TY - JOUR
AU - , van Delft, Joost H. M.
AB - Abstract Assessing the potential carcinogenicity of chemicals for humans represents an ongoing challenge. Chronic rodent bioassays predict human cancer risk at only limited reliability and are simultaneously expensive and long lasting. In order to seek for alternatives, the ability of a transcriptomics-based primary mouse hepatocyte model to classify carcinogens by their modes of action was evaluated. As it is obvious that exposure will induce a cascade of gene expression modifications, in particular, the influence of exposure time in vitro on discriminating genotoxic (GTX) carcinogens from nongenotoxic (NGTX) carcinogens class discrimination was investigated. Primary mouse hepatocytes from male C57Bl6 mice were treated for 12, 24, 36, and 48 h with two GTX and two NGTX carcinogens. For validation, two additional GTX compounds were studied at 24 and 48 h. Immunostaining of γH2AX foci was applied in order to phenotypically verify DNA damage. It confirmed significant induction of DNA damage after treatment with GTX compounds but not with NGTX compounds. Whole-genome gene expression modifications were analyzed by means of Affymetrix microarrays. When using differentially expressed genes from data sets normalized by Robust Multi-array Average, the two classes and various compounds were better separated from each other by hierarchical clustering when increasing the treatment period. Discrimination of GTX and NGTX carcinogens by Prediction Analysis of Microarray improved with time and resulted in correct classification of the validation compounds. The present study shows that gene expression profiling in primary mouse hepatocytes is promising for discriminating GTX from NGTX compounds and that this discrimination improves with increasing treatment period. primary mouse hepatocytes, transcriptomics, carcinogens, class discrimination, time The carcinogenic potential of chemical compounds is currently evaluated through application of chronic rodent bioassays. This strategy, however, generates a high false-positive rate and is therefore not quite reliable (Ennever and Lave, 2003) while simultaneously being expensive and time consuming as it requires the use of many animals and large quantities of the test compound during what is usually a 2-year period of study. Therefore, the usefulness of in vitro assays for predicting human carcinogenicity in vivo is increasingly explored (Dambach et al., 2005). As the liver is the main organ for metabolism of many compounds, including procarcinogens, and also represents a major target organ for chemical carcinogens in vivo, in vitro models based on hepatocyte systems are frequently considered for the prediction of toxicity. Among these, primary hepatocytes, precision cut liver slices, and hepatic cell lines are well-established in vitro models (Butterworth et al., 1989; Dambach et al., 2005; Groneberg et al., 2002). Combining such in vitro models with powerful genomic-based methods might provide high-throughput screening methods for predicting toxic risks in humans. The application of DNA microarray technology enables examining differential gene expression for many genes simultaneously. These DNA microarray technologies have the potential to both enhance our understanding of the mechanism underlying a compound's carcinogenic effects and identify a characteristic set of genes from a database of reference profiles, which may allow the prediction of an unknown compound's mode of action (Eun et al., 2008; Le Fevre et al., 2007). Based on their mechanisms of action, chemical carcinogens are classified as genotoxic (GTX) or nongenotoxic (NGTX) carcinogens (Ashby, 1992; Silva Lima and Van der Laan, 2000). A GTX compound may covalently bind with DNA and cause direct damage to DNA by adduct formation. These lesions may not or incorrectly be repaired, which leads to mutations and ultimately to the formation of tumors (Eun et al., 2008). A NGTX compound, on the other hand, lacks the ability to induce DNA damage directly or indirectly. Because NGTX compounds have different features from GTX compounds, it may be hypothesized that GTX and NGTX induce distinct gene expression profiles, which consequently may be used for mechanism-based classifying unknown compounds as GTX or NGTX (Ashby, 1992; Silva Lima and Van der Laan, 2000). Length of exposure has a major influence on gene expression profiles in cells in vitro, and this appears to have a higher impact than compound's dose (Crump et al., 2008; Hockley et al., 2006; Kruse et al., 2007; Lambert et al., 2009). Therefore, it may be hypothesized that the exposure period affects class discrimination based on transcriptomics data. The current study therefore aimed to investigate whether incorporating time dependency as a parameter improves mechanism-based classification models to discriminate GTX and NGTX carcinogens. For this, we apply whole-genome gene expression analysis to sandwich-cultured primary mouse hepatocytes. Primary mouse hepatocytes are metabolic competent and are of additional advantage because of the availability of the complete sequence of the mouse genome (Waterston et al., 2002) and of transgenic mouse models, which allow relevant mechanistic studies. Gene expression profiles generated at four time points from primary mouse hepatocytes treated with four model compounds were used in the training phase. From these, a classification model was derived, which was subsequently validated by using a set of two additional compounds. The functionality of the resulting classifiers was analyzed by MetaCore. Because histone H2AX is phosphorylated on serine-139 by ataxia telangiectasia mutated kinase in response to DNA damage, particularly double-strand (ds) breaks (Koike et al., 2008; Rogakou et al., 1998), we analyzed the formation of phosphorylated H2AX (γH2AX) foci in order to confirm GTX properties of the compounds. MATERIALS AND METHODS Chemicals Dulbecco's modified Eagle's medium, fetal calf serum, Hanks’ calcium- and magnesium-free buffer, Alexa fluor 488 goat anti-mouse IgG antibody, insulin, and Trizol were obtained from Invitrogen (Breda, The Netherlands). Glucagon, hydrocortisone, collagenase type IV, benzo(a)pyrene (BaP), aflatoxin B1 (AFB1), cyclosporin A (CsA), dimethylnitrosamine (DMN), mitomycin C (MitC), trypan blue, dimethyl sulfoxide (DMSO), bovine serum albumin (BSA), 4′,6-diamidino-2-phenylindole (DAPI), and Tween-20 were purchased from Sigma-Aldrich (Zwijndrecht, The Netherlands) and 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) from C.N. Schmidt BV (Amsterdam, The Netherlands). Triton X-100, NaCl, Na2HPO4.2H2O, and NaH2PO4 were obtained from Merck (Darmstadt, Germany) and paraformaldehyde from ICN biomedicals (Auroro, OH). The anti-phospho-Histone H2A-X (ser139) Clone JBW 301 antibody was purchased by Upstate Biotechnology (Lake Placid, NY). Vectashield Mounting Medium was supplied by Vector Labs (Burlingame, CA). Collagen Type I Rat Tail was obtained from BD BioSciences (Bedford, MA). The iScript cDNA Synthesis kit en iQ SYBR Green Supermix were purchased from Bio-Rad (Hercules, CA). Primers were designed using Primer Express 2.0 from Applied Biosystems (Foster City, CA) and purchased from Operon (Cologne, Germany). The RNeasy minikit was obtained from Qiagen, Westburg B.V. (Leusden, The Netherlands). The 5× MegaScript T7 Kit was obtained from Ambion (Austin, TX). The GeneChip Expression 3′-Amplification Two-Cycle cDNA Synthesis Kit and Reagents, the Hybridization, Wash, and Stain Kit and the Mouse Genome 430 2.0 Arrays were purchased from Affymetrix (Santa Clara, CA). Animals Permission for isolating primary hepatocytes from mice was obtained from the Animal Ethical Committee. Adult male C57Bl/6 mice, weighing 20–25 g, were obtained from Charles River GmbH, Sulzfeld, Germany. This mouse strain was chosen because it is frequently used in toxicological and pharmacological investigations, and it is a common background for transgenic mouse strains. The animals were housed in macrolon cages with sawdust bedding at 22°C and 50–60% humidity. The light cycle was 12-h light/dark. Feed and tap water were available ad libitum. Isolation of hepatocytes. Hepatocytes were isolated by a two-step collagenase perfusion method according to Seglen (1976), with modifications as described before (Mathijs et al., 2009). Cell Culture and Treatment Cell suspensions with cell viability > 85%, determined by trypan blue exclusion, were brought into culture in a collagen-collagen sandwich formation as described before (Mathijs et al., 2009; Beken et al., 1998). Dead cells are removed during this procedure. Prior to treatment, primary cultures of mouse hepatocytes were allowed to recover for 40–42 h at 37°C in a humidified chamber with 95%/5% air/CO2 in serum-free culture medium supplemented with insulin (0.5 U/ml), glucagon (7 ng/ml), hydrocortisone (7.5 μg/ml), and 2% penicillin/streptomycin (5000 U/ml penicillin and 5000 μm/ml streptomycin). Culture medium was refreshed every 24 h. After the recovery period, the culture medium was replaced by culture medium containing one of the selected four compounds or with vehicle control (Table 1). Cells were incubated for 12, 24, 36, or 48 h before being used for detecting dsDNA breaks or harvested for RNA isolation by adding Trizol reagent. These time points were chosen as they are regularly used in genotoxicity tests, including the unscheduled DNA synthesis test in primary hepatocytes, and because metabolic stability of the cells until 48 h has been demonstrated previously (Mathijs et al., 2009). Three independent replicate biological experiments were conducted for each compound. TABLE 1 Overview for the Compounds Used in Primary Mouse Hepatocyte Exposures and the Number of Differentially Expressed Genes for the Treatments per Time Point Chemical Abbreviation Solvent and dose (vol/vol, %) Dose PAM training/test Differentially expressed genes 12 h 24 h 36 h 48 h GTX carcinogens     Benzo(a)pyrene BaP DMSO, 0.5 30μM Training 83 1407 1523 2189     Aflatoxin B1 AFB1 DMSO, 0.5 15μM Training 563 2343 1773 1701     Dimethylnitrosamine DMN Culture medium 2mM Test 3073 3679     Mitomycin C MitC Ethanol, 0.5 5μM Test 5276 8222 NGTX carcinogens     Cyclosporin A CsA DMSO, 0.5 1μM Training 311 857 272 158     2,3,7,8-Tetrachlorodibenzo-p-dioxin TCDD DMSO, 0.5 1nM Training 376 834 403 551 PAM 1070 3910 3084 3616 Chemical Abbreviation Solvent and dose (vol/vol, %) Dose PAM training/test Differentially expressed genes 12 h 24 h 36 h 48 h GTX carcinogens     Benzo(a)pyrene BaP DMSO, 0.5 30μM Training 83 1407 1523 2189     Aflatoxin B1 AFB1 DMSO, 0.5 15μM Training 563 2343 1773 1701     Dimethylnitrosamine DMN Culture medium 2mM Test 3073 3679     Mitomycin C MitC Ethanol, 0.5 5μM Test 5276 8222 NGTX carcinogens     Cyclosporin A CsA DMSO, 0.5 1μM Training 311 857 272 158     2,3,7,8-Tetrachlorodibenzo-p-dioxin TCDD DMSO, 0.5 1nM Training 376 834 403 551 PAM 1070 3910 3084 3616 Open in new tab TABLE 1 Overview for the Compounds Used in Primary Mouse Hepatocyte Exposures and the Number of Differentially Expressed Genes for the Treatments per Time Point Chemical Abbreviation Solvent and dose (vol/vol, %) Dose PAM training/test Differentially expressed genes 12 h 24 h 36 h 48 h GTX carcinogens     Benzo(a)pyrene BaP DMSO, 0.5 30μM Training 83 1407 1523 2189     Aflatoxin B1 AFB1 DMSO, 0.5 15μM Training 563 2343 1773 1701     Dimethylnitrosamine DMN Culture medium 2mM Test 3073 3679     Mitomycin C MitC Ethanol, 0.5 5μM Test 5276 8222 NGTX carcinogens     Cyclosporin A CsA DMSO, 0.5 1μM Training 311 857 272 158     2,3,7,8-Tetrachlorodibenzo-p-dioxin TCDD DMSO, 0.5 1nM Training 376 834 403 551 PAM 1070 3910 3084 3616 Chemical Abbreviation Solvent and dose (vol/vol, %) Dose PAM training/test Differentially expressed genes 12 h 24 h 36 h 48 h GTX carcinogens     Benzo(a)pyrene BaP DMSO, 0.5 30μM Training 83 1407 1523 2189     Aflatoxin B1 AFB1 DMSO, 0.5 15μM Training 563 2343 1773 1701     Dimethylnitrosamine DMN Culture medium 2mM Test 3073 3679     Mitomycin C MitC Ethanol, 0.5 5μM Test 5276 8222 NGTX carcinogens     Cyclosporin A CsA DMSO, 0.5 1μM Training 311 857 272 158     2,3,7,8-Tetrachlorodibenzo-p-dioxin TCDD DMSO, 0.5 1nM Training 376 834 403 551 PAM 1070 3910 3084 3616 Open in new tab Detection of dsDNA breaks. dsDNA breaks were detected by immunostaining of the phosphorylated histone H2AX (γH2AX) as described before (Hamer et al., 2003). Hepatocytes, used for the detection of dsDNA breaks, were cultured in a sandwich configuration on top of a cover slip. After desired exposure times, medium was aspirated and the cells were rinsed with 1× PBS before fixation in 4% paraformaldehyde for 10 min at ambient temperature. The cells were washed again with 1× PBS and made permeable with 0.2% Triton X-100 at 4°C for 5 min. Thereafter, cells were washed two times with 1× PBS for 5 min and blocked during 1 h at 37°C with blocking buffer (1% BSA + 0.5% Tween-20 in 1× PBS). Samples were washed two times with 1× PBS for 5 min and incubated with the first antibody (Mouse monoclonal γH2AX [Ser-139]) 1:1000 in dilution buffer (0.5% BSA + 0.5% Tween-20 in 1× PBS) during 2 h at 37°C. After two washing steps with 1× PBS, cells were incubated with the second antibody (Alexa 488 labeled) 1:1000 in dilution buffer during 1 h at 37°C and washed again two times with 1× PBS. Samples were incubated with DAPI for 15 min at room temperature, dehydrated in series of ethanol solutions (70, 90, and 100%) for 3 min, and air dried. Vectashield Mounting medium was used to stick the cover slips onto microscopic glass slides. Microscopic photographs were taken and samples were scored manually. A damage score, ranging from 0 to 4, determined by the grade of damage was given to 100 cells per sample. The damage scores were multiplied by the amount of cells with that score and all together calculated to get a total percentage of damage in each sample. For the statistical analysis, the t-test was used with a significance level of 5%. RNA isolation. Total RNA was isolated from cultured mouse hepatocytes using Trizol and the RNeasy kit according to the manufacturer's protocol. RNA concentrations were measured by means of a spectrophotometer and the quality of each RNA preparation was determined using a bio-analyzer (Agilent Technologies, Amstelveen, The Netherlands). Extracted RNA was stored at −80°C until further analysis. Whole-genome gene expression analysis. Targets were prepared according to the Affymetrix protocol. The complementaryRNA targets were hybridized according to the manufacturer's recommended procedures on high-density oligonucleotide gene chips (Affymetrix Mouse Genome 430 2.0 GeneChip arrays). The gene chips were washed and stained using an Affymetrix fluidics station and scanned by means of an Affymetrix GeneArray scanner. A total of 60 GeneChips was run (one chip per RNA sample; 60 RNA samples were generated from three experiments with four time points for four compounds and solvent control DMSO). Quality controls (including scaling factors, average intensities, present calls, background intensities, noise, and raw Q values) were within acceptable limits, according to the manufacturer, for all chips. Hybridization controls, BioB, BioC, BioD, and CreX, were identified on all chips and yielded the expected increases in intensities. Data Analysis Selection of differentially expressed genes. Data sets from 60 GeneChips were obtained in this study. Raw data were imported into ArrayTrack (Tong et al., 2003) and normalized using Robust Multi-array Average (integrated into ArrayTrack) (Irizarry et al., 2003). Present-Marginal-Absent calls were used to identify and omit genes with probe sets of poor quality or with low expression values (Affymetrix, 2002). Probe sets are a collection of probes designed for a given gene. Subsequently, the total remaining genes for each time point (12 h: 22,986; 24 h: 23,606; 36 h: 23,811; 48 h: 24,358; and 26,666 in total) were logarithmically (base 2) transformed and corrected for their vehicle control. For each time, genes were then filtered for those for which expression was in at least one compound upregulated or downregulated by a minimum of 1.2-fold in at least two of three experiments with expressions altered in the same direction in all replicate and with a mean fold upregulated or downregulated of 1.5 (Shi et al., 2006). The generated list with differentially expressed genes (log2 ratios) was used for hierarchical clustering analysis (HCA) with the Ward's minimum variance method (Ward, 1963) and prediction analysis of microarray (PAM). The gene expression data discussed in this publication have been deposited in ArrayExpress (accession number: E-MEXP-2209), the European Bioinformatics Institute database (http://www.ebi.ac.uk/arrayexpress/). Correlation Analysis Genes that correlated with γH2AX foci formation were identified with the gene expression profile analysis suite (GEPAS 4.0; CIPF, Valencia, Spain). The values for γH2AX foci formation, corrected by subtracting the values for the DMSO from the values for the exposed samples and log (base 2) transformed, were correlated with the log (base 2) transformed expression changes of the filtered genes (24 h: 23,606 and 48 h: 24,358). Spearman correlation coefficients were calculated and correlating genes were selected by p < 0.05 for PAM software. Pathway Analysis The complete data set of genes from all compounds and all experiments (log2 of the ratios for treated control) were uploaded in T-profiler (Boorsma et al., 2005) to identify transcriptional regulation of biochemical pathways and biological processes in the complete data set of genes without any preselection of genes. T-profiler uses the t-test to score the difference between the mean expression level of predefined groups of genes and that of all other genes without any preselection of modulated genes (Boorsma et al., 2005). Significance was determined by generating an E value, a Bonferroni corrected p value. Pathways and processes were significant when E values were below 0.05 and used for hierarchical clustering GenePattern (Reich et al., 2006) using Pearson correlation with pairwise complete linkage. Class Discrimination and Functional Analyses of Classifiers The PAM software was used for identifying genomic classifiers and for discrimination of GTX and NGTX carcinogens (Tibshirani et al., 2002). PAM uses gene expression data to identify a subset of genes that best characterize each class by using the method of “nearest shrunken centroids” (Tibshirani et al., 2002). For this analysis, the gene list with differentially expressed genes (Table 1) was used. Tenfold cross validation was applied in all time points, using each treatment as an independent experiment. For each time point, a set of genes (classifiers) was generated by using the smallest estimated misclassification error rate and a > 80% test probability. Additionally, two well-known GTX compounds were used to validate the classification. Genes correlating with γH2AX foci formation at 24 and 48 h were also used for PAM to investigate if this improves the classification. The classifiers from the differentially expressed genes from the each time point were further analyzed for functional annotation by MetaCore using the shortest path algorithm (GeneGo, San Diego, CA). Again, data derived at 12 h were omitted. The total list of genes, without genes of poor quality, were selected for each time point as described before (24 h: 23,606; 36 h: 23,811; and 48 h: 24,358) and used as background list in MetaCore. A false discovery rate of 0.2 was used to select significant MetaCore maps. MetaCore was also used to generate biological networks. RESULTS Time-Dependent Gene Expression Profiles We selected four compounds in order to examine time-dependent differences in class discrimination performance by mouse hepatocyte gene expression changes, which might be induced by GTX or NGTX carcinogens. Only noncytotoxic doses as determined by the (3-(4,5-Dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide, a tetrazole) assay (∼80% viability) were used (Table 1). Only one dose was investigated, as several studies have shown that in in vitro models, treatment time has a much larger impact on chemically induced gene expression modifications than dose (Hockley et al., 2006; Lambert et al., 2009). In total, 60 GeneChips were run, with no failures (one chip per RNA sample; 60 RNA samples were generated from three experiments with four time points for four compounds and solvent control DMSO). No overt outliers appeared between the three replicate experiments, as investigated by HCA. For each compound and time point, the differentially expressed genes were retrieved (Table 1, Supplementary Data). Differential expression was not validated by real time (RT)-PCR, as the micro array quality control consortium demonstrated an excellent correlation quantitative RT-PCR data with Affymetrix GeneChip data for hundreds of genes (Canales et al., 2006). To visualize variability and time-dependent effects on gene expression profiles from GTX and NGTX compounds, HCA for each time point was done by including only modulated gene expressions from that treatment period (Fig. 1). For all compounds, the clustering of replicates improved with treatment time: At 12 h, it was quite poor, while at 48 h, for all compounds, replicates grouped together. Furthermore, only at 48 h, the GTX and NGTX classes were completely separated from each other. FIG. 1. Open in new tabDownload slide Hierarchical clustering per time point of treatments using genes differentially expressed in primary mouse hepatocytes after exposure to BaP, AFB1, TCDD, or CsA for (A) 12, (B) 24, (C) 36, and (D) 48 h (all in triplicate experiments). FIG. 1. Open in new tabDownload slide Hierarchical clustering per time point of treatments using genes differentially expressed in primary mouse hepatocytes after exposure to BaP, AFB1, TCDD, or CsA for (A) 12, (B) 24, (C) 36, and (D) 48 h (all in triplicate experiments). Detection of dsDNA Breaks The generation of phosphorylation of γH2AX (forming of γH2AX foci), which is a signature for dsDNA damage (Fernandez-Capetillo et al., 2004; Rogakou et al., 1998), after 24- and 48-h exposure to the four compounds is shown in Figure 2. Only one dose and two time points were investigated. At 48 h, levels of γH2AX foci in cells treated with GTX compounds were significantly increased as compared to DMSO controls, while at 24 h, only AFB1 had induced a significant increase compared to DMSO controls. Cells treated with the NGTX compounds, on the other hand, showed no changes in levels of γH2AX foci compared to the controls, except for CsA at 48 h when a significant decrease compared to the DMSO controls was shown. Time had only a significant effect on the cells treated with BaP, where more γH2AX foci were detected as exposure time elongated. FIG. 2. Open in new tabDownload slide Levels of γH2AX foci, a measure for dsDNA breaks, in primary mouse hepatocytes exposed to BaP, AFB1, TCDD, CsA, or solvent for 24 or 48 h (average with SD of three experiments). *Significant (p value < 0.05) compared to solvent controls; #significant (p value < 0.05) over time. FIG. 2. Open in new tabDownload slide Levels of γH2AX foci, a measure for dsDNA breaks, in primary mouse hepatocytes exposed to BaP, AFB1, TCDD, CsA, or solvent for 24 or 48 h (average with SD of three experiments). *Significant (p value < 0.05) compared to solvent controls; #significant (p value < 0.05) over time. Pathway Analysis Pathways and processes significantly altered in hepatocytes in at least two experiments and altered in the same direction in all three replicates were considered significant. For each compound and each time point, significantly altered pathways were selected (in total 471 pathways; Supplementary Data). Only five pathways were significantly altered at 12 h. At 24 h, 215 pathways were significantly altered, 175 at 36 h and 382 pathways at 48 h. More pathways were significantly changed after treatment of hepatocytes with a GTX compound. Hierarchical clustering of these pathways with GenePattern (Reich et al., 2006) is presented in Figure 3, showing distinct patterns for groups of pathways. Processes related to macromolecules, cellular organization, and regulation are mostly upregulated by all compounds at 24 h, and signal transduction is only upregulated by the NGTX compounds at 24 h. Immune responses; extracellular, membrane, and cytoplasm processes; lipid-related processes; and metabolic and biosynthetic processes are downregulated by GTX compounds and upregulated by NGTX compounds at later time points. p53-related gene groups are upregulated by GTX compounds at 36 h. Cytochrome P450–related pathways were clearly triggered by treatment with BaP and TCDD. FIG. 3. Open in new tabDownload slide Hierarchical clustering of all significantly altered pathways after treatment with AFB1, BaP, TCDD, or CsA for 12, 24, 36, and 48 h based on the t-values generated by T-profiler. Yellow and blue mean upregulation and downregulation, respectively. FIG. 3. Open in new tabDownload slide Hierarchical clustering of all significantly altered pathways after treatment with AFB1, BaP, TCDD, or CsA for 12, 24, 36, and 48 h based on the t-values generated by T-profiler. Yellow and blue mean upregulation and downregulation, respectively. Class Discrimination For each time point, the log2 of the ratio's for the differentially expressed genes upon treatments with BaP, AFB1, TCDD, or CsA was uploaded into PAM in order to identify molecular classifiers, e.g., a subset of genes, that can best discriminate GTX from NGTX carcinogens. PAM generates a misclassification error rate curve based on cross-validation reducing the number of genes (data not shown). The misclassification error rate reached < 0.1 for all time points, except for 12 h where the misclassification was always > 0.4. Therefore, data sampled at 12 h were not further analyzed. For each time point, a set of classifiers was generated by using a threshold with a misclassification error rate of 0 and a probability for classification of > 80% for the training set data. This resulted in 15 genes after 24 h of exposure, 16 genes after 36 h, and 26 genes after 48 h: scores generated by PAM for all classifiers are presented in Table 2 (Supplementary Data). Classifiers upregulated in GTX compounds were downregulated in NGTX compounds and vice versa. TABLE 2 Genes Selected for Class Discrimination and Their GTX and NGTX Scores (=log10 [class probability: positive/class probability: negative]) as Obtained by the PAM Software GENEBANKACC Gene name Differentially expressed genes Genes correlating with γH2AX 24 h 36 h 48 h 24 h 48 h GTX score NGTX score GTX score NGTX score GTX score NGTX score GTX score NGTX score GTX score NGTX score NM_008522 Ltf 0.051 −0.051 0.333 −0.333 0.347 −0.347 0.158 −0.158 0.512 −0.512 AK005731 1700007K13Rik 0.319 −0.319 0.252 −0.252 0.227 −0.227 0.417 −0.417 0.339 −0.339 BI651416 Cdc42bpg 0.253 −0.253 0.165 −0.165 0.138 −0.138 0.369 −0.369 0.284 −0.284 BE956581 Cpt1c 0.160 −0.160 0.095 −0.095 0.261 −0.261 X58876 Mdm2 0.012 −0.012 0.090 −0.090 0.107 −0.107 0.252 −0.252 BB463610 4632434I11Rik 0.018 −0.018 0.183 −0.183 BB043558 9230114K14Rik 0.114 −0.114 0.051 −0.051 0.169 −0.169 AF297615 Ggta1 0.039 −0.039 0.164 −0.164 AK004719 Mdm2 0.114 −0.114 AF335325 Ddit4l 0.111 −0.111 BM198879 Ercc5 0.091 −0.091 0.182 −0.182 0.111 −0.111 BC008105 Polk 0.101 −0.101 BC003267 BC003267 0.097 −0.097 AV327248 Zfp365 0.093 −0.093 AV246296 Eda2r 0.074 −0.074 BC021352 Plod2 0.069 −0.069 AK014608 4632434I11Rik 0.066 −0.066 AF033112 Siva 0.060 −0.060 AK007854 1810053B23Rik 0.058 −0.058 BG797099 Ddit4l 0.017 −0.017 AV273409 Wnt9a 0.000 0.000 NM_030697 Ankrd47 0.191 −0.191 0.155 −0.155 0.294 −0.294 AV377066 9130221J18Rik 0.110 −0.110 0.172 −0.172 BM200015 Hsdl2 0.054 −0.054 0.138 −0.138 NM_013743 Pdk4 0.043 −0.043 0.134 −0.134 BM230508 A030007D23Rik 0.081 −0.081 NM_011316 Saa4 0.001 −0.001 0.074 −0.074 NM_011388 Slc10a2 0.066 −0.066 BB414224 Tmprss2 0.049 −0.049 NM_019976 Psrc1 0.021 −0.021 BC025198 Rap2a 0.021 −0.021 AI451920 C530044N13Rik 0.014 −0.014 AV009441 Ivl 0.004 −0.004 BB821363 Scfd2 −0.177 0.177 0.051 −0.051 BC026831 Fads1 −0.001 0.001 BC024358 Tpm2 −0.002 0.002 AA561726 Phgdh −0.009 0.009 NM_008935 Prom1 −0.022 0.022 BB454777 4930579E17Rik −0.015 0.015 −0.030 0.030 AK003186 Tpm2 −0.031 0.031 AV213379 Oxct1 −0.032 0.032 NM_024188 Oxct1 −0.045 0.045 BB458460 Chchd6 −0.054 0.054 BB246912 1700112E06Rik −0.061 0.061 BC024120 Tat −0.067 0.067 AF055573 Fhit −0.108 0.108 −0.112 0.112 AI747296 Gmds −0.040 0.040 −0.244 0.244 −0.141 0.141 BI247584 Fdps −0.142 0.142 NM_053122 Immp2l −0.158 0.158 −0.156 0.156 L21027 Phgdh −0.157 0.157 AF000969 Cadps2 −0.215 0.215 −0.585 0.585 −0.287 0.287 BC017615 Slc24a3 −0.114 0.114 −0.351 0.351 AI596632 Ptprg −0.245 0.245 −0.161 0.161 AK010447 Smyd3 −0.039 0.039 AK019979 2610528E23Rik −0.098 0.098 NM_012006 Acot1 0.056 −0.056 NM_026517 Rpl22l1 −0.001 0.001 GENEBANKACC Gene name Differentially expressed genes Genes correlating with γH2AX 24 h 36 h 48 h 24 h 48 h GTX score NGTX score GTX score NGTX score GTX score NGTX score GTX score NGTX score GTX score NGTX score NM_008522 Ltf 0.051 −0.051 0.333 −0.333 0.347 −0.347 0.158 −0.158 0.512 −0.512 AK005731 1700007K13Rik 0.319 −0.319 0.252 −0.252 0.227 −0.227 0.417 −0.417 0.339 −0.339 BI651416 Cdc42bpg 0.253 −0.253 0.165 −0.165 0.138 −0.138 0.369 −0.369 0.284 −0.284 BE956581 Cpt1c 0.160 −0.160 0.095 −0.095 0.261 −0.261 X58876 Mdm2 0.012 −0.012 0.090 −0.090 0.107 −0.107 0.252 −0.252 BB463610 4632434I11Rik 0.018 −0.018 0.183 −0.183 BB043558 9230114K14Rik 0.114 −0.114 0.051 −0.051 0.169 −0.169 AF297615 Ggta1 0.039 −0.039 0.164 −0.164 AK004719 Mdm2 0.114 −0.114 AF335325 Ddit4l 0.111 −0.111 BM198879 Ercc5 0.091 −0.091 0.182 −0.182 0.111 −0.111 BC008105 Polk 0.101 −0.101 BC003267 BC003267 0.097 −0.097 AV327248 Zfp365 0.093 −0.093 AV246296 Eda2r 0.074 −0.074 BC021352 Plod2 0.069 −0.069 AK014608 4632434I11Rik 0.066 −0.066 AF033112 Siva 0.060 −0.060 AK007854 1810053B23Rik 0.058 −0.058 BG797099 Ddit4l 0.017 −0.017 AV273409 Wnt9a 0.000 0.000 NM_030697 Ankrd47 0.191 −0.191 0.155 −0.155 0.294 −0.294 AV377066 9130221J18Rik 0.110 −0.110 0.172 −0.172 BM200015 Hsdl2 0.054 −0.054 0.138 −0.138 NM_013743 Pdk4 0.043 −0.043 0.134 −0.134 BM230508 A030007D23Rik 0.081 −0.081 NM_011316 Saa4 0.001 −0.001 0.074 −0.074 NM_011388 Slc10a2 0.066 −0.066 BB414224 Tmprss2 0.049 −0.049 NM_019976 Psrc1 0.021 −0.021 BC025198 Rap2a 0.021 −0.021 AI451920 C530044N13Rik 0.014 −0.014 AV009441 Ivl 0.004 −0.004 BB821363 Scfd2 −0.177 0.177 0.051 −0.051 BC026831 Fads1 −0.001 0.001 BC024358 Tpm2 −0.002 0.002 AA561726 Phgdh −0.009 0.009 NM_008935 Prom1 −0.022 0.022 BB454777 4930579E17Rik −0.015 0.015 −0.030 0.030 AK003186 Tpm2 −0.031 0.031 AV213379 Oxct1 −0.032 0.032 NM_024188 Oxct1 −0.045 0.045 BB458460 Chchd6 −0.054 0.054 BB246912 1700112E06Rik −0.061 0.061 BC024120 Tat −0.067 0.067 AF055573 Fhit −0.108 0.108 −0.112 0.112 AI747296 Gmds −0.040 0.040 −0.244 0.244 −0.141 0.141 BI247584 Fdps −0.142 0.142 NM_053122 Immp2l −0.158 0.158 −0.156 0.156 L21027 Phgdh −0.157 0.157 AF000969 Cadps2 −0.215 0.215 −0.585 0.585 −0.287 0.287 BC017615 Slc24a3 −0.114 0.114 −0.351 0.351 AI596632 Ptprg −0.245 0.245 −0.161 0.161 AK010447 Smyd3 −0.039 0.039 AK019979 2610528E23Rik −0.098 0.098 NM_012006 Acot1 0.056 −0.056 NM_026517 Rpl22l1 −0.001 0.001 Open in new tab TABLE 2 Genes Selected for Class Discrimination and Their GTX and NGTX Scores (=log10 [class probability: positive/class probability: negative]) as Obtained by the PAM Software GENEBANKACC Gene name Differentially expressed genes Genes correlating with γH2AX 24 h 36 h 48 h 24 h 48 h GTX score NGTX score GTX score NGTX score GTX score NGTX score GTX score NGTX score GTX score NGTX score NM_008522 Ltf 0.051 −0.051 0.333 −0.333 0.347 −0.347 0.158 −0.158 0.512 −0.512 AK005731 1700007K13Rik 0.319 −0.319 0.252 −0.252 0.227 −0.227 0.417 −0.417 0.339 −0.339 BI651416 Cdc42bpg 0.253 −0.253 0.165 −0.165 0.138 −0.138 0.369 −0.369 0.284 −0.284 BE956581 Cpt1c 0.160 −0.160 0.095 −0.095 0.261 −0.261 X58876 Mdm2 0.012 −0.012 0.090 −0.090 0.107 −0.107 0.252 −0.252 BB463610 4632434I11Rik 0.018 −0.018 0.183 −0.183 BB043558 9230114K14Rik 0.114 −0.114 0.051 −0.051 0.169 −0.169 AF297615 Ggta1 0.039 −0.039 0.164 −0.164 AK004719 Mdm2 0.114 −0.114 AF335325 Ddit4l 0.111 −0.111 BM198879 Ercc5 0.091 −0.091 0.182 −0.182 0.111 −0.111 BC008105 Polk 0.101 −0.101 BC003267 BC003267 0.097 −0.097 AV327248 Zfp365 0.093 −0.093 AV246296 Eda2r 0.074 −0.074 BC021352 Plod2 0.069 −0.069 AK014608 4632434I11Rik 0.066 −0.066 AF033112 Siva 0.060 −0.060 AK007854 1810053B23Rik 0.058 −0.058 BG797099 Ddit4l 0.017 −0.017 AV273409 Wnt9a 0.000 0.000 NM_030697 Ankrd47 0.191 −0.191 0.155 −0.155 0.294 −0.294 AV377066 9130221J18Rik 0.110 −0.110 0.172 −0.172 BM200015 Hsdl2 0.054 −0.054 0.138 −0.138 NM_013743 Pdk4 0.043 −0.043 0.134 −0.134 BM230508 A030007D23Rik 0.081 −0.081 NM_011316 Saa4 0.001 −0.001 0.074 −0.074 NM_011388 Slc10a2 0.066 −0.066 BB414224 Tmprss2 0.049 −0.049 NM_019976 Psrc1 0.021 −0.021 BC025198 Rap2a 0.021 −0.021 AI451920 C530044N13Rik 0.014 −0.014 AV009441 Ivl 0.004 −0.004 BB821363 Scfd2 −0.177 0.177 0.051 −0.051 BC026831 Fads1 −0.001 0.001 BC024358 Tpm2 −0.002 0.002 AA561726 Phgdh −0.009 0.009 NM_008935 Prom1 −0.022 0.022 BB454777 4930579E17Rik −0.015 0.015 −0.030 0.030 AK003186 Tpm2 −0.031 0.031 AV213379 Oxct1 −0.032 0.032 NM_024188 Oxct1 −0.045 0.045 BB458460 Chchd6 −0.054 0.054 BB246912 1700112E06Rik −0.061 0.061 BC024120 Tat −0.067 0.067 AF055573 Fhit −0.108 0.108 −0.112 0.112 AI747296 Gmds −0.040 0.040 −0.244 0.244 −0.141 0.141 BI247584 Fdps −0.142 0.142 NM_053122 Immp2l −0.158 0.158 −0.156 0.156 L21027 Phgdh −0.157 0.157 AF000969 Cadps2 −0.215 0.215 −0.585 0.585 −0.287 0.287 BC017615 Slc24a3 −0.114 0.114 −0.351 0.351 AI596632 Ptprg −0.245 0.245 −0.161 0.161 AK010447 Smyd3 −0.039 0.039 AK019979 2610528E23Rik −0.098 0.098 NM_012006 Acot1 0.056 −0.056 NM_026517 Rpl22l1 −0.001 0.001 GENEBANKACC Gene name Differentially expressed genes Genes correlating with γH2AX 24 h 36 h 48 h 24 h 48 h GTX score NGTX score GTX score NGTX score GTX score NGTX score GTX score NGTX score GTX score NGTX score NM_008522 Ltf 0.051 −0.051 0.333 −0.333 0.347 −0.347 0.158 −0.158 0.512 −0.512 AK005731 1700007K13Rik 0.319 −0.319 0.252 −0.252 0.227 −0.227 0.417 −0.417 0.339 −0.339 BI651416 Cdc42bpg 0.253 −0.253 0.165 −0.165 0.138 −0.138 0.369 −0.369 0.284 −0.284 BE956581 Cpt1c 0.160 −0.160 0.095 −0.095 0.261 −0.261 X58876 Mdm2 0.012 −0.012 0.090 −0.090 0.107 −0.107 0.252 −0.252 BB463610 4632434I11Rik 0.018 −0.018 0.183 −0.183 BB043558 9230114K14Rik 0.114 −0.114 0.051 −0.051 0.169 −0.169 AF297615 Ggta1 0.039 −0.039 0.164 −0.164 AK004719 Mdm2 0.114 −0.114 AF335325 Ddit4l 0.111 −0.111 BM198879 Ercc5 0.091 −0.091 0.182 −0.182 0.111 −0.111 BC008105 Polk 0.101 −0.101 BC003267 BC003267 0.097 −0.097 AV327248 Zfp365 0.093 −0.093 AV246296 Eda2r 0.074 −0.074 BC021352 Plod2 0.069 −0.069 AK014608 4632434I11Rik 0.066 −0.066 AF033112 Siva 0.060 −0.060 AK007854 1810053B23Rik 0.058 −0.058 BG797099 Ddit4l 0.017 −0.017 AV273409 Wnt9a 0.000 0.000 NM_030697 Ankrd47 0.191 −0.191 0.155 −0.155 0.294 −0.294 AV377066 9130221J18Rik 0.110 −0.110 0.172 −0.172 BM200015 Hsdl2 0.054 −0.054 0.138 −0.138 NM_013743 Pdk4 0.043 −0.043 0.134 −0.134 BM230508 A030007D23Rik 0.081 −0.081 NM_011316 Saa4 0.001 −0.001 0.074 −0.074 NM_011388 Slc10a2 0.066 −0.066 BB414224 Tmprss2 0.049 −0.049 NM_019976 Psrc1 0.021 −0.021 BC025198 Rap2a 0.021 −0.021 AI451920 C530044N13Rik 0.014 −0.014 AV009441 Ivl 0.004 −0.004 BB821363 Scfd2 −0.177 0.177 0.051 −0.051 BC026831 Fads1 −0.001 0.001 BC024358 Tpm2 −0.002 0.002 AA561726 Phgdh −0.009 0.009 NM_008935 Prom1 −0.022 0.022 BB454777 4930579E17Rik −0.015 0.015 −0.030 0.030 AK003186 Tpm2 −0.031 0.031 AV213379 Oxct1 −0.032 0.032 NM_024188 Oxct1 −0.045 0.045 BB458460 Chchd6 −0.054 0.054 BB246912 1700112E06Rik −0.061 0.061 BC024120 Tat −0.067 0.067 AF055573 Fhit −0.108 0.108 −0.112 0.112 AI747296 Gmds −0.040 0.040 −0.244 0.244 −0.141 0.141 BI247584 Fdps −0.142 0.142 NM_053122 Immp2l −0.158 0.158 −0.156 0.156 L21027 Phgdh −0.157 0.157 AF000969 Cadps2 −0.215 0.215 −0.585 0.585 −0.287 0.287 BC017615 Slc24a3 −0.114 0.114 −0.351 0.351 AI596632 Ptprg −0.245 0.245 −0.161 0.161 AK010447 Smyd3 −0.039 0.039 AK019979 2610528E23Rik −0.098 0.098 NM_012006 Acot1 0.056 −0.056 NM_026517 Rpl22l1 −0.001 0.001 Open in new tab To investigate whether gene expression changes correlating with γH2AX foci formation after exposure to BaP, AFB1, CsA, or TCDD are also capable of discriminating GTX from NGTX compounds, Spearman correlation coefficients were calculated and correlating genes were selected at p < 0.05. This resulted in 683 correlating genes after 24 h of exposure and 374 genes after 48 h of exposure. These genes were also uploaded into PAM. A misclassification error rate of 0 was reached after exposure periods of 24 and 48 h. Classifiers were again generated by using a threshold with a misclassification error rate of 0 and a classification probability of > 80% for the training set data, resulting 19 classifiers after 24 h of exposure and 22 classifiers after 48 h, presented in Table 2 (Supplementary Data). From these classifiers, six genes were common between exposure periods of 24 and 48 h. From the classifiers found after 24 h of exposure, 10 were common with classifiers found from the differentially expressed genes. After 48 h of exposure, eight were common with classifiers found from the differentially expressed genes. For the purpose of validating class discrimination profiles, primary hepatocytes were treated with two additional genotoxic compounds, DMN and MitC, or vehicle control for the apparently optimal exposure periods of 24 or 48 h. A sampling time of 12 h was not included as the classification was not successful at that early time point. Also after 36 h of exposure, no analysis was performed as classifiers found at this particular time point were barely different from those at 24 and 48 h. The classifiers generated by PAM from the differentially expressed genes and from the significantly modulated pathways, by the training set, were then applied for classifying these two additional GTX compounds. After both exposure periods, all independent triplicate treatments of both compounds were classified correctly as GTX with a predicted test probability of 100%. Additionally, the list of classifiers generated from the genes correlating with γH2AX foci formation by PAM was validated by evaluating these new compounds. Only after 48 h of exposure, however, the independent triplicate treatments of both compounds were classified correctly as GTX, with a predicted test probability > 90%. After 24 h, the predicted test probability was lower but both compounds were again classified correct (data not shown). Functional Analyses of Classifiers Because the classifiers generated by PAM after 36 h of exposure were barely different from those at 24 and 48 h, only classifiers after 24 and 48 h of exposure were used for overrepresentation analysis in MetaCore maps. Classifiers appeared to have a predominant role in transcription, responses to extracellular stimuli, DNA damage, cell cycle and apoptosis, and survival and also in aminoacid, carbohydrate, lipid, nucleotide, and steroid metabolism (Supplementary Data). One classifier, namely murine double minute (Mdm2), was present in a large number of MetaCore maps. A network was generated using shortest path algorithm to visualize possible interactions between the major classifiers based on literature text mining (Fig. 4). Again, this network demonstrated the pivotal role of Mdm2. Mdm2 appeared upregulated in GTX and downregulated in NGTX compounds. Fragile histidine triad gene (FHIT), which was downregulated in GTX compounds, had an inhibiting role on Mdm2, while XPG, as shown in Figure 4, was indirectly upregulated in GTX and was induced by Mdm2. Esr1, inhibited by Mdm2, had an inhibiting role on C10orf11 and on NCKX3, which were both downregulated in GTX compounds. Esr1 had a stimulating role on C3orf26, lactoferrin, and PDK4 but only C3orf26 was downregulated in GTX compounds. Mdm2 had also an inhibiting role on p53 and indirectly on SERA and MRCKγ. But only SERA appeared downregulated in GTX compounds. This network plays a role in cellular proliferation and tumor suppression. FIG. 4. Open in new tabDownload slide Biological network generated in MetaCore to map interactions between the classifiers selected in primary mouse hepatocytes after 24-, 36-, and 48-h exposure to BaP, AFB1, TCDD, or CsA. The legend for the biological network is provided in Supplementary Data. FIG. 4. Open in new tabDownload slide Biological network generated in MetaCore to map interactions between the classifiers selected in primary mouse hepatocytes after 24-, 36-, and 48-h exposure to BaP, AFB1, TCDD, or CsA. The legend for the biological network is provided in Supplementary Data. DISCUSSION With the aim to find alternatives for expensive and long-lasting chronic rodent bioassays, primary mouse hepatocytes were used to evaluate the time dependence of this in vitro assay to classify carcinogens by their modes of action using gene expression profiling. Evaluation of γH2AX foci formation confirmed the correct selection of GTX and NGTX compounds. Induction of γH2AX foci is a measure for the formation of dsDNA breaks in human cell lines and in the liver of mice (Koike et al., 2008; Rogakou et al., 1998), but this has, to our knowledge, not previously been demonstrated in primary mouse hepatocytes. The two GTX carcinogens appeared clearly capable of inducing γH2AX foci formation, while the two NGTX carcinogens, on the other hand, were not able to cause γH2AX foci formation. These results are in line with findings on cell lines done by Zhou et al. (2006), who found comparable results in human amnion FL and Chinese hamster CHL cells after 24 h of exposure to BaP and other GTX compounds. NGTX compounds CsA and azathioprine did not show an induction in γH2AX foci formation in human amnion FL and Chinese hamster CHL cells in the study of Zhou et al. (2006). To discriminate GTX and NGTX compounds based on gene expression profiles, we used a supervised clustering approach, namely PAM. Because length of exposure has been shown to have a major influence on gene expression profiles in cells in vitro (Hockley et al., 2006; Lambert et al., 2009), it may be hypothesized that exposure period affects class discrimination based on transcriptomics data. In the current study, class discrimination is indeed dependent on exposure time. At 12 h, the two classes could not be discriminated in the training set. After longer exposure times, however, class discrimination was successful. At the 12-h time point, a high misclassification error rate was shown. At the other time points, misclassification errors were zero, and therefore, these incubation periods all seem suited for class discrimination. The lists of classifiers sampled at 24, 36, and 48 h of exposure were compared. Only one classifier at 36 h of exposure was unique, all the other classifiers at 36 h were also present in the list of classifiers at 24 or 48 h. Six classifiers were present in the lists for 24 and 48 h. Consequently, transcriptomic response at 36 h of exposure have not been evaluated in the subsequent validation. The PAM classification models based on classifiers generated at 24 and 48 h of exposure were validated in a study with two additional compounds, DMN and MitC. Based on their gene expression profiles, both compounds were correctly classified as GTX, demonstrating that the obtained classifiers following exposure to carcinogens may indeed be able to discriminate between GTX and NGTX classes. At both time points, classification was perfect, although the classifiers lists were quite different. Gene expression modifications correlating with γH2AX foci formation as induced by the four model compounds were also used in PAM as a training set. Generated classifiers were validated again by using the additional compounds and both compounds were classified correct only at 48 h but with lower test probabilities as compared to class discrimination with all differentially expressed genes. This suggests that DNA damage–related genes are an important but not the only contributor in discriminating GTX from NGTX. The applicability of class discrimination for carcinogens by gene expression profiling was shown in several mouse and rat in vivo studies (Ellinger-Ziegelbauer et al., 2007; Eun et al., 2008; Iida et al., 2005; Nioi et al., 2008). Furthermore, the reliability of classification methods similar to the one we used were previously successfully demonstrated in in vivo and in vitro (HepG2 cells) systems to discriminate GTX from NGTX carcinogens (Eun et al., 2008; Uehara et al., 2008; van Delft et al., 2004, 2005). Most of these studies did not focus on time-dependent discrimination but evaluated changes at only one time point. The study from Uehara et al. (2008) showed time-dependent increase in discrimination scores by PAM in rats treated with NGTX. The in vivo study from Ellinger-Ziegelbauer et al. (2005) demonstrated that more differences between GTX and NGTX compounds were observed after treatment for 7 days compared to 1 or 3 days. Since it is known that DNA damage is induced by GTX compounds, it is not surprising that many of the classifiers for GTX are involved in cell cycle checkpoints, repair mechanism, and response to DNA damage. The induction of γH2AX foci formation confirms that also in these primary hepatocytes cultures, the GTX compounds induce DNA damage. Predominantly, the formation of these foci is a result of dsDNA damage, which can be caused by repair mechanisms deleting DNA adducts (Fernandez-Capetillo et al., 2004; Zhou et al., 2006). Analysis by MetaCore demonstrated that the Mdm2 oncogene is the most prominent classifier, which is a regulator of p53, but also has some p53-independent activities (Daujat et al., 2001). The Mdm2 oncoprotein binds to p53 and thereby blocks its ability to inhibit cellular proliferation or to induce cell death (Momand et al., 2000). Mdm2 mediates the ubiquitination of p53 and promotes the degradation of p53 by the proteasome (Haupt et al., 1997). Overexpression of Mdm2 is related to the increase of tumorigenic potential and can overcome the suppression of growth activity of p53 (Bueso-Ramos et al., 1993; Finlay, 1993; Schlott et al., 1999). This p53 is related to DNA damage responses as levels of p53 are increasing vary fast in cells sustaining DNA damage (Lakin and Jackson, 1999). Mdm2 and p53 form a feedback loop that plays a role in cell cycle arrest and apoptosis (Jiang et al., 2003). GTX compounds might therefore be able to induce overexpression of this Mdm2 oncogene. In our study, treatment with GTX compounds clearly increases the Mdm2 gene expression. Furthermore, the gene expression of the Mdm2 oncogene is suppressed by FHIT, which is a tumor suppressor gene (Schlott et al., 1999). Treatment with GTX compounds caused a downregulation of this FHIT gene in our study, leading to stimulation of Mdm2. Other classifiers linked to Mdm2 gene in the network from MetaCore are involved in response to DNA damage and metabolism. Furthermore, Mdm2 was previously detected as a classifier that could discriminate between GTX and NGTX in rat livers (Uehara et al., 2008). Although the other classifiers found in that study are different from our classifiers, the affected responses to DNA damage and cell cycle regulation–related processes are comparable. More in vivo class discrimination studies found classifiers involved in metabolism, cell cycle–related and DNA damage responses (Nioi et al., 2008; Iida et al., 2005). An HepG2 in vitro study detected classifiers mainly involved in apoptosis-related pathways; however, also some p53-related DNA damage responses were detected (van Delft et al., 2004). Further investigation of the network generated by MetaCore showed that some of these classifiers (Mdm2, lactoferrin, Sera, MRCKgamma, Pdk4, Esr1, and E2F1) share a common transcriptional activator, namely SP1. This SP1 is mostly involved in cell cycle regulation and might play a role in DNA damage repair (Grinstein et al., 2002; Iwahori et al., 2008; Olofsson et al., 2007). Pathway analysis was used to visualize compound- and time-related effects. At 12 h of treatment, hardly any effects are observed on pathways. This is in agreement with the effects on genes, indicating that this time point is not suitable for further transcriptomics experiments. The biggest effects on pathways, however, are observed at 48 h of treatment. At 24 h of treatment, an upregulation of most gene sets related to macromolecules, cellular organization, and regulation by all compounds is observed. These processes might be naturally occurring in primary hepatocytes in culture as it was shown before in untreated hepatocytes that cellular organization and macromolecule related gene groups are mainly upregulated during culture (Mathijs et al., 2009). Besides these common processes triggered by all compounds, differences between the GTX and NGTX compounds are observed. Signal transduction gene groups are only upregulated by the NGTX compounds at 24 h. Gene groups involved in immune responses; extracellular, membrane, and cytoplasm processes; lipid-related processes; and metabolic and biosynthetic processes are mainly downregulated by GTX compounds and upregulated by NGTX compounds at later time points. The few p53-related gene groups are mainly upregulated by GTX compounds, especially at 36 h. Treatment of primary mouse hepatocytes with CsA triggered only a small number of pathways in these cells, and more pathways were significantly altered at 48 h of treatment. CsA is a well-known human carcinogen but showed no tumor development before in wild-type male C57Bl/6 mice (van Kreijl et al., 2001); this might explain the observed minor effects on pathways in this mouse hepatocyte model. Also treatment of primary mouse hepatocytes with TCDD led to a small number of significantly modulated pathways, with most pathways modulated at 48 h of treatment, showing that both NGTX compounds have less pronounced effects on primary mouse hepatocytes and their effects are observed at a later stage as compared to GTX compounds. The major differences between GTX and NGTX compounds at the level of pathways were detected in p53-related processes and immune responses. Furthermore, cytochrome P450–related gene groups are mostly upregulated by BaP and TCDD at 36 and 48 h. Most likely, this is due to the activation of the aryl hydrocarbon receptor by these compounds (Levin et al., 1982; Bock, 1994). In summary, the present study demonstrates that in primary mouse hepatocytes, exposure length is of major importance for classifying GTX and NGTX compounds based on gene expression profiling. Twelve hours of exposure appeared too short, and a treatment period of 48 h seems optimal. These findings may be used to further develop an in vitro assay of mouse primary hepatocytes using gene expression profiling to predict possible carcinogens and thus that the use of mouse primary hepatocytes is a specific and promising in vitro model. FUNDING Financial support was provided by the Netherlands Genomics Initiative (NGI), the Netherlands Organisation for Scientific Research (NWO) and the CARCINOGENOMICS FP6 project sponsored by the European Union (PL037712). References Affymetrix , Statistical Algorithms Description Document , 2002 Technical report. Affymetrix, Inc., Santa Clara, CA Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Ashby J . Use of short-term tests in determining the genotoxicity or nongenotoxicity of chemicals , IARC Sci. Publ. , 1992 , vol. 116 (pg. 135 - 164 ) OpenURL Placeholder Text WorldCat Beken S , Vanhaecke T , De Smet K , Pauwels M , Vercruysse A , Rogiers V . Collagen-gel cultures of rat hepatocytes: collagen-gel sandwich and immobilization cultures , Methods Mol. Biol. , 1998 , vol. 107 (pg. 303 - 309 ) Google Scholar PubMed OpenURL Placeholder Text WorldCat Bock KW . Aryl hydrocarbon or dioxin receptor: biologic and toxic responses , Rev. Physiol. Biochem. Pharmacol. , 1994 , vol. 125 (pg. 1 - 42 ) Google Scholar PubMed OpenURL Placeholder Text WorldCat Boorsma A , Foat BC , Vis D , Klis F , Bussemaker HJ . T-profiler: scoring the activity of predefined groups of genes using gene expression data , Nucleic Acids Res. , 2005 , vol. 33 (pg. W592 - W595 ) Google Scholar Crossref Search ADS PubMed WorldCat Bueso-Ramos CE , Yang Y , deLeon E , McCown P , Stass SA , Albitar M . The human MDM-2 oncogene is overexpressed in leukemias , Blood , 1993 , vol. 82 (pg. 2617 - 2623 ) Google Scholar PubMed OpenURL Placeholder Text WorldCat Butterworth BE , Smith-Oliver T , Earle L , Loury DJ , White RD , Doolittle DJ , Working PK , Cattley RC , Jirtle R , Michalopoulos G , et al. Use of primary cultures of human hepatocytes in toxicology studies , Cancer Res. , 1989 , vol. 49 (pg. 1075 - 1084 ) Google Scholar PubMed OpenURL Placeholder Text WorldCat Canales RD , Luo Y , Willey JC , Austermiller B , Barbacioru CC , Boysen C , Hunkapiller K , Jensen RV , Knight CR , Lee KY , et al. Evaluation of DNA microarray results with quantitative gene expression platforms , Nat. Biotechnol. , 2006 , vol. 24 (pg. 1115 - 1122 ) Google Scholar Crossref Search ADS PubMed WorldCat Crump D , Chiu S , Egloff C , Kennedy SW . Effects of hexabromocyclododecane and polybrominated diphenyl ethers on mRNA expression in chicken (Gallus domesticus) hepatocytes , Toxicol. Sci. , 2008 , vol. 106 (pg. 479 - 487 ) Google Scholar Crossref Search ADS PubMed WorldCat Dambach DM , Andrews BA , Moulin F . New technologies and screening strategies for hepatotoxicity: use of in vitro models , Toxicol. Pathol. , 2005 , vol. 33 (pg. 17 - 26 ) Google Scholar Crossref Search ADS PubMed WorldCat Daujat S , Neel H , Piette J . MDM2: life without p53 , Trends Genet. , 2001 , vol. 17 (pg. 459 - 464 ) Google Scholar Crossref Search ADS PubMed WorldCat Ellinger-Ziegelbauer H , Gmuender H , Bandenburg A , Ahr HJ . Prediction of a carcinogenic potential of rat hepatocarcinogens using toxicogenomics analysis of short-term in vivo studies , Mutat. Res , 2007 , vol. 637 (pg. 23 - 39 ) Google Scholar Crossref Search ADS PubMed WorldCat Ellinger-Ziegelbauer H , Stuart B , Wahle B , Bomann W , Ahr HJ . Comparison of the expression profiles induced by genotoxic and nongenotoxic carcinogens in rat liver , Mutat. Res. , 2005 , vol. 575 (pg. 61 - 84 ) Google Scholar Crossref Search ADS PubMed WorldCat Ennever FK , Lave LB . Implications of the lack of accuracy of the lifetime rodent bioassay for predicting human carcinogenicity , Regul. Toxicol. Pharmacol. , 2003 , vol. 38 (pg. 52 - 57 ) Google Scholar Crossref Search ADS PubMed WorldCat Eun JW , Ryu SY , Noh JH , Lee MJ , Jang JJ , Ryu JC , Jung KH , Kim JK , Bae HJ , Xie H , et al. Discriminating the molecular basis of hepatotoxicity using the large-scale characteristic molecular signatures of toxicants by expression profiling analysis , Toxicology , 2008 , vol. 249 (pg. 176 - 183 ) Google Scholar Crossref Search ADS PubMed WorldCat Fernandez-Capetillo O , Lee A , Nussenzweig M , Nussenzweig A . H2AX: the histone guardian of the genome , DNA Repair (Amst.) , 2004 , vol. 3 (pg. 959 - 967 ) Google Scholar Crossref Search ADS PubMed WorldCat Finlay CA . The mdm-2 oncogene can overcome wild-type p53 suppression of transformed cell growth , Mol. Cell. Biol. , 1993 , vol. 13 (pg. 301 - 306 ) Google Scholar Crossref Search ADS PubMed WorldCat Grinstein E , Jundt F , Weinert I , Wernet P , Royer HD . Sp1 as G1 cell cycle phase specific transcription factor in epithelial cells , Oncogene , 2002 , vol. 21 (pg. 1485 - 1492 ) Google Scholar Crossref Search ADS PubMed WorldCat Groneberg DA , Grosse-Siestrup C , Fischer A . In vitro models to study hepatotoxicity , Toxicol. Pathol. , 2002 , vol. 30 (pg. 394 - 399 ) Google Scholar Crossref Search ADS PubMed WorldCat Hamer G , Roepers-Gajadien HL , van Duyn-Goedhart A , Gademan IS , Kal HB , van Buul PP , de Rooij DG . DNA double-strand breaks and gamma-H2AX signaling in the testis , Biol. Reprod. , 2003 , vol. 68 (pg. 628 - 634 ) Google Scholar Crossref Search ADS PubMed WorldCat Haupt Y , Maya R , Kazaz A , Oren M . Mdm2 promotes the rapid degradation of p53 , Nature , 1997 , vol. 387 (pg. 296 - 299 ) Google Scholar Crossref Search ADS PubMed WorldCat Hockley SL , Arlt VM , Brewer D , Giddings I , Phillips DH . Time- and concentration-dependent changes in gene expression induced by benzo(a)pyrene in two human cell lines, MCF-7 and HepG2 , BMC Genomics , 2006 , vol. 7 pg. 260 Google Scholar Crossref Search ADS PubMed WorldCat Iida M , Anna CH , Holliday WM , Collins JB , Cunningham ML , Sills RC , Devereux TR . Unique patterns of gene expression changes in liver after treatment of mice for 2 weeks with different known carcinogens and non-carcinogens , Carcinogenesis , 2005 , vol. 26 (pg. 689 - 699 ) Google Scholar Crossref Search ADS PubMed WorldCat Irizarry RA , Hobbs B , Collin F , Beazer-Barclay YD , Antonellis KJ , Scherf U , Speed TP . Exploration, normalization, and summaries of high density oligonucleotide array probe level data , Biostatistics , 2003 , vol. 4 (pg. 249 - 264 ) Google Scholar Crossref Search ADS PubMed WorldCat Iwahori S , Yasui Y , Kudoh A , Sato Y , Nakayama S , Murata T , Isomura H , Tsurumi T . Identification of phosphorylation sites on transcription factor Sp1 in response to DNA damage and its accumulation at damaged sites , Cell. Signal. , 2008 , vol. 20 (pg. 1795 - 1803 ) Google Scholar Crossref Search ADS PubMed WorldCat Jiang Y , Yang W , Zhou Y , Ma L . Up-regulation of murine double minute clone 2 (MDM2) gene expression in rat brain after morphine, heroin, and cocaine administrations , Neurosci. Lett. , 2003 , vol. 352 (pg. 216 - 220 ) Google Scholar Crossref Search ADS PubMed WorldCat Koike M , Mashino M , Sugasawa J , Koike A . Histone H2AX phosphorylation independent of ATM after X-irradiation in mouse liver and kidney in situ , J. Radiat. Res. (Tokyo) , 2008 , vol. 4 (pg. 445 - 449 ) Google Scholar Crossref Search ADS WorldCat Kruse JJ , Svensson JP , Huigsloot M , Giphart-Gassler M , Schoonen WG , Polman JE , Jean Horbach G , van de Water B , Vrieling H . A portrait of cisplatin-induced transcriptional changes in mouse embryonic stem cells reveals a dominant p53-like response , Mutat. Res. , 2007 , vol. 617 (pg. 58 - 70 ) Google Scholar Crossref Search ADS PubMed WorldCat Lakin ND , Jackson SP . Regulation of p53 in response to DNA damage , Oncogene , 1999 , vol. 18 (pg. 7644 - 7655 ) Google Scholar Crossref Search ADS PubMed WorldCat Lambert CB , Spire C , Claude N , Guillouzo A . Dose- and time-dependent effects of phenobarbital on gene expression profiling in human hepatoma HepaRG cells , Toxicol. Appl. Pharmacol. , 2009 , vol. 234 (pg. 345 - 360 ) Google Scholar Crossref Search ADS PubMed WorldCat Le Fevre AC , Boitier E , Marchandeau JP , Sarasin A , Thybaud V . Characterization of DNA reactive and non-DNA reactive anticancer drugs by gene expression profiling , Mutat. Res. , 2007 , vol. 619 (pg. 16 - 29 ) Google Scholar Crossref Search ADS PubMed WorldCat Levin W , Wood A , Chang R , Ryan D , Thomas P , Yagi H , Thakker D , Vyas K , Boyd C , Chu SY , et al. Oxidative metabolism of polycyclic aromatic hydrocarbons to ultimate carcinogens , Drug Metab. Rev. , 1982 , vol. 13 (pg. 555 - 580 ) Google Scholar Crossref Search ADS PubMed WorldCat Mathijs K , Kienhuis AS , Brauers KJ , Jennen DG , Lahoz A , Kleinjans JC , van Delft JH . Assessing the metabolic competence of sandwich-cultured mouse primary hepatocytes , Drug Metab. Dispos. , 2009 , vol. 37 (pg. 1305 - 1311 ) Google Scholar Crossref Search ADS PubMed WorldCat Momand J , Wu HH , Dasgupta G . MDM2–master regulator of the p53 tumor suppressor protein , Gene , 2000 , vol. 242 (pg. 15 - 29 ) Google Scholar Crossref Search ADS PubMed WorldCat Nioi P , Pardo ID , Sherratt PJ , Snyder RD . Prediction of non-genotoxic carcinogenesis in rats using changes in gene expression following acute dosing , Chem. Biol. Interact. , 2008 , vol. 172 (pg. 206 - 215 ) Google Scholar Crossref Search ADS PubMed WorldCat Olofsson BA , Kelly CM , Kim J , Hornsby SM , Azizkhan-Clifford J . Phosphorylation of Sp1 in response to DNA damage by ataxia telangiectasia-mutated kinase , Mol. Cancer Res. , 2007 , vol. 5 (pg. 1319 - 1330 ) Google Scholar Crossref Search ADS PubMed WorldCat Reich M , Liefeld T , Gould J , Lerner J , Tamayo P , Mesirov JP . GenePattern 2.0 , Nat. Genet. , 2006 , vol. 38 (pg. 500 - 501 ) Google Scholar Crossref Search ADS PubMed WorldCat Rogakou EP , Pilch DR , Orr AH , Ivanova VS , Bonner WM . DNA double-stranded breaks induce histone H2AX phosphorylation on serine 139 , J. Biol. Chem. , 1998 , vol. 273 (pg. 5858 - 5868 ) Google Scholar Crossref Search ADS PubMed WorldCat Schlott T , Ahrens K , Ruschenburg I , Reimer S , Hartmann H , Droese M . Different gene expression of MDM2, GAGE-1, -2 and FHIT in hepatocellular carcinoma and focal nodular hyperplasia , Br. J. Cancer , 1999 , vol. 80 (pg. 73 - 78 ) Google Scholar Crossref Search ADS PubMed WorldCat Seglen PO . Preparation of isolated rat liver cells , Methods Cell Biol. , 1976 , vol. 13 (pg. 29 - 83 ) Google Scholar PubMed OpenURL Placeholder Text WorldCat Shi L , Reid LH , Jones WD , Shippy R , Warrington JA , Baker SC , Collins PJ , de Longueville F , Kawasaki ES , Lee KY , et al. The MicroArray Quality Control (MAQC) project shows inter- and intraplatform reproducibility of gene expression measurements , Nat. Biotechnol. , 2006 , vol. 24 (pg. 1151 - 1161 ) Google Scholar Crossref Search ADS PubMed WorldCat Silva Lima B , Van der Laan JW . Mechanisms of nongenotoxic carcinogenesis and assessment of the human hazard , Regul. Toxicol. Pharmacol. , 2000 , vol. 32 (pg. 135 - 143 ) Google Scholar Crossref Search ADS PubMed WorldCat Tibshirani R , Hastie T , Narasimhan B , Chu G . Diagnosis of multiple cancer types by shrunken centroids of gene expression , Proc. Natl. Acad. Sci. U.S.A. , 2002 , vol. 99 (pg. 6567 - 6572 ) Google Scholar Crossref Search ADS PubMed WorldCat Tong W , Cao X , Harris S , Sun H , Fang H , Fuscoe J , Harris A , Hong H , Xie Q , Perkins R , et al. ArrayTrack—supporting toxicogenomic research at the U.S. Food and Drug Administration National Center for Toxicological Research , Environ. Health Perspect. , 2003 , vol. 111 (pg. 1819 - 1826 ) Google Scholar Crossref Search ADS PubMed WorldCat Uehara T , Hirode M , Ono A , Kiyosawa N , Omura K , Shimizu T , Mizukawa Y , Miyagishima T , Nagao T , Urushidani T . A toxicogenomics approach for early assessment of potential non-genotoxic hepatocarcinogenicity of chemicals in rats , Toxicology , 2008 , vol. 250 (pg. 15 - 26 ) Google Scholar Crossref Search ADS PubMed WorldCat van Delft JH , van Agen E , van Breda SG , Herwijnen MH , Staal YC , Kleinjans JC . Discrimination of genotoxic from non-genotoxic carcinogens by gene expression profiling , Carcinogenesis , 2004 , vol. 25 (pg. 1265 - 1276 ) Google Scholar Crossref Search ADS PubMed WorldCat van Delft JH , van Agen E , van Breda SG , Herwijnen MH , Staal YC , Kleinjans JC . Comparison of supervised clustering methods to discriminate genotoxic from non-genotoxic carcinogens by gene expression profiling , Mutat. Res. , 2005 , vol. 575 (pg. 17 - 33 ) Google Scholar Crossref Search ADS PubMed WorldCat van Kreijl CF , McAnulty PA , Beems RB , Vynckier A , van Steeg H , Fransson-Steen R , Alden CL , Forster R , van der Laan JW , Vandenberghe J . Xpa and Xpa/p53+/- knockout mice: overview of available data , Toxicol. Pathol. , 2001 , vol. 29 Suppl. (pg. 117 - 127 ) Google Scholar Crossref Search ADS PubMed WorldCat Ward JH . Hierarchical grouping to optimize an objective function , J. Am. Stat. Assoc. , 1963 , vol. 58 (pg. 236 - 244 ) Google Scholar Crossref Search ADS WorldCat Waterston RH , Lindblad-Toh K , Birney E , Rogers J , Abril JF , Agarwal P , Agarwala R , Ainscough R , Alexandersson M , An P , et al. Initial sequencing and comparative analysis of the mouse genome , Nature , 2002 , vol. 420 (pg. 520 - 562 ) Google Scholar Crossref Search ADS PubMed WorldCat Zhou C , Li Z , Diao H , Yu Y , Zhu W , Dai Y , Chen FF , Yang J . DNA damage evaluated by gammaH2AX foci formation by a selective group of chemical/physical stressors , Mutat. Res. , 2006 , vol. 604 (pg. 8 - 18 ) Google Scholar Crossref Search ADS PubMed WorldCat © The Author 2009. Published by Oxford University Press on behalf of the Society of Toxicology. All rights reserved. For permissions, please email: journals.permissions@oxfordjournals.org
TI - Discrimination for Genotoxic and Nongenotoxic Carcinogens by Gene Expression Profiling in Primary Mouse Hepatocytes Improves with Exposure Time
JO - Toxicological Sciences
DO - 10.1093/toxsci/kfp229
DA - 2009-12-01
UR - https://www.deepdyve.com/lp/oxford-university-press/discrimination-for-genotoxic-and-nongenotoxic-carcinogens-by-gene-A07HUIS8qV
SP - 374
EP - 384
VL - 112
IS - 2
DP - DeepDyve
ER -