Skip Navigation


ToxSci Advance Access originally published online on March 22, 2007
Toxicological Sciences 2007 97(2):595-613; doi:10.1093/toxsci/kfm065
This Article
Right arrow Abstract Freely available
Right arrow FREE Full Text (PDF) Freely available
Right arrow Supplementary Material
Right arrow All Versions of this Article:
97/2/595    most recent
kfm065v1
Right arrow Alert me when this article is cited
Right arrow Alert me if a correction is posted
Services
Right arrow Email this article to a friend
Right arrow Similar articles in this journal
Right arrow Similar articles in ISI Web of Science
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Add to My Personal Archive
Right arrow Download to citation manager
Right arrow Search for citing articles in:
ISI Web of Science (3)
Right arrowRequest Permissions
Right arrow Disclaimer
Google Scholar
Right arrow Articles by Martin, M. T.
Right arrow Articles by Dix, D. J.
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Martin, M. T.
Right arrow Articles by Dix, D. J.
Social Bookmarking
Add to CiteULike   Add to Connotea   Add to Del.icio.us  
What's this?

Published by Oxford University Press 2007.

Toxicogenomic Study of Triazole Fungicides and Perfluoroalkyl Acids in Rat Livers Predicts Toxicity and Categorizes Chemicals Based on Mechanisms of Toxicity

Matthew T. Martin*, Richard J. Brennan{ddagger}, Wenyue Hu{ddagger}, Eser Ayanoglu{ddagger}, Christopher Lau{ddagger}, Hongzu Ren{dagger}, Carmen R. Wood{dagger}, J. Christopher Corton{dagger}, Robert J. Kavlock* and David J. Dix*,1

* National Center for Computational Toxicology {dagger} National Health and Environmental Effects Research Laboratory, Office of Research and Development, U.S. Environmental Protection Agency, Research Triangle Park, North Carolin 27711 {ddagger} Iconix Biosciences, Mountain View, California 94043

1 To whom correspondence should be addressed at National Center for Computational Toxicology (D343-03), U.S. Environmental Protection Agency, Research Triangle Park, NC 27711. Fax: (919) 541-1194. E-mail: dix.david{at}epa.gov.

Received October 24, 2006; accepted March 20, 2007


   ABSTRACT
TOP
 ABSTRACT
INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 SUPPLEMENTARY DATA
 REFERENCES
 
Toxicogenomic analysis of five environmental chemicals was performed to investigate the ability of genomics to predict toxicity, categorize chemicals, and elucidate mechanisms of toxicity. Three triazole antifungals (myclobutanil, propiconazole, and triadimefon) and two perfluorinated chemicals [perfluorooctanoic acid (PFOA) and perfluorooctane sulfonate (PFOS)] were administered daily via oral gavage for one, three, or five consecutive days to male Sprague-Dawley rats at single doses of 300, 300, 175, 20, or 10 mg/kg/day, respectively. Clinical chemistry, hematology, and histopathology were measured at all time points. Gene expression profiling of livers from three rats per treatment group at all time points was performed on the CodeLink Uniset Rat I Expression array. Data were analyzed in the context of a large reference toxicogenomic database containing gene expression profiles for over 630 chemicals. Genomic signatures predicting hepatomegaly and hepatic injury preceded those results for all five chemicals, and further analysis segregated chemicals into two distinct classes. The triazoles caused similar gene expression changes as other azole antifungals, particularly the induction of pregnane X receptor (PXR)-regulated xenobiotic metabolism and oxidative stress genes. In contrast, PFOA and PFOS exhibited peroxisome proliferator–activated receptor {alpha} agonist-like effects on genes associated with fatty acid homeostasis. PFOA and PFOS also resulted in downregulation of cholesterol biosynthesis genes, matching an in vivo decrease in serum cholesterol, and perturbation of thyroid hormone metabolism genes matched by serum thyroid hormone depletion in vivo. The concordance of in vivo observations and gene expression findings demonstrated the ability of genomics to accurately categorize chemicals, identify toxic mechanisms of action, and predict subsequent pathological responses.

Key Words: myclobutanil; propiconazole; triadimefon; PFOA; PFOS; genomic signatures.


   INTRODUCTION
TOP
 ABSTRACT
 INTRODUCTION
MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 SUPPLEMENTARY DATA
 REFERENCES
 
Microarray technology reveals changes in gene expression simultaneously across thousands of genes. These gene expression changes can be mined with novel mathematical and statistical approaches to derive expression profiles that predict outcomes and elucidate mechanisms of toxicity. Using a reference database comprised of a large collection of gene expression profiles from compound-treated rats and a collection of diagnostic and predictive gene expression signatures for a variety of pharmacological and toxicological end points mined from these data, expression profiles of chemicals of interest can be quickly investigated and mechanisms of toxicity elucidated (Natsoulis et al., 2005Go). In the current study, the DrugMatrix reference toxicogenomics database contains data from 1738 experiments in which liver gene expression profiles were collected following treatments with over 300 chemicals, primarily pharmaceuticals. The database also contains 166 gene expression–based signatures derived by querying the liver profiling data for classifiers of pharmacological or toxicological end points.

The objective of the present study was to evaluate the ability of gene expression data to predict toxicity, classify chemicals, and identify relevant biological pathways and mechanisms of toxicity. To achieve this goal, we carried out a comparative analysis among five test chemicals (myclobutanil, propiconazole, triadimefon, perfluorooctanoic acid [PFOA], and perfluorooctane sulfonate [PFOS]) and performed a contextual analysis across the large-scale toxicogenomic database. Body and organ weights, serum hormone levels, clinical chemistry, hematology, and histopathology were assessed in order to evaluate general or specific tissue/organ effects. This project represents part of the U.S. Environmental Protection Agency's (EPA) research program exploring the potential of genomics for regulatory and risk assessment applications (Dix et al., 2006Go).

Triazoles are a class of fungicides used agriculturally on fruit, vegetables, cereals, and seeds and in pharmaceutical applications for the treatment of local and systemic fungal infections. The three agricultural fungicides selected for this study all contain the 1,2,4-triazole moiety: myclobutanil, propiconazole, and triadimefon. Toxicological profiles of these three triazole fungicides are available from the Integrated Risk Information System (http://www.epa.gov/iris/). In the livers of male rats from reproductive studies, myclobutanil caused hepatocyte hypertrophy at 9.84 mg/kg/day and more generalized hepatotoxicity at 46.4 mg/kg/day. In chronic studies, myclobutanil increased liver mixed function oxidase (at 39.2 mg/kg/day) and caused hepatocyte hypertrophy at 125 mg/kg/day. Propiconazole was hepatotoxic in rats at 25 mg/kg/day, and triadimefon was also hepatotoxic in rats, reducing body weight and increasing liver weight at 25 mg/kg/day, both in chronic studies.

Two derivatives of perfluoroalkyl acid (PFAA), PFOA, and PFOS were also chosen for this study. PFAA surfactants have wide applications in industrial and consumer products. These chemicals are stable and persist in the environment and, until recently, were considered to be biologically inactive. Biomonitoring studies now detect widespread prevalence of PFOS and PFOA in humans and wildlife, and preliminary studies have suggested reproductive and developmental toxicities of these chemicals in laboratory animals (Lau et al., 2004Go). Repeat-dose studies in rats demonstrated that the liver is a primary target organ with low-dose effects on liver weights, clinical chemistry parameters, and pathology. In a subchronic study (Palazollo, 1993), male rats exposed to 0.64 mg/kg/day of PFOA experienced absolute and relative liver weight increases with hepatocyte hypertrophy. In a chronic study (Sibinski, 1987), male rats exposed to PFOA at 14.2 mg/kg/day resulted in increased liver weights, alanine aminotranferase (ALT) increases, and hepatocellular necrosis. In a chronic study (3M, 2002Go), male rats exposed to PFOS at 1 mg/kg/day also resulted in increased liver weight with hepatocyte hypertrophy.

Although the five chemicals chosen for this study have demonstrated liver toxicity, there is little mechanistic information in the public domain explaining how the structurally similar triazoles and, separately, the PFAAs result in rodent hepatotoxicity. In the present study, we used gene expression profiling to identify biological pathways affected by these environmental chemicals. Furthermore, genomic signatures and affected biological pathways were used to infer mechanisms of toxicity and categorize the chemicals. Finally, the genomic signatures from this short-term in vivo toxicogenomic study were shown to accurately predict subsequent hepatotoxicity.


   MATERIALS AND METHODS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
RESULTS
 DISCUSSION
 SUPPLEMENTARY DATA
 REFERENCES
 
Test materials.
Myclobutanil (CAS# 88671-89-0; 95.8% purity) was obtained from Dow AgroSciences (Indianapolis, IN). Propiconazole (CAS# 60207-90-1; 94.2% purity) was obtained from Syngenta Crop Protection (Greensboro, NC). Triadimefon (CAS# 43121-43-3; 96.7% purity) was obtained from Bayer CropScience (Research Triangle Park, NC). PFOA ammonium salt (CAS# 3825-26-1; > 98% purity) and PFOS potassium salt (CAS# 2795-29-3; > 98% purity) were obtained from Fluka (Steinheim, Switzerland). Each test material was dissolved in an aqueous solution of 15% Alkamuls EL-620 (CAS# 61791-12-6), a gift of Rhodia Inc.,West Point, GA. The test articles were stored at room temperature. Since the test articles are only stable in solution at room temperature for 4 days, dosing solutions were prepared in two batches; the first batch to cover days 1–3 and the second batch to cover days 4–5. The five test chemicals and vehicle were provided by EPA to Iconix Pharmaceuticals and the chemicals' identities, structures, and putative pharmacological targets were blinded to Iconix until after data analysis was completed. 1,2-[13C]-PFOA was purchased from Perkin-Elmer (Wellesley, MA) and used as the PFOA internal standard for quantitative analytical chemistry. [18O2]-perfluorooctane sulfonate ammonium salt was purchased from RTI International (Research Triangle Park, NC) and used as the PFOS internal standard.

Animals and treatments.
Ten-week-old male Sprague-Dawley (Crl:CD(SD)IGS BR) rats were obtained from Charles River Laboratories (Hollister, CA) and acclimated for approximately 5–9 days prior to dosing. Animals were housed individually in metabolic study cages. A 12-h light/dark photoperiod was maintained. Room temperature and relative humidity was maintained between 19°C–25°C (67°F–77°F) and 30–70%, respectively. Food (Lab Diet Certified Rodent Chow No. 5002) and water were available ad libitum. All aspects of the study were conducted in facilities certified by the American Association for Accreditation of Laboratory Animal Care in compliance with the guidelines of that association and the EPA/National Health and Environmental Effects Research Laboratory Institutional Animal Care and Use Committee.

A total of 90 male Sprague-Dawley rats were randomly assigned to treatment groups and sequentially ordered. The animals were divided into groups of five individuals, each group treated with a different test chemical formulation or vehicle-control by oral gavage at a volume of 10 ml/kg for one, three, or five consecutive days. Animals were necropsied without fasting on days 2, 4, or 6 (24 h after final respective dose). Myclobutanil, propiconazole, and triadimefon groups received 300, 300, and 175 mg/kg/day, respectively. PFOA and PFOS groups received 20 and 10 mg/kg/day, respectively. The doses used were estimated from an initial dose range–finding study to represent a maximum tolerated dose (MTD) sufficient to cause a reduction in body weight gain without causing mortality, severe clinical signs, or pathological findings over a 5-day, daily dosing regimen. All results from test chemicals were quantitatively and statistically assessed versus the vehicle-control treatment group.

PFAA levels in liver and serum.
Liver and serum levels of PFOA and PFOS were analyzed by a modified method of Hansen et al. (2001)Go. In brief, frozen liver samples were thawed and homogenized in water (wt:vol = 1:6). Twenty-five microliters of the homogenate were added to 1 ml 0.5M tetrabutylammonium hydrogen sulfate (pH 10) and 2 ml 0.25M sodium carbonate and mixed by vortex for 20 min. Three hundred microliters of this mixture were added to 20 µl of the 13C-PFOA internal standard (1 ng/µl) or 25 µl of the 18O-PFOS internal standard (1 ng/µl) and 5 ml methyl tertiary butyl ether. The mixture was extracted by vortex for 20 min and centrifuged at 2100 x g for 3 min. One milliliter of the organic layer was evaporated to dryness with nitrogen at 45°C. The residue was reconstituted in 400 µl 2mM ammonium acetate-acetonitrile (1:1), and an aliquot was analyzed by high-performance liquid chromatography-electrospray tandem mass spectrometry. Serum samples (25 µl) were analyzed in a similar manner.

Clinical and postmortem evaluation.
Gross necropsy observations, organ and body weights, clinical chemistry, and hematology results were collected as described previously (Ganter et al., 2005Go). For histopathological analysis, the caudal lobe of liver from all animals was preserved in 10% neutral buffered formalin. The medial lobe of each liver of all animals was dissected and snap frozen for RNA extraction.

Hormone assays.
Blood samples were collected by cardiac puncture. After clotting on ice, whole-blood samples were centrifuged at 4°C for 30 min at 1185 x g. Serum was collected and stored frozen at – 80°C. Serum testosterone, free and total T4, and total T3 levels were assayed using Testosterone, Free T4, Total T4, and Total T3 125Iodine Coat-A-Count radioimmunoassay kits (Diagnostic Products Co., Los Angeles, CA) according to the manufacturer's instructions. All hormone levels of treatment groups were compared to controls by ANOVA and subsequent t-tests.

Microarray expression profiling.
Gene expression profiling of livers from three rats per treatment group at all time points was performed on the CodeLink Uniset Rat I Expression (RU1) microarrays. The first three animals, sequentially, in each treatment group were chosen for expression profiling, barring statistically significant differences in body weight from the other animals within the treatment group or other anomalous toxicological observations. Effects on gene expression from all test chemical treatment groups were compared to the vehicle-control group. Effects of vehicle-control versus untreated animals were not determined. Comparisons of gene expression, clinical chemistry, hematology, and histopathology between individual animals were not performed. The CodeLink RU1 rat arrays were purchased from Amersham Biosciences (Piscataway, NJ, now part of GE Healthcare). The CodeLink RU1 rat array contains 9911 unique 30mer probes, and 8565 probes were used for primary data analysis (Ganter et al., 2005Go). Gene expression results for Cyp2b15 based on probes GE20381 and GE22156 (legacy probe names NM_017156_PROBE1 and X63545_PROBE1, respectively) were also reported. The specificity of both these probe sequences for Cyp2b15 (RefSeq NM_017156 [GenBank] ) was confirmed using National Center for Biotechnology Information's BLAST (Altschul et al., 1997Go). Complete gene annotation of the CodeLink RU1 rat array is available as Supplementary Data as well as the complete primary microarray data. Gene expression profiling, data processing, and quality control procedures were performed as previously described (Ganter et al., 2005Go). Briefly, gene expression profiles from liver RNA extract were analyzed using the RU1 microarray. Log-transformed signal data for all probes were arraywise normalized using Array Qualifier (Novation Biosciences, Palo Alto, CA), a nonlinear normalization procedure adapted for the RU1 microarray. Log10 ratios for each gene were computed for each experimental group as the difference between the average of the logs of the normalized probe signals within each treatment group and the average of the logs of the normalized signals for all nine vehicle-control animals in the study. Differentially expressed genes filtered at p ≤ 0.05 for all subsequent analysis of pathways. Unfiltered log10 ratios were used for analysis against genomic signatures.

Genomic signatures.
Genomic signatures were derived using published methodologies (Natsoulis et al., 2005Go). Using the sparse linear programing algorithm (El Ghaoui, 2003Go), 166 nonredundant, robust signatures were developed for the liver and compiled into the DrugMatrix reference database (Iconix Pharmaceuticals, Mountain View, California). The signatures have been further categorized by signature type and include the following classifications: structure activity, clinical pathology, pharmacology, histopathology, literature, route of administration, and body and organ weights. The majority of these signatures defined a toxic mode or mechanism of action. Each gene in a given signature has an associated weight and bias term, and the sum of the product of the log10 ratios and the gene weight, minus the bias term, determines the scalar product (SP) score (Natsoulis et al., 2005Go, Fielden et al., 2005Go). Signatures are binary classifiers, i.e., a positive SP score indicates that the query expression pattern matches the expected pattern for the positive class of the signature, while a negative SP indicates that the expression pattern does not match the positive class. Each signature has an estimated false-positive and false-negative rate, determined from cross-validation testing during signature derivation (Natsoulis et al., 2005Go). The higher the value of the SP score, the less likely a match to the signature is a false positive; positive SP scores were therefore divided into three categories: strong (SP ≥ 1), moderate (1 > SP ≥ 0.5), and weak (0.5 > SP ≥ 0), reflecting the strength of the correlation or match to the signature. Signatures were derived using at least three different chemicals from three different structure-activity classes, with the exception of signatures based on structure-activity or activity class annotations where a minimum of three individual chemicals from the same class are required. The use of multiple chemicals and activity classes ensures that compound-specific genomic signatures are avoided. The signature SP scores, pathway maximum impact, and individual gene log10 ratios were considered both independently and in combination for chemical classification and differentiation. Hierarchical clustering (UPGMA, Pearson correlation) and scatter plots were used to identify similar profiles of gene expression and/or toxicological responses, and statistical methods (Fisher exact test) used to identify enrichment of specific classes of compound or treatment among groups of treatments with similar profiles.

Pathway analysis.
DrugMatrix currently contains 135 curated biological pathways, consisting of lists of genes associated with a particular biological process (e.g., Nrf2-mediated oxidative stress response, xenobiotic metabolism) and a visual representational map of the pathway illustrating the component proteins, metabolites, and signaling molecules. Pathway perturbations were visualized by overlaying a visual transformation of the log10 ratio (usually filtered for significance) for each probe representing a given gene in the pathway.

The impact of a given treatment on a particular pathway was calculated as the probability that the list of genes perturbed in the query treatment (either upregulated genes, downregulated genes, or both up- and downregulated genes) overlapped with the genes present in each curated pathway, determined by the hypergeometric distribution. The input parameters included the universe set of genes, the query subset of genes, the subset of genes in each pathway, and the intersection of the two gene subsets. The universe was defined as the genes on the relevant platform (in this case, the CodeLink RU1 rat array) with pathway association. A multiple testing correction was applied to the calculated p value since the impacts across > 100 pathways were calculated simultaneously.

The Pathway Response Analysis tool was used to identify and quantify the relative impact of a test experiment against the curated pathway genes. By comparing the observed changes induced by a test compound relative to those induced by reference compounds, the statistical and biological significance of gene expression changes can be inferred. The impact of a test experiment across the genes in the pathway was quantified by summing the weighted log ratios. This "Sum of Log ratio" score was ranked across all experiments within a tissue. The larger the "Sum of Log ratio" score, the more the biological process is impacted. A "Percentile," a "Rank," and a "Percent Max" score was calculated for each test and reference experiment relative to the experiment achieving the highest "Sum of Log ratio" score. By doing so, the biological significance of a test experiment can be put into perspective relative to that induced by reference compounds in the reference database.

Quantitative PCR.
Measurement of select gene transcripts was carried out by quantitative PCR (qPCR) of reverse-transcribed cDNAs. All the RNA samples used were verified DNA free by PCR and quantified by Ribogreen Quantitation Kit (Invitrogen, Carlsbad, CA; R11490 [GenBank] ). RNA was reverse transcribed (Applied Biosystems [AB], Foster City, CA; cDNA Archive Kit 4322171) and 25 ng equivalent cDNA was amplified in a 20-µl volume using AB TaqMan Gene Expression Assays (see genes [assays] below) and AB Universal Master Mix 4304437. Amplification, data acquisition, and data analysis were carried out on an AB model 7900HT sequence detection system. All samples were run in duplicate. Beta-2-microglobulin was used as the endogenous control since it showed no significant change among all samples with a standard deviation of Ct < 0.5 (p = 0.273). The genes (AB assays; RefSeq; Gene ID) selected for confirmatory analysis or independent testing were Cyp3a3 (Rn01640761_gH; NM_013105 [GenBank] .1; 25642), Nqo1 (Rn00566528_m1; NM_017000 [GenBank] .2; 24314), type III iodothyronine deiodinase (Dio3) (Rn00568002_s1; NM_017210 [GenBank] .1; 29475), Cyp4a14 (Rn00598411_m1; NM_175760 [GenBank] .2; 298423), Cpt1a (Rn00580702_m1; NM_031559 [GenBank] .1; 25757), Serpina1 (Rn00574670_m1; NM_022519 [GenBank] .2; 24648), and Cyp2b3 (Rn01476085_m1; NM_173294 [GenBank] .1; 286953).


   RESULTS
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
DISCUSSION
 SUPPLEMENTARY DATA
 REFERENCES
 
Decreased Body Weight Gain and Increased Organ Weights
The triazoles and PFOS did not significantly alter mean body weight after five consecutive days of treatment (p > 0.05 compared to average weight gain in the control animals of 3.9 ± 4.1%). In contrast, PFOA decreased mean body weight relative to control animals by 7.9% (p = 0.032). Relative liver weight significantly increased following exposure to the three triazoles and PFAAs at days 3 and 5 for propiconazole and PFOA and at day 5 for myclobutanil, triadimefon, and PFOS (p < 0.01). Significant increases in relative kidney weight were observed with myclobutanil after 1 day of dosing, with triadimefon, PFOA, and PFOS after 3 days and with all five chemicals after 5 days (p < 0.05) (data not shown).

PFAA Accumulation in Liver and Serum
Only background residuals of PFOA and PFOS are detected in liver and serum of control rats (Table 1). The tissue accumulation of chemicals increased with repeated daily treatment. The hepatic PFOA appeared to reach saturation at 250 ppm after three treatments, with similar levels in serum. In contrast, the hepatic PFOS levels continued to increase with repeated treatments, reaching 400 ppm after five daily doses. Consistent with previous findings (Thibodeaux et al., 2003Go), serum PFOS levels were about one-fourth of that in liver.


View this table:
[in this window]
[in a new window]

  		TABLE 1 Liver and Serum Levels of PFOA and PFOS in the Rat after Receiving Daily Oral Gavage

 
Hepatic Histopathology
Hepatocyte eosinophilia and hepatocellular hypertrophy were noted in rats given any of the five chemicals (Table 2). Nonzonal macrovesicular lipid accumulation (steatosis) was observed with all five chemicals after three or five daily treatments. Nonzonal necrosis was observed only with PFOA after day 5. Generally, pathological events were less prevalent at 1 day than at later time points.


View this table:
[in this window]
[in a new window]

  		TABLE 2 Summary of Histopathology Findings for Triazole Fungicides and PFAAs.

 
Testosterone, Thyroid, and Clinical Chemistry Biomarkers of Hepatotoxicity
Propiconazole treatment for 5 days significantly increased serum cholesterol, whereas PFOA and PFOS significantly decreased serum cholesterol (Fig. 1). PFOA treatment for 3 and 5 days decreased serum cholesterol by more than 50%. Serum hormone levels were also significantly affected by triazoles and PFAAs. Myclobutanil significantly increased serum testosterone after 1 day of dosing (Fig. 1). In addition, there was an upward trend in serum testosterone levels after day 1 for all triazole exposures, with subsequent depression of serum testosterone levels below controls on days 3 and 5, although not statistically significant. PFOA significantly decreased serum testosterone after 3 and 5 days (Fig. 1). Myclobutanil significantly decreased total T4 after 3 and 5 days. PFAAs substantially decreased total and free T4 at all time points, to approximately one-fourth of the control level after 1 day (PFOA) and 3 days (PFOS) of dosing. Myclobutanil and triadimefon decreased total T3 at day 3, PFOS decreased T3 only on day 5, while PFOA substantially decreased total T3 at all time points (Fig. 1). Clinical chemistry and hematology findings were minor overall. Significant increases in neutrophil levels along with decreases in lymphocyte levels were observed following propiconazole and triadimefon exposures. PFAA exposure resulted in significant increases in lymphocyte levels and decreases in mean corpuscular hemoglobin concentration. No significant changes in ALT or alkaline phosphatase (ALP) were observed in any of the treatment groups.


Figure 1
View larger version (39K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		FIG. 1. Serum cholesterol, testosterone, total T4, free T4 and total T3 levels following 1-, 3-, or 5-day treatment of myclobutanil (Myclo), propiconazole (Propi), triadimefon (Triad), PFOA, and PFOS. Treatment groups significantly different from the controls (ANOVA) were designated using asterisks.

 
Microarray-Based Gene Expression Profiling
Following exposure to the three triazoles and two PFAAs, gene expression was profiled using CodeLink RU1 DNA microarrays and compared to the liver gene expression profiles of 1738 drug-dose-time combinations profiled in DrugMatrix. Propiconazole and PFOA on days 5 and 3, respectively, caused the largest number of significantly altered transcript levels in this study and were in the 95th percentile for the number of significantly altered transcripts across all 1738 reference liver experiments. Day 1 treatments generally resulted in fewer gene changes than later time points, although matches to genomic signatures (Tables 3 and 4) and effects on genes in key pathways such as oxidative stress (Table 5) demonstrate that expression changes related to key compound-induced findings were observed at both early and later time points.


View this table:
[in this window]
[in a new window]

  		TABLE 3 Summary of Genomic Signature Matches with Triazole Fungicides

 

View this table:
[in this window]
[in a new window]

  		TABLE 4 Summary of Genomic Signature Matches with PFAAs

 

View this table:
[in this window]
[in a new window]

  		TABLE 5 Effect of Triazoles and PFFAs on Genes Involved in Oxidative Stress Response

 
Genomic Signatures Identifying Mode of Action
The ability of the five compounds to induce the pharmacological and toxicological properties indicated by the 166 genomic signatures in the reference database were measured via determination of SP scores for each of the treatments at individual compound-time point combinations (Natsoulis et al., 2005Go). Positive SP scores can be divided into three categories: strong (SP ≥ 1), moderate (1 > SP ≥ 0.5), and weak (0.5 > SP ≥ 0), reflecting the strength of the correlation or match to the signature. The liver toxicity-related SP scores for myclobutanil, propiconazole, and triadimefon suggested strong, moderate, and mild hepatotoxicity, respectively (Table 3). Myclobutanil, 3-day treatment group, demonstrated the largest number of and strongest signature matches compared to 1- and 5-day myclobutanil treatment groups and the other two triazole groups at any time point. Myclobutanil is predicted to cause hepatocellular toxicity, reflected by matches to ALT elevation, hepatic necrosis, and cholestatic-related signatures. The triazoles matched multiple ALT and ALP elevation signatures, although these were not corroborated by clinical chemistry findings in the current study. All three triazoles showed weak to moderate matches to the pregnane X receptor (PXR) activation signature at varying time points and moderate to strong matches to the hepatic leukocytosis signature at later time points. Weak to moderate matches to hepatomegaly signatures for all three chemicals is consistent with the relative liver weight increases at the later time points. Notably, myclobutanil and propiconazole matched one or more hepatomegaly signatures on day 1 and triadimefon on day 3. In these cases, the signature matches were predictive, preceding a significant increase in relative liver weight by at least 2 days (4 days for propiconazole). All time points of PFOA treatment demonstrated strong matches to hepatotoxicity-related genomic signatures (Table 4). Multiple hepatocellular hypertrophy- and increased mitotic nuclei-related signature matches suggest hepatomegaly due to both hypertrophy and hyperplasia, consistent with relative liver weight increase and histopathological findings. At later time points, the strong hepatomegaly signature matches (signatures derived from body and organ weight data) were consistent with the preceding signature matches and histopathological findings. Matches to hypocholesterolemia, hypolipidemia, and peroxisome proliferator signatures were demonstrated with PFOA treatments at various time points, indicating potential effects on lipid and cholesterol metabolism, consistent with the dramatic decreases (> 50%) in serum cholesterol level observed following PFOA treatment. Although yielding fewer and weaker signature matches, PFOS treatment also was predicted to cause mild hepatotoxicity. The moderate matches for hepatomegaly and hepatocellular hypertrophy signatures were consistent with the relative liver weight increase and hypertrophic histopathology findings.

Although all five test chemicals resulted in similar in vivo findings, distinct classifications formed when SP scores for the PXR activation and hepatomegaly signatures were compared (Fig. 2). Seven of the nine triazole fungicide treatment groups matched both PXR activation and hepatomegaly signatures, as also observed with 121 other treatments in the reference database. The positive region for both PXR activation and hepatomegaly was significantly enriched for azole antifungals (p < 4.9E-11). There are seven unique pharmaceuticals in the azole antifungal chemical class with expression data in DrugMatrix, representing 60 experiments. Six of these seven pharmaceuticals contain imidazole rings; only itraconazole bears a triazole like the tested agrichemicals myclobutanil, propiconazole, and triadimefon. Of these reference experiments, 23/60 matched both PXR activation and hepatomegaly signatures. Within this activity class, econazole and miconazole exhibited dose-dependent matches to both the PXR activation and hepatomegaly signatures with 8/8 at the high dose and 1/8 at the low dose matching both signatures. Clotrimazole in 11 of 12 experiments at varying doses and durations matched both signatures. Itraconazole, oxiconazole, and sulconazole treatments were not associated with either the PXR activation or hepatomegaly signature. The structure-activity classes associated with the azole antifungal chemicals further demonstrated enrichment for specific azole antifungal chemicals with 13/20 and 9/22 treatments classified as sterol 14-demethylase inhibitors and miconazole-like sterol 14-demethylase inhibitors, respectively, matching both signatures. Ketoconazole-like sterol 14-demethylase inhibitors matched both signatures in only 2/18 experiments.


Figure 2
View larger version (33K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		FIG. 2. Scatter plot of the SP scores of the hepatomegaly signature versus the PXR activation signature. Positive SP scores indicate a match with the specific genomic signature. The nine samples from the triazole treatment group and six samples from the PFAA treatment group (black diamond) are identified within separate circles. Scatter plot regions containing treatments that matched both hepatomegaly and PXR activation signatures or matched only hepatomegaly signature are outlined separately, and the enrichment for DrugMatrix features (structure-activity class and drug signature training class) is calculated using the hypergeometric test.

 
PFOA 3- and 5-day and PFOS 5-day treatment groups matched the hepatomegaly signature without matching PXR activation. This was observed with 89 other treatments in the reference database in a set of treatments enriched for peroxisome proliferators, fibric acid peroxisome proliferator–activated receptor (PPAR) {alpha} agonists (p < 2.1E-07), and fibrate ester PPAR{alpha} agonists (p < 0.0019) (Fig. 2). The other, earlier PFOA and PFOS treatment groups did not match either the PXR activation or hepatomegaly signatures.

Pathway Analysis of Gene Expression
Gene expression for the five chemical treatment groups was analyzed across 128 literature-curated biological pathways of relevance to pharmacological and toxicological processes. There was significant induction of cytochrome P450 (CYP) genes for all five chemicals at all time points. The three triazole treatments groups perturbed CYP genes known to be regulated by constitutive androstane receptor (CAR)/PXR nuclear receptors (i.e., Cyp3a1, Cyp3a3, and Cyp3a9); while the PFAAs induced PPAR{alpha} nuclear receptor–regulated genes (i.e., Cyp4a14, Cyp7a1, Cyp7b1, Cyp8b1, and Cyp17a1). Figure 3 demonstrates the similarities in CYP gene induction between the triazole fungicides and known CAR/PXR agonists (phenobarbital, clotrimazole, and dexamethasone) and between the PFAAs and known PPAR{alpha} agonists (bezafibrate, clofibric acid, and fenofibrate). Hierarchical cluster analysis (UPGMA Pearson correlation) of 50 phase I, phase II, and phase III xenobiotic metabolism enzyme (XME) genes across the 1738 liver experiments in the reference database demonstrated the similarity of the triazole fungicides to aromatase inhibitors (p < 1.03E-4), NSAIDs Cox-2/1 inhibitors (p < 7.5E-4), and azole antifungals (p < 0.0166), whereas PFOA and PFOS clustered with a set of treatments statistically enriched with PPAR{alpha} agonists (data not shown).


Figure 3
View larger version (42K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		FIG. 3. Heat map depicting differentially expressed P450 genes under control of nuclear receptors AhR, PXR/CAR, and PPAR{alpha} in rat liver in response to triazoles and PFAAs and classic AhR, PXR/CAR, and PPAR{alpha} agonists.

 
Myclobutanil, propiconazole, and triadimefon significantly induced numerous genes associated with other phases of xenobiotic metabolism. The response of the 3-day myclobutanil treatment on the xenobiotic metabolism pathway (p < 0.025) included induction of phase I (ALDH1a1 and Alas1), phase II (Ugt1a1 and Tpmt), and phase III (Mrp3) drug metabolism genes, a pattern consistent with activation of PXR. The response on the xenobiotic metabolism pathway from 1- and 3-day propiconazole treatment (p < 0.04) and from 1- and 5-day triadimefon treatment (p < 0.003) further established the highly similar pattern of expression among the triazole chemicals.

Myclobutanil, propiconazole, and triadimefon all induced genes associated with the oxidative stress response via Nrf2 pathway (Table 5). The response of the 3-day myclobutanil treatment on the oxidative stress response pathway (p < 0.0014) included the significant induction (p < 0.05) of epoxide hydrolase 1, glutathione S-transferase-{alpha}, glutathione S-transferase-µ, glutathione reductase, glutamate-cysteine ligase, heme oxygenase 1, malic enzyme 1, NAD(P)H dehydrogenase, quinone oxidoreductase 1, thioredoxin reductase 1, and thioredoxin 2.

Myclobutanil and propiconazole were ranked in the 90th percentile for the induction of genes associated with the Nrf2-mediated oxidative stress pathway, based on the normalized distribution of the sum of the gene perturbation values (log10 ratios) across all 1738 reference liver experiments. Triadimefon was ranked in the 80th percentile. A significant correlation (linear regression, p < 0.0001) was observed between SP scores for the Hepatic Leukocytosis signature and the percent maximum induction of oxidative stress pathway genes. Both triazoles and PFAAs matched the Leukocytosis signature and the triazoles exhibited greater impact on the oxidative stress pathway. However, there was not enough differentiation between expression of the oxidative stress pathway genes following treatment of the triazoles and PFAAs to indicate that PFAAs induced hepatic leukocytosis through a mechanism distinct from the triazoles.

PFOA treatments at all time points were ranked in the 90th percentile of all 1738 liver experiments based on the percent maximum impact on genes in the PPAR{alpha} and fatty acid metabolism pathway. Altered gene expression in the PPAR{alpha} and fatty acid metabolism pathway (p < 0.0001) supports the characterization of the PFAAs as PPAR{alpha} agonists and the in vivo findings of steatosis (Fig. 4).


Figure 4
View larger version (37K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		FIG. 4. Diagram showing gene expression effects in the PPAR{alpha} and fatty acid metabolism pathway following PFOA 5-day treatment, in which a red spot indicates upregulated genes and a green spot represents downregulated genes. Significantly changed genes at p < 0.05 are indicated with a colored diamond.

 
HMG-CoA reductase is a rate-limiting enzyme involved in cholesterol biosynthesis. HMG-CoA reductase inhibitors (statins) strongly induce other cholesterol biosynthesis pathway genes, reflecting a negative feedback mechanism. In contrast, the PFAAs significantly suppressed numerous cholesterol biosynthesis pathway genes, with the exception of the upregulation of HMG-CoA reductase (Fig. 5). The suppression of the cholesterol biosynthesis pathway genes correlates to the significant decrease in serum cholesterol levels following PFOA (all time points) and PFOS (days 3 and 5) treatment. The cholesterol biosynthesis gene expression pattern following PFOA and PFOS suggests that PFOA and PFOS reduce serum cholesterol level through a mechanism distinct from that of statins.


Figure 5
View larger version (28K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		FIG. 5. Diagram showing gene expression effects in the cholesterol biosynthesis pathway following PFOA 3-day treatment, in which a red spot indicates upregulated genes and the green spot represents downregulated genes. Significantly changed genes at p < 0.05 are indicated with a colored diamond.

 
PFOA treatment significantly lowered thyroid hormone (total T4 and total T3) levels at all time points. Through hierarchical clustering (UPGMA Pearson correlation) of genes associated with the thyroid hormone release and synthesis pathway, PFOA affected the pathway genes similar to PPAR{alpha} agonists and thyroperoxidase inhibitors, the latter known for its pharmacological effects on repression of thyroid hormone synthesis. Dio3 and type I iodothyronine deiodinase (Dio1) play key roles in the regulation of thyroid hormones and were significantly perturbed by PFOA (confirmed by qPCR; Table 7), supporting the dramatic decrease in thyroid hormone levels following PFOA treatment (Fig. 1). Dio3 catalyzes the inactivation of the active thyroid hormone T3, while Dio1 deiodinates prohormone thyroxine (T4) to bioactivate T3. Numerous other experiments in the reference database significantly repressed Dio1 while upregulating Dio3, including a number of experiments with peroxisome proliferators (p < 0.03). However, the 3-day PFOA treatment was the 10th most powerful repressor of Dio1 and the 20th most powerful inducer of Dio3 relative to the 1738 liver experiments in the reference database. Experiments that exhibited Dio3 induction and Dio1 suppression are also significantly enriched for compounds causing body weight decrease (p < 1.5E-06), testosterone agonism (p < 8.4E-05), and severe liver apoptosis (p < 0.00018; Fig. 6), suggesting a correlation between the causative effects of body weight decrease and cellular injury and thyroid hormone release and synthesis. Dio3 and Dio1 expression were not significantly changed in diet-restricted animals (data not shown), indicating that these enzymes are not affected as a downstream response to reduced food consumption. PFOS caused significant Dio1 repression and Dio3 induction on day 5 of exposure, concordant with significant total T3 decreases only on day 5 and less severe depression of total and free T4 than PFOA.


View this table:
[in this window]
[in a new window]

  		TABLE 7 Comparison of qPCR and Microarray Results for Gene Expression in Liver following Treatment of Myclobutanil, Propiconazole, Triadimefon, PFOA, and PFOS

 

Figure 6
View larger version (22K):
[in this window]
[in a new window]
[Download PowerPoint slide]
  		FIG. 6. Scatter plot of log10 ratio gene expression values for DIO1 versus DIO3 genes filtered by p < 0.05. The scatter plot region containing treatments that significantly induced DIO3 but repressed DIO1 is outlined and the enrichment for structure-activity class and drug signature training class is calculated using the hypergeometric test.

 
All five chemicals affected serum testosterone homeostasis. The increasing trend in serum testosterone level and subsequent depression following exposure to the triazoles and PFOA was further evaluated using gene expression of steroid metabolism and testosterone homeostasis genes (Cyp2c, Cyp17a1, Srd5a1, Cyp3a3, and Hsd3b). There was significant modulation of Cyp3a3, Srd5a1 (myclobutanil and propiconazole), and Cyp17a1 (propiconazole and triadimefon). PFAAs significantly induced the steroid hormone biosynthesis pathway genes, Hsd17b2 and Cyp17a1, and suppressed Srd5a1. The gene expression patterns supported the observed modulation in serum testosterone homeostasis, but liver gene expression data alone were unable to elucidate a specific mechanism of action.

Comparative Analysis of Gene Expression Across Chemicals
Similar gene expression profiles were observed with all three triazole treatments, especially between myclobutanil and propiconazole (correlation coefficient = 0.69). Propiconazole exhibited the greatest number of significant gene perturbations (p < 0.05) of the triazoles and was in the top 5% of all 1738 experiments in DrugMatrix. Myclobutanil demonstrated the highest intensity and greatest number of hepatotoxicity-related signature matches. All three triazoles had similar gene expression profiles and hepatotoxicity-related signature matches compared to other azole antifungal pharmaceuticals in the database. However, the three triazole agrichemicals did not match antifungal pharmacological signatures (Table 6). Besides PXR activation signature matches, and a CAR activator match with the 5-day myclobutanil treatment, the triazole chemicals did not share a common genomic signature profile with phenobarbital treatments of the same duration. Furthermore, Cyp2b15, a gene regulated by CAR in rats (Slatter et al., 2006Go), was only perturbed in the triadimefon day 3 treatment group (p < 0.05) and was not affected by the other two triazoles in this study.


View this table:
[in this window]
[in a new window]

  		TABLE 6 Signature Match SP Scores across the Three Triazoles Compared to the Signature Profiles of Azole-like Fungicides and Phenobarbital, Respectively (only positive matches shown)

 
The two different PFAAs had low concordance at the individual gene expression level. PFOA treatments at all time points had roughly 40% greater gene perturbations than PFOS treatments, and there was low correlation in gene response (correlation coefficient = 0.52). PFOA, however, showed high correlation between its average gene response and the average gene response of chemicals classified as peroxisome proliferators (correlation coefficient = 0.73). PFOS on the other hand was poorly correlated with peroxisome proliferators in global gene expression patterns (correlation coefficient = 0.26). PFOS also showed weak matches with hepatotoxicity-related signatures and weak correlation to PPAR{alpha} agonist treatments. PFOA did not match the PPAR{alpha} agonist signature, although there were weak matches to the peroxisome proliferator and hepatotoxicity signatures, corresponding with PPAR{alpha} agonists in the reference database.

qPCR Validation of Microarray Results
Validation and supplementation of microarray results was accomplished by qPCR analysis for seven genes across all treatment groups. For a total of 90 treatment-time-gene combinations corresponding to microarray results, qPCR confirmed 25 of the significant gene perturbations across all treatment groups and 33 of the unaffected treatment-time-gene combinations for a total of 58 confirmations. Only two of the 32 discrepancies between microarray and qPCR measurements indicated significant expression changes in opposite direction (Table 7). These two discrepancies were both with Cyp3a3 gene expression at day 1, for myclobutanil and propiconazole. However, microarray and qPCR results were concordant at days 3 and 5, indicating upregulation of Cyp3a3 by all three triazoles. The supplemental Cyp2b3 interrogation demonstrated significant downregulation at day 3 for all five compounds and day 5 for Triadimefon, PFOA, and PFOS.


   DISCUSSION
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
SUPPLEMENTARY DATA
 REFERENCES
 
Several years ago, it was asserted that applying toxicogenomics to predictive toxicology would require reference databases of definitive chemical toxicities and corresponding gene expression profiles, and that these databases would need to span appropriate chemical and toxicological space encompassing the unknown chemicals (Hamadeh et al., 2002Go). Since that time, several commercial and noncommercial entities have created genomic databases for data management and storage, microarray data analysis, and data interpolation and extrapolation. Public genomic databases store and analyze data from many sources with broad experimental design parameters (e.g., ArrayExpress, ArrayTrack, and Gene Expression Omnibus). Unfortunately, these public databases include limited quantities of standardized toxicology studies. The private sector has developed reference toxicogenomic databases with more uniform experimental design and data analysis methodologies [e.g., Iconix Pharmaceuticals (http://www.iconixpharm.com); Gene Logic Inc. (http://www.GeneLogic.com), Gaithersburg, MD]. A recent National Research Council workshop on the application of genomic signatures to toxicology determined that reference databases were necessary tools for the proper interpretation of toxicogenomic data (http://dels.nas.edu/emergingissues/toxicogenomics_meet11.shtml) and that developing such a database and populating it with standardized data sets was a worthy goal. As a step in that direction, the present study was designed to assess the utility of the Iconix reference toxicogenomics database. DrugMatrix stores in vivo toxicological information and associated gene expression changes, allowing prediction of toxicological outcomes and providing detailed mechanistic information (Ganter et al., 2005Go).

The DrugMatrix database was critical in the present study for comparative toxicogenomics using genomic signatures to predict toxicological outcomes and delineate mechanisms of action. The standardized microarray data were analyzed in a multitiered analytical environment comparing gene expression and signature matches across significant chemical and biological space. The temporal predictive ability of genomic signatures from this approach has previously been demonstrated, indicating that gene expression responses can be a useful early signature of pathological responses (Fielden et al., 2005Go; Tugendriech et al., 2006). Similar predictive power was demonstrated in the present study. For example, the triazoles matched multiple ALT and ALP elevation signatures. Although these predictions were not corroborated by clinical chemistry findings in the current study, ALT or ALP elevation was significantly increased by all three triazoles following subchronic or chronic exposures at doses comparable to this short-term study (Bomhard et al., 1991Go; Hunter et al., 1982Go; O'hara et al., 1984Go). Matches to hepatomegaly signatures for all three triazoles on days 1 or 3, consistent with relative liver weight increases at later time points, were also observed. For PFOA, matches to hepatotoxicity-, hepatocellular hypertrophy–, and increased mitotic nuclei–related signatures predicted subsequent hepatomegaly and histopathological findings.

Specific, nonredundant genomic signatures derived from the reference database were used for classifying chemicals and inferring mechanisms of action. PXR activation and hepatomegaly signatures accurately categorized the blinded chemicals into two distinct groups, azole antifungals and PPAR{alpha} agonists. The agrichemical triazoles in the present study were more precisely categorized by negative matches to the therapeutic antifungal-related signatures common to pharmaceutical conazoles. Further distinctions from phenobarbital were based on negative matches to anticonvulsant/sedative therapeutic-related genomic signatures.

Differential expression of CAR- and PXR-regulated genes and matches to the PXR activation signature were common results across the three triazoles. These results are also common to the pharmaceutical conazoles in DrugMatrix and consistent with strong affects on xenobiotic-metabolizing pathways. In a prior 14-day toxicogenomic rat study with these same triazoles, similar differential expression was observed for genes regulated by the xenobiotic-sensing nuclear receptors CAR and PXR (Tully et al., 2006Go). Furthermore, conazole fungicides, including propiconazole, have been identified as ligands for human PXR in transactivation cell assays (Lemaire et al., 2006Go). The differential expression of multiple isoforms of Cyp1a, 2b, 2c, 3a, and other XME genes regulated by CAR and PXR following triazole exposure is likely significant in the hepatotoxicity and perturbation of hormone homeostasis reported with these chemicals (Goetz et al., 2007Go).

A complete understanding of triazole effects on xeno-sensing nuclear receptors from the present study is not possible, in part due to limited gene and transcript representation on the CodeLink microarray. However, the current gene expression data did elucidate mechanisms of toxicity and generate hypotheses testable in subsequent studies. For example, low scoring for the CAR activator signature, no significant induction of phenobarbital-inducible Cyp2b15, and significant downregulation of the noninducible Cyp2b3 indicated a mode of action distinct from that of the CAR activator phenobarbital. Other Cyp2b genes, however, which are known responders to xenobiotic-sensing nuclear receptors such as CAR, were not measured by the CodeLink array. Testing expression of the additional five rat Cyp2b genes and other XMEs will be critical in future triazole studies. The importance of delineating the role of CAR in triazole toxicity is based on the CAR-dependent hepatomegaly induced by cyproconazole (Peffer et al., 2006Go, 2007Go) and phenobarbital-induced liver tumors (Yamamoto et al., 2004Go) that have been demonstrated in CAR-null mice. It is possible that the response of CAR and other nuclear receptors to phenobarbital, triazole fungicides, and other chemicals may explain species differences between mice, rats, and humans. Development and improvement of genomic signatures for nuclear receptor signatures including liver X receptor (LXR), farnesoid X receptor (FXR), retinoid X receptor (RXR), PPAR, CAR, and PXR could further characterize a broad range of biological pathways and mechanisms relevant to rodent and human toxicology (Carlberg and Dunlop, 2006Go; Lala, 2005Go). Overall, the current data support the current mechanistic understanding of triazole toxicity based on XME induction and PXR activation in the literature. However, through comparisons within a reference genomic database, the present study was able to add additional mechanistic insights based on the toxicological signatures shared by the three triazoles versus pharmacological signatures of the reference compounds phenobarbital and pharmaceutical azoles. Future studies of triazole toxicity will focus on delineating the role of nuclear receptors in these putative toxicological mechanisms indicated by these genomic signatures.

Repeated exposure to high doses of PFOA and PFOS led to a substantial accumulation of the fluorochemicals in the liver and serum. In fact, the levels of PFOA might have reached a steady state after only three daily treatments, supporting results from the preliminary dose-finding study that the dose regimens chosen for the PFAA microarray experiments represented MTD. PFOA is a recognized peroxisome proliferator exerting morphological and biochemical effects characteristic of other PPAR{alpha} agonists (Kennedy et al., 2004Go; Vanden Heuvel et al., 2006Go). PFAA effects on expression of fatty acid and steroid metabolism and immunomodulatory genes in the present study are consistent with other PPAR{alpha} agonists (Anderson et al., 2004Go; Yang et al., 2002Go). Other effects are likely PPAR{alpha} independent, such as the downregulation of cholesterol biosynthesis genes and the dramatic decrease in serum cholesterol level following PFOA and PFOS treatments. There is no evidence from the literature of coordinate downregulation of these genes by other peroxisome proliferators. In fact, chemicals in the DrugMatrix reference database that decrease serum cholesterol level and suppress cholesterol synthesis genes are highly enriched for estrogen receptor agonists. The exceptional response of HMG-CoA reductase to PFAA was consistent with the upregulation by PPAR{gamma} agonists (Iida et al., 2002Go), and the overall inhibition of cholesterol biosynthesis, thus supporting PFAA as a partial PPAR{gamma} agonists (Vanden Heuvel et al., 2006Go). Other PPAR{alpha}-independent effects included the activation of CAR- and PXR-regulated Cyp3a genes not consistently regulated by other peroxisome proliferators in DrugMatrix. However, it must be noted that the PFAA-induced PXR gene expression was minor compared to that induced by phenobarbital or the three triazoles in this study. Phthalates are also peroxisome proliferators that increase the expression of PXR-regulated genes (Fan et al., 2004Go; Hurst and Waxman, 2004Go). Like phthalates, and triazoles, both PFOA and PFOS are likely inducing hepatomegaly independently of PPAR{alpha} (Yang et al., 2002Go).

In addition to effects on lipid metabolism genes, peroxisome proliferator exposure leads to alteration of genes involved in steroid metabolism. PFOA decreased levels of serum testosterone, similar to the phthalate ester plasticizers di-(2-ethyl)hexyl phthalate (DEHP) (Jones et al., 1993Go; Kim et al., 2003Go) and di-n-octyl phthalate (Jones et al., 1993Go) in rats and WY-14643 and DEHP in mice (Gazouli et al., 2002Go). These decreases in serum testosterone could be due to changes in expression of steroid metabolizing enzyme genes, as has been reported in male rats and mice following exposure to other peroxisome proliferators leading to alterations in serum levels of testosterone and estrogen (Akingbemi et al., 2001Go, 2004Go; Eagon et al., 1994Go; Gazouli et al., 2002Go; Liu et al., 1996Go).

Serum thyroid hormone levels were also reduced by PFOA and PFOS treatment. These hormonal changes were consistent with the actions of PPAR agonists (Lehotay et al., 1984Go; Miller et al., 2001Go), thus suggesting a possible linkage between PFAA, PPAR, and thyroid hormone homeostasis. Indeed, decreased serum T3 and T4 has been mechanistically linked to increased hepatocyte proliferation after exposure to peroxisome proliferators (Miller et al., 2001Go), presumably due to increased metabolism of thyroid hormones. The linkage between thyroid effects and the PPAR agonism of PFOA and PFOS is further supported by enrichment for peroxisome proliferators in the DrugMatrix compounds that repress Dio1 and induce Dio3 expression. It is possible that the effects of decreased thyroid hormone levels are exacerbated by PFAA activation of PPAR, and subsequent competitive recruitment of the heterodimerizing RXR, resulting in reduced activity of thyroid receptors (Hunter et al. 1996Go).

Through class prediction methods, the pharmaceutical industry has been using toxicogenomics to predict chemicals toxicities early in the drug discovery process (Maggioli et al., 2006Go). In the present study, toxicological outcomes and chemical class were correctly predicted by genomic signatures and affected pathways following exposure to five blinded environmental chemicals. The combination of a reference database and gene expression-based signatures clearly segregated the five chemicals into two distinct sets of chemicals. Further analysis elucidated key pathways involved in hypothesized toxic mechanisms of action for both the triazoles and PFAAs. The in vivo and histopathological findings for the triazoles and PFAA were similar, with myclobutanil and PFOA having slightly more severe hepatotoxic observations. The toxicogenomic data corroborated these findings with greater and more intense hepatotoxicity-related genomic signature matches with myclobutanil and PFOA. The hypocholesterolemia and thyroid hormone depletion following PFAA exposures were supported by the changes in expression of cholesterol biosynthesis and thyroid hormone synthesis and release pathway genes. In general, the interpretation of the toxicogenomic data was concordant with the in vivo data from this study and from other short-term in vivo studies. This concordance was achieved despite blinding the data analysts to compound identities prior to the findings being reported to EPA and the structural dissimilarity of PFOA and PFOS to other compounds in the reference database.

In the current study, toxicogenomics identified modes and mechanisms of action for the five environmental chemicals, including XME induction by nuclear receptors PXR by triazoles, and PPAR{alpha} by PFAA that ultimately lead to hepatotoxicity. Additional toxicological signatures common to the three agrichemical triazoles distinguished them from reference compounds such as phenobarbital and the pharmaceutical azoles. For PFOA and PFOS, affected pathways and signatures extended beyond just PPAR{alpha} activation, with evidence of PPAR{gamma} agonism and perturbation of cholesterol and thyroid hormone homeostasis. In considering how toxicogenomics might be more generally applied to the classification of environmental chemicals, the present study serves as an example of the value of a gene expression database linked to toxicological outcomes for comparative analyses. It also demonstrates the utility of developing and using genomic signatures from a reference database and pathway-based analysis methods. Following this example, toxicogenomics will likely be useful in environmental risk assessments for evaluating mechanisms of action and predicting toxicity.


   SUPPLEMENTARY DATA
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 SUPPLEMENTARY DATA
REFERENCES
 
Supplementary data are available online at http://toxsci.oxfordjournals.org/.


   NOTES
 
Disclaimer: The information in this document has been funded wholly by the EPA. This work was reviewed by EPA and approved for publication but does not necessarily reflect official Agency policy. Mention of trade names or commercial products does not constitute endorsement or recommendation by EPA for use.


   ACKNOWLEDGMENTS
 
The authors wish to thank Drs Robert D. Zehr and Mark A. Strynar for their analytical chemistry support. Conflicts of Interest: None declared.


   REFERENCES
TOP
 ABSTRACT
 INTRODUCTION
 MATERIALS AND METHODS
 RESULTS
 DISCUSSION
 SUPPLEMENTARY DATA
 REFERENCES
 
3M. 104-Week Dietary Chronic Toxicity and Carcinogenicity Study with Perfluorooctane Sulfonic Acid Potassium Salt (PFOS; T-6295) in Rats. Final Report, 3M T-6295 (Covance study no.: 6329-183) (2002) Vol. I–IX:4068. 2 January 2002. 3M, St Paul, MN.

Akingbemi BT, Ge R, Klinefelter GR, Zirkin BR, Hardy MP. Phthalate-induced Leydig cell hyperplasia is associated with multiple endocrine disturbances. Proc. Natl. Acad. Sci. USA (2004) 101:775–780.[Abstract/Free Full Text]

Akingbemi BT, Youker RT, Sottas CM, Ge R, Katz E, Klinefelter GR, Zirkin BR, Hardy MP. Modulation of rat Leydig cell steroidogenic function by di(2-ethylhexyl)phthalate. Biol. Reprod. (2001) 65:1252–1259.[Abstract/Free Full Text]

Altschul SF, Madden TL, Schäffer AA, Zhang J, Zhang Z, Miller W, Lipman DJ. Gapped BLAST and PSI-BLAST: A new generation of protein database search programs. Nucleic Acids Res. (1997) 25:3389–3402.[Abstract/Free Full Text]

Anderson SP, Dunn C, Laughter A, Yoon L, Swanson C, Stulnig TM, Steffensen KR, Shandraratna RA, Gustafsson JA, Corton JC. Overlapping transcriptional programs regulated by the nuclear receptors peroxisome proliferators-activated receptor alpha, retinoid X receptor, and liver X receptor in mouse liver. Mol. Pharmacol. (2004) 66:1440–1452.[Abstract/Free Full Text]

Bomhard E, Schilde B. MEB 6447: Chronic Toxicity and Cancerogenicity Studies on Wistar Rats with Administration in Diet Over a Period of 105 Weeks: Lab Project (1991) Number: 20774: 101922. Unpublished study prepared by Bayer Ag. Department of Toxicology. (Environmental Protection Agency MRID No. 42153901).

Carlberg C, Dunlop TW. An integrated biological approach to nuclear receptor signaling in physiological control and disease. Crit. Rev. Eukaryot. Gene