ToxSci Advance Access originally published online on November 17, 2006
Toxicological Sciences 2007 96(1):40-46; doi:10.1093/toxsci/kfl171
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
A Comparison of Transcriptomic and Metabonomic Technologies for Identifying Biomarkers Predictive of Two-Year Rodent Cancer Bioassays
* CIIT Centers for Health Research, Research Triangle Park, North Carolina 27709-2137
Division of Molecular Pharmaceutics, School of Pharmacy, University of North Carolina, Chapel Hill, North Carolina 27599
SAS Institute Inc., Cary, North Carolina 27513-2414
1 To whom correspondence should be addressed at CIIT Centers for Health Research, 6 Davis Drive, Research Triangle Park, NC 27709-2137. Fax: (919) 558-1300. E-mail: rthomas{at}ciit.org.
Received September 25, 2006; accepted November 14, 2006
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
Two-year rodent bioassays play a central role in evaluating the carcinogenic potential of both commercial products and environmental contaminants. The bioassays are expensive and time consuming, requiring years to complete and costing $2–4 million. In this study, we compare transcriptomic and metabonomic technologies for discovering biomarkers that can efficiently and economically identify chemical carcinogens without performing a standard two-year rodent bioassay. Animals were exposed subchronically to two chemicals (one genotoxic and one nongenotoxic) that were positive for lung and liver tumors in a standard two-year bioassay, two chemicals that were negative, and two control groups. Microarray analysis performed on liver and lung tissues identified multiple biomarkers in each tissue that could discriminate between carcinogenic and noncarcinogenic treatments. The discriminating biomarkers shared a common expression profile among carcinogenic treatments despite different genotoxicity categories and potential modes of action, suggesting that they reflect underlying cellular changes in the transition toward neoplasia. Statistical classification analysis exhibited 100% accuracy in both tissues when the number of genes was less than 5000. Additional genes reduced the predictive accuracy of the model. Serum samples were analyzed by 1H nuclear magnetic resonance (NMR) spectroscopy, and chemical-specific metabolites were removed from the spectra. The statistical classification analysis of the endogenous serum metabolites showed relatively low predictive accuracy with few metabolites in the model, but the accuracy increased to a maximum of 94% when all metabolites were added. These results suggest that individual endogenous metabolites are relatively poor biomarkers, but the metabolite profile as a whole is altered following carcinogen treatment.
Key Words: genomics; metabonomics; biomarkers; rodent cancer bioassays.
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
The two-year rodent bioassay is widely used to assess the carcinogenic potential of chemical, biological, and physical agents. Current regulatory standards require select agents to be tested for carcinogenic activity prior to commercial release, including pharmaceuticals, food additives, and pesticides. In the interest of public safety, other important commercial, industrial, or environmental chemicals are tested by the federal government through the National Toxicology Program to identify potential hazards and generate limited information on dose-response behavior for chemical risk assessments.
For both the business and risk assessment communities, there is significant motivation for developing efficient and economical methods to identify carcinogenic chemicals. From a business perspective, each bioassay requires hundreds of animals and $2–4 million per chemical (NTP, 1996
). Typically, the bioassays are performed late in the developmental pipeline after commitment of substantial resources in product development. A positive result can delay release of the product until the potential carcinogenic risks can be addressed through further study or may even result in discontinuation of the product. Thus, identifying potential carcinogens earlier in the development pipeline could provide substantial monetary savings. From a risk assessment perspective, there are approximately 80,000 chemicals registered for commercial use in the United States and 2000 more added each year (NTP, 2001). Since most have not been tested for carcinogenic activity, a more economical method to identify potential carcinogens would allow more chemicals to be tested for long-term health effects prior to human exposure.
The utility of applying transcriptomic and metabonomic technologies to identify biomarkers associated with toxicological end points has been the subject of considerable research. Thus far, most toxicology studies employing these technologies have focused on identifying biomarkers associated with relatively acute end points, such as hepatotoxicity and nephrotoxicity (Amin et al., 2004; Fielden et al., 2005; Nicholls et al., 2001; Thomas et al., 2001; Waring et al., 2001). Fewer studies have utilized these technologies to identify biomarkers that predict chronic end points, such as cancer (Ellinger-Ziegelbauer et al., 2005; Kramer et al., 2004; Nie et al., 2006). The success of these studies highlights the potential for genomic and metabonomic technology to identify biomarkers that can predict the carcinogenic activity of a specific chemical after a subchronic exposure.
The objectives of the present study were to (1) compare transcriptomic and metabonomic technologies for their ability to identify predictive biomarkers related to these chemicals and (2) demonstrate that biomarkers collected following a subchronic exposure to a chemical have the potential to predict liver and lung tumor formation observed in a two-year rodent bioassay.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
Animals and treatment.
Thirty female B6C3F1 mice were obtained from Charles River Laboratories (Raleigh, NC). Upon receipt, the mice were randomized by weight and divided into six treatment groups (Table 1). Animal treatment was initiated at 5 weeks of age. Mice were housed five per cage in polycarbonate cages in a temperature- and humidity-controlled environment with a standard 12-h light/dark cycle. All animals were given access to food (National Institutes of Health-07 ground meal; Zeigler Brothers, Gardners, PA) and water ad libitum. Animal use in this study was approved by International Animal Use and Care Committee of CIIT Centers for Health Research and was conducted in accordance with the National Institutes of Health guidelines for the care and use of laboratory animals. Animals were housed in fully accredited American Association for Accreditation of Laboratory Animal Care facilities. Pentachloronitrobenzene (PCNB) (8187 ppm), N-(1-naphthyl)ethylenediamine dihydrochloride (NEDD) (2000 ppm), and 1,5-naphthalenediamine (NAPD) (2000 ppm) were administered 7 days/week via feeding. Benzofuran (BFUR) (240 mg/kg) was administered 5 days/week via gavage in a corn oil vehicle. A feeding control (FCON) and corn oil vehicle control (CCON) were also included. All chemicals were purchased at the highest purity available (Sigma-Aldrich, St Louis, MO).
|
Following 13 weeks of exposure, the mice were anesthetized with a lethal ip dose of sodium pentobarbital (Abbott Laboratories, Chicago, IL). Blood was drawn by cardiac puncture and placed in a serum separator Microtainer tube (Benton Dickinson, Franklin Lakes, NJ), and the serum was isolated by centrifugation. The four right lung lobes were isolated by suturing and were removed and minced together in RNAlater (Ambion, Austin, TX). The left lung lobe was inflated with 10% neutral buffered formalin and stored in 10% formalin. The right, caudate, and median liver lobes were minced in RNAlater. The left liver lobe was removed and placed in 10% formalin. For histology, the formalin-fixed lung and liver tissues were embedded in paraffin blocks, sectioned at 5 µm, and stained with hematoxylin and eosin.
Gene expression microarray analysis.
Microarray analysis was performed on the lungs and livers from three animals per treatment group. A total of 18 animals were analyzed. Total RNA was isolated from the lung and liver tissue using Trizol reagent (Invitrogen, Carlsbad, CA) and further purified using RNeasy columns (Qiagen, Valencia, CA). Integrity of the RNA was verified with the Agilent 2100 Bioanalyzer (Agilent, Palo Alto, CA). The double-stranded cDNA was synthesized using the one-cycle cDNA synthesis kit (Affymetrix, Santa Clara, CA), and biotin-labeled cRNA was transcribed using the GeneChip IVT Labeling Kit (Affymetrix). Labeled cRNA was fragmented and hybridized to Affymetrix Mouse Genome 430 2.0 arrays. Microarray data were processed using Robust Multi-array Average (RMA) with a log2 transformation (Irizarry et al., 2003). The gene expression results have been deposited in the National Center for Biotechnology Information Gene Expression Omnibus (accession no.: GSE5127 and GSE5128).
Serum NMR analysis.
NMR analysis was performed on the serum from three animals per treatment group. A total of 18 animals were analyzed. NMR samples were prepared by diluting serum samples to a final volume of 600 µl with a solution of D2O, containing 2,2-dimethyl-2-silapentane-5-sulfonate sodium salt (5mM final) and sodium azide (0.02% w/v final). The 1H spectra were obtained at 399.80 MHz on a Varian Inova 400 MHz NMR spectrometer using a Varian 5-mm pulsed field gradient, inverse detection probe. The spectra were acquired with 256 scans, using a 2-s solvent presaturation period and a 200-ms Carr-Purcell-Meiboom-Gill (CPMG) filter to reduce the signals from the protein and lipid components. The total recycle time for each scan was 4.8-s. Spectral interpretation was aided by two-dimensional 1H-13C gradient heteronclear single quantum correlation (gHSQC) correlation experiments on selected samples.
Data were processed using ACD software (Advanced Chemistry Development, Toronto, Ontario). A 0.1-Hz exponential line broadening was applied to the data. The spectra were phased, baseline corrected, integrated using the ACD intelligent binning protocol, and normalized based on total bin area. The region around the residual water signal from 4.6 to 6 ppm was excluded from the analysis. To avoid inclusion of toxicant or exogenous metabolite peaks, the entire region above 7.0 ppm was excluded, as well as peaks associated with pentobarbital, propylene glycol, and lactate.
Basic statistical and annotation analysis of tissue gene expression and serum NMR data.
To obtain an overall sense of key differences in the data, gene expression measurements were analyzed using a linear model (Smyth, 2005) with a contrast between the carcinogenic (NAPD and BFUR) and noncarcinogenic treatments (NEDD, PCNB, FCON, and CCON). Genes identified as statistically significant were subject to an additional filter by selecting only those that exhibited a 
1.5-fold change. Serum NMR measurements were analyzed using two-sample t-tests. Probability values for both gene expression and NMR measurements were adjusted for multiple comparisons using a false discovery rate of 5% (Reiner et al., 2003). For the significant gene lists, a gene ontology (GO) analysis was conducted using GO Tree Machine (Zhang et al., 2004).
Statistical classification analysis of tissue gene expression and serum NMR data.
Classification analysis was performed using a combination of the Golub algorithm (Golub et al., 1999) for feature selection and a support vector machine model for classification (radial basis function kernel, C = 31.6, 
= 0.0001 for metabolite prediction; radial basis function kernel, C = 1000, = 0.001 for gene expression prediction). To assess the predictive accuracy of the model on the current data set, sixfold cross-validation was performed. The cross-validation process is outlined in Figure 1 and consisted of first randomly dividing all 18 animals into six equally sized groups (i.e., three animals per group). Five of the groups were then lumped together to use as a training set (15 animals) and the remaining group was used as the test set (three animals). The data for the animals in the test set was set aside as if we had never observed them. Feature selection was then performed on the training set using the Golub algorithm (Golub et al., 1999), and the genes or NMR spectral bins with the largest Golub statistic were used to build a support vector machine classification model. The model was then used to predict the classes for the three animals in the test set that were held out at the beginning of the process. The cross-validation process was repeated 100 times to obtain a good estimate of the predictive accuracy. Accuracy was calculated by dividing the number of correct predictions in the test set by the total number of predictions. Different numbers of genes were evaluated in the feature selection process to assess the change in predictive accuracy with gene number. The classification analysis was performed using the PCP software program (Buturovic, 2006).
|
|
RESULTS AND DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
Gross histological examination of the target tissues identified lung lesions in the NAPD treatment group, with morphological changes in all five animals. Lesions were limited to the bronchiolar epithelial cells, which exhibited karyomegaly and karyorrhexis. There was occasional peribronchiolar infiltration by neutrophils and mononuclear cells. Bronchiolar epithelial cell morphology was suggestive of regenerative hyperplasia. Liver lesions were only observed in BFUR-treated animals, with relatively minor single-cell necrosis. Given the absence of lung lesions in the BFUR treatment group and the absence of liver lesions in the NAPD treatment group, histological changes alone following a 90-day exposure to these known carcinogens were not predictive of tumor formation observed in a two-year bioassay. This result is consistent with a previous study that reported the poor predictive properties of histological lesions (Allen et al., 2004
). To identify potential transcriptional biomarkers that may be more predictive of results from a two-year rodent bioassay, gross statistical comparisons were performed between animals treated with chemicals positive in a two-year bioassay (NAPD and BFUR) and animals treated with chemicals negative in a two-year bioassay plus the vehicle controls (NEDD, PCNB, FCON, and CCON). In the lung, 187 genes were significantly upregulated in the carcinogenic chemicals and 23 genes were significantly downregulated. In the liver, 464 genes were significantly upregulated in the carcinogenic chemicals and 101 genes were significantly downregulated. A total of 33 altered genes were shared between the lung and liver. The gene expression differences between the carcinogenic chemicals and the noncarcinogenic chemicals are depicted in Figures 2A and B, and a complete list is provided as supplemental material. A subset of the significant gene expression changes was also verified using quantitative reverse transcriptase–polymerase chain reaction (qRT-PCR) (Supplemental Figs. 1 and 2).
|
Based on the general statistical comparison, there were a number of highly discriminating gene expression changes that were shared among the carcinogenic chemicals despite the diversity in genotoxicity status and tumor types observed in the two-year cancer bioassay. For genotoxicity, NAPD was positive in the Ames test while BFUR was negative (Table 1). For differences in tumor type, NAPD caused a significant increase in hepatocellular carcinomas, hepatocellular adenomas, and the combined count of alveolar/bronchiolar adenomas and carcinomas (NTP, 1978
). In contrast, BFUR caused only a significant increase in hepatocellular adenomas and alveolar/bronchiolar adenomas (NTP, 1989). The common gene expression changes among the carcinogenic chemicals despite these differences suggest that not all of the gene expression changes following a 90-day exposure are related to a specific chemical mechanism or a specific path to tumorigenesis. Rather, the gene expression alterations shared between the carcinogenic chemicals may reflect some common cellular changes related to the underlying carcinogenic process. Given that many nongenotoxic carcinogens cause a transitory increase in cell proliferation followed by a period of negative selection (Andersen et al., 1995), performing these studies following shorter exposures would likely identify biomarkers that are less robust with a transient predictive window. In addition, biomarkers from shorter exposures would probably be more chemical specific rather than those that reflect the underlying molecular changes in the carcinogenic process.
A comparison of the genes identified in our study with those identified in a previous study of nongenotoxic carcinogens in the rat liver (Nie et al., 2006) showed some common changes. In the liver, nine genes were found in common including aldehyde dehydrogenase 1a1 (Aldh1a1), aldehyde dehydrogenase 1a7 (Aldh1a7), complement component 9 (C9), peripheral benzodiazepine receptor (Bzrp), early growth response 1 (Egr1), microsomal epoxide hydrolase 1 (Ephx1), glutathione-S-transferase µ1 (Gstm1), chaperonin 10 (Hspe1), and transketolase (Tkt). No similar studies were found for comparing the gene expression changes in the lung.
A GO analysis of significant gene expression changes showed enrichment in multiple categories (Table 2). In both the lung and liver, enrichment was observed in glutathione metabolism and biosynthesis. These changes in glutathione-related processes are consistent with a variety of known biomarkers in both rodent and human tumorigenesis (Balendiran et al., 2004). Other significant GO categories in the lung were related to nitric oxide signaling, fatty acid oxidation, electron transport, and a variety of xenobiotic and endogenous metabolic processes. In the liver, significant enrichment was observed in vitamin metabolism, cytoskeletal processes, ion homeostasis, carboxylic acid metabolism, and complement activation. Apart from indicating changes in biological processes, the GO analysis identified a total of 48 genes in the liver and 33 genes in the lung classified as "extracellular" that have the potential to be noninvasive biomarkers.
|
For the serum metabonomics data, only one NMR spectral bin (
3.00–3.06) showed statistically significant changes between animals treated with the carcinogenic chemicals and animals treated with the noncarcinogenic chemicals plus the vehicle controls (Fig. 2D; Supplemental Table 2). The spectral bin was significantly reduced in the carcinogen-treated samples and was assigned to both creatine and oxidized glutathione (GSSG). These assignments were supported by the 1H-13C gHSQC spectra. Notably, a decrease in GSSG is consistent with the gene expression results, which showed a significant upregulation of glutathione reductase in both the lung and the liver of carcinogen-treated animals. Other metabolites showed treatment-specific changes, but few discriminating markers were consistently altered among the carcinogenic treatments.
A statistical classification analysis demonstrated that the tissue gene expression profiles were capable of predicting tumor formation observed in a two-year bioassay with 100% accuracy and when the number of genes used in the model was less than 5000 (Fig. 3A). As more genes were added, the predictive accuracy declined. The decline in the predictive accuracy with increasing gene numbers has been reported previously and was due to the addition of genes that are treatment specific and not related to the predicted toxicological end point (Thomas et al., 2001). The top five potential gene expression biomarkers based on selection by the statistical classification model are listed in Figure 3B. In the lung, two xenobiotic metabolizing enzymes (Gstm1 and Ephx1), an enzyme involved in cholesterol esterification (Ces1), a key kinase involved in NF
B signaling (Ikbkg), and a gene involved in the degradation of medium-chain fatty acids (Acsm1) were among the most predictive genes. Although multiple studies have examined the relationships between Gstm1 and Ephx1 polymorphisms and lung cancer, few studies have examined the relationship with respect to gene expression. Those that have examined expression changes in lung tumors were generally negative for an association (e.g., Coller et al., 2001; Spivack et al., 2003). No information was found on the expression of the remaining genes in lung tumors. In the liver, three of the top five gene expression biomarkers had no known function (E130013N09Rik, 4922503N01Rik, and AI427122). The remaining two genes were a serine protease inhibitor (Itih1) and an enzyme involved in the conversion of glucose to glucuronate (Ugdh). No reports of increased expression of these genes have been reported in liver tumors.
|
In contrast to the gene expression measurements, the statistical classification analysis of the NMR spectral bins showed relatively low predictive accuracy with few metabolites in the model and increasing accuracy as more bins were added. With all bins in the model, the predictive accuracy was 94% with a sensitivity and specificity of 100% and 83%, respectively. Efforts were made to remove all chemical-specific metabolites so that only changes in the endogenous metabolites were used in the analysis. These results suggest that individual endogenous metabolites make relatively poor biomarkers, but the metabolite profile as a whole is altered following carcinogenic treatment and may accurately predict the two-year bioassay results. Given the chemicals used in this study produce both lung and liver tumors, it is unknown what changes in the serum metabolite profile are attributed to each target organ.
The primary purpose of this study was to compare and contrast transcriptomic and metabonomic technologies for identifying biomarkers that can predict a two-year rodent cancer bioassay. The results of the study demonstrate that both transcriptional and metabonomic biomarkers collected following a subchronic exposure to a chemical have the potential to predict liver and lung tumor formation observed in a two-year rodent bioassay. The gene expression biomarkers appear to be more accurate than the serum metabolite markers, but the increased accuracy may be offset by the invasive nature and the need to develop gene expression biomarkers for each tissue. Despite these informative results, the size of the training set of chemicals was limited in this investigation and additional studies are needed to verify these observations on a larger set of chemicals. In addition, future studies will be needed to address questions related to how these biomarkers behave with dose and whether gene expression changes from nontarget tissue (e.g., kidney and blood) can predict tumor formation in another tissue.
|
SUPPLEMENTARY DATA |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
The supporting material includes (1) a complete list of the genes identified as being significantly different between animals treated with chemicals positive in the bioassay (NAPD and BFUR) and animals treated with chemicals negative in the bioassay plus the vehicle controls (NEDD, PCNB, FCON, and CCON) (Supplemental Table 1); (2) a complete list of the metabolite bins and statistical differences between animals treated with chemicals positive in the bioassay (NAPD and BFUR) and animals treated with chemicals negative in the bioassay plus the vehicle controls (NEDD, PCNB, FCON, and CCON) (Supplemental Table 2); (3) qRT-PCR validation of a subset of the differentially expressed genes in the lung (Supplemental Fig. 1); and (4) qRT-PCR validation of a subset of the differentially expressed genes in the liver (Supplemental Fig. 2). These supplementary data are available online at http://toxsci.oxfordjournals.org/.
|
ACKNOWLEDGMENTS |
|---|
Research was supported by the American Chemistry Council's Long Range Research Initiative under the Improved Methods Focus Area. The metabonomics studies were supported in part by the General Clinical Research Center Grant RR000046 from the National Institutes of Health.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTS AND DISCUSSIONSUPPLEMENTARY DATAREFERENCES |
|---|
Allen DG, Pearse G, Haseman JK, Maronpot RR. (2004) Prediction of rodent carcinogenesis: An evaluation of prechronic liver lesions as forecasters of liver tumors in NTP carcinogenicity studies. Toxicol. Pathol. 32:393–401.[CrossRef][ISI][Medline]
Amin RP, Vickers AE, Sistare F, Thompson KL, Roman RJ, Lawton M, Kramer J, Hamadeh HK, Collins J, Grissom S, et al. (2004) Identification of putative gene based markers of renal toxicity. Environ. Health Perspect. 112:465–479.[ISI][Medline]
Andersen ME, Mills JJ, Jirtle RL, Greenlee WF. (1995) Negative selection in hepatic tumor promotion in relation to cancer risk assessment. Toxicology 102:223–237.[CrossRef][ISI][Medline]
Balendiran GK, Dabur R, Fraser D. (2004) The role of glutathione in cancer. Cell Biochem. Funct. 22:343–352.[CrossRef][ISI][Medline]
Buturovic LJ. (2006) PCP: A program for supervised classification of gene expression profiles. Bioinformatics 22:245–247.
Coller JK, Fritz P, Zanger UM, Siegle I, Eichelbaum M, Kroemer HK, Murdter TE. (2001) Distribution of microsomal epoxide hydrolase in humans: An immunohistochemical study in normal tissues, and benign and malignant tumours. Histochem. J. 33:329–336.[CrossRef][ISI][Medline]
Ellinger-Ziegelbauer H, Stuart B, Wahle B, Bomann W, Ahr HJ. (2005) Comparison of the expression profiles induced by genotoxic and nongenotoxic carcinogens in rat liver. Mutat. Res. 575:61–84.[ISI][Medline]
Fielden MR, Eynon BP, Natsoulis G, Jarnagin K, Banas D, Kolaja KL. (2005) A gene expression signature that predicts the future onset of drug-induced renal tubular toxicity. Toxicol. Pathol. 33:675–683.[CrossRef][ISI][Medline]
Golub TR, Slonim DK, Tamayo P, Huard C, Gaasenbeek M, Mesirov JP, Coller H, Loh ML, Downing JR, Caligiuri MA, et al. (1999) Molecular classification of cancer: Class discovery and class prediction by gene expression monitoring. Science 286:531–537.
Irizarry RA, Bolstad BM, Collin F, Cope LM, Hobbs B, Speed TP. (2003) Summaries of Affymetrix GeneChip probe level data. Nucleic Acids Res. 31:e15.
Kramer JA, Curtiss SW, Kolaja KL, Alden CL, Blomme EA, Curtiss WC, Davila JC, Jackson CJ, Bunch RT. (2004) Acute molecular markers of rodent hepatic carcinogenesis identified by transcription profiling. Chem. Res. Toxicol. 17:463–470.[CrossRef][ISI][Medline]
Nicholls AW, Holmes E, Lindon JC, Shockcor JP, Farrant RD, Haselden JN, Damment SJ, Waterfield CJ, Nicholson JK. (2001) Metabonomic investigations into hydrazine toxicity in the rat. Chem. Res. Toxicol. 14:975–987.[CrossRef][ISI][Medline]
Nie Ay, McMillian M, Brandon Parker J, Leone A, Bryant S, Yieh L, Bittner A, Nelson J, Carmen A, Wan J, et al. (2006) Predictive toxicogenomics approaches reveal underlying molecular mechanisms of nongenotoxic carcinogenicity. Mol. Carcinog. 45:914–933.[CrossRef][ISI][Medline]
National Toxicology Program (NTP). (1978) Bioassay of 1,5-Naphthalenediamine for Possible Carcinogenicity(U.S. Department of Health and Human Services National Toxicology Program, Washington, DC).
NTP. (1989) Toxicology and Carcinogenesis Studies of Benzofuran in F344/N Rats and B6C3F1 Mice(U.S. Department of Health and Human Services National Toxicology Program, Washington, DC).
NTP. (1996) Annual Plan for Fiscal Year 1996(National Toxicology Program, Washington, DC).
NTP. (2001) Annual Plan for Fiscal Year 2001(National Toxicology Program, Washington, DC).
Reiner A, Yekutieli D, Benjamini Y. (2003) Identifying differentially expressed genes using false discovery rate controlling procedures. Bioinformatics 19:368–375.
Smyth GK. (2005) Limma: Linear models for microarray data. In Gentleman R, Carey V, Dudoit S, Irizarry RA, Huber W (Eds.). Bioinformatics and Computational Biology Solutions Using R and Bioconductor(Springer, New York).
Spivack SD, Hurteau GJ, Fasco MJ, Kaminsky LS. (2003) Phase I and II carcinogen metabolism gene expression in human lung tissue and tumors. Clin. Cancer Res. 9:6002–6011.
Thomas RS, Rank DR, Penn SG, Zastrow GM, Hayes KR, Pande K, Glover E, Silander T, Craven MW, Reddy JK, et al. (2001) Identification of toxicologically predictive gene sets using cDNA microarrays. Mol. Pharmacol. 60:1189–1194.
Waring JF, Jolly RA, Ciurlionis R, Lum PY, Praestgaard JT, Morfitt DC, Buratto B, Roberts C, Schadt E, Ulrich RG. (2001) Clustering of hepatotoxins based on mechanism of toxicity using gene expression profiles. Toxicol. Appl. Pharmacol. 175:28–42.[CrossRef][ISI][Medline]
Zhang B, Schmoyer D, Kirov S, Snoddy J. (2004) GOTree machine (GOTM): A web-based platform for interpreting sets of interesting genes using Gene Ontology hierarchies. BMC Bioinformatics 5:16.[CrossRef][Medline]
CiteULike 
Connotea 
Del.icio.us What's this?
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||