism cage availability) the toxicants were administered in 2 separate studies. The hepatotoxicants, ANIT and CCl4, were administered in the first study, and the nephrotoxicants; BEA and PAP along with the diuretic FURO, were administered in the second study. The design of these studies is summarized in Table 1Go. ANIT and CCl4 were prepared in corn oil with ANIT administered as a single oral dose by gavage (10 ml/kg) and CCl4 administered as a single ip dose (10 ml/kg). FURO, BEA and PAP were prepared in 0.9% saline and administered ip in a dose volume of 10 ml/kg. Limited availability of appropriately sized metabolism cages necessitated the use of older (339–425 g) rats for the first experiment relative to the second experiment (246–322 g rats). The ages for each group of animals are indicated in Table 1Go. Each group was euthanized 4 days after dosing, except for one high-dose group of CCl4 and BEA-treated animals, along with their concurrent controls, which were euthanized10 days after dosing to better assess reversibility. For the purposes of this experiment, study Day refers to the 24-h period following dosing, during which urine was collected, with subsequent 24-h periods constituting subsequent days. Clinical pathology samples were collected at the end of the 24-h periods on their respective days.


View this table:
[in this window]
[in a new window]
 
TABLE 1 Study Design
 
Clinical pathology.
Blood was collected from each animal on Days 1 and 2 (Days 1 and 4 for Groups 1, 6, 7, and 14) via the jugular vein, under light isoflurane anesthesia, and from all animals via cardiac puncture at termination. Activities of serum alanine aminotransferase (ALT), aspartate aminotransferase (AST), alkaline phosphatase (AP), urea nitrogen (BUN), creatinine and creatinine kinase (CK) were determined on all samples. Additionally, serum bilirubin and gamma glutamyl transferase activities were determined on samples from Groups 1–6. All measurements were performed using standard automated clinical chemistry instrumentation.

Urine collection.
Twenty-four h prior to dosing, animals were placed in plastic metabolism cages (Harvard Apparatus, Holliston, Mass.) where they remained for the next 4 days. Food and water were available ad libitum. Urine was collected into cups with 1 ml of 1% sodium azide. Total urine volume was recorded each day for Groups 7–14. Collecting cups were maintained at 0°C using water-jacketed holders attached to refrigerated water circulators. Preliminary experiments demonstrated that both cold collection and sodium azide were necessary to prevent bacterial contamination from interfering with the NMR analyses. Urine samples were collected from all animals immediately prior to dose (pretest) and daily on Days 1–4. Animals in reversal groups were then returned to standard stainless-steel caging until Day 9 when they were returned to metabolism cages to collect a 24-h, Day-10 sample. After sample collection, urine was diluted approximately 60:40 urine:buffer (0.2 M sodium phosphate, pH 7.2) to minimize pH variations. Additionally, 7.5 volume percent of D2O and 2,2',3,3'-deuterotrimethylsilylproprionic acid (TSP) was added to a final concentration of 0.1 mM to provide an internal frequency lock and chemical shift reference, respectively.

Necropsy.
At termination, rats were euthanized by carbon dioxide inhalation. Liver and kidney samples were fixed in 10% formalin. Liver samples from Groups 1–6 and kidney samples from Groups 7–14 were further processed for light microscopic evaluation. One section from the left lateral lobe of each liver and kidney section, cut in the sagittal plane, were stained with hematoxylin and eosin and evaluated in a non-blinded fashion.

NMR analyses.
NMR free-induction decays (FIDs) were acquired on a Varian Inova 600 running VNMR software version 6.1B and equipped with a 1H-{15N, 13C} flow cell (120 µl active volume) and Varian automated sample transport (VAST) accessory. Using the VAST accessory, a total of 520 µl of the prepared urine samples was withdrawn from either foil-sealed polypropylene 96-deep-well plates or 2-ml septum-capped glass vials and pumped into the flow cell. Two complete cell washes with an approximately isotonic phosphate buffer were executed between each sample injection.

Final spectra were accumulations of 64 individual FIDs. Each FID was induced using a nonselective, 90-degree excitation pulse (4.3 µs @ 61 dB) following a selective soft pulse (1.5 s @ 3 dB) set on the water resonance and digitized into 32 K complex data points. A total inter-excitation pulse delay of 3.0 s was used and was initiated by sequential x and y trim pulses (1.0 ms @ 61 dB) to destroy residual transverse magnetization. A spectral sweep width of 6982.6 Hz resulted in an FID acquisition time of 4.679 s for a total recycle time of 7.7 s. Including 2 dummy scans that were discarded at the beginning of each acquisition, and approximately 3 min to change samples and wash the cell, the total experiment time was ~11 min/sample. NMR data were transferred to a Silicon Graphics Indigo workstation where they were processed with XWINNMR(V2.5, Bruker Instruments). The Varian data files were converted to Bruker format and the FIDs were multiplied by an exponential decay function (LB = 0.3) and converted to frequency domain using fast Fourier transformation (FFT). After FFT, the data were individually phased and subjected to polynomial baseline correction using the BASL routine in XWINNMR. Finally, the data were reduced by summing the intensity of all data points (without scaling) over 0.04 ppm (approximately 113 data points) regions using AMIX (V2.5, Bruker Instruments). Regions devoid of endogenous peaks at either end of the spectrum and the region 6.0–4.5 ppm containing urea and water resonances were excluded from data reduction. The resulting ~212 integrals/spectrum were output into an ASCII file for later statistical analysis.

Pattern recognition analyses.
The principles of pattern recognition techniques used in NMR spectroscopy are well described by El-Deredy et al. (1997) and the practical application of these techniques in metabonomic investigations is well documented (Anthony et al., 1994aGo; Beckwith-Hall et al., 1998Go; Holmes et al. 1992aGo Holmes et al. 1998). Briefly, the pattern recognition technique used in this study, principal component analysis (PCA), is an unsupervised multivariate statistical method useful for reducing multidimensional data (such as multiple NMR spectra) down to 2 or 3 dimensions that can readily be comprehended. The graphical representations presented utilize the first 2 or 3 principal components as the axes. In the graphical representations, toxicity- or physiologic-induced variations in spectral patterns are indicated by separation from the control region or centroid. The magnitude of response, severity in the case of toxicity, is proportional to the distance from the control centroid within any one PCA, but not necessarily across PCAs, for different compounds. Additionally, separation of centroids within the PCA is indicative of different NMR spectral patterns, which may involve differences in one or more analytes.

Reduced NMR spectra were labeled as to sample origin and assigned to class types (based on the compound and dose that the animal was given) and imported into Pirouette (V 2.6, Informetrix, Inc., Woodenville, WA). NMR spectra were normalized against the average total integrated intensity of all spectra within the same PCA analysis. A principal components analysis was then obtained on individual groups with controls, or on data from the entire study at once. The integral data were first either weighted against the mean or auto-scaled and then subjected to PCA with normalization transform (normalization = 100). For most results comparing groups within a drug treatment with controls, the first two principal components, PC1 and PC2, were sufficient to distinguish treated from control. Raw spectra were subjectively evaluated for the presence of parent compound or metabolites as determined by chemical shifts consistent with the structure of the compound or potential metabolite. In no case was the contribution of such peaks considered to have affected the PCA. However, minor contributions of xenobiotic metabolites to the PCA cannot be ruled out, and no attempt was made to separate or suppress the contribution of such metabolites in the overall PCA.

Statistics.
Clinical chemistry data were analyzed using one-way ANOVA with pairwise comparisons to vehicle control. The criterion for statistical significance was set at p < 0.05.


    RESULTS
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
Clinical Chemistry
Marked changes in clinical chemistry parameters were restricted to the high dose of each toxicant (Tables 2 and 3GoGo). High dose CCl4 toxicity was characterized by markedly elevated (over 30-fold) transaminases with 1 animal (d) having a moderately elevated bilirubin (1.4–1.5 mg/dL) on Days 1 and 2. ANIT, on the other hand, was characterized by markedly elevated total bilirubin (> 15-fold) in addition to elevated transaminases and, to a lesser extent, elevated GGT and AP. High doses of PAP and BEA produced moderate (up to 3-fold) elevations in both BUN and creatinine. Serum chemistry parameters for all toxicants were maximal within 24–48 h of dose, returning towards control levels by Day 4. One notable exception was a single CCl4-treated animal (a) that had normal ALT and AST levels on Days 1 and 2 (< 70 and 140, respectively) but had marked elevations in both parameters by Day 4 (484 and 693, respectively). No effects on serum chemistry parameters were noted on Day 10. High dose PAP and BEA treatments were associated with a profound diuresis as evidenced by 2–3-fold increases in daily urine output as early as Day 1. Unlike serum chemistry parameters, urine output remained elevated until termination (Day 4 for PAP and Day 10 for BEA). Two animals given 10 mg/kg ANIT had slightly elevated AP (265–275 IU/L) on Day 1. An additional animal had a slightly elevated bilirubin on Day 1 (0.8 mg/dl). These changes produced statistically significant elevations in AP and bilirubin on Day 1 in the 10 mg/kg ANIT-treated group, with both parameters returning to control levels subsequently. No other statistically significant changes were observed in low-dose-treated groups. Neither dose of FURO produced any effect on any serum chemistry parameter and urine volume was unaffected.


View this table:
[in this window]
[in a new window]
 
TABLE 2 Effect of Hepatotoxicants on Selected Clinical Chemistry Parameters (Mean ± SE)
 

View this table:
[in this window]
[in a new window]
 
TABLE 3 Effect of Nephrotoxicants on Selected Clinical Chemistry Parameters (Mean ± SE)
 
Histopathology
ANIT.
ANIT administered at 10 mg/kg resulted in minimal bile duct proliferation in 2 of 4 animals on Day 4 (Fig. 1bGo). In the 100 mg/kg ANIT group, moderate bile duct proliferation with an increase in Kupffer cells and fibrosis in portal areas was evident in all animals (Fig. 1bGo). In 2 of 4 animals, there was necrosis of periportal hepatocytes.



View larger version (140K):
[in this window]
[in a new window]
 
FIG. 1. Liver histopathology after ANIT and CCl4 administration. Control liver (a). ANIT (10 mg/kg) produced minimal bile duct proliferation 4 days after dosing as evidenced by increased numbers of basophilic cells (b). When administered at 100 mg/kg, ANIT produced moderate bile duct proliferation with hepatic necrosis and fibrosis in portal areas on Day 4 (c). Centrilobular region of liver, 4 days after treatment with 0.1 ml/kg CCl4, showing low incidence of minimal hepatocellular vacuolation and single cell necrosis (d). Centrilobular region of liver 4 days after administration of 0.5 mL/kg CCl4. Changes included centrilobular hepatocellular necrosis and vacuolation with Kupffer cell proliferation (e).

 
CCl4.
CCl4 at 0.1 ml/kg resulted in minimal hepatocellular vacuolation and single cell necrosis in 1 of 4 animals euthanized on Day 4 (Fig. 1dGo). This change could not be clearly defined as treatment related. At 0.5 mL/kg, hepatocellular changes ranged from mild to marked centrilobular hepatocellular necrosis and vacuolation and Kupffer cell proliferation on Day 4 that reversed by Day 10 (Fig. 1eGo).

PAP.
One of 4 animals administered PAP at 15 mg/kg had tubular basophilia in the renal cortex on Day 4 (Fig. 2bGo). All animals administered PAP at 150 mg/kg had tubular basophilia in the outer medulla and cortex with cellular casts and evidence of epithelial regeneration or reepithelization (Fig. 2cGo). Two of 4 high-dose animals also had tubular cellular swelling and necrosis in the outer medulla and cortex.



View larger version (170K):
[in this window]
[in a new window]
 
FIG. 2. Renal histopathology after PAP and BEA administration. Renal cortex of control kidney (a). PAP administered at 15 mg/kg induced a low incidence of tubular basophilia in the renal cortex 4 days after dosing. Tubular basophilia in this animal was confined to subcapsular area (b). All rats administered 150 mg/kg PAP had tubular epithelial degeneration and necrosis was associated cellular casts within tubules in the cortex and outer medulla 4 days after dosing (c). Renal papilla of control kidney (d). BEA (150 mg/kg) produced minimal dilatation of tubules in the renal papilla 4 days after dosing (e) with extensive renal papillary necrosis noted 10 days after dosing (f).

 
BEA.
One of 4 animals given BEA at 150 mg/kg had minimal dilatation of tubules in the renal papilla on Day 4 (Fig. 2eGo). No pathology was noted in animals given 15 mg/kg. Pathologic changes increased in incidence and severity with time. By Day 10, 2 of 4 high-dose animals had extensive renal papillary necrosis (Fig. 2fGo). The remaining 2 animals had no pathologic changes.

NMR
Although it is beyond the scope of this work to provide a detailed analysis of all of the NMR spectra utilized, the following highlights were noted:

ANIT.
The NMR spectra shown in Figure 3Go illustrate the changes in endogenous components of urine observed over the course of the study with a single rat dosed with 100 mg/kg ANIT. The predose urine spectrum is representative of naïve animals. In the 24 h period after dosing, there are clear reductions in intensity of resonances from 2-oxoglutarate, citrate and hippurate. By Day 2, these signals have continued to decrease, along with reduced levels of succinate and creatinine and an increase in creatine. Spectra returned towards pretest by Day 3 and on Day 4 were similar to pretest.



View larger version (27K):
[in this window]
[in a new window]
 
FIG. 3. The aliphatic region of the 600 MHz 1H NMR urine spectra from a single rat dosed with 100 mg/kg ANIT, showing changes in urine NMR spectrum with time. Major analytes that were affected by the treatment are labeled.

 
CCl4.
Similar to the case of ANIT, all of the animals showed a decrease in several Krebs cycle intermediates and hippurate within 24 h of dose; however, greater animal to animal variation in NMR spectra was evident. One rat (a) showed a muted, but gradual decrease in Krebs components over the course of 4 days, with little recovery noted. A second rat (d) appeared to be more severely affected, showing complete disappearance of citrate, succinate, and hippurate on Day 1, with slow recovery through Day 4. Correspondingly, PCA flagged these two individuals as "outliers" (see below).

PAP.
In contrast to the hepatotoxicants, PAP had no substantial effect on citrate, succinate, or 2-oxoglutarate, but hippurate and creatinine were reduced on Day 1, along with a significant increase in both glucose and several amino acid levels. Glucose and amino acid levels remained high on Day 2 and gradually returned to normal by Day 4.

BEA.
Animals treated with BEA consistently displayed large reductions in urine taurine, 2-oxoglutarate, hippurate, and succinate on Day 1. By Day 2, succinate and taurine levels had recovered, and moderate reductions in citrate were observed. High creatine levels were observed in all animals on Day 2 only.

Pattern Recognition
The results of pattern recognition analyses are presented graphically in Figures 4–10GoGoGoGoGoGoGo. A total of 296 samples were analyzed from which 23 were excluded from PCA analysis. Of these, 20 were excluded due to bacterial contamination with the other 3 excluded due to spectral acquisition difficulties. Data for Figures 4–8GoGoGoGoGo and Figure 10Go are presented using only the first 2 principal components, since that was sufficient to demonstrate pattern separation and made for comprehensible visual representation. Addition of the third principal component as a third dimension (as done in Fig. 9Go) enhanced separation of patterns, but became too complex to easily visualize. Despite assessing data in only 2 dimensions, it was clear that all 4 toxicants produced distinct patterns of toxicity as determined by PCA analyses.



View larger version (25K):
[in this window]
[in a new window]
 
FIG. 4. PCA analysis of urine spectra from rats treated with ANIT (100 mg/kg except as noted). Data are plotted as the first 2 principal components (PC1 and PC2). Number designations indicate sample days and the letter indicates an individual animal. Individual animal data from the same sample day are joined by shaded polygons. Pretest and Day 1 data from animals treated with 10 mg/kg ANIT (open and closed circles, respectively) are also included.

 


View larger version (28K):
[in this window]
[in a new window]
 
FIG. 5. PCA analysis of urine spectra from rats treated with CCl4 (0.5 mL/kg). Data are presented as in Figure 3Go. Circled data are from a single animal (animal d) with unusually high ALT levels and elevated serum bilirubin, which was not characteristic of other animals. Boxed data are from an animal (animal a) with delayed onset (Days 3 and 4) of ALT elevation relative to other animals. Corresponding ALT levels (when available) are indicated next to the circles or boxes.

 


View larger version (24K):
[in this window]
[in a new window]
 
FIG. 6. PCA analysis of urine spectra from rats treated with PAP (150 mg/kg). Data are presented as in Figure 3Go. Circled data are from a single animal that had minimal serum creatinine elevations on Day 1 (0.7 mg/dl) with normal levels (<= 0.6 mg/dl) at subsequent time points.

 


View larger version (24K):
[in this window]
[in a new window]
 
FIG. 7. PCA analysis of urine spectra from rats treated with BEA (150 mg/kg). Data are presented as in Figure 3Go. Pretest and Day 1 data from animals treated with 15 mg/kg (open and closed circles respectively) are also included.

 


View larger version (10K):
[in this window]
[in a new window]
 
FIG. 8. PCA analysis of urine spectra from rats treated with FURO (5 mg/kg). Data are presented as in Figure 3Go, except polygons are omitted for clarity.

 


View larger version (36K):
[in this window]
[in a new window]
 
FIG. 9. Combined PCA analysis of all toxicant treatments on Day 2 (filled symbols), excluding outliers (see text), and all untreated samples, both pretest samples and controls, at all time points (open circles). Toxicant data are as follows: Squares = CCl4, hexagons = ANIT, diamonds = PAP and triangles = BEA. Data are presented in 3 principal component dimensions (PC1, PC2 and PC3).

 


View larger version (22K):
[in this window]
[in a new window]
 
FIG. 10. Combined PCA analysis of all untreated samples (both controls and pretest). Filled squares = data from 8-week-old rats. Open circles = data from 13-week-old rats. Insets: within each age group, there was a suggestion of separation between naïve pretest rats (open symbols) and vehicle treated controls (filled symbols).

 
ANIT.
PCA evaluation of urine from ANIT-treated rats revealed relatively tight clustering of sample results on each sample day (Fig. 4Go). Severity, as determined by distance from pretest samples, was maximal on Day 2, corresponding to the day of maximum serum bilirubin elevation. Subsequently, reversal of sample trajectory towards pretest was associated with reduction in group-mean serum bilirubin. At the 10 mg/kg dose, a pattern shift was also evident on Day 1. The trajectory of the change was similar to that observed for Day 1 at 100 mg/kg. Subsequent data fell within the perimeter of pretest and control samples (data not shown).

CCl4.
Of the 4 compounds studied, CCl4 (0.5 ml/kg) had the greatest dispersion of PCA data. Yet, a distinct pattern was clearly evident with maximal effect, as determined by distance from pretest, on Days 1 and 2. By Day 10, all 4 animals had reversed trajectory completely (Fig. 5Go). These results were in agreement with clinical chemistry data, where the greatest transaminase elevations were noted on Days 1 and 2 (Table 3Go) with rapid regression to pretest levels by Day 4. One animal (animal d in Fig. 5Go) had unusually high ALT and AST elevations on Day 1. It also had a corresponding elevation in serum bilirubin on Days 1 and 2 of 1.4 and 1.5 mg/dl, respectively, which returned to normal (0.4 mg/dl) by Day 4. Serum bilirubin for all other CCl4-treated animals did not exceed 0.4 mg/dl at any time point. This animal had a PCA pattern that was clearly distinct from all other samples on Days 1, 2, and 3 (circled data points in Fig. 5Go). All animals, with one exception, had maximal transaminase elevations on Day 1. One animal (rat a in Fig. 5Go) had normal ALT levels on Days 1 and 2 (67 mg/dl), but markedly elevated ALT and AST on Day 4 (484 and 693, respectively). Correspondingly, the PCA pattern from that animal showed maximum separation from pretest on Days 3 and 4. No definitive PCA pattern, distinct from pretest or control, was evident in animals administered 0.1 ml/kg CCl4 (data not shown).

PAP.
PAP (150 mg/kg) produced a distinct PCA pattern that was most severe (as determined by distance from pretest) on Day 1, with little change for 3 of 4 rats noted in the PCA pattern on Day 2 (Fig. 6Go). Subsequent samples returned towards pretest. Similarly, maximal elevations of BUN, creatinine, and urine volume were noted during Days 1 and 2, with regression of severity noted by Day 4. PCA analyses of animal b in Figure 6Go revealed an initial marked pattern shift on Day 1 followed by rapid regression towards control, with the Day 3 and Day 4 results similar to pretest. Correspondingly, this animal only had slightly elevated Day 1 creatinine (0.7 mg/dl) and BUN (34 mg/dl), which returned to control levels (<=0.6 mg/dl and <=23 mg/dl, respectively) by Day 2. Compared to pretest or control samples, no distinct PCA pattern was evident in the 15 mg/kg PAP-treated group, which corresponded with normal serum chemistry parameters.

BEA.
BEA produced the tightest clustering of PCA data on each of the individual sampling days. Severity, as determined by distance from pretest, was maximal on Days 1 and 2, with subsequent regression towards control by Day 10 corresponding to maximum BUN and creatinine elevations and urine volumes on Day 1, with subsequent regression towards pretest (Fig. 7Go). The PCA pattern of the 15 mg/kg BEA-treated rats on Day 1 also appeared to shift in the same direction as the 150 mg/kg group, although some overlap was evident with the pretest pattern. Subsequent samples were indistinguishable from control or pretest.

FURO.
Furosemide did not produce any distinguishable pattern at either dose; these results correspond to its lack of effect on urine volume, BUN, and creatinine. Evidently the FURO doses employed were too low to produce the anticipated physiologic effect (Fig. 8Go).

Combined PCA Analysis.
PCA analyses were conducted on the combined data set of control (including pretest) and Day 2 samples (excluding outliers indicated above). Only Day 2 samples were used, to simplify the analysis, since it was the time point of maximum severity with all treatments considered together, and effects from any potential xenobiotic metabolites would be minimized relative to Day 1. The combined analysis revealed that all 4 compounds clustered in localized regions within metabolic space defined by the first 3 principal components, and furthermore, these regions were clearly separated from combined pretest and control samples from both studies (Fig. 9Go).

Control samples.
When concurrent controls from the 2 studies were merged for PCA analysis, a distinct separation in PCA pattern between the 2 studies was noted that was evident even in naïve (pretest) animals (Fig. 10Go). Therefore, the difference was not due to the vehicle alone (corn oil for Experiment 1 and saline for Experiment 2). The only other difference between the two sets of control animals was the age (and hence weight) of the animals. Even within the same-aged animals, there was a tendency for the PCA patterns from vehicle-treated rats to be right-shifted relative to pretest (Fig. 10Go, insets). Comparison of the NMR spectra of the 2 groups of control rats revealed that their pattern separation was due to subtle differences in levels of creatinine and taurine, although contributions from other constituents cannot be ruled out.


    DISCUSSION
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
The data generated from this study fully support use of the metabonomic approach for development of rapid-throughput in vivo toxicology screening. Analysis, after sample collection, can be automated using robotic liquid handling and NMR-flow probe technology. Manual inspection of the spectra represents the most labor-intensive aspect of the analysis, taking an average of 2–3 min per spectrum. Although the PCA analyses and evaluation reported here were done manually, they could be automated as well. Subsequent to these studies, instrument analysis time was reduced from 11 to 7 min. Therefore, analysis of up to 200 samples per day is feasible.

A detailed comparison of NMR spectra obtained in this study, compared to previously published work, is beyond the scope of this investigation. However, in general, biochemical changes noted in NMR spectra in our investigations were similar to those published previously (Beckwith-Hall et al., 1998Go; Gartland et al., 1989aGo,bGo; Holmes et al., 1992aGo). Differences in methodologies prevent direct comparison of PCA data; however, it was clear that our data are in agreement with the earlier work. This is true despite the fact that only simple PCA analyses on mean-center weighted data were used to evaluate results, and no attempt was made to suppress contributions from xenobiotic metabolites.

In general, all 4 compounds produced clinical and microscopic pathology consistent with previous studies. (Anthony et al., 1994aGo, 1995Go; Holmes et al., 1992aGo; Kossor et al., 1995Go; Pirmohamed and Park, 1996Go; Rao et al., 1997Go; Roth and Jean, 1995; Shao and Tarloff, 1996Go) and descriptions of their effects will not be further elaborated. The PCA data were in good agreement with the clinical chemistry data; however, the data also suggest that metabonomics technology is more than just an expensive clinical chemistry analysis method. This conclusion was demonstrated with the 150 mg/kg BEA-treated animals, where only 5 of 8 had Day 1 BUN or creatinine levels outside the normal range, while all 8 treated animals exhibited diuresis (1.5–3.5-fold increases in urine volume relative to pretest). PCA patterns were clearly indicative of a consistent effect in all 8 animals. Although it could be argued that more sensitive clinical markers of nephrotoxicity could readily have determined the presence of kidney lesions in all animals on Day 1, this work clearly demonstrates that utilization of the metabonomics approach eliminates the need for a priori assay selection and is at least as effective as standard clinical chemistry indices. The shift in PCA pattern, noted on Day 1 with low doses of ANIT or BEA, in the absence of any clinical chemistry findings suggests that, at least in some instances, the metabonomics approach may be more sensitive than traditional clinical chemistry. This raises the question of whether the technology may be too sensitive for a screening methodology. Obviously any screen that produces a 100% hit rate is useless. However, the furosemide data served as an unintentional negative control, since, associated with the absence of any PCA effect, toxicologic or physiologic findings were not noted at either dose. The selectivity of the technique was highlighted by the fact that in both the CCl4- and PAP-treated groups, the rats that were clearly outliers in either time or severity of response, were readily identified by the technique.

This study also demonstrates that PCA analyses were more closely associated with function than morphology, in that the NMR data aligned more closely with clinical chemistry data than with histopathology. This was most amply demonstrated in the high-dose BEA data, where morphology was clearly most disrupted at Day 10 (at least in 2 animals) while both clinical chemistry data and PCA data were returning toward pretest levels. This is somewhat an artifact of the timing of the microscopic analyses since specimens were obtained only on Days 4 and 10, and a full temporal microscopic evaluation would have provided a better correlation. Regardless, it was clear that the PCA data were more consistent than either the clinical chemistry or pathology data with regard to temporal toxicity, which has implications as to the predictive power of PCA at a given point in time, relative to traditional assessment. In all treatments only 1 low-dose-treated animal (administered 15 mg/kg PAP) had a definitive toxicant-induced lesion that was not picked up by PCA. This probably has more to do with the simple pattern recognition approach employed in our studies, rather than any inherent insensitivity of metabonomics technology.

Figure 9Go provides a glimpse of where the future of metabonomic screening technology lies, as proposed by Nicholson et al. (1999). Drugs producing toxicity via discrete pathophysiologic mechanisms will produce distinct patterns of toxicity. Theoretically a 3 (or more) dimensional model of metabolic space could be established, which could then be used for predicting target organs. Metabonomics has already been used to develop a precise model for correct classification of hepatic, renal, and reproductive toxicity via urine samples (Anthony et al., 1995Go). This type of approach could be largely automated such that numerous urine samples (> 200/day) could be screened and characterized. The caveat to such an approach is that a comprehensive spectral database, covering all or most potential target organs will need to be developed. This represents a fairly Herculean task and establishment of a multicompany consortium to generate such a database is under consideration (Car and Robertson, 1999Go).

The metabonomics approach allows for a wealth of toxicity data to be collected non-invasively using a relatively simple study protocol. The practical difficulties associated with the technology have been previously discussed (Robertson and Bulera, 2000Go). Although screening techniques have been emphasized in this report, the NMR spectra themselves can be used to obtain even more significant information which could be used to identify mechanisms of toxicity (Gartland et al., 1989aGo; Holmes et al., 1992bGo) or to identify novel biomarkers of toxicity (Holmes et al., 1998bGo). We used a simple PCA pattern recognition approach to assess our data. However, numerous chemometric approaches are available for even more sophisticated analyses of NMR spectra (Anthony et al., 1995Go; Holmes et al., 1994Go Holmes et al., 1998; Spraul et al., 1994Go). The advantages of obtaining such a vast quantity of information from a simple urine sample are obvious. However, this strength is also its greatest weakness. The problem revolves around the inability to condense such a vast quantity of information into a concise, comprehensible metric of toxicity. This will be an absolute requirement for any screening methodology. Additionally, physiologic effects (age, estrus, diet, etc), and pharmacologic effects (renal xenobiotic clearance, induction, diuresis, hypotension, etc) will need to be separated from toxicity. These concerns were exemplified by the distinctly different PCA patterns observed between 8- and 13-week-old rats in this study. Assuming the PCA shift was truly age-related, the ability to resolve such a difference via urine is fascinating; however, the ability to distinguish animals varying in age by only 4 weeks will need to be confirmed. Even more remarkable, though certainly more speculative, was the observation of an apparent pattern shift induced by the vehicle. Although the PCA shifts due to age and vehicle were clearly less profound than toxicity-induced changes in PCA patterns, the finding raises a cautionary note about the necessity of proper control and interpretation of metabonomics data. A metabonomic database obtained from rats under various physiologic and pharmacologic conditions, analogous to the database of know toxicants discussed earlier, will aid in distinguishing toxic from homeostatic or therapeutic responses. However, the problems facing metabonomics are no different from the issues facing toxicogenomic or proteomic initiatives and in some ways are more easily addressed. With metabonomics, specific biochemical markers can be readily associated with individual pattern changes allowing for generation of specific testable mechanistic hypotheses. However, if the biochemical changes are secondary or tertiary to the mechanistic event, the problem becomes somewhat more intractable.

In conclusion, the results of this study support the utilization of metabonomics technology to generate an in vivo screening method for assessment of toxicity. The results also highlight the need for more extensive testing and validation of the approach, with both known and novel toxicants as well as under varying physiological and environmental conditions.


    ACKNOWLEDGMENTS
 
The authors wish to thank Dr. William Pool for bringing the subject of NMR in toxicology and the work of Imperial College to their attention. The authors are also grateful to Dr. Elaine Holmes of Imperial College for her aid with pattern recognition and PCA analyses.


    NOTES
 
1 To whom correspondence should be addressed at the Department of Pathology and Experimental Toxicology, Parke-Davis Pharmaceutical Research, 2800 Plymouth Rd., Ann Arbor, MI 48106–1047. Fax: (734) 622-3478. E-mail: donald.robertson{at}wl.com. Back


    REFERENCES
 TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 REFERENCES
 
Anthony, M. L., Gartland, K. P., Beddell, C. R., Lindon, J. C., and Nicholson, J. K. (1994a). Studies of the biochemical toxicology of uranyl nitrate in the rat. Arch. Toxicol. 68, 43–53.[ISI][Medline]

Anthony, M. L., Rose, V. S., Nicholson, J. K., and Lindon, J. C. (1995). Classification of toxin-induced changes in 1H NMR spectra of urine, using an artificial neural network. J. Pharm. Biomed. Anal. 3, 205–211.

Anthony, M. L., Sweatman, B. C., Beddell, C. R., Lindon, J. C., and Nicholson, J. K. (1994b). Pattern recognition classification of the site of nephrotoxicity based on metabolic data derived from proton nuclear magnetic resonance spectra of urine. Mol. Pharmacol. 46, 199–211.[Abstract]

Beckwith-Hall, B. M., Nicholson, J. K., Nicholls, A. W., Foxall, P. J., Lindon, J. C., Connor, S. C., Abdi, M., Connelly, J., and Holmes, E. (1998). Nuclear magnetic resonance spectroscopic and principal components analysis investigations into biochemical effects of three model hepatotoxins. Chem. Res. Toxicol. 11, 260–272.[ISI][Medline]

Car, B. D., and Robertson, R. T. (1999). Commentary: Discovery toxicology—a nascent science. Toxicol. Pathol. 27, 481–483.[ISI][Medline]

El-Deredy, W. (1997). Pattern recognition approaches in biomedical and clinical magnetic resonance spectroscopy: A review. NMR Biomed. 10, 99–124.[ISI][Medline]

Gartland, K. P., Bonner, F. W., and Nicholson, J. K. (1989a). Investigations into the biochemical effects of region-specific nephrotoxins. Mol. Pharmacol. 35, 242–250.[Abstract]

Gartland, K. P., Bonner, F. W., Timbrell, J. A., and Nicholson, J. K. (1989b). Biochemical characterisation of para-aminophenol-induced nephrotoxic lesions in the F344 rat. Arch. Toxicol. 63, 97–106.[ISI][Medline]

Gray, J. A., Nicholson, J. K., and Timbrell, J. A. (1986). Creatinuria as an early indicator of cadmium-induced testicular damage. Hum. Toxicol. 5, 402A.

Holmes, E., Bonner, F. W., Sweatman, B. C., Lindon, J. C., Beddell, C. R., Rahr, E., and Nicholson, J. K. (1992a). Nuclear magnetic resonance spectroscopy and pattern recognition analysis of the biochemical processes associated with the progression of and recovery from nephrotoxic lesions in the rat induced by mercury(II) chloride and 2-bromoethanamine. Mol. Pharmacol. 42, 922–930.[Abstract]

Holmes, E., Foxall, P. J., Nicholson, J. K., Neild, G. H., Brown, S. M., Beddell, C. R., Sweatman, B. C., Rahr, E., Lindon, J. C., Spraul, M., and Neidig, P. (1994). Automatic data reduction for analysis of 1H nuclear magnetic resonance spectra of human urine from normal and pathological states. Anal. Biochem. 220, 284–296.[ISI][Medline]

Holmes, E., Nicholls, A. W., Lindon, J. C., Ramos, S., Spraul, M., Neidig, P., Connor, S. C., Connelly, J., Damment, S. J., Haselden, J., and Nicholson, J. K. (1998a). Development of a model for classification of toxin-induced lesions using 1H NMR spectroscopy of urine combined with pattern recognition. NMR Biomed. 11, 235–244.[ISI][Medline]

Holmes, E, Nicholson, J. K., Bonner, F. W., Sweatman, B. C., Beddell, C. R., Lindon, J. C., and Rahr, E. (1992b). Mapping the biochemical trajectory of nephrotoxicity by pattern recognition of NMR urinanalysis. NMR Biomed. 5, 368–372.[ISI][Medline]

Holmes, E., Nicholson, J. K., Nicholls, A. W., Lindon, J. C., Connor, S. C., Polley, S., and Connelly, J. (1998b). The identification of novel biomarkers of renal toxicity using automatic data reduction techniques and PCA of proton NMR spectra of urine. Chemomet. Intel. Lab. Syst. 44, 245–255.

Holmes, E., and Shockcor, J. P. (2000). Accelerated toxicity screening using NMR and pattern recognition-based methods. Curr. Opin. Drug Discovery Dev. 3, 72–78.

Jean, P. A., and Roth, R. A. (1995). Naphthylisothiocyanate disposition in bile and its relationship to liver glutathione and toxicity. Biochem. Pharmacol. 50, 1469–1474.[ISI][Medline]

Kossor, D. C., Goldstein, R. S., Ngo, W., DeNicola, D. B., Leonard, T. B., Dulik, D. M., and Meunier, P. C. (1995) Biliary epithelial cell proliferation following {alpha}-naphthylisothiocyanate (ANIT) treatment: Relationship to bile duct obstruction. Fundam. Appl. Toxicol. 26, 51–62.[ISI][Medline]

Lindon, J. C., Nicholson, J. K., and Everett, J. R. (1999). NMR spectroscopy of biofluids. Ann. Rep. NMR Spec. 38, 1–88.

Nicholson, J. K., Lindon, J. C., and Holmes, E. (1999). "Metabonomics": Understanding the metabolic responses of living systems to pathophysiological stimuli via multivariate statistical analysis of biological NMR spectroscopic data. Xenobiotica 29, 1181–1189.[ISI][Medline]

Pirmohamed, M., and Park, B. K. (1996). Mechanisms of hypertransaminemia. In Drug-induced Hepatotoxicity (R. G. Cameron, G. Feuer, and F. A. de la Iglesia, Eds.), pp. 424–425. Springer, New York.

Rao, P. S., Mangipudy, R. S., and Mehendale, H. M. (1997). Tissue injury and repair as parallel and opposing responses to CCl4 hepatotoxicity: A novel dose response. Toxicol. 118, 181–193.[ISI][Medline]

Robertson, D. G., and Bulera, S. J. (2000). High-throughput toxicology: practical considerations. Curr. Opin. Drug Discovery Dev. 3, 42–47.

Shao, R., and Tarloff, J. B. (1996). Lack of correlation between para-aminophenol toxicity in vivo and in vitro in female Sprague-Dawley rats. Fundam. Appl. Toxicol. 31, 268–278.[ISI][Medline]

Spraul, M., Neidig, P., Klauck, U., Kessler, P., Holmes, E., Nicholson, J. K., Sweatman, B. C., Salman, S. R., Farrant, R. D., Rahr, E., Beddell, C. R., and Lindon, J. C. (1994). Automatic reduction of NMR spectroscopic data for statistical and pattern recognition classification of samples. J. Pharm. Biomed. Anal. 12, 1215–1225.[ISI][Medline]