Prediction of Nephrotoxicant Action and Identification of Candidate Toxicity-Related Biomarkers

Abstract

A vast majority of pharmacological compounds and their metabolites are excreted via the urine, and within the complex structure of the kidney, the proximal tubules are a main target site of nephrotoxic compounds. We used the model nephrotoxicants mercuric chloride, 2-bromoethylamine hydrobromide, hexachlorobutadiene, mitomycin, amphotericin, and puromycin to elucidate time- and dose-dependent global gene expression changes associated with proximal tubular toxicity. Male Sprague–Dawley rats were dosed via intraperitoneal injection once daily for mercuric chloride and amphotericin (up to 7 doses), while a single dose was given for all other compounds. Animals were exposed to 2 different doses of these compounds and kidney tissues were collected on day 1, 3, and 7 postdosing. Gene expression profiles were generated from kidney RNA using 17K rat cDNA dual dye microarray and analyzed in conjunction with histopathology. Analysis of gene expression profiles showed that the profiles clustered based on similarities in the severity and type of pathology of individual animals. Further, the expression changes were indicative of tubular toxicity showing hallmarks of tubular degeneration/regeneration and necrosis. Use of gene expression data in predicting the type of nephrotoxicity was then tested with a support vector machine (SVM)-based approach. A SVM prediction module was trained using 120 profiles of total profiles divided into four classes based on the severity of pathology and clustering. Although mitomycin C and amphotericin B treatments did not cause toxicity, their expression profiles were included in the SVM prediction module to increase the sample size. Using this classifier, the SVM predicted the type of pathology of 28 test profiles with 100% selectivity and 82% sensitivity. These data indicate that valid predictions could be made based on gene expression changes from a small set of expression profiles. A set of potential biomarkers showing a time- and dose-response with respect to the progression of proximal tubular toxicity were identified. These include several transporters (Slc21a2, Slc15, Slc34a2), Kim 1, IGFbp-1, osteopontin, α-fibrinogen, and Gstα.

Introduction

The process of drug discovery and development is cumbersome and expensive. It is estimated that 99% of new drug candidates fail during the drug development process. The development of 30% of total new drug candidates is halted due to unforeseen toxicity profiles and side effects in clinical studies (Garrett and Workman, 1999; Lesko and Atkinson, 2001). Thus, the ability to assess potential toxicity of drug candidates early in the drug discovery process could save time and money. An emerging approach to achieve this objective is the use of genomic technologies that are impacting many areas of drug discovery and development (Afshari et al., 1999; Petricoin et al., 2002; Ulrich and Friend, 2002; Suter et al., 2004). Of these technologies, the analysis of gene expression during a toxic response holds promise in identifying toxicological pathways and liabilities (Nuwaysir et al., 1999). Indeed, gene expression profiling has been successfully applied to identify compound-specific and toxicity-specific gene changes in both liver (Waring et al., 2001a, 2001b; Hamadeh et al., 2002a, 2002b) and kidney (Huang et al., 2001; Yoshida et al., 2002; Amin et al., 2004).

We have applied a toxicogenomics approach to understand the mechanisms of region-specific nephrotoxicity and to identify potential novel markers of toxicity. The kidney is especially vulnerable to toxic insult by various drugs and xenobiotics because it receives nearly one quarter of the cardiac output, and transports, metabolizes and concentrates a variety of potentially toxic substances within its parenchyma (Toback, 1992; Bennett, 1997). Specifically, renal tubular epithelial cells are most susceptible to toxic insults. A major challenge is to provide measurements, both qualitative and quantitative, that give an early indication of the initial site of renal damage before gross deterioration in kidney function has occurred. Traditional markers of kidney toxicity such as blood urea nitrogen (BUN) and creatinine for toxicological evaluation have limitations (Duarte and Preuss, 1993). First, they are not region specific and second, significant changes may not occur until 30–50% damage has occurred (determined by histopathology assessment and/or functional tests). Thus, it is intriguing to determine whether genomics evaluation could lead to the elucidation of markers that may provide additional sensitivity or earlier detection of proximal tubular damage. The ability to possibly detect subtle kidney damage in acute exposure as a prelude to chronic exposure could impact the attrition rate of new chemical entities due to nephrotoxicity.

The prerequisite for developing biomarkers of an organ-specific toxicity is to first understand molecular pathways governing the toxicity in vivo. This knowledge allows differentiation of toxic response pathways from those required for the intended activity (i.e., pharmacological action). Subsequently, candidate biomarkers from the toxicity pathways that show a dose response and predict the toxicity endpoint may be selected (Bailey and Ulrich, 2004; Roberts et al., 2003; Tugwood et al., 2003). A biomarker discovery effort requires experimental designs that encompass several compounds of diverse chemical nature, causing the same toxic endpoint. In addition, multiple time points and doses are necessary to allow the teasing out of gene expression changes that are early indicators of severity and progression of lesion.

In the current study, we used mercuric chloride (MC), 2-bromoethylamine hydrobromide (2-BH), hexachlorobutadiene (HEX), and puromycin aminonucleoside (PA) to study gene expression patterns associated with proximal tubular toxicity in male Sprague-Dawley rats. The main effects of MC, 2-BH, and HEX are on but not limited to the proximal tubular toxicity. PA causes both glomerular and proximal tubular toxicity (Amin et al., 2004). The study was conducted at 2 dose levels and kidney tissues were collected at 3 time points. Kidney expression data in conjunction with histopathology data were analyzed to identify gene changes associated with specific lesions. Gene expression profiles were consistent with the type and severity of lesion. We report potential markers of proximal tubular toxicity, that are undergoing further validation.

Materials and Methods

Test Animals, Test Articles and Study Design

Male Sprague–Dawley (SD) rats (300–350 g) approximately 10–11-weeks-old were purchased from Charles River Laboratories (Hollister, CA) and housed in metabolic cages. Certified rodent chow (Teklad 7012C, Madison, WI) and drinking water were available ad libidum except for overnight fasting prior to the day of necropsy. The rats were kept in a controlled (60–80°F) temperature and humidity (30–70%) with a 12-hour light/dark cycle. Animals were acclimated up to 4 days prior to dosing. During the study period, all animals were observed at least twice daily for mortality and moribundity. The experiments were performed in accordance with Amgen’s Institutional Animal Care and Use Committee. All animals were randomly assigned into treatment and control groups (4 rats/group/timepoint). MC, HEX, 2-BH, PA, Amphotericin B (AB), and Mitomycin C (MTC) were purchased from Sigma chemical company (St. Louis, MO) and administered via intraperitoneal (IP) injection at 10 ml/kg body weight. MC and AB were dissolved in 5% dextrose solution and administered once daily up to 7 doses. MC was dosed at 0.4 mg/kg or 1.5 mg/kg and AB was dosed at 4 mg/kg or 15 mg/kg. HEX was dissolved in corn oil and administered once at a dose of 30 mg/kg or 100mg/kg. 2-BH was dissolved in 0.9% saline and administered once at a dose of 50 mg/kg or 150 mg/kg. PA and MTC were dissolved in 0.9% saline solution and administered once. PA was dosed at 38 mg/kg or 150 mg/kg and MTC was dosed at 0.625 mg/kg or 2.5 mg/kg. Time-matched control animals received corresponding quantities (10 ml/kg body weight) of the vehicles. Dosing for all compounds was initiated on a day prior to the first necropsy time point. Terminal necropsies were then performed on days 1, 3, and 7 postdosing.

Tissue Collection

At necropsy, each rat was euthanized with CO₂ and the blood was collected from the posterior vena cava for clinical chemistry tests. Upon the removal of the left and right kidneys, the longitudinal half of the right kidney and an approximately 3 mm cross section from the center of the left kidney were fixed in 10% neutral buffered formalin for routine histological processing. The remainders of the kidneys were chopped into smaller pieces (~2 mm³), snap frozen in liquid nitrogen and kept at −70°C until RNA isolation. All animals received complete postmortem examinations for gross abnormalities during collection of tissues for histopathologic evaluation and RNA extraction.

Clinical Chemistry and Histology

Blood for clinical chemistry measurements was collected at terminal necropsies on days 1, 3, and 7 postdosing. The BUN and creatinine were measured using standard clinical chemistry analysis. For histology, formalin-fixed tissues were embedded in paraffin blocks (cut into sections of approximately 5 μm each) and stained with hematoxylin and eosin. Histology slides were evaluated in-house by a board-certified veterinary pathologist (L. Healy).

Microarray Preparation and cDNA Probe Hybridization

Custom high-density cDNA microarrays were generated in-house using 17,000 rat cDNA clones purchased from Research Genetics (Huntsville, AL). All clones were sequenced verified and their identities confirmed using BLAST 〈http://ncbi.nlm.nih.gov.BLAST/〉. The clone inserts were then amplified by the polymerase chain reaction (PCR). After verifying their quality on the agarose gels, the amplified products were arrayed on GAPS II coated slides (Corning, Corning, NY). The total RNA for microarrays was isolated from kidneys using the RNeasy Maxi Kit (Qiagen, Valencia, CA) according to the manufacturer’s protocol. The RNA was quantified using a Eppendorf biophotometer (Brinkman, Westbury, NY), and RNA quality determined using a Bioanalyzer (Agilent Technologies, Palo Alto, California). Equal amounts of RNA from each animal (n = 4) in the time-matched control vehicle group were pooled. The pooled control RNA was used in a hybridization paired with RNA from each treated animal from the same time group. Twenty-five μg of total RNA was reverse-transcribed using Superscript II and oligo dT (Invitrogen, Carlsbad, CA) in the presence of 0.1 mM Cy-dye (Cy3 or Cy5) labeled dCTP (Amersham Biosciences, Piscataway, NJ). The resulting labeled cDNAs were hybridized to cDNA arrays.

All hybridizations were performed using fluor reversal and in duplicate to minimize the bias due to dye incorporation. The hybridizations where poor incorporation of either dye was observed were discarded and repeated. The spotting, denaturation, hybridization, and washing of the arrays were conducted using protocols similar to the GAPS II instruction manual. Fluorescent images of the dried slides were obtained using a GenePix Scanner 4000 (Axon Instruments, Union City, CA), and feature extraction was performed using GenePix software. These data were then imported into Resolver v3.2 (Rosetta Biosoftware, Kirkland, WA) for analysis where the ratio of intensity of the individual treated animals to the pooled control animal values. The profiles were combined into ratio experiments in Resolver.

Statistical Analyses

Agglomerative 2-dimensional hierarchical cluster analysis of gene expression data was carried out using the Resolver package (Rosetta Biosoftware, Kirkland, WA). Principal component analysis (PCA) and 2-Way analysis of variance (ANOVA) were carried out using Partek-Pro software version 5.1 (St. Charles, Missouri). PCA was used to reduce the dimensionality of the data for visualization and ANOVA was used to elucidate gene sets that discriminate between pathological classes. The model based on gene expression for predicting pathology of the animals was built using the Support Vector Machine (SVM) program written by Brown et al. (2000). The SVM model was built using normalized intensities from the individual microarray hybridizations. Z-tests were performed to select candidate biomarkers from the gene expression data. The intensity of each gene from individual profiles was compared to the mean intensity and standard deviation of control animals. Then, 307 control hybridizations were used for this comparison. Control data was obtained from 150 animals using Cy3 and Cy5 dyes for each animals (some were repeat hybridizations). The distributions of all of the hybridizations (both control and experimental) were normalized to the same distribution using a quantile normalization scheme. Genes having a Z-score of 3 or more (3 standard deviations away from the mean of the controls) in at least 3 out of the 4 nephrotoxic compounds were selected.

Branched DNA (bDNA) Signal Amplification Assay

The bDNA assay is a signal amplification technique to measure cellular mRNA (Pachl et al., 1995). We employed the bDNA technique to measure and confirm the mRNA levels of select genes of interest obtained from microarray analysis. The gene sequences were accessed from GenBank, and multiple oligonucleotide probe sets (containing capture probes, label probes, and blocker probes) were designed using Probe-Designer software (Bayer Diagnostics, East Walpole, MA). All probe sequences were submitted to the National Center for Biotechnological Information (NCBI) for nucleotide comparison by BLASTn (NCBI, Bethesda, MD) to ensure minimal cross-reactivity with other known rat sequences and expressed sequence tags (ESTs). All probes were synthesized on an ABI 384 oligo-synthesizer (Applied Biosystems, Foster City, CA). The probe sets were validated and the linear ranges of the mRNA detection for each gene were established using serial dilutions. The bDNA assay was performed with Quantigene bDNA Kits following manufacturer’s instructions (Bayer Diagnostics, East Walpole, MA). Ten μg total RNA were used for each reaction, and the assays were performed in duplicate in 96-well plate. The luminescence for each well was recorded as relative light units per 10 μg of total RNA.

Results

Clinical Chemistry

The BUN and creatinine findings for the MC, PA, HEX, and 2-BH treatments at the indicated low and high doses on day 1, 3, and 7 postdosing are presented in Figure 1. Increases in blood-urea nitrogen (BUN) and creatinine were observed following the treatments with PA, MC, and 2-BH, at high dose levels. In 2-BH and MC the BUN and creatinine levels were highly variable among animals of the same treatment group. For PA, a single high-dose injection caused significant increases in BUN and creatinine levels compared to the time-matched vehicle controls at day 7. Further, the high dose of PA resulted in 2.5-fold reduction in serum albumin by day 7 (data not shown), indicating loss of protein through the glomerulus and nephrotic syndrome (Duarte and Preuss, 1993). None of these compounds resulted in significant changes in the ALT levels at the tested doses and times (data not shown). However, there was a significant drop in ALP and AST levels in PA treated animals. The high dose of PA resulted in a 2-fold decrease in ALP at day 3 and day 7, and AST at day 7 (data not shown).

Histopathology

A summary of the significant histopathological findings at high doses of tested compounds is presented in Table 1. There were no histopathological changes in kidneys with administration of the vehicle control solutions at any of the time points examined. The PA-induced lesions were observed only at day 7. A single injection of the high dose of PA in rats induced significant renal lesions that included intratubular protein, degeneration and/or regeneration of tubular epithelium, and multifocal atrophy of glomeruli. The high dose MC-treated rats had evidence of renal tubular necrosis by day 3. These animals showed tubular degeneration/regeneration and casts (protein, granular and cellular). Degeneration/regeneration was used as a diagnosis when degenerative cells (apoptosis, swelling, or sloughing) were observed in addition to regenerative tubules. The 7-day, high-dose group animals went through necropsy on day 5 since 2 of the animals in this group became lethargic. At this time, rats showed severe tubular necrosis with renal lesions extending to the cortex, and protein in Bowman’s space.

The high dose of HEX induced tubular degeneration and necrosis in all animals by day 1. In addition, 3/4 rats showed an increase in cytoplasmic droplets within the proximal tubule epithelium. By day 3, 2/4 rats in the high-dose group showed the tubular degeneration and necrosis. By day 7, lesions started to resolve, showing only basophilia and dilation of tubules. A single injection of a high dose of 2-BH induced minimal to mild protein loss in the tubules and necrosis in the renal pelvis by day 1. By day 3, 2 of the 4 rats in the high dose group showed moderate necrosis of the renal pelvis, and 3/4 rats showed protein in the renal tubules. On day 7, animals in both dose groups showed minimal to moderate degeneration/regeneration of tubules. In general, rats treated with MC, PA, 2-BH, and HEX developed renal lesions as expected based on the reported literature. The localized tubular injury and cell proliferation caused by MC, 2-BH, and HEX reflected their segment-specific nephrotoxicity.

Microarray

Custom rat cDNA microarrays containing 17K elements were used to analyze the gene expression profiles of rat kidneys associated with the PA, MC, HEX, and 2-BH treatments. Gene expression changes associated with each chemical exposure at 2 dose levels and three time points postdosing were analyzed using agglomerative 2-dimensional hierarchical clustering (Eisen et al., 1998) in Rosetta Resolver. The statistical criteria for clustering analysis were ≥1.5-fold change and p ≤ 0.01. Figure 2 shows the hierarchical clustering maps for grouping of each of the animals from the 4 compound treatments based on the genes that met the selection criteria. This analysis facilitated the visualization of groupings of animals based on gene expression changes, potentially reflecting underlying molecular mechanisms of toxicity. The expression profiles generally clustered in a dose and time-dependent manner. However, certain animals did not cluster with their respective time-dose group. Interestingly, histopathology indicated that animals generally clustered with other animals having similar type and severity of lesion.

The PA experiments formed two major clusters with the 7-day high dose group animals forming their own cluster (Figure 2A). All the time points and doses were differentiated into subclusters. It was interesting to note that animal #851 (day 3, high-dose) branched off separately from the other three animals in the same group. A review of its histopathology indicated that this animal had low-severity pyelonephritis. Hierarchical clustering of MC experiments formed groups that correlated with the severity of renal lesion (Figure 2B). The expression profile groupings consisted of treatments associated with no pathology, tubular degeneration/regeneration, and necrosis. Individual animals clustered according to their severity of pathology. For example, day-1, high-dose animal #8014 grouped with animals having necrosis. Hierarchical clustering of HEX experiments also showed 3 distinct groupings: no pathology, tubular dilation, and tubular necrosis (Figure 2C). Day 3 high-dose animals clustered into different groups but matched perfectly with the severity of pathology. For example day-3, high-dose animals (#5137 and #5138) did not show tubular lesions and thus grouped with animals having no pathology. The hierarchical clustering divided 2-BH experiments into 2 major nodes (Figure 2D). The analysis of histopathology and gene expression data revealed the group of profiles showing the highest gene changes associated with severe and diverse pathology (tubular regeneration, tubular protein, and pelvic necrosis). The other main group had subclusters, which separated animals with no pathology and regenerative basophilia/tubular protein. Although most of the animals from the same dose-time tightly clustered, several animals did not cluster with their dose-time group, including animals 5173, 5174, 5147, and 5141. These animals grouped based on the severity of their pathology. A day-3, low-dose animal (#5141) clustered with the 3-day high dose group due to a more severe pathology.

Mechanisms of Tubular Toxicities

To separate mechanistic changes associated with tubular toxicity from compound-associated changes, a combined 2-D hierarchical clustering analysis was performed. Using a statistical criteria of ≥ 1.5-fold change at p≤ 0.01, 2400 informative genes were identified (Figure 3). Similar to the results obtained from clustering analysis of individual compounds, it was noted that the expression profiles associated with a similar toxicity endpoint grouped together. Stated in general terms, expression profiles grouped into 3 toxicity categories: no pathology, tubular degeneration/regeneration, and tubular necrosis. In close proximity to the tubular necrosis group, 4 expression profiles from the 7-day, high-dose group of PA-treated rats formed a separate cluster. This group of animals manifested both glomerular and tubular lesions. Similar grouping of animals based on gene expression changes and on pathology suggest that the gene changes were indicative of the underlying mechanism of toxicity.

Based on this paradigm, we attempted to narrow the 2400 gene set to select key genes associated with tubular toxicity. Expression profiles were categorized according to their observed pathology, and key gene changes that discriminated between a lack of pathology and degeneration/regeneration were obtained using ANOVA ( p ≤ 0.01). Similarly, genes discriminating between a lack of pathology and necrosis were obtained. Major functional categories of derived gene changes associated with necrosis and regeneration/degeneration are shown in Table 2. Only genes showing ≥1.5-fold change are described. The gene changes associated with tubular necrosis reflected the underlying physiological and molecular changes. Expressions of several key genes for xenobiotic drug metabolism were down regulated, indicating possible failure of detoxification mechanisms or loss of cells that normally express metabolism controlling genes. Other down-regulated categories of genes included renal epithelial transporters of proteins, cations, anions, and phosphate as well as mediators of energy metabolism/lipid transport. Functional categories of genes that were upregulated included inflammation/lytic enzymes and tissue remodeling. In contrast to necrosis, fewer genes were modulated in regeneration/ degeneration of tubular epithelium. These data indicate that transport, energy generation, and xenobiotic detoxification mechanisms were most likely left intact.

SVM Prediction

The utility of gene expression data to predict nephrotoxicity endpoints was evaluated using a SVM-based prediction module. The SVM-based approach uses salient features from datasets of known categories (training set) to build a classifier for predicting categories of an unclassified dataset. The ability of the trained model to correctly predict classes is tested with naïve samples of known categories (test set). This serves as a measure of robustness and potential usefulness of the data set as a prognostic tool. A SVM prediction model was trained using 120-expression profiles (generated from PA, MC, HEX, AB, and MTC treatments) and divided into four classes based on the severity of microscopic lesions. These categories were: 1, no pathology; 2, mild severity (tubular degeneration/regeneration); 3, medium severity (degeneration/regeneration and glomerular damage); and 4, high severity (tubular necrosis). The third category contained only 4 expression profiles. Although MTC and AB treatments did not cause toxicity, their expression profiles were included in the training set to increase the sample size and the predictive power of the SVM model. Using this classifier, all of the 28 expression profiles of 2-BH excluded from the training set were tested by the model for assignment of categories. The results showed the SVM-module predicted the type of pathology with 82% accuracy (Table 3). All profiles associated with no pathology class (category 1) were predicted with 100% accuracy. Of the 5/28 (18%) incorrect predictions, 3 were 1 factor off in predicting the severity and may result from the subjective nature of histopathology reading versus more objective criteria used for the microarray experiment. The predicted lesions for animals BH5148 and BH5121 were markedly different from the observed lesions. BH5148 was predicted to have no pathology (1) while it had highly severe necrosis (4). BH5121 was predicted to have high severity tubular necrosis (4) while it had only mild severity tubular regeneration degeneration (2). There are several possibilities for these discrepancies in the predicted and the observed pathology. One possible reason is that although the model predicted with high accuracy using a small training set, it was not enough to build the classifier to predict with 100% accuracy. Other likely reasons could be the variations at the levels of tissue collection or microarray hybridization. The parts of the kidney used for gene expression and microscopic observations may not have similar level of damage. Lastly, microarrays for these samples may not have recorded accurate expressions for one or more genes critical in making a prediction.

Selection of Candidate Biomarkers and Confirmation of their Expression by bDNA Assay

SVM modeling based on the microarray data demonstrated that valid histopathological predictions could be made based on gene expression changes from a compound not included in the training set. The number of genes was then narrowed to select a candidate biomarker set with ANOVA that could accurately distinguish the four pathological classes [1, no pathology; 2, mild severity (tubular degeneration/regeneration); 3, medium severity (degeneration/regeneration and glomerular damage); and 4, high severity (tubular necrosis)]. A reduced set of 49 genes was obtained that could still distinguish the region of toxicity and severity groups as demonstrated by PCA (Figure 4).

To verify the reliability of our microarray data for this gene set, we selected 10 genes by the class criteria from 4 groups described below, and examined their expression by bDNA assay. Branched DNA assay for each gene was performed on multiple samples and expressed as a fold change relative to the mean value of time-matched controls. Four samples were chosen from each of the following 3 classes: no pathology class (Amphotericin B—4 mg/kg, 7 day and MC—0.04 mg/kg, 1 day), degeneration/ regeneration (2-BH—150 mg/kg 1day), and tubular necrosis (MC—1.5 mg/kg, 7 day). Expression levels of 4 of these genes (IGFBP-1, fibrinogen A α, Slc15a2, and Slc34a2) measured by bDNA assay and microarray are shown in Figure 5. These results, which are demonstrative of the entire 10-gene subset, suggested that the expression patterns of all tested genes measured by both techniques were similar. However, the microarray data had a 2- to 4-fold compression in the amplitude of differential signal when compared to bDNA.

Discussion

The analysis of histopathology data revealed that MC, 2-BH, and HEX treatments manifested epithelial damage in the cells of the proximal tubule. Microscopic damage in proximal tubules reflected the severity of the insult, with lethal necrosis being observed only at high doses. The sublethal damage at low doses and early time was recorded as degeneration/regeneration. However, BUN and creatinine levels did not correlate well with the microscopic alterations observed in the kidneys. Although serum creatinine and BUN are commonly used in assessing toxic insult to the kidney, they are not very sensitive and do not indicate the region affected by the toxic insult (Duarte and Preuss, 1993). Usually, by the time changes in BUN and creatinine levels are detected significant damage to nephrons has already occurred. Thus, there is a clear need for biomarkers of nephrotoxicity that could indicate when damage begins, assess the level of damage, and also help to localize the lesion.

We used the cDNA microarrays to identify gene expression changes in the kidney following proximal tubular damage and to select potential molecular biomarkers of this nephrotoxicity. We showed that tubular damage and its severity were easily recognized with the associated gene changes. Clustering analysis showed that the profiles grouped according to pathology rather than compound, dose or time. Interestingly, even very low-severity microscopic observations were captured by microarray analysis. In general, the low doses of nephrotoxicants used in this study resulted in degeneration/regeneration of tubular epithelium whereas high doses caused necrosis of tubular epithelium. Using specialized statistical tools we were able to select gene changes associated with tubular degeneration/regeneration and necrosis to elucidate some underlying effects. Commensurate with the damage, epithelial necrosis was associated with a greater abundance of gene expression changes and down regulation of a large number of cellular pathways.

The gene expression changes associated with tubular necrosis indicated alterations in cellular, structural, and physiological components characteristic of necrotic damage to the renal epithelial cells. Gene expression changes reflected perturbation in phosphate-calcium homeostasis, energy metabolism, cell survival, tissue remodeling, and most notably, renal transport. The proximal tubule of the kidney is rich in transporters and transport mechanisms that contribute to drug clearance and drug-drug interactions (Lee and Kim, 2004). As a result of their concentrative capacity, proximal tubules are often the victim of toxicity. We observed pronounced down-regulation of gene expression of several transporters during tubular necrosis.

These included anion transporters (Slc21a1, Slc21a4, and Slc21a7); cation transporters (Slc22a2, Slc22a5, Slc22a6, and Slc22a10), the peptide transporter (Slc15a2), the mono-carboxylate cotransporter (Slc16A7), mitochondria carrier Slc25a11, the fatty acid cotransporter Slc27a2, and the Na-H exchanger (Slc38a3). The majority of the transporters whose expression was reduced during toxicity were the organic anion (Slc21) or cation transporters (Slc22). The Slc21/22 transporter families participate in the absorption and/or excretion of drugs, xenobiotics, and endogenous compounds in the intestine, liver, and kidney (You, 2004). The reduction in the expression of the transporters may have resulted either due to loss of tubular epithelial cells during necrosis or generalized down regulation of transport mechanism or a combination of both. We observed that even during the sublethal damage to proximal tubules (degeneration/regeneration of tubular epithelium), expressions of some of the transporters were down regulated (Slc15a2, Slc21a1, Slc21a4).

While most of the transporters were down-regulated during kidney toxicity, the sodium-phosphate co-transporter (Slc34a2) was up-regulated both during degeneration/ regeneration of tubules and tubular necrosis. Members of Slc34 family are involved in transepithelial transport of phosphate in the proximal tubules and are required in maintaining Ca-Pi homeostasis (Murer et al., 2004). We observed that concomitant with the up regulation of Slc34a2 expression, PTHR expression level was down-regulated. It is known that during calcium overload, PTH-signaling is inhibited in a feedback mechanism and is associated with increased inorganic phosphate (Pi) uptake. Further, PTH-signaling negatively regulates expression of sodium-phosphate co transporters (Kempson et al., 1995). It is thus possible that the lower expression of PTHR resulted in the alleviation of negative regulation of Slc34a2. Consistent with these indications of calcium overload, increased urinary calcium was observed in animals that manifested tubular necrosis (data not shown). In addition, there was a significant decrease in expression of SMP30 (regucalcin), a regulator of intracellular calcium, indicating perturbation of calcium homeostasis. Decrease in SMP30 expression was previously observed during cisplatin (Huang et al., 1991) and cephaloridine (Kurota and Yamaguchi, 1995) induced kidney toxicities.

Since renal secretion is an active process, the severely compromised transport machinery reflects attenuation of ATP generative processes during tubular necrosis. We also observed reduced expression of genes associated β-oxidation of fatty acids, electron transfer, fatty acid transport, and nitrogen metabolism. Besides the effect on transport mechanisms, the prolonged depletion of ATP initiates alterations in cytoskeletal structure (For review see Lien et al., 2003).

The integrity of the cytoskeleton is crucial to several aspects of renal tubule epithelial cell biology, including the polarity of the cells, permeation of solutes, and intercellular communication (Atkinson and Molitoris, 2001; Levine and Lieberthal, 2001). We observed gene changes during necrosis that suggested compromised intercellular communication and cellular polarity. Expression of the gap junction gene Connexin 26 (Cx26) was attenuated, indicating that reduced intercellular communication was a possible consequence of necrosis. Inhibition of gap junctional intercellular communication and reduced Cx26 expression has been previously shown following cadmium and MeHg induced toxicity after the rise in intracellular calcium (Yoshida et al., 1998; Jeong et al., 2000; Fukumoto et al., 2001). The expression of tight junction genes (occludin; claudin 10) was also down modulated. Tight junctions are required for the permeation of solutes across epithelia and maintaining polarity of the apical and basolateral surfaces of the cell (Fanning et al., 1999). Their inhibition initiates a back-leak of glomerular filtrate, including BUN and creatinine, and glomerular feedback mechanisms. Initiation of a glomerular feedback mechanism during necrosis was also evident from increased expression of the adenosine receptor, ADORA1. Activation of ADORA1 is known to cause reduction in the glomerular filtration rate (Munger and Jackson, 1994). The inhibition of its activity or gene deletion in mice results in an increase in the renin-angiotensin activity and abolishes tubuloglomerular feedback (Brown et al., 2001; Sun et al., 2001).

Following lethal and sublethal injuries, tissues undergo structural remodulation commensurate with the extent of damage. Consistent with the extensive remodeling of cellular structure following necrosis, we observed an increase in expression of a large number of genes related to this category. The expressions of heat shock genes HSP-27 and HSP86 were up-regulated. HSPs act as intracellular chaperones and are generally up-regulated following ischemia-reperfusion injury. They assist in restitution of cell polarity and the repair of actin microfilaments during tissue remodeling (Van Why and Siegel, 1998; Van Why et al., 2003). Additional genes associated with tissue remodeling included osteopontin, TIMP1, IGFBP1, all of which have been associated with kidney remodeling following acute renal failure (Eddy and Giachelli, 1995; Lee et al., 1997; Bohe et al., 1998; Persy et al., 1999; Huang et al., 2001). While IGFBP1 expression was upregulated, IGF1, IGFBP3, and IGFBP5 were down regulated. Similar expression pattern for IGF1 and IGFBPs were also observed following folic acid-induced proximal tubular damage (Hise et al., 1995).

In this study, we identified several genes that were mechanistically related to tubular toxicity and whose expression levels correlated to the severity levels. Expression of a subset of these genes during toxicity was also confirmed by bDNA assay. However, further work is needed to adapt these genes into a toxicity screen by validating their reversibility and developing robust and convenient assays. Recently there has been substantial debate on defining criteria for developing good biomarkers (Lesko and Atkinson, 2001; Roberts et al., 2003, Tugwood et al., 2003).

A key feature of a toxicity biomarker is that it should precede but track with the pathology (i.e., its expression level should rise or drop with an increase in severity of pathology and reverse to the normal level with recovery from toxicity). The reversibility of a biomarker is best studied by designing experiments wherein animals are allowed to recover after chronic dosing. Although we did not conduct reversibility experiments, there was an indication that the expression levels of the chosen potential markers reversed to normal levels with the recovery of toxicity (data not shown). In the case of 2-BH, which was a single-dose treatment, a high severity of toxicity was manifested in the high dose group on day 3 post-dosing but reversed by day 7. We observed that expression of several candidate biomarkers (kim-1, IGFbp-1, and Slc34a2, osteopontin, α-fibrinogen) followed an expression course that tracked with progression from necrosis to tissue regeneration (data not shown). However, recovery in expression levels of these genes may have resulted from regeneration of tubules.

An additional criterion for kidney toxicity biomarkers is to provide qualitative and quantitative measurements that give an early indication of the initial site of renal damage before gross deterioration in kidney function has occurred. Recently, several investigators have proposed Kim-1 as an early biomarker of kidney toxicity (Bailly et al., 2002; Han et al., 2002; Amin et al., 2004). We also observed that RNA expression of Kim-1 was highly induced (>5-fold) early during injury (Table 2). KIM1 is a type1 transmembrane protein that is cleaved upon kidney injury and can be detected in the urine before significant increases in BUN and creatinine levels in the serum (Bailey et al., 2002; Han et al., 2002). Similarly, ostepontin, clusterin, TweakR, and lipocalin 2 have also been proposed as potential markers of kidney toxicity (Amin et al., 2004). On our microarray chip only osteopontin and TweakR were present. We also saw consistent induction of osteopontin expression (>5-fold), both during tubular degeneration/regeneration and necrosis, but induction of TweakR was not consistent across all compounds. While these potential biomarkers were previously identified they have yet to be validated for robustness, early detection and reversibility. Before initiating time consuming and expensive validation studies, it is prudent to choose potential biomarkers that can be reliably measured and distinguished from control measurements.

Our approach to select candidate biomarkers for validation from this study was to identify genes with significant change in expression at an early time or low dose as compared to their expressions in control experiments. Expression of each gene across various treatments of all 4 toxicants was thus compared to the expression of the gene in a large set of 307 control profiles. The 307 control profiles were generated from kidney RNA of approximately 150 control animals using 2 dyes (Cy3 and Cy5). Only genes that gave a significant Z score in treatments of at least 3 out of 4 compounds have been selected for further validation. With this criterion we were able to select several genes whose expression associated with toxicity at an early time point and/or low dose before the advent of significant toxicity. This approach is exemplified with MC-induced expression of Slc34, osteopontin, and IGFBP1 (Figure 6).

Last, a predictive toxicity biomarker should also be easily measured by robust inexpensive procedures, which can be carried out on a routine basis to assess safety and prioritize lead compounds. Currently, we are exploring whether the initial safety of the compound, or a series of compounds, can be effectively assessed with a small battery of tests in cell culture models utilizing both cell culture and cytotoxicity endpoints. The safety of the desirable compounds selected from this initial screen can then be confirmed in animal models. Non-invasive biomarkers are also desirable and coupling of protein or metabolite endpoints that extend tissue observed gene expression changes are being explored. Certainly individual markers such as KIM-1 (Han et al., 2003) or GST-α have protein components that are detectable in the urine and their value as sensitive biomarkers of toxicity are currently being validated by the toxicogenomics field, including the ILSI-HESI Committee on Biomarkers. However, without further validation studies, it is too early to state whether a single or a set of biomarkers would be would be required for prediction of toxicity.

In conclusion, using DNA microarray we have identified genes whose expression correlate with the proximal tubular damage. Our future efforts are focused on the validation of potential early biomarkers of tubular damage and the expansion of our toxicogenomics database to include other region-specific toxicities of the kidney.