Abstract
Drug-induced nephrotoxicity is a major concern, since many pharmacological compounds are filtered through the kidneys for excretion into urine. To discover biochemical biomarkers useful for early identification of nephrotoxicity, metabolomic experiments were performed on Sprague-Dawley Crl:CD (SD) rats treated with the nephrotoxins gentamicin, cisplatin, or tobramycin. Using a combination of gas chromatography/mass spectrometry (GC/MS) and liquid chromatography/mass spectrometry (LC/MS), a global, nontargeted metabolomics analysis was performed on urine and kidney samples collected after one, five, and twenty-eight dosing days. Increases in polyamines and amino acids were observed in urine from drug-treated rats after a single dose, and prior to observable histological kidney damage and conventional clinical chemistry indications of nephrotoxicity. Thus, these metabolites are potential biomarkers for the early detection of drug-induced nephrotoxicity. Upon prolonged dosing, nephrotoxin-induced changes included a progressive loss of amino acids in urine, concomitant with a decrease in amino acids and nucleosides in kidney tissue. A nephrotoxicity prediction model, based on the levels of branched-chain amino acids in urine, distinguished nephrotoxin-treated samples from vehicle-control samples, with 100%, 93%, and 70% accuracy at day 28, day 5, and day 1, respectively. Thus, this panel of biomarkers may provide a noninvasive method to detect kidney injury long before the onset of histopathological kidney damage.
Introduction
The kidney is one of the primary sites of drug toxicity. It is especially vulnerable to toxic insults by drugs because of the high renal blood flow and because it transports, metabolizes, and concentrates chemicals present in the tubular fluid (Bennett 1997). Drugs are known that damage the kidney glomerulus (puromycin and adriamycin), papilla (NSAIDs), and various segments of the proximal tubule (aminoglycosides, cyclosporine, and cisplatin) (Werner et al. 1995).
It is estimated that the development of 30% of new drug candidates is halted because of unforeseen toxicity profiles and side effects in clinical studies (Thukral et al. 2005). However, despite the kidneys being affected by many drugs, the tests available to detect toxicity and early kidney injury are either invasive and difficult to quantify, or noninvasive but nonspecific and insensitive. Routinely used measures of renal function, such as levels of blood urea nitrogen (BUN) and serum creatinine, have severe limitations. They are not region-specific and they increase significantly only after substantial kidney injury occurs, generally after a loss of two thirds or greater of nephron functional capacity (Davis and Kramer 2006; Hewitt et al. 2004; Thukral et al. 2005). Thus it has been recognized that better biomarkers for kidney injury are needed both for animal studies and for use in humans, where early detection of kidney injury will influence therapy and potentially morbidity and mortality (Holmes et al. 2000; Niemann and Serkova 2007). In addition, the availability of biomarkers for the early detection of nephrotoxicity during preclinical drug development programs would provide significant savings in cost, time, and animal experiments.
In the past decade, several efforts have been undertaken to identify better and earlier markers of nephrotoxicity using genomics and proteomics approaches (Amin et al. 2004; Devarajan 2008; Kramer et al. 2004; Thukral et al. 2005). Based on much of this genomics and proteomics work, the US Food and Drug Administration (FDA) and the European Medicines Agency (EMEA), together with the C-Path Predictive Safety Testing Consortium (PSTC), recently accepted seven new biomarkers to evaluate kidney damage. The new biomarkers are KIM-1, Albumin, Total Protein, β2-Microglobulin, Cystatin C, Clusterin, and Trefoil Factor-3. Those new markers are more sensitive and can detect damage earlier than BUN and creatinine levels.
Metabolomics has recently attracted increasing interest in the field of toxicology, since it has proven to be a fast and reproducible method directly reflecting biological events (Dieterle et al. 2006). Metabolomics involves the determination of changes in the levels of endogenous or exogenous metabolites in biological samples, owing to physiological stimuli or genetic modification (Nicholson et al. 1999). The power of metabolomics lies in the global determination of metabolites, or patterns of biomarkers that increase or decrease as the result of an observed drug toxicity or in relation to a particular disease (Schnackenberg et al. 2007). The use of peripheral fluids for metabolomic analysis, such as urine and plasma, makes it an attractive method for studying the toxic effects of different drugs. It is therefore intriguing to determine whether metabolomics could lead to the discovery of biomarkers that may provide additional sensitivity or earlier detection of kidney damage than classical analytical techniques or histopathology evaluation.
A few recent studies have been reported on the metabolomic analysis of nephrotoxins, with the goal of identifying new metabolic biomarkers of nephrotoxicity, or to better understand their individual mode of action. Jia et al. (2008) studied chronic renal failure through a metabolomic analysis of human serum and identified seven potential biomarkers of renal failure: creatinine, tryptophan, phenylalanine, kynurenine, and three lysophosphatidylcholines. Xu et al. (2008) investigated urine and kidney transcriptomic profiles of rats treated with two nephrotoxins, which induced Fanconi-like syndromes. Gartland et al. (1989) and Anthony et al. (1994) tested the utility of metabolomics for classifying nephrotoxins based on the specific region of the kidney that is targeted by a given compound. A further application was investigated by Dieterle et al. (2006), who reported a metabolomics analysis in a compound ranking study, which revealed increased excretion of choline in rat urine. Several other reports have been published on the metabolomic analysis of individual nephrotoxins, such as cadmium chloride (Griffin et al. 2001), mercury chloride (Holmes et al. 2000), gentamicin (Lenz et al. 2005), cisplatin (Portilla et al. 2006), and aristolochate (Ni et al. 2007), among others. Finally, a large comprehensive study was reported from the Consortium on Metabonomic Toxicology (COMET) (Ebbels et al. 2007), in which many nephrotoxins were analyzed, to build prediction and classification models for drug toxicity. However, the report did not mention any metabolites by name.
Gentamicin, cisplatin, and tobramycin are three well-characterized nephrotoxins. Administration of gentamicin, the prototype aminoglycoside antibiotic, induces proximal tubular necrosis in rodents (Houghton et al. 1986). Tobramycin belongs to the aminoglycoside antibiotics, as well, and induces tubular necrosis, although generally less severe than gentamicin (Kahlmeter et al. 1984; Kepczyk et al. 1990; Schentag et al. 1981). Cisplatin is an antineoplastic agent, used in the treatment of a variety of solid tumors, that induces severe tubular toxicity (Dobyan et al. 1980).
In the present study, we used a combination of gas chromatography/mass spectrometry (GC/MS) and liquid chromatography/mass spectrometry (LC/MS) to monitor the biochemical changes induced in laboratory Sprague-Dawley Crl:CD (SD) rats after one, five, and twenty-eight days of daily exposure to the model nephrotoxicants gentamicin, cisplatin, and tobramycin. The aims of the study were to (1) analyze the metabolic changes in urine and kidney tissue after dosing with three proximal tubule nephrotoxins; (2) determine which metabolites showed early versus late-stage changes; and (3) identify a panel of biomarkers for early detection of proximal tubule nephrotoxicity. In addition to the metabolomic analysis, plasma and kidney tissue were analyzed using clinical chemistry and histopathology so that metabolite changes could be correlated with kidney tissue damage. We report novel metabolic biomarkers of nephrotoxicity, which can detect damage to the kidney tubules earlier than clinical chemistry and histopathology.
Materials and Methods
Test Animals, Study Design, and Sampling
The animal study was performed at the IDRI (IVAX Drug Research Institute Ltd., Budapest, Hungary) and was sponsored by Teva Pharmaceutical Industries Ltd. (Netanya, Israel). The approval of the Animal Ethics Committee of IDRI was obtained prior to the commencement of the study. The study was conducted according to the Guide for the Care and Use of Laboratory Animals, NRC, 1996, and in compliance with the principles of the Hungarian Act 1998: XXVIII regulating animal protection. Ninety SD rats (forty-five males and forty-five females) were involved in the study. Animals were purchased from Charles River Hungary Ltd., at an approximate age of four to six weeks. Animals were acclimated for six days and housed individually. The experimental groups contained three animals of each sex, except for the control group, which contained six animals of each sex. The animals were treated with one of three compounds or saline (vehicle). Each compound and the vehicle were administered intraperitoneally to the animals by daily dosing for one, five, or twenty-eight days. The drug dosages were as follows: gentamicin (40 mg/kg), cisplatin (0.5 mg/kg), and tobramycin (40 mg/kg). The animals in the control group received saline (vehicle). Gentamicin was purchased from Sandoz (Hungary); cisplatin was purchased from Pharmachemie BV (Netherlands); tobramycin was provided by Teva Pharmaceutical Industries Ltd. (Hungary). All three compounds were diluted in Salsol A (saline) obtained from Teva Pharmaceutical Works Company Ltd. (Hungary). Urine and kidney tissue samples were collected after one, five, and twenty-eight days. Subjects were fasted overnight prior to kidney tissue collection, and urine was collected overnight for approximately sixteen hours. Kidney samples for metabolomics analysis were taken sterile from the animals, frozen in liquid nitrogen, and stored at −80°C, together with urine samples. Kidneys for histopathology evaluation were preserved in 4% buffered formaldehyde solution.
Clinical Chemistry and Histopathology
Blood was collected from the retro-orbital plexus before autopsy under isoflurane anesthesia. Blood clinical chemistry parameters were measured by a Konelab 20i. Qualitative examination of urine was performed on all animals (appearance, turbidity, pH, specific gravity, and presence of protein, blood/hemoglobin, glucose, ketones, urobilinogen (UBG), and bilirubin). Kidneys were weighed and subjected to gross pathological and histopathological examinations. After processing for routine histology, kidney sections were stained with hematoxylineosin and examined by light microscopy. For metabolomics, several kidney sections of about 2–3 mm thickness (approximately 100 mg) were prepared. These samples were placed into cryo-vials, and after snap freezing in liquid nitrogen, they were preserved at −80°C.
Sample Preparation for Metabolomics
At the time of analysis, one of each sample was thawed and extracts prepared as previously described (Lawton et al. 2008). For kidney, approximately 70 mg of tissue was homogenized in an 8:1 ratio of water to tissue, using zirconium beads with a Geno/Grinder 2000 (Glen Mills Inc.). The mixture was shaken for two minutes at 1350 strokes/min. For urine, osmolality measurements were collected for each sample. Proteins were precipitated from 100 μL of homogenized tissue and 100 μL of urine using methanol, which contained four standards to monitor extraction efficiency, using an automated liquid handler (Hamilton LabStar). The supernatant from the precipitated extract was split into four aliquots and dried in a vacuum. For LC/MS analysis, the dried extract was reconstituted in 50 μL 0.1% formic acid in 10% methanol. For GC/MS analysis, the dried extract was derivatized using equal parts bistrimethyl-silyl-trifluoroacetamide and solvent mixture acetonitrile:di-chloromethane:cyclohexane (5:4:1) with 5% triethylamine at 60°C for one hour to a final volume of 50 μL. In addition to the above samples, pooled samples of homogenized kidney and pooled samples of urine were extracted four independent times per day. These samples served as technical replicates throughout the data set to assess process variability. Also, 100 μL of water was extracted five independent times per day to serve as process blanks. Every sample analyzed was spiked with standards to monitor and evaluate instrument and extraction performance. Standards to monitor extraction were d6-cholesterol, fluorophenylglycine, and tridecanoic acid. A standard to monitor GC/MS derivatization was 2-tert-butyl-6-methylphenol (BHT). LC/MS standards to monitor LC and MS performance were d3-leucine, chloro and bromo-phenylalanine, d2-maleic acid, amitriptyline, and d10-benzophenone. GC/MS standards to monitor GC and MS performance were C5–C18 alkylbenzenes.
LC/MS and GC/MS Metabolic Profiling of Urine and Kidney Tissue
The samples from the different treatment groups were equally distributed over the run-days. Then, within each run-day the samples were completely randomized.
LC/MS was carried out using a Surveyor HPLC (Thermo-Electron Corporation, San Jose, CA, USA) with an electrospray ionization source coupled to an LTQ mass spectrometer (ThermoElectron Corporation). The mobile phase consisted of 0.1% formic acid in H2O (solvent A) and 0.1% formic acid in methanol (solvent B). The extract was loaded onto a 100 x 2.1 mm, 3 μm particle, Aquasil column (ThermoElectron Corporation) via a CTC autosampler (LeapTechnologies, Carrboro, NC, USA) and gradient eluted (0% B, four minutes; 0%–50% B, two minutes; 50%–80% B, five minutes, 80%–100% B, one minute; maintain 100% B, two minutes) directly into the mass spectrometer at a flow rate of 200 μL/min. The LTQ took full-scan mass spectra (99–1500 m/z) while switching polarity, 4.25 kV spray voltage for positive ion scan, and 3.75 kV spray voltage for negative ion scan to monitor both negative and positive ions. Each scan was composed of two microscans with a max ion trap fill time of 500 msec. The inlet capillary and sheath gas were maintained at 275°C and 40°C, respectively, throughout the analyses.
The derivatized samples for GC/MS were analyzed on a Thermo-Finnigan Trace DSQ fast-scanning single-quadrupole mass spectrometer operated at unit mass resolving power. The GC column was 20 m x 0.18 mm with a 0.18 μm film phase consisting of 5% phenyldimethyl silicone. The temperature program started with an initial oven temperature of 60°C that was ramped to 340°C, with helium as the carrier gas. The mass spectrometer was operated using electron impact ionization with a 50–750 amu scan range and was tuned and calibrated daily for mass resolution and mass accuracy.
Metabolite Identification
Metabolites present in the urine and kidney tissue sample extracts were identified by automated comparison to Metabolon’s reference library entries, using Metabolon’s proprietary software. This software matches experimentally derived spectra to a library database of spectra created from known chemical entities. Peaks that elute from either the LC or GC method are compared to the library at a particular retention time and its associated spectra for that metabolite. Internal standards are primarily used in both the GC and LC methods to calibrate retention times of metabolites across all of the samples in the study and for quality control of each instrument run.
Data Normalization
Each metabolite from urine was normalized to correct for differences in osmolality between samples. Missing ion intensity values were assumed to result from areas falling below the limits of detection. For each metabolite, the missing values were imputed with the observed minimum for that metabolite.
Statistical Analyses
Welch’s two-sample t test was performed for comparing each of the three drug groups versus the control group, at each time point separately, and for each matrix separately. The t tests were performed with R, an open-source software package (http://cran.r-project.org/). For all t tests, data were log-transformed. The level of the test was equal to 0.05. To account for multiple testing, false discovery rates (FDRs) were computed for each comparison (Benjamini and Hochberg 1995). The FDRs were estimated using the q value method (Storey and Tibshirani 2003). For predictive modeling, CART (Breiman et al. 1984) and logistic regression analysis were performed using JMP version 7 (May 2007) (SAS Institute Inc., Cary, NC, USA).
Results
Histopathology and Clinical Chemistry
After twenty-eight days of treatment, histopathological and clinical chemistry evaluation was performed and nephrotoxicity was observed in cisplatin-, gentamicin-, and tobramycin-treated animals (Table 1). Based on histopathological examination, cisplatin caused the most prominent nephrotoxicity and tobramycin the least. Males showed more abnormal histopathology in response to the nephrotoxins than did females. At autopsy, kidneys from the cisplatin-, gentamicin-, and tobramycin-treated animals weighed more and were enlarged and pale (Table 1). Changes indicative of nephrotoxicity were also observed in the clinical chemistry results. Higher urea and creatinine values were found in all females and in one male treated with cisplatin for twenty-eight days. In gentamicin- and tobramycin-treated animals, granular cylinders were observed in the urine sediment of both sexes.
After five days of treatment, minimal focal tubular epithelial cell necrosis was observed in the medulla of the kidneys in one of three males and one of three females treated with cisplatin. Transitional epithelial cells were found in the urine sediment of tobramycin-treated males and females. No other significant effects were observed.
After a single day of treatment, minimal focal tubular epithelial cell necrosis was noted in one of three females treated with cisplatin. Transitional epithelial cells were found in the urine sediment of tobramycin-treated males. No other significant effects were observed.
Metabolites Detected in Kidney Tissue and Urine
Metabolomic analysis was performed on 180 rat kidney tissue and urine samples, collected from four treatment groups (gentamicin, cisplatin, tobramycin, or vehicle control) at three time points (days 1, 5, and 28). The analytical signatures of the chromatography peaks present in the samples were matched against an in-house database containing thousands of metabolites. Full data analysis confirmed the detection of 547 metabolites in kidney tissue, and 657 metabolites in urine samples. In kidney tissue, approximately 30% of the metabolites were detected using GC/MS and 70% by LC/MS. For urine, approximately 46% of the metabolites were detected using GC/MS and 54% by LC/MS. Several metabolites were detected on both platforms. About 35% of the entities corresponded to identifiable chemical metabolites, and the remaining entities represented currently unnamed structural identities. For the remainder of this paper, we focus our discussion on named metabolites that match our library of authentic chemical standards. Metabolites detected but without an associated chemical standard are not further discussed. Named metabolites detected in this study included amino acid metabolites, lipid metabolites, carbohydrates, nucleotides, energy metabolites, vitamins, cofactors, short peptides, and xenobiotics.
Identification of Candidate Biomarkers for Nephrotoxicity
A Welch’s t test was performed on the urine and kidney tissue metabolite data to compare each of the nephrotoxic drug treatment groups versus the vehicle group at each of the three time points. The number of significantly changed metabolites (p ≤ .05, q ≤ .2) for each of the comparisons is shown in Table 2. In general, day 28 showed the highest number of statistically changed metabolites. Among the three nephrotoxicants, cisplatin induced the highest number of significant differences to the metabolome at day 28.
To identify biomarkers indicative of drug-induced nephrotoxicity, criteria were applied for selecting metabolites that were significantly and unidirectionally changed across the different drug groups and time points. To be included on the candidate biomarker list, the metabolite must be significantly changed by each of the three drugs, or in the absence of a strong p value must display a consistent change across the three drugs. Metabolites that were increased with one of the drugs and decreased or unchanged with the other drugs were not selected. By applying those stringent selection criteria, thirty-eight putative nephrotoxicity biomarkers were identified from urine, and thirty-seven biomarkers were discovered from kidney tissue (Tables 3 and 4).
The selected putative nephrotoxicity biomarkers included metabolites belonging to amino acid, peptide, lipid, nucleotide, carbohydrate, energy, vitamin, and xenobiotic metabolism (Tables 3 and 4). In urine, the majority of metabolites increased in the drug-treated samples compared to vehicle samples, whereas in kidney tissue the opposite effect was observed. We observed clear distinctions between individual metabolites in the timing of the effect to the three nephrotoxic drugs. This finding allowed for grouping of the metabolites as early, late, or continuous nephrotoxicity marker metabolites. Metabolites that showed a very early day 1 response to the three nephrotoxicants, and which were the main focus of this study, are labeled yellow in Tables 3 and 4. The early markers in urine included polyamines, several amino acids, glycylproline, glucosamine, 1,5-anhydroglucitol, monoethanolamine, and phosphate (Table 3, Figure 1). In kidney tissue, metabolites responding at day 1 included sorbitol, glucose, and 5-methyltetrahydrofolate (Table 4). Metabolites that showed a much later response to the nephrotoxins at day 28 included carnitine and riboflavin from urine (Figure 1) and 3-indoxyl sulfate from kidney.
However, the large majority of putative nephrotoxicity markers identified in this study showed a progressive increase or decrease over time, starting at day 1 or day 5, likely mirroring the increasingly toxic effects of the drugs on the kidneys with prolonged dosing. Examples of this class of biomarker identified in kidney tissue were nucleosides such as 2’-deoxyinosine, guanosine, cytidine, uridine, and the majority of amino acids and dipeptides. Examples of such progressive marker metabolites from urine included 3-hydroxyphenylacetate, most amino acids, N-acetylneuraminate, mannose, glucose, myoinositol, 3-hydroxybutyrate, 3-hydroxy-3-methylglutarate, and hippurate. Two interesting metabolites from this category were 3-hydroxyphenylacetate and hippurate; both showed decreased excretion in the drug groups, as opposed to the general trend of increased excretion for the majority of metabolites. In fact, 3-hydroxyphenylacetate was reduced up to seven-fold in the urine of rats treated with the three nephrotoxins (Table 3, Figure 1). This decrease was dramatic and highly significant at both the day 5 and day 28 time points. However, slight decreases in urinary 3-hydroxyphenylacetate excretion were present already at the early day 1 time point. Another metabolite that followed a similar trend in urine was hippurate. Hippurate was decreased as much as ten-fold in urine of the nephrotoxic drug groups at day 28, but decreases were clearly present at the earlier day 5 and day 1 time points as well. Another hippurate-related metabolite, 2-hydroxyhippurate, showed a similar pattern in urine.
Aminoaciduria
One of the strongest and most significant responses induced by the three nephrotoxins was a dramatic increase in many amino acids and dipeptides in urine, concomitant with a decrease in kidney tissue (Tables 3 and 4 ). Amino acids that showed significant increases in urine in the different drug groups were serine, threonine, alanine, asparagine, glutamine, histidine, lysine, tryptophan, isoleucine, leucine, valine, cysteine, methionine, and arginine (Table 3). Other amino acids showed similar trends, but they either lacked significance or had less dramatic fold-changes. The increased excretion of amino acids in urine was as high as 18.5-fold for valine at day 28 with cisplatin. Many other amino acids showed a range of 1.5- to 15-fold increased excretion (Table 3). Interestingly, the increased excretion of amino acids and dipeptides was already significant at the early day 1 and day 5 time points, when no histopathological kidney damage yet was present.
In kidney tissue, seventeen amino acids were significantly and/or dramatically decreased in the three drug groups at day 28 (Table 4). Decreases ranged from 1.5- to 3-fold. Glycine, glutamine, and cysteine were decreased as well, but not as convincingly. Several dipeptides such as prolylleucine and gamma-glutamylphenylalanine showed significant decreases as well in the drug groups at day 28. Many of the amino acids in kidney tissue were decreased already at the day 5 and day 1 time points, although not significantly (p values > .05). Figure 2 shows the significant increase of histidine and glycylproline in urine and the significant decrease of citrulline and prolylleucine in kidney tissue upon treatment with all three nephrotoxins.
Polyamine Excretion as an Early Nephrotoxicicty Marker
Several polyamines, including cadaverine, putrescine, agmatine, and spermidine, were increased up to 4.4-fold in the urine of the three nephrotoxicant-treated animals compared to the vehicle animals (Table 3, Figure 3). The increases were most significant and dramatic at the very early time point (day 1), and the effect was less pronounced upon prolonged dosing. It is striking that all four polyamines detected in urine showed the exact same trend, and therefore this early effect is likely to be an important finding.
Purine and Pyrimidine Nucleosides Decreased in Kidney Tissue
Treatments with gentamicin, cisplatin, and tobramycin caused a significant decrease in the levels of purine and pyrimidine nucleosides in kidney tissue, including inosine, 2’-deoxyinosine, adenosine, guanosine, cytidine, and uridine (Table 4, Figure 4).
The changes were most pronounced at day 28, with decreases as high as five-fold. However, many of the nucleosides were already decreased at the early day 1 and day 5 time points, with the effect gradually becoming more severe upon prolonged dosing. In contrast, most nucleosides were not significantly altered in urine from the nephrotoxic drug groups (data not shown).
Early Nephrotoxicity Prediction Modeling
For prediction modeling of early nephrotoxicity effects, we fitted a model for the urine metabolomics data at day 28 and used the model to predict the outcome at days 1 and 5. Since the data for days 1 and 5 were not used to build the model, there was no danger of overfitting the data. The animals that received the nephrotoxins gentamicin, cisplatin, and tobramycin were treated as one group (6 animals/drug), designated the “tox group,” and the animals that received the vehicle (twelve animals) were designated the “control group.”
CART was used to build classification trees at day 28, based on the list of putative nephrotoxicity markers from urine (Table 3). The model was fitted with JMP. The top five classification trees were single-split trees that either completely separated the nephrotoxicant-treated group from the control group or misclassified only one observation. The top five classifiers at day 28 were leucine, hippurate, isoleucine, glucose, and valine. Using these top five metabolites, the classification of day 5 urine samples to the tox or control group reached a prediction accuracy of 90%, 83%, 83%, 70%, and 90%, respectively, based on each biomarker. For day 1 urine samples, the prediction accuracies were 70%, 57%, 63%, 57%, and 70%. Therefore, the branched-chain amino acids (BCAAs) leucine, isoleucine, and valine had the best predictive power for days 1 and 5. Using the average of the three BCAAs (which had 100% accuracy for day 28), a slight improvement was gained in predictive ability over the top predictor leucine: 93% for day 5 and 70% for day 1.
Logistic regression was also performed on the day 28 urine metabolomics data using JMP. Using the average value for the three BCAAs, the predictive ability was the same as using CART: 93% for day 5 and 70% for day 1. Adding additional variables such as glucose did not improve the predictions. After building a random forest (Breiman 2001) with the BCAAs and glucose, or the three BCAAs alone, values close to the above were obtained as well.
Discussion
The analysis of histopathology data revealed that gentamicin, cisplatin, and tobramycin treatments manifested damage to the rat kidneys after twenty-eight days of daily dosing. Lymphocytic inflammation, tubular dilatation, and basophilia in the cortex and medulla were present at day 28, and cisplatin also caused enlargement and necrosis in the tubular epithelium. Pale and enlarged kidneys were also observed at day 28 in all drug groups. However, at the earlier time points day 5 and day 1, very few significant and mostly very mild histopathological effects were observed, except for a few minor and isolated effects in cisplatin- and tobramycin-treated animals. These findings were consistent with the initial rationale behind this study, to select doses of each drug that were expected to cause no histological changes after single dose administration, but that induced significant kidney toxicity after twenty-eight days of daily treatment.
In addition to the histopathology, clinical chemistry measurements found that higher BUN and serum creatinine were present in all females and in one male treated with cisplatin at day 28. However, no other treatment groups at day 28 had elevated BUN or creatinine. This finding is not surprising, since the commonly used serum creatinine and BUN measurements are not very sensitive markers for nephrotoxicity (Duarte and Preuss 1993; Davis and Kramer 2006; Hewitt et al. 2004; Thukral et al. 2005). Usually by the time changes in BUN and creatinine levels are detected, significant damage to the nephrons has already occurred. Thus, better and earlier biomarkers for kidney injury are clearly needed. Such early biomarkers would indicate when the damage begins, assess the level of damage, and be useful to assess nephrotoxicity of therapeutic agents in preclinical drug development and phase I drug trials. In human subjects with renal failure, these early biomarkers that can be evaluated using noninvasive methods would offer the prospect of earlier diagnosis, which would greatly enhance the chance of a successful treatment outcome.
We used metabolomics to identify changes in metabolites from two different matrices (kidney and urine) upon dosing with three strong nephrotoxins and one vehicle control. We showed that kidney damage and its severity were easily recognized from the metabolite changes at day 28. One of the strongest nephrotoxin-induced responses was a dramatic increase in amino acids in urine, concomitant with a decrease in kidney tissue at day 28 (Tables 3 and 4 ). These results are consistent with earlier reports describing increased amino acid excretion upon dosing with cisplatin, aminoglycosides, and other nephrotoxins (Ghiculescu and Kubler 2006; Kim et al. 1995; Macpherson et al. 1991; Niemann and Serkova 2007; Portilla et al. 2006; Xu et al. 2008). Under normal conditions, only minimal amounts of amino acids are excreted in urine (Harper and Doolan 1963; Silbernagl 1988; Tizianello et al. 1980; van de Poll et al. 2004). In most mammals, about 99% of filtered amino acids are reabsorbed in the proximal tubule. However, under nephrotoxic conditions, amino acid excretion often increases because of impaired reabsorption by the renal tubules, increased cellular turnover, or increased permeability of the glomerular membranes (Macpherson et al. 1991; van de Poll et al. 2004). This increase in urinary amino acids is also known as aminoaciduria. The increase in amino acids in urine was the most dramatic at day 28, but interestingly the increase was already significant at the earlier time points of day 5 and day 1, during which no histopathological kidney damage was yet present. Therefore the increase in amino acid excretion is an excellent potential marker for very early kidney malfunction and nephrotoxicity.
We also found that many sugars such as glucose, mannose, fucose, N-acetylneuraminate, and gluconate were elevated at day 28 in the urine of the drug-treated groups. Similarly, as for the amino acids, many of the sugars were already significantly increased at the earlier time points of day 5 and day 1. In fact, some sugars such as 1,5-anhydroglucitol (1,5-AG) and glucosamine showed a preferential response at the earliest day 1 time point, whereas others such as mannose, glucose, and gluconate showed a more continuous response over time. 1,5-AG is a naturally occurring polyol that is normally filtered and completely reabsorbed in the kidneys (Buse et al. 2003). During normal kidney function, carbohydrates are excreted in the urine only when there are abnormally elevated levels such as is the case during diabetes. However, during nephrotoxicity, in which there is improper functioning of the renal tubules, glucose and other carbohydrates are eliminated in the urine (Gartland et al. 1989; Portilla et al. 2006). The condition in which glucose is excreted by the kidneys is also known as glycosuria, and is part of the Fanconi syndrome. The increase we observed in glucose excretion is consistent with earlier glycosuria reports for cisplatin, aminoglycosides, and other nephrotoxins (Ghiculescu and Kubler 2006; Kim et al. 1995; Niemann and Serkova 2007; Portilla et al. 2006; Xu et al. 2008). In addition, our findings show that an increase in carbohydrates in urine can be a very early metabolic indicator of renal malfunction and toxicity.
One of the earliest and strongest metabolic changes induced by the three nephrotoxins was a significant increase (up to 4.4-fold) in the excretion of polyamines (Table 3, Figure 3). This effect was most pronounced after a single dose and decreased gradually at the later time points. It is striking that all four polyamines detected in the urine showed the same trend, suggesting this early effect is an important finding. The polyamines are directly involved in cell proliferation and viability and can also reflect alterations in protein synthesis and breakdown rates (Seiler and Raul 2005; Wallace and Niiranen 2007). Increased excretion of polyamines has been documented in patients with renal failure, trauma, or increased cell death (Igarashi et al. 2006; Jeevanandam et al. 1989; Pöyhönen et al. 1993; Russell 1983). Our results indicate that increased excretion of polyamines in urine at the early time points may be an excellent predictive marker of early nephrotoxicity.
Another metabolite with an early pattern very similar to the polyamines was phosphate. Increased excretion of phosphate, as a result of nephrotoxicity, has been documented (Berndt and Knox 1992; Ghiculescu and Kubler 2006; Harper and Doolan 1963; Kim et al. 1995). Other markers of very early nephrotoxicity were 3-hydroxybutyrate, 3-hydroxyphenylacetate, mono-ethanolamine, and hippurate from urine (Table 3), all of which showed significant changes by day 5 in all drug groups. In kidney, 5-methyltetrahydrofolate (5MeTHF) was sharply decreased at days 1 and 5. 5MeTHF is used to recycle homo-cysteine back to methionine. The changes in 5MeTHF may be related to altered synthesis or breakdown in the folate pathway or alterations in the S-adenosylmethionine (SAM) cycle as a result of the toxic effect of the drugs in the kidney. The importance of folate metabolism in renal failure has been described by Schaefer et al. (2002). Also, several kidney osmolytes, such as myoinositol and sorbitol, showed very early responses to the three nephrotoxins, consistent with earlier literature on changes in osmolytes during renal stress (Niemann and Serkova 2007).
A major observation in kidney tissue was the significant decrease in many nucleosides in all three drug groups (Table 4, Figure 4). Those changes were most pronounced at day 28, with decreases as high as five-fold. However, many of the kidney tissue nucleosides were already decreased at the early day 1 and day 5 time points, with the effect becoming gradually more severe upon prolonged dosing. Nucleosides, as metabolites and precursors of nucleotides, are vital to nucleic acid synthesis. Nucleosides tend to have minimal binding to plasma proteins and hence are completely filtered into nephron tubules (Elwi et al. 2006). Several mechanisms such as selective reabsorption, secretion, and passive filtering seem to be in place for renal handling of nucleosides (Elwi et al. 2006). The decrease in the levels of many nucleosides in kidney tissue observed in this study may reflect altered activity of renal transporters (Thukral et al. 2005), altered synthesis/breakdown of nucleotides, or altered filtering by the kidneys resulting from the nephrotoxic drug action.
Finally, we built a general kidney toxicity prediction model to detect the very early stages of nephrotoxicity using noninvasive urine samples before it can actually be detected through histopathology, BUN, and creatinine measurements. We first selected all metabolites from urine, which showed significant and/or convincing unidirectional fold-changes in all three drug groups, at one or more time points. By selecting metabolites that showed only changes common to the three drug effects, we avoided including biomarkers that were specific to a single drug. By taking this approach we identified thirty-nine putative nephrotoxicity biomarkers in urine. We fitted a CART model for the urine metabolomics data at day 28 and used that model to predict the outcome at days 1 and 5. The BCAAs leucine, isoleucine, and valine provided the best predictive power for correctly classifying samples to either the drug-treated group or the vehicle group. Using the average of the three BCAAs, we were able to separate drug-treated samples from vehicle-treated samples with 100%, 93%, and 70% accuracy at day 28, day 5, and day 1, respectively. Therefore, measurement of the average levels of the three BCAAs in the urine of rats dosed with nephrotoxins could become a future test for the very early detection of drug-induced nephrotoxicity after further validation.
In conclusion, by using metabolomics, we identified very early candidate biomarkers of nephrotoxicity that can detect tubular kidney damage much earlier than histopathology and BUN or creatinine measurements. Many of the identified markers are novel in their association with early nephrotoxicity, such as polyamines, 1,5-AG, monoethanolamine, phosphate, glycylproline, glucosamine, sorbitol, 5MeTHF, and others. In addition, our analysis validated some of the earlier published markers of nephrotoxicity, such as amino acids, glucose, and osmolytes, but it highlighted that many of these metabolites are in fact very early indicators of kidney malfunction. Our future efforts will be focused on the validation of those potential biomarkers, and expansion of the test panel of nephrotoxins to include other mechanisms of action. Also, correlation of our metabolite markers with the newly announced seven-member protein nephrotoxicity panel from the FDA may provide further insight and validation of early nephrotoxicity markers.