ToxSci Advance Access originally published online on January 3, 2008
Toxicological Sciences 2008 102(2):291-309; doi:10.1093/toxsci/kfm313
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
Toxicogenomic Analysis of Gender, Chemical, and Dose Effects in Livers of TCDD- or Aroclor 1254–Exposed Rats Using a Multifactor Linear Model
* General Electric Company, Global Research Center, Environmental Technology Laboratory, One Research Circle, Niskayuna, New York 12309
General Electric Company, Global Research Center, Applied Statistics Laboratory, One Research Circle, Niskayuna, New York 12309
W. Harry Feinstone Center Genomic Research, University of Memphis, Memphis, Tennessee 38512
1 To whom the correspondence should be addressed at General Electric Company, Global Research Center, Environmental Technology Laboratory, One Research Circle, Niskayuna, NY 12309. Fax: 518-387-6972. E-mail: silkworth{at}crd.ge.com.
Received October 16, 2007; accepted December 21, 2007
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMETHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAFUNDINGREFERENCES |
|---|
Chronic exposure of Sprague–Dawley (SD) rats to either 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) or Aroclor 1254 results in female-selective induction of hepatic tumors. The relative potency of dioxins and polychlorinated biphenyl mixtures, such as Aroclor 1254, is often estimated using the internationally endorsed toxic equivalency (TEQ) approach. Comparing the genome wide changes in gene expression in both genders following exposure to TEQ doses of these chemicals should identify critical sets of early response genes while further defining the concept of the TEQ of halogenated aromatic hydrocarbons. Aroclor 1254 at 0.6, 6.0, and 60 mg/kg body weight and TEQ doses of TCDD (0.3 and 3.0 µg/kg), calculated to match the top two Aroclor 1254 doses, were orally administered to SD rats for three consecutive days. Day 4 gene expression in hepatic tissue was determined using microarrays. A linear mixed-effects statistical model was developed to analyze the data in relation to treatment, gender, and gender * treatment (G*T) interactions. The genes most changed included 54 genes with and 51 genes without a significant model G*T term. The known aryl hydrocarbon receptor (AHR) battery genes (Cyp1a1, Cyp1a2, Cyp1b1, Aldh3a1), and novel genes, responded in a TEQ dose-dependent manner in both genders. However, an important observation was the apparent disruption of sexually dimorphic basal gene expression, particularly for female rats. Because many of these genes are involved in steroid metabolism, exposure to either TCDD or Aroclor 1254 could disrupt proliferative signals more in female rats as a possible consequence of altered estrogen metabolism. This study extends the findings of previous rodent bioassays by identifying groups of genes, other than the well-characterized AHR response genes, whose disruption may be important in the tumorigenic mechanism in this rat strain.
Key Words: TCDD; Aroclor 1254; PCB; microarray; liver; toxic equivalency factor.
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAFUNDINGREFERENCES |
|---|
Chronic exposure of rats to halogenated aromatic hydrocarbons (HAHs) such as dioxins, furans, and polychlorinated biphenyls (PCBs) results in hepatic tumors. Over three decades of research have indicated that activation of the aryl hydrocarbon receptor (AHR) pathway is a necessary first step in this process. However, the tumorigenic events that follow AHR activation remain elusive. Interestingly, in some studies, female rats appear to be far more susceptible than males for induction of liver tumors following exposure to 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) (Kociba et al., 1979
; National Toxicology Program, 2006c; Walker et al., 2006). This gender-selective tumorigenicity of HAH compounds in rats has also been observed for industrial mixtures of PCBs, including Aroclor 1254 (Mayes et al., 1998).
Because there is evidence that HAH toxicity involves AHR pathway activation, it has been proposed that the ability of an HAH compound to alter the expression of AHR-regulated genes (e.g., cytochrome P450 1A1, Cyp1a1) may be used to differentiate the toxic potencies of individual HAH congeners or mixtures containing HAH congeners. The toxic equivalency factor (TEF) is an expression of the potency of individual HAH congeners relative to TCDD, the most potent inducer of AHR-mediated genes. Thus, the sum of TEF-adjusted concentrations of the HAHs present in a mixture, that is, the toxic equivalent (TEQ) (Eadon et al., 1986; Safe, 1997), can be used to estimate the total AHR-dependent toxicity of a mixture. To standardize the TEQ methodology, the World Health Organization (WHO) published "expert"-opinion TEF values for several HAH compounds (Van den Berg et al., 1998, 2006).
Several aspects of the TEF/TEQ approach may limit its ability to accurately describe the actual potencies of various HAH mixtures when estimating potential human health risks. For example, the TEQ approach does not account for AHR partial agonism and/or AHR antagonism observed for various HAH compounds that could lead to substantially lower TEQ values for a mixture (Aarts et al., 1995; Bannister et al., 1987; Chen and Bunce, 2004; Morrissey et al., 1992; Suh et al., 2003). Furthermore, this approach assumes parallelism of dose–response across endpoints and AHR ligands. If correct, such parallelism should presumably be evident for any genes involved in toxic pathways (Starr et al., 1999). In addition, TEFs/TEQs neither address the notable gender differences in HAH toxicity demonstrated in rodent models, nor assess potential chemical-specific and/or AHR-independent mechanisms. Finally, several studies have demonstrated that significant differences in individual TEFs exist among different animal species including humans (Peters et al., 2004; Silkworth et al., 2005; Vamvakas et al., 1996; Wiebel et al., 1996; Xu et al., 2000; Zeiger et al., 2001). These uncertainties need to be addressed if the TEQ approach is to continue to be applied in human health risk assessment.
Building on earlier methods such as differential hybridization, the recent advent of microarray technology has allowed toxicologists to assess genome wide responses to chemical compounds in a relatively unbiased manner. Thus, novel molecular pathways affected by chemical exposure can be revealed without prior experimental evidence (Sutter, et al., 1991). A growing number of studies have applied microarray technology to address HAH toxicity. The transcriptome response to TCDD has been investigated using various cell/tissue types from several animal species including humans (Ahn et al., 2005; Frueh et al., 2001; Hanlon et al., 2005; Martinez et al., 2002; Puga et al., 2000; Thackaberry et al., 2005). In addition, microarrays have been used to probe HAH-induced gene expression in light of differences between/among animal strain/species (Boutros et al., 2004; Boverhof et al., 2006; Pastorelli et al., 2006), experimental protocols (Dere et al., 2006), time points following exposure (Boverhof et al., 2005), dose (Boverhof et al., 2005; Fletcher et al., 2005), and chemical compounds (Vezina et al., 2004; Yang et al., 2006). Recently, Hayes et al. (2007) used microarrays to distinguish primary transcriptional events from downstream indicators of TCDD toxicity. Interestingly, even though gender appears to be a primary determinant of HAH tumorigenicity, there has yet to be a report of a microarray study investigating gender differences in gene expression following exposure.
The current study evaluates chemical-, dose-, and gender-specific responses in hepatic gene expression in rats acutely exposed to TEQ-equivalent doses of TCDD and Aroclor 1254. A probe-level, backward selecting, linear mixed-effects statistical model was developed to analyze the microarray data in relation to chemical treatment, gender, and gender * treatment interactions (G*T). This model also evaluated the inherent biological and technical variability of the data. The results provide new insights into the likely role of gender in the hepatic tumorigenic responses of Sprague–Dawley (SD) rats to treatment with TCDD or Aroclor 1254, while also revealing that the TEQ provides only partial insight into the genome wide response to these chemicals.
|
METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAFUNDINGREFERENCES |
|---|
Chemicals.
TCDD was obtained from Accustandard (New Haven, CT; catalog no. D404N; Lot no. 970401R-AC; 99.1% pure). The single contaminant was identified as a pentachloro-hydroxydiphenyl ether determined by GC/MS. Aroclor 1254, lot no. 122-078, was the same lot of material used in an earlier chronic bioassay conducted for the General Electric Company (Mayes et al., 1998
). The calculated Aroclor 1254 WHO-1998 TEQ based on the 12 PCB congeners with an assigned TEFs (Van den Berg et al., 1998) is 47 ppm. Following completion of the current experimental study, the WHO updated the TEF values of several PCB congeners. The new WHO-2005 TEQ (Van den Berg et al., 2006) for the Aroclor 1254 lot used is 
21 ppm.
Animals.
Twenty-four female (171–194 g) and 24 male (190–207 g) SD rats (Crl:CD(SD)IGS BR, Charles River Laboratories, Wilmington, MA) were randomly assigned to 6 groups of four rats per sex. No group adjustments based on body weights were necessary. Animals were acclimated for 8 days.
Dose selection and treatment.
Administered concentrations were chosen to provide comparable WHO-1998 TEQ dosages of Aroclor 1254 and TCDD for the high and medium dose levels. The medium Aroclor 1254 dose, 6 mg/kg/day, was chosen to approximate the highest adjusted daily intake reported in the bioassay (Mayes et al., 1998). At the estimated Aroclor 1254 WHO-1998 TEQ of 47 ppm, the medium TCDD dose was approximately 0.3 µg/kg/day. To attain TCDD liver concentrations in this short-term study comparable with those observed in the chronic ongoing National Toxicology Program (NTP) TCDD studies, and based on TCDD modeling efforts, a dose level ten times higher than this was also chosen (Nigel Walker, personal communication, NTP). The highest dose of Aroclor 1254 was estimated to be WHO-1998 TEQ-comparable with the highest TCDD dose. The lowest Aroclor 1254 dose was 10 times lower than the medium dose to assure that the response of highly inducible genes would not be saturated. TCDD and Aroclor 1254 were dissolved in warm corn oil. Test chemicals were administered by gastric intubation in corn oil (5 ml/kg) at dose levels of 0.6, 6.0, or 60 Aroclor 1254 mg/kg, or 0.3 or 3.0 TCDD µg/kg for 3 consecutive days. Control groups for each gender received corn oil alone. On Day 4 the animals were euthanized by CO2 overdose and the livers were removed and dissected into 10 sections. Each section was wrapped in foil, frozen in liquid nitrogen within two minutes of death, and stored at –70°C until processed for RNA. Animals were fed Certified Rodent Diet, No 5002 (PMI Nutrition International, St Louis, MO). Food and water were provided ad libitum. All animals were treated in accordance with the Animal Welfare Act. Rats were observed twice daily for signs of overt toxicity and weighed pretest and before necropsy. No treatment related effects were observed.
RNA extraction and microarray processing.
Total RNA was isolated from 200 mg of tissue using the RNA-STAT-60 (Tel-Test, Friendswood, TX) procedure. The RNA was dissolved in RNAse-free water and quantified by spectrophotometry. Quality of RNA isolations was determined using the Agilent BioAnalyzer 2100 Agilent, Palo Alto, CA). Three RNA samples that failed were re-isolated. RNA levels were measured using Affymetrix GeneChip RG-U34A arrays, containing 8799 probe sets, according to the standard protocol. One RNA sample from each rat was hybridized to a single microarray. In addition, technical replicates were generated from re-extracted RNA of randomly chosen rats within selected groups, whereas in other cases existing extracts were hybridized to two additional chips. Initially, a total of 60 chips were analyzed for this study. An additional 16 chips were added following quality control (QC) analysis (see below). All microarray data has been submitted to the Gene Expression Omnibus (GEO; http://www.ncbi.nlm.nih.gov/geo) as series GSE9838
[NCBI GEO]
.
QC analysis.
Raw GeneChip data (.CEL files) were fit to probe-level models using the fitPLM function of affyPLM package version 1.12.0 (Bolstad et al., 2005) of Bioconductor (Gentleman et al., 2004) as implemented in R version 2.4.1 (R Development Core Team, 2006). For specific arguments within the fitPLM function, background.method = "GCRMA" and normalize.method = "quantile" were selected. Normalized unscaled standard errors (NUSE) were calculated and a cutoff of a median NUSE value of > 1.05 was used to exclude poor quality data. This procedure identified eight chips of poor quality. The NUSE cutoff value was further justified by the observance that chips with a NUSE > 1.05 appeared to be poorly stained in scanned images, had lower median perfect match (PM) signal-to-noise ratios and lower median PM signal intensities, and higher scaling factors relative to all the chips analyzed in this study. QC data for each GeneChip, including median and interquartile range NUSE values, are reported in Supplementary Table 1. It is important to note that this data processing procedure was used only for QC analysis.
Because the QC analysis removed data that were essential to this study, RNA was re-isolated from frozen liver tissues for all experimental control rats and for those exposure groups most affected by the data removal. The re-isolated RNA was then used to hybridize an additional 16 RG-U34A genechips. Thus, this additional genechip data required the incorporation of a "scanner effect" component into all downstream microarray analyses (see next section).
Multifactor statistical model.
Background-corrected .CEL files were natural logarithm-transformed, then quantile-normalized using the bg.adjust.gcrma function of gcrma version 2.8.0 (Wu et al., 2004) and normalize.AffyBatch.quantiles function of affy version 1.12.2 (Irizarry et al., 2003) as implemented in R version 2.4.1 (R Development Core Team, 2006). Each probe set was modeled independently using a linear mixed-effects model (Pinheiro and Bates, 2004) to account for fixed treatment, gender, probe affinity, and scanner effects, as well as animal, RNA extraction, and hybridization random effects. A model selection procedure was implemented to choose the most parsimonious model that describes the variability in measured intensities. Starting from a complex model containing all estimable two-way interactions between fixed effects, terms were tested for removal one-by-one using hypothesis tests based on sums of squared residuals.
The linear mixed-effects model applied to each probe set can be written
 | (Model 1) |
Second, the function f() contains all the fixed effects in the model. Main effects and two-way interactions are considered between treatment, gender, probe, and scanner. The Greek alphabet is used to denote the parameters themselves. For instance, 
t is the treatment effect for treatment t, and (
)tg is the two-way interaction between treatment t and gender g. Other parameters are defined analogously. Note that the experimental design does not allow for estimation of the interaction between scanner and treatment, so this term is not included in the arguments of f(). The actual form of the function f() will be discussed further below.
Third, the function r() contains all the random effects in the model. Two levels of grouping are assumed in this error model. The ba random effect accounts for animal-to-animal variability, and the bae random effect accounts for RNA extraction-to-extraction variability from the same animal. Throughout this paper, the function r() is always assumed to take the form
 |
Finally, the residual error between the predictor f() + r() and the observed ytgpsaeh is denoted 
tgpsaeh, and is assumed to be normally distributed with mean 0 and variance v3, and independent of the two levels of random effects.
The indices in the linear model have the following ranges: there are five treatments plus the baseline corn oil treatment, so t ranges from 1 to 6 (t = 1 indicates corn oil). There are two genders, g = 1 or 2, where g = 1 indicates males. There are between 11 and 20 probes in any given probe set, p = 1, ..., [11–20]. Two scanners were used, s = 1 or 2. Forty-six animals were included, a = 1, ..., 46. Up to three RNA extractions were performed per animal, e = 1, ..., [1–3]. And up to 3 hybridizations were performed for each RNA extract, h = 1, ..., [1–3].
The first, and largest, model considered in the model selection procedure was one containing all estimable two-way interactions between the fixed effects. The f() function was defined
 | (1) |
The baseline condition was defined to be males given corn oil only (g = 1 and t = 1). Treatment contrasts were used for the gender and treatment factors with the constraint that 1 = 0 and 1 = 0. Because the object of primary interest is the average expression from a probe set on an average scanner, not any particular probe on either individual scanner, probe effects 
p and scanner effects 
s were constrained to sum to zero: 
and 
. The interaction terms were constrained similarly: ()tg = 0 for g = 1 and for t = 1; ()tp = 0 for t = 1; ()gp = 0 for g = 1; ()sg = 0 for g = 1; 
for all t; 
for all g; 
for all s; 
for all p; and 
for all g.
Under this choice of parameterization, the treatment and gender coefficients express gene expression differences relative to a baseline defined as the response of male rats under the corn oil treatment. The model parameters then have the following interpretations: 
represents the average probe intensity in corn oil exposed male rats, which is also averaged over the two scanners; t is the treatment effect for treatment t applied to male rats; 2 is the gender effect comparing females with males in the corn oil treatment; p is the probe affinity effect for probe p; ()tg is the interaction between treatment t and gender g; ()tp is the interaction between the treatment t and the affinity of probe p; ()gp is the interaction between gender g and the affinity of probe p; ()sp is the interaction between scanner and probe affinity; and ()sg is the interaction between scanner and gender.
It is expected that in many probe sets to be analyzed, some of the main effects and interactions given in equation (1) will not be necessary to fully describe the variability in observed intensities. To remove nonstatistically significant interactions from Model 1, a standard backward selection process was used (Venables and Ripley, 2002). First, each of the five two-way interactions was considered for deletion from the model one-by-one. For each deletion under consideration, a single F statistic was calculated comparing the sum of squared residuals from Model 1 with the residuals from the smaller model not containing the term under consideration, where in each case the F statistic was calculated conditional on a maximum likelihood estimate of the variance components v1, v2, and v3. The R function anova.lme contained in the nlme library (version 3.1–78) was used for these calculations (see Pinheiro and Bates, 2004, section 2.4.2, for details).
Once all five F statistics and their p values were tabulated, the term with largest p value was removed from the model if its p value was larger than 0.05 or the F statistic itself was smaller than 3. If the term was removed, then the four remaining F statistics were recalculated and examined in the same way. Once all remaining two-way interactions have p values less than 0.05 and F statistics larger than 3, the four main effects were next considered for deletion in the same manner. Using this procedure, the final model was guaranteed to have all terms be statistically significant by our definition (p < 0.05 and F > 3). The threshold of 3 on the F statistic has the effect of making the algorithm more strict about including interactions involving the probe affinities, because those F tests contain large degrees of freedom in the numerator. This is desirable because including a probe affinity interaction reduces the statistical power to detect fold change due to the large number of parameters in the interaction, for example, 11–20 parameters for a probe/gender interaction and 55–100 parameters for a probe/treatment interaction.
Under the parameter constraints given above, treatment-based fold changes were calculated relative to males given corn oil. Several other sets of parameter constraints (also called contrasts) were implemented to calculate, for instance, fold changes of females given the five chemical treatments versus females on corn oil, and comparisons of WHO-1998 TEQ-equivalent doses of the two chemicals. To deal with the multiplicity of tests performed, the false discovery rate (FDR) was determined by generating q values as described by Storey (2003) via the R qvalue package version 1.8.0.
Identification of discovery genes set.
The model output was further processed to obtain a "discovery" gene set consisting of those genes most changed by exposure. Initial filtering involved removal of probe sets lacking a significant model treatment term, t (i.e., no overall treatment effects). Contrasts were then made for each of the 10 chemical treatment groups versus corn oil control for both male and female rats. It is important to note that probe sets lacking a significant G*T term, ()tg, have the same fold-change values for both genders within a specific treatment group. A cutoff was established to identify those probe sets most altered by exposure by only including probe sets whose differential expression ratio was 
2.718- or 
0.368-fold change (p value 0.0001). This fold-change cutoff was arbitrarily chosen because the estimates derived from the model output were in natural logarithm form (i.e., loge(2.718) and loge(0.368) equal 1 and –1, respectively). When the p values were adjusted to control for FDR, q values for all treatments where a probe set met both these aforementioned criteria were q 0.00043.
The "discovery" gene set was then subjected to further annotation to identify probe sets representing "redundant" transcripts. Briefly, probe sets were first given annotations found in the NetAffx RG-U34A annotation file release 21. Each probe set was then manually checked for alignment and overlap with other probe sets in the discovery gene set using the Rat Genome November 2004 Assembly, version 3.4. Probe sets clearly representing redundant transcripts were removed from the resulting discovery set by including only the redundant probe set with the maximum (induced or repressed) fold-change value across all the treatment groups. Attempts were made to include probe sets representing potential alternative transcripts. In addition, the gene symbol annotations of some probe sets were altered to better represent the genome alignment. The discovery gene list was further segmented into two groups based upon the presence or absence of a G*T interaction term (i.e., ()tg) in the model.
Gene clustering and ontological analyses.
Nonredundant discovery probe sets representing genes of interest were subjected to a hybrid clustering method, Hierarchical Ordered Partitioning and Collapsing Hybrid (HOPACH) version 1.8.0 (van der Laan and Pollard, 2003) implemented in R. Gene dissimilarities in expression were calculated via cosine angle distance metric (i.e., "cosangle" option) and the first level of the tree below which the mean split silhouette increases (i.e., "greedy" option) was utilized to differentiate gene clusters. The resulting gene clusters were then further assessed for stability via nonparametric bootstrap resampling (1000 iterations). The effects of several other clustering methods (i.e., hierarchical, K-means, PAM, DIANA, etc.) were also determined but not used in the final analyses because only the HOPACH method produced a final ordering of genes, distinct cluster boundaries, and statistical estimation of cluster memberships for each probe set.
Gene ontology analyses were conducted using the functional annotation tool of the Database for Annotation, Visualization and Integrated Discovery (Dennis et al., 2003). An EASE score of 0.05 was used as a cutoff for overrepresented ontologies and the rat RG-U34A genechip served as the background. Functional classes analyzed were the gene ontology categories "biological process" and "molecular function," as well as the Kyoto encyclopedia of genes and genomes (KEGG) pathways.
Analysis of GEO series GSE5789.
Microarray data, from several NTP 2-year rat carcinogenicity studies, had previously and independently been submitted to the GEO (http://www.ncbi.nlm.nih.gov/geo) as series GSE5789
[NCBI GEO]
. In those studies, female Harlan SD rats were gavaged (5 days/week) with TEQ-equivalent doses of TCDD (100 ng/kg), 2,3,4,7,8-pentachlorodibenzofuran (PeDF; 200 ng/kg), PCB 126 (1000 ng/kg), PCB 153 (1000 µg/kg), or a binary mixture of PCB 126 (1000 ng/kg) and PCB 153 (1000 µg/kg) (Abe et al., 2006; National Toxicology Program, 2006a, b, c, d, e). At 13 weeks, liver RNA was extracted and analyzed with Affymetrix RG-U34A microarrays. This GEO series consists of 21 CEL files, which have been analyzed previously (Ovando et al., 2006; Vezina et al., 2004). For the current analysis, NTP data was background-corrected and quantile-normalized using the gcrma package version 2.8.0 in R (Wu et al., 2004) prior to the development of a standard linear model using the limma R package version 2.5.0 (Smyth, 2005). Briefly, contrasts between control and treatment groups were made using the contrasts.fit function prior to variance shrinkage using the eBayes function. q values were then determined for each contrast via the decideTests function with arguments method = "global" and adjust.method = "BH." Resulting fold-change values (treatment vs. corn oil control) were converted to natural log and further analyzed in respect to the current discovery gene data set. It is important to note that the NTP data was not modeled together with the current experimental data described in this paper (i.e., the GE study). Thus, only patterns of gene expression, and not the actual magnitude of changes, should be compared across the two studies.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAFUNDINGREFERENCES |
|---|
Multifactor Model Output
Model outputs for estimates, contrasts, and p and q values for all 8,799 probe sets were submitted to the GEO database (Series GSE9838 [NCBI GEO] ; http://www.ncbi.nlm.nih.gov/geo) in the supplementary file Linear_model_Output.xls. However, for illustration purposes, the estimates for each parameter for three sample genes are shown in Table 1. The estimates for the male baselines (
) represent the x-intercepts for each gene. Depending on the contrast in question, estimates of the constituent parameters that survived the model iterations are also shown. These estimates represent fold change from the baseline. For example, for Female Treatment, which compares females (gender 2) exposed to 3 µg TCDD/kg/day (treatment 3) with females on corn oil, one adds the treatment main effect, 3, and the treatment gender interaction, ()32, to the baseline, yielding the estimated fold change. As another example, the G*T interaction for the thyroxine-binding globulin (TBG) gene was significant, whereas the G*T term for the cytochrome P450 2b (Cyp2b) and Arnt2 genes were deemed insignificant and subsequently left out of the model. Accordingly, significant downregulation of TBG was observed in female rats exposed to highest dose of TCDD (estimate 0.27; p < 10–6), whereas no significant differential expression was seen in males given the same treatment (estimate 0.94; p = 0.8). Although no significant treatment effect was seen for Cyp2b for high-dose TCDD exposure in either gender (p = 0.8), a highly significant Cyp2b induction by Aroclor 1254 was seen when comparing the highest doses of Aroclor 1254 and TCDD (p < 10–21). Analysis of the Gender Difference in expression for rats receiving just corn oil demonstrated that both TBG and Cyp2b displayed sexually dimorphic basal expression. The AHR nuclear translocator 2 (Arnt2) gene had no significant treatment or gender effects, thus the final model did not contain any parameters estimates.
|
Table 2 gives the estimated residual standard deviation for the three levels of the error model (animal, RNA extraction, and hybridization). In all three cases the residual error due to hybridization was larger than animal and extraction error, likely reflecting both hybridization and date of hybridization, that is, batch effects. In both genes where a significant treatment effect was demonstrated, all three levels of the error model explained at least a moderate portion of residual variance (> 13%). In Arnt2, the animal-to-animal contribution to total residual variance was small (2%). Residual error estimates for all genes are presented in GEO Series GSE9838 [NCBI GEO] .
|
Estimation of HAH-TEQ Dose Comparability Using Gene Expression
The two higher doses of Aroclor 1254 (6.0 and 60 mg/kg/day) and the two doses of TCDD (0.3 and 3.0 µg/kg/day) administered daily in this study were specifically chosen to be comparably potent (i.e., equal TEQs) based upon the WHO-1998 TEFs for coplanar PCBs (Van den Berg et al., 1998
). Because the 0.3 and 3.0 µg/kg/day doses of TCDD correspond to the 6.0 and 60 mg/kg/day doses of Aroclor, these doses are subsequently referred to as "medium" and "high," respectively for both chemicals. The 0.6 mg/kg/day (i.e., low dose) of Aroclor 1254 did not have a TCDD equivalent dose in the current study. Figures 1a and 1b depict the fold-change values for representative sets of genes, with or without a G*T interaction, responsive to both TCDD and Aroclor. The x axis is expressed as log10 WHO-1998 TEQ dose to emphasize the initially estimated chemical equivalency. The WHO has recently updated its TEF values for several PCBs used derive the estimated Aroclor 1254 TEQ initially used in the current study (Van den Berg et al., 2006), and the relationships of these new values (WHO-2005 TEFs) to the altered genes in Figures 1a and 1b are reflected in Figures 1c and 1d .
|
Identification of "Discovery" Subset of Genes Most Responsive to HAHs
Removal of probe sets lacking a significant model treatment term,
t (i.e., no overall treatment effects), created an initial subset of 3413 probe sets (out of 8799 probe sets on the RG-U34A chip) for further analysis. Model outputs for estimates, contrasts, and p and q values for all probe sets with a treatment term are presented in Supplementary Table 2. Following the application of the cutoff procedure described in the "Materials and Methods" section, the resulting discovery set consisted of 233 highly responsive probe sets or 173 unique genes. Subsequent removal of redundant probe sets and segmentation into 2 groups based upon the presence or absence of a G*T interaction term (i.e., ()tg) in the model, resulted in the identification of 54 and 51 nonredundant probe sets with or without a significant G*T model term, respectively. It is important to note that because redundant probe sets were removed from the discovery gene set there is a possibility that alternatively spliced transcripts might not be fully represented. However, an attempt was made to avoid missing unique transcripts by manually verifying redundant probe sets using the rat genome alignment. Complete data for both the redundant and nonredundant discovery gene sets are given in Supplementary Table 3.
Clustering of Potentially Coregulated Genes
The discovery gene set was subjected to hybrid clustering separately for genes with or without a G*T term across the 10 or 5 treatment groups, respectively. Figure 2 displays a heatmap with 8 gene clusters with a G*T term (numbered 1–8 in the leftmost column) predicted using the HOPACH program. Clusters of 2–12 genes were generated using the expression data depicted in columns under the heading "Treatment versus Corn Oil." Genes (rows) are rank-ordered such that genes near the adjacent cluster margin demonstrate greater similarity to the nearby cluster than those genes further away from the cluster margin. Nonparametric bootstrap analysis (i.e., fuzzy clustering) demonstrated that HOPACH cluster memberships were robust for most gene assignments following 1000 bootstrap resamplings. Detailed HOPACH and bootstrapping results are given in Supplementary Table 3. Figure 3 depicts the clustering results for discovery genes lacking a G*T term. Five unisex treatment columns are displayed under the "Treatment versus Corn Oil" heading. Genes without a G*T term formed five discernable clusters of 4–28 genes each (labeled A–E in the leftmost column).
|
|
For discovery genes with G*T term, Clusters 1, 5, 6, 7, and 8 consisted of induced genes, whereas Clusters 2–4 were primarily made up of repressed genes (Fig. 2). Most genes in Cluster 1 displayed induction at the low and medium doses of Aroclor 1254 in female rats only. A subset of genes in Cluster 2 (e.g., ornithine aminotransferase [Oat], glutamine synthase [Glul], alcohol dehydrogenase 1 [Adh1], ribonuclease A family 4 [Rnase4]) was repressed in male rats exposed to the high dose of Aroclor 1254 (Fig. 2). Cluster 3 and 4 genes were repressed primarily in female rats exposed to either chemical. Genes induced by low and medium doses (but not at the high dose) of Aroclor 1254 dominated Cluster 5 (e.g., apolipoprotein B [ApoB]). Genes in Cluster 6 were induced only in female rats exposed to Aroclor 1254 (cytochrome P450 3a [Cyp3a]), whereas 2 genes induced by Aroclor 1254 in only male rats made up cluster 7 (e.g., cytochrome P450 2c37 [Cyp2c37] and UDP-glucuronosyltransferase phenobarbitol-inducible form [Udpgtr2]). Finally, Cluster 8 genes were primarily induced by both chemicals, in at least 1 treatment, in either gender (e.g., cytochrome P450 1b1 [Cyp1b1] and aldehyde dehydrogenase 3a1 [Aldh3a1]).
For discovery genes lacking a G*T term, genes in Cluster A were repressed primarily in rats exposed to the high dose of Aroclor 1254 (Fig. 3). Genes induced by Aroclor 1254 only made up Cluster B (e.g., Cyp2b and aldehyde dehydrogenase 1a4 [Aldh1a4]). The majority of the genes in Cluster C were induced by both chemicals (e.g., cytochromes P450 1a1 and 1a2 [Cyp1a1, Cyp1a2]). Cluster D genes were repressed by the high doses of both chemicals (e.g., transforming growth factor beta 1i4 [Tgfb1i4]), whereas Cluster E genes were repressed primarily only by the high dose of Aroclor 1254 (e.g., carbonic anhydrase 3 [Ca3]).
Discovery Gene Ontologies
Gene clusters were initially screened for enrichment in gene ontology categories and KEGG pathways. Various functional classes that were significantly (p 0.05) enriched in any one cluster are listed in Table 3. Functional analysis of individual clusters in the current study was somewhat limited due to the small number of genes found in many clusters. Therefore, an additional functional analysis was performed to reveal gene ontologies/KEGG pathways represented in the entire discovery gene subset, independent of clusters, but separately based upon presence of a G*T model term (Table 4). This latter analysis revealed significant (p 0.05) functional categories that may have spanned several gene clusters. Overall, and as expected, the discovery gene set was enriched with genes involved in xenobiotic/steroid/lipid metabolism, both within and across gene clusters. For example, Clusters 6, 7, and C displayed significant enrichment for enzymes involved in the KEGG pathway "metabolism of xenobiotics by cytochrome P450" (Table 3). In addition, genes involved in this pathway were found in to be significantly enriched in discovery gene subsets with or without a G*T model term (i.e., Clusters 2, 6–8, and A–D) (Table 4).
|
|
G*T Interactions and Gender Differences in Baseline Expression
The G*T interaction coefficients, (
)tg, in the multifactor model represent the difference between female and male responses to any one treatment (i.e., Aroclor 1254 low dose) corrected for baseline (corn oil control) differences in expression between the sexes (g). Because the estimate of ()tg must be analyzed in the context of the direction of fold change (i.e., induced or repressed), the absolute value of the natural log fold change represents the magnitude of the difference in response between the two sexes. Thus, the columns under the heading "gender * treatment" in Figure 2 display the relative natural log fold difference between genders for discovery genes with a G*T term within each treatment group. Although all clusters for genes with a G*T term possessed subsets of genes with significant (p 0.01) gender effects (Fig. 2), Clusters 1, 3, 4, 5, and 6 had an overall mean absolute G*T difference 2.7 fold for at least 1 treatment group (data not shown).
Table 5 summarizes the discovery genes with the largest G*T differences (i.e., 2.7-fold and p 0.01) for high-dose exposure to either Aroclor 1254 or TCDD. Cytochrome P450 2c (Cyp2c), Cd36, TBG, and ectonucleotide pyrophosphatase/phosphodiesterase 2 (Enpp2) all possessed large G*T interactions ( 2.7-fold) in both Aroclor 1254 and TCDD high-dose exposure groups. Overall, the vast majority of genes with a G*T term 2.7 fold were altered to a greater extent in female rats than males. Exceptions included male-specific induction of elastin (Eln) by TCDD, male-specific repression of Oat by Aroclor, and the approximately 7 fold greater induction of Aldh3a1 in male rats than females by Aroclor 1254 (Table 5). Discovery genes with a significant (p 0.01) G*T interaction in at least one exposure group were found in all clusters (1–8), although the actual fold difference may have been < 2.7 (Fig. 2).
|
Differences in baseline (corn oil control) gene expression between the genders (
g) were also of interest when analyzing the discovery gene set. Thus, the natural log fold difference estimate of the model component g is represented in Figures 2 and 3 under the column heading "Baseline Gender." Positive values for "Baseline Gender" indicate greater baseline expression for females and genes with negative values have a greater baseline expression in male rats. It was evident that some clusters were largely made up of genes with higher baseline expression in either female (Clusters 3, 4, and 7) or male (Clusters 1 and 6) rats, whereas the other clusters, including all of the clusters for discovery genes lacking a G*T term, failed to exhibit clear patterns of gender baseline effects (Figs. 2 and 3). Supplementary Table 3 lists the values for g, ()tg, fold-change estimates, p, and q values for all probe sets in the discovery gene set.
Chemical Differences in Genomic Response
Elucidation of chemical differences in the genomic response to TCDD and Aroclor 1254 was another major objective of this study. Although the multifactor model does not directly differentiate between chemical types, it does separate each treatment group to allow for calculation of recontrasts representing the difference in fold change (and associated p and q values) between any two gender-specific treatments. Thus, contrasts were made to examine the "chemical effect" as the difference in the natural log fold-change response between Aroclor 1254 and TCDD for males and females treated with WHO-1998 TEQ-equivalent doses of each compound (i.e., medium and high doses). The absolute values of the "chemical effect" natural log fold difference for each discovery gene are given under the column heading "Aroclor versus TCDD" in Figures 2 and 3.
Table 6 highlights genes displaying the largest chemical effects at the highest dose level of each compound (i.e., 2.7-fold difference and p 0.01). Genes with large chemical effects for male rats responded to Aroclor 1254 only and not TCDD at the highest dose. In addition, all five genes without a G*T term in Table 6 were significantly altered by the high dose of Aroclor 1254 only. In female rats, three genes (M24875, glucose-6-phosphatase [G6pc], and insulin growth factor binding protein 1 [Igfbp1]) responded to TCDD more than Aroclor, whereas an additional three genes were induced in an Aroclor-specific manner (Udpgtr2, Cyp3a, and Cyp2c).
|
Comparison with the NTP Subchronic HAH Exposure Microarray Study
Figure 4 depicts a heatmap of the multifactor model output for discovery genes determined in the current rat study aligned to the linear model output for these same probe sets using Series GSE5789 [NCBI GEO] microarray data submitted to the GEO database. Because the subchronic NTP study utilized only female rats, only female rat data is given for discovery genes with a G*T model term. Although the magnitude of fold change is most probably not directly comparable due to obvious experimental and data analysis differences, similar trends of HAH-induced induction/repression are certainly apparent. For instance, most Cluster C genes were significantly induced following acute exposure to TCDD and Aroclor 1254 in the current study and by TCDD, PeDF, PCB 126, and the binary mixture of PCB 126 and PCB 153 in the subchronic NTP study. Other gene clusters demonstrating a strikingly similar gene expression pattern for both studies include Clusters 3, 8, B, D, and E (Fig. 4).
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAFUNDINGREFERENCES |
|---|
The current study was designed to elucidate the initial molecular events that may responsible for Aroclor 1254– and TCDD-induced liver tumors in rats. To accomplish this, we characterized liver gene expression relative to various experimental factors, including chemical, dose, gender, and TEQ. A statistical model was developed to determine the influence of each factor, or interactions of these factors, on each gene. The major questions addressed were: (1) are there gender-specific gene responses that explain the differential susceptibility of female rats to HAH tumorigenicity; (2) does the TEQ approach accurately characterize the broad genomic response of TEQ-equivalent chemicals; and (3) can microarray analyses of multifactor experimental designs provide useful information regarding a chemical's mode of action?
Multifactor Linear Model: Justification and Performance
Several studies have investigated the relative contributions of biological variability and the technical measurement process to observed expression variability (Klebanov and Yakovlev, 2007; van den Huevel et al., 2005; Zakharkin et al., 2005). Probe-level affinity effects were incorporated directly into the current model because large variation in probe affinities exist within single probe sets (Li and Wong, 2001). Typically, probe affinity is added as a blocking factor in a multiway analysis of variance because individual probes demonstrate similar binding affinity across multiple chips. Methods which incorporate probe-level affinities have been shown to be more sensitive and specific in validation studies than methods based on probe set summaries (Barrera et al., 2004; Lemieux, 2006; Liu et al., 2006).
Results vary as to which components of the entire measurement system contribute most to total variability, but it is clear that biological and at least one level of technical variability should be modeled separately if the level of replication in a designed experiment allows for them. The current study allows explicit estimation of the contributions of animal, RNA extraction, hybridization, and scanner to total RNA expression variability. By accounting for all pertinent sources of variability, proper statistical inference can be made on chemical and gender effects and their statistical confidence levels. Table 2 clearly demonstrates that for all three of these representative probe sets hybridization error, which includes batch effect, contributes the most variability to the total residual error. This is also true in the vast majority of the probe sets analyzed (Supplementary Table 2). Thus, technical replication, particularly at the hybridization level, is strongly justified in this type of experiment.
A model selection procedure was used to find the most parsimonious predictor of observed intensities based on the covariates under consideration: treatment, gender, probe affinity, and scanner. A main-effects-only model did not adequately describe the data for many probe sets. For instance, some genes were upregulated in females at high doses of TCDD, but were not upregulated in males. This is a critical finding which can only be detected by a model which includes a G*T interaction. However, it is unwise to model every probe set with an overly complicated model containing all interactions. In many cases these interactions are not supported by the data and therefore should not be included in the predictive model. Because no single model described all the probe sets, a widely accepted model selection procedure, the backwards selection algorithm (Venables and Ripley, 2002), was performed independently for each probe set.
Assessment of AHR-Mediated Response at TEQ-Equivalent Doses
The medium and high doses of Aroclor 1254 and TCDD were designed to be of equal potency according to WHO-1998 TEFs (Van den Berg et al., 1998). Recalculation of the Aroclor 1254 TEQ using the recently updated TEFs (Van den Berg et al., 2006) resulted in a lower overall TEQ (from 47 ppm to 21 ppm total TEQ) due to reduced TEF values for most of the mono–ortho PCB congeners. Nevertheless, the relative magnitude of fold changes observed for both chemicals at the two TEQ-equivalent doses used in the current study were generally very close for the most responsive AHR genes (Fig. 1). This supported the comparison of the responses to the two chemicals. Interestingly, for many genes (e.g., TBG), gender was the predominant determinant of dose comparability rather than which WHO-TEFs were used to calculate the TEQ.
Known AHR "battery" genes (Nebert et al., 2000) fell into two gene clusters (Clusters 8 and C) either with (i.e., Cyp1b1, Aldh3a1) or without (i.e., Cyp1a1, Cyp1a2, Nqo1) a significant G*T interaction model term. The sensitivity of the current analysis to detect significant G*T interactions is supported by the identification of Cyp1b1. An earlier study of the effect of TCDD on the expression of Cyp1b1 in SD rats clearly demonstrated that the induction of Cyp1b1 was much greater in female than in male rats (Walker et al., 1995). However, due to the limited number of TEQ doses employed in the current study, it could not be determined which version of PCB TEF values (i.e., 1998 or 2005) was more accurate. Because the TEF/TEQ approach does not attempt to account for sexually dimorphic responses, it is somewhat disconcerting that a whole cluster of genes responding in a manner consistent with AHR regulation (i.e., Cluster 8) displayed significant gender differences in expression. This diminishes the value of TEQ for gender-dependent evaluations.
Many of the genes in Clusters 8 and C are involved in xenobiotic/steroid/lipid metabolism, consistent with the main function of AHR battery genes. Examination of microarray data from NTP cancer studies (Ovando et al., 2006, GEO series GSE5789
[NCBI GEO]
; Vezina et al., 2004) revealed that 6 out of 12 Cluster 8 genes and 22 out of 28 Cluster C genes were also significantly induced (q 0.05) at 13 weeks of exposure of female SD rats to either TCDD, PeDF, PCB 126, or a binary mixture of PCBs 126 and 153 (Fig. 4). Furthermore, 83% of Cluster 8 genes and 79% of Cluster C genes were also found to be induced 2-fold by a single oral gavage of 40 µg TCDD/kg body weight (BW) in male SD rats (see Supplementary Table 4 of Boverhof et al., 2006; Fletcher et al., 2005). Many Cluster 8 and Cluster C genes, such as Igfbp1, epoxide hydrolase 1 (Ephx1), malic enzyme 1 (Me1), and cytochrome P450 oxidoreductase (Por), possess phylogenetically conserved (between mouse and rat) core dioxin responsive elements (DREs) in their promoter regions (E. A. Carlson, unpublished data; Boverhof et al., 2006); suggesting that these genes are likely also AHR-regulated.
Recently, Ovando et al. (2006) reported several genes whose expression appears to be downregulated by subchronic and acute exposure of rodents to TCDD, PeDF, and PCB 126 in an AHR-dependent manner. In the current study, several discovery genes (Clusters 2, 3, 4, A, D, and E) were HAH-repressed genes, including many of the repressed genes previously described by Ovando et al. (2006) such as TBG (a.k.a. Serpina7), Cyp3a13, solute carrier organic anion transporter 1a4 (Slco1a4), and steroid 5 alpha-reductase 1 (Srd5a1). However, unlike the induced genes in Clusters 8 and C, many of the downregulated gene clusters appeared inconsistent with the TEF/TEQ approach. Specifically, repressed genes in Clusters 3 and 4 were more responsive in female rats. Furthermore, some clusters displayed significant differences in response to TCDD and Aroclor 1254 (i.e., Clusters 2 and 4). Although Cluster A and E were repressed in the current study, many of these same genes were not significantly repressed by subchronic exposure of female SD rats to various HAHs (Ovando et al., 2006; Vezina et al., 2004), suggesting that these genes may represent an acute, short-term response to HAHs only. Finally, Cluster D genes appeared to exhibit a pattern of repression most consistent with AHR regulation. Indeed, a dependence of TCDD-mediated Cyp3a13 suppression upon a functional AHR has been previously demonstrated for the murine ortholog of Cyp3a13 using AHR knockout mice (Ovando et al., 2006; Tijet et al., 2006). AHR-dependent downregulation has also been demonstrated for murine orthologs of the following rat genes: TBG, phosphoenolpyruvate carboxylkinase 1 (Pck1), Srd5a1, and Slco1a4 (Ovando et al., 2006; Tijet et al., 2006).
Gender-Specific Gene Expression Patterns
Chronic exposure of SD rats to TCDD or Aroclor 1254 induced hepatocellular adenomas primarily in female rats (Kociba et al., 1979; Mayes et al., 1998), although longevity was increased and mammary tumor counts were reduced. Previous studies on HAH hepatic tumor promotion have indicated that the presence of estrogen may be, in part, responsible for the apparent hyper-responsiveness of female rats (Lucier et al., 1991; Wyde et al., 2001, 2002). Furthermore, estrogen itself is a tumor promoter in female rats under laboratory conditions (Campen et al., 1990) and may also be a complete hepatocarcinogen (Dombrowski et al., 2006; Yager and Liehr, 1996). The exact mechanisms for the apparent HAH-estrogen interaction in female rat adenoma induction are unknown, but one possible sequence of events includes an AHR-mediated increase in estrogen metabolism, production of reactive oxygen species (ROS) by redox cycling of reactive estrogen metabolites, and increased rates of proliferation of spontaneously initiated cells mediated by ROS signal transduction (Brown et al., 2007; Klaunig and Kamendulis, 2004; Knerr and Schrenk, 2006).
This study was designed to characterize early, gender-specific gene expression patterns following acute HAH exposure. One obvious observation was the lack of a strong female-specific induction of potential AHR-regulated genes. Although the model predicted a G*T component for Cluster 8 genes, the G*T interaction was not consistently biased toward greater induction in females (e.g., Aldh3a1, Eln). One notable exception was the significantly (p 0.01) greater induction of the metastatic tumor progression marker Enpp2 (a.k.a. Autotaxin) in female rats by the high dose of TCDD and Aroclor 1254 (Table 5). However, a high-dose exposure of male SD rats to TCDD (40 µg/kg BW) did reveal that Enpp2 could be significantly induced in males (Boverhof et al., 2006; Fletcher et al., 2005). Cluster 6 genes, significantly induced by Aroclor 1254 and not TCDD, appeared to respond only in female rats. Significant induction of Cyp3a (Cluster 6) by TCDD was observed for male SD rats by Fletcher et al. (2005), suggesting that the gender differences observed for Enpp2 and Cyp3a may be related to slight differences in dose–response between the sexes rather than entirely female-specific induction.
Cluster 1 genes appeared to be induced in only female rats (primarily by Aroclor 1254 only), however, this induction was limited to the lower doses and not observed in the high-dose exposure groups. Cluster 1 genes were also not acutely induced in male SD rats exposed to 40 µg TCDD/kg BW by gavage in a previous study (Boverhof et al., 2006; Fletcher et al., 2005). Analysis of GEO series GSE5789
[NCBI GEO]
microarray data (Ovando et al., 2006; Vezina et al., 2004) revealed that subchronic exposure to a binary mixture of PCBs 126 and 153 induced several Cluster 1 genes (Fig. 4; e.g., actin beta [Actb], 3-hydroxy-3-methylglutaryl-coenzyme A reductase [Hmgcr], and lanosterol synthase [Lss]). The role of these female-responsive genes that are involved in actin polymerization/depolarization (e.g., Actb and profilin1 [Pfn1]) and biosynthesis of steroids (e.g., Lss and Hmgcr), in the gender-specific hepatic tumorigenicity of HAHs (Kociba et al., 1979; Mayes et al., 1998) deserves future attention.
Unlike HAH-induced genes, both TCDD and Aroclor 1254 repressed two clusters of genes (Clusters 3 and 4) in a female-specific manner primarily in the lower dose exposure groups. However, many of these same genes were also repressed by a relatively higher dose TCDD (40 µg/kg BW) in male SD rats (Fletcher et al., 2005), questioning whether these responses are truly female-specific. Notable exceptions include the female-specific repression of carboxypeptidase A2 (Cpa2_pred) and TBG at the high dose of TCDD and Aroclor 1254 in the current study that were not observed in the previous TCDD study using male SD rats (Fletcher et al., 2005). Both TBG (a.k.a. Serpina7) and Cpa2_pred were similarly suppressed by subchronic exposure to TCDD, PCB 126, and a binary mixture of PCBs 126 and 153 (Fig. 4) (Ovando et al., 2006; Vezina et al., 2004). Ovando et al. (2006) also demonstrated that TCDD-induced repression of TGB was AHR-dependent using AHR knockout mice. Thus, the current study did reveal that a few genes/clusters were more responsive in female rats compared with males, particularly at the lower HAH doses. Other than the exception of TBG, determination if the AHR pathway directly mediates these gender-specific patterns will require further investigation.
Analysis of discovery genes displaying significant differences in basal expression between genders revealed that HAH exposure significantly alters many sexually dimorphic genes. Although the clustering approach used in the current study did not incorporate baseline gender differences in gene expression, some clusters were primarily made up of either male-dominant (i.e., Clusters 1 and 6) or female-dominant (i.e., Clusters 3, 4, and E) genes. The control of sexually dimorphic gene expression patterns by rodent hormones has been well documented (Mode and Gustafsson, 2006). Growth hormone (GH) is released from the rat pituitary in a sexually dimorphic pattern that subsequently controls the gender-specific expression of many genes, particularly steroid metabolism enzymes (Waxman and O'Connor, 2006). Several microarray studies have analyzed the rat genomic response to disruption of gender-specific GH release by hypophysectomy and continuous GH introduction in male rats (Ahluwalia et al., 2004; Flores-Morales et al., 2001; Stahlberg et al., 2005).
A major observation in the current study is that HAHs significantly disrupt many genes that display sexually dimorphic patterns of expression and this disruption is more evident for female rats. Discovery genes from the current study that were previously found to be repressed by female GH secretion patterns (i.e., male-dominant genes) in these studies include Cyp2c (a.k.a., Cyp2c11), Cyp3a, Enpp2, Cdo1, Gstm1, Ca3, and Tgfb1i4. Female-dominant discovery genes whose high expression is induced by continuous GH include Sult2a1, Alcam, Hal, Cd36, Cyp2c37 (a.k.a., Cyp2c12), and glutathione S transferase Yc2 subunit (Yc2) (Ahluwalia et al., 2004; Flores-Morales et al., 2001; Stahlberg et al., 2005). In this study, there is an obvious tendency for male-dominant genes to be induced in females only (Clusters 1 and 6) or both sexes (i.e., Enpp2, Akr7a3, Ephx1, Ugt1a6, Mettl7b, Gstm1, and Cluster B). In addition, several clusters containing female-dominant genes were downregulated by HAH exposure in females only (i.e., Clusters 3 and 4) or both sexes (i.e., Clusters D and E). Clearly, HAH exposure is opposing baseline gene expression levels disproportionately in female rats.
It is tempting to speculate that HAHs may specifically disrupt GHR signaling in female rats. Indeed, exposure of females to both TCDD and Aroclor 1254 did result in significant repression of Growth hormone receptor (Ghr) expression. Furthermore, previously studies in mice have demonstrated that 3-methylcholanthrene represses Ghr in an AHR-dependent manner (Nukaya et al., 2004). Thus, HAH disruption of sexually dimorphic gene expression patterns observed in the current study likely involves the interaction of several mechanisms, including AHR-mediated induction/repression and, possibly, altered GHR signaling. Because many of the altered genes encode enzymes directly involved in steroid metabolism and others are, in part, regulated by GHR, a reasonable hypothesis is that both altered estrogen metabolism and the disruption of GHR regulated genes produce a condition in the female rat hepatocyte that strongly favors tumorigenesis. Because gender is such a critical factor, any considerations based on the mode of action should incorporate this factor.
Chemical-Specific Gene Expression Patterns
Three gene clusters had significantly greater responses to Aroclor 1254 than TCDD at the TEQ-equivalent high dose (i.e., Clusters 6, 7, and B). Two additional clusters displayed Aroclor-specific responses at the medium dose (i.e., Clusters 1 and 5). Aroclor-specific responses were expected because this PCB mixture contained ligands for both the pregnane X receptor (PXR) and constitutive androstane receptor (CAR) pathways (Connor et al., 1995; Schuetz et al., 1998; Tabb et al., 2004). Cluster B contains the prototypical CAR-responsive gene Cyp2b, whereas Cyp3a in Cluster 6 exemplifies PXR-mediated signal transduction. However, recent studies have found significant overlap in the gene expression patterns induced through activation of both the PXR and CAR pathways in rodents (Maglich et al., 2002). Furthermore, a subset of rodent genes (mainly metabolic enzymes) was determined to be responsive to various ligands of the AHR, CAR, and PXR pathways including the current discovery genes Cyp1a2, Cyp1a1, Enpp2, Ephx1, Gadd45a, Me1, Nqo1, Ugt1a6, and Trib3 (Slatter et al., 2006). Indeed, the classic PXR-responsive gene Cyp3a was also found to be induced by subchronic exposure to AHR ligands TCDD and PCB 126, as well as the CAR/PXR ligand PCB 153 (Fig. 4) (Ovando et al., 2006; Vezina et al., 2004). Possible interactions are complicated by the fact that some predominant Aroclor 1254 congeners activate multiple pathways such as the AHR/CAR ligand PCB 118 (11% wt/wt of Aroclor 1254) and the CAR/PXR ligand PCB 153 (4.7% wt/wt) (Connor et al., 1995; Tabb et al., 2004; Van den Berg et al., 2006). However, it was clear in the current study that Aroclor 1254 induced distinct subsets of genes at TEQ doses that were nonresponsive to TCDD. These Aroclor-specific genes appeared to respond in both gender-dependent (Clusters 1, 6, and 7) and gender-independent (i.e., Cluster B and E) manners, and included both male-dominant (i.e., Cyp2b) and female-dominant (i.e., Cyp2c37) genes. For the TEF/TEQ methodology to remain valid for assessment of potential risk, future studies are needed to determine the relative contribution, if any, of such AHR-independent responses to the overall toxicity of PCB mixtures. At present, the reliance on TEQ may greatly over- or underestimate the gender-dependent toxicity, depending on the TEQ chemical makeup.
Conclusions
The utility of probe-level, mixed effect linear models for investigating complex microarray studies involving both biological and technical replication was clearly demonstrated. This multifactor model revealed an expanded subset of genes that are potentially regulated by the AHR pathway. Many HAH-responsive genes had significant G*T interactions and displayed sexually dimorphic basal expression patterns that were disrupted by HAH exposure. Because a significant number of the affected genes are important in steroid metabolism, the acute changes observed here were consistent with the hypothesis that HAHs modify hepatic estrogen metabolism and that this is a primary determinant of female-specific hepatocellular adenoma formation in SD rats chronically dosed with high levels of HAH. Considering that CYP1A induction in both genders was essentially the same, this study clearly demonstrates that one cannot rely on the responsiveness of a single gene or gene family to predict long-term effects in rats, even when it is the most responsive gene. This implies that other factors are also involved in the development of such complex outcomes as cancer. Our approach using genome wide analysis has provided valuable information and a new approach for further research in this area.
|
SUPPLEMENTARY DATA |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAFUNDINGREFERENCES |
|---|
Supplementary data are available online at http://toxsci.oxfordjournals.org/.
|
FUNDING |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAFUNDINGREFERENCES |
|---|
General Electric Company.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMETHODSRESULTSDISCUSSIONSUPPLEMENTARY DATAFUNDINGREFERENCES |
|---|
Aarts JM, Denison MS, Cox MA, Schalk MA, Garrison PM, Tullis K, de Haan LH, Brouwer A. Species-specific antagonism of Ah receptor action by 2,2',5,5'-tetrachloro- and 2,2',3,3'4,4'-hexachlorobiphenyl. Eur. J. Pharmacol. (1995) 293:463–474.[CrossRef][Web of Science][Medline]
Abe Y, Sinozaki H, Takagi T, Minegishi T, Kokame K, Kangawa K, Uesaka M, Miyamoto K. Identification of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD)-inducible genes in human amniotic epithelial cells. Reprod. Biol. Endocrinol. (2006) 4:27.[CrossRef][Medline]
Ahluwalia A, Clodfelter KH, Waxman DJ. Sexual dimorphism of rat liver gene expression: Regulatory role of growth hormone revealed by deoxyribonucleic Acid microarray analysis. Mol. Endocrinol. (2004) 18:747–760.
Ahn NS, Hu H, Park JS, Park JS, Kim JS, An S, Kong G, Aruoma OI, Lee YS, Kang KS. Molecular mechanisms of the 2,3,7,8-tetrachlorodibenzo-p-dioxin-induced inverted U-shaped dose responsiveness in anchorage independent growth and cell proliferation of human breast epithelial cells with stem cell characteristics. Mutat. Res. (2005) 579:189–199.[Web of Science][Medline]
Bannister R, Davis D, Zacharewski T, Tizard I, Safe S. Aroclor 1254 as a 2,3,7,8-tetrachlorodibenzo-p-dioxin antagonist: Effects on enzyme induction and immunotoxicity. Toxicology (1987) 46:29–42.[CrossRef][Web of Science][Medline]
Barrera L, Benner C, Tao YC, Winzeler E, Zhou Y. Leveraging two-way probe-level block design for identifying differential gene expression with high-density oligonucleotide arrays. BMC Bioinformatics (2004) 5:42.[CrossRef][Medline]
Bolstad BM, Collin F, Brettschneider J. Quality assessment of afffymetrix genechip data. In: Bioinformatics and Computational Biology Solutions Using R and Bioconductor—Gentleman R, Carey V, Huber W, Irizarry R, Dudoit S, eds. (2005) New York: Springer. 42–47.
Boutros PC, Moffat ID, Franc MA, Tijet N, Tuomisto J, Pohjanvirta R, Okey AB. Dioxin-responsive AHRE-II gene battery: Identification by phylogenetic footprinting. Biochem. Biophys. Res. Commun. (2004) 321:707–715.[CrossRef][Web of Science][Medline]
Boverhof DR, Burgoon LD, Tashiro C, Chittim B, Harkema JR, Jump DB, Zacharewski TR. Temporal and dose-dependent hepatic gene expression patterns in mice provide new insights into TCDD-Mediated hepatotoxicity. Toxicol. Sci. (2005) 85:1048–1063.
Boverhof DR, Burgoon LD, Tashiro C, Sharratt B, Chittim B, Harkema JR, Mendrick DL, Zacharewski TR. Comparative toxicogenomic analysis of the hepatotoxic effects of TCDD in Sprague Dawley rats and C57BL/6 mice. Toxicol. Sci. (2006) 94:398–416.
Brown JF Jr, Mayes BA, Silkworth JB, Hamilton SB. PCB-modulated tumorigenesis in Sprague-Dawley rats: Correlation with mixed function oxidase (MFO) activities and superoxide (O2•–) formation potentials and implied mode of action. Toxicol. Sci. (2007) 98:375–394.
Campen D, Maronpot R, Lucier G. Dose-response relationships in promotion of rat hepatocarcinogenesis by 17 alpha-ethinylestradiol. J. Toxicol. Environ. Health (1990) 29:257–268.[Web of Science][Medline]
Chen G, Bunce NJ. Interaction between halogenated aromatic compounds in the Ah receptor signal transduction pathway. Environ. Toxicol. (2004) 19:480–489.[CrossRef][Web of Science][Medline]
Connor K, Safe S, Jefcoate CR, Larsen M. Structure-dependent induction of CYP2B by polychlorinated biphenyl congeners in female Sprague-Dawley rats. Biochem. Pharmacol. (1995) 50:1913–1920.[CrossRef][Web of Science][Medline]
Dennis G Jr., Sherman BT, Hosack DA, Yang J, Gao W, Lane HC, Lempicki RA. DAVID: Database for Annotation, Visualization, and Integrated Discovery. Genome Biol. (2003) 4:3.[CrossRef]
Dere E, Boverhof DR, Burgoon LD, Zacharewski TR. In vivo - in vitro toxicogenomic comparison of TCDD-elicited gene expression in Hepa1c1c7 mouse hepatoma cells and C57BL/6 hepatic tissue. BMC Genomics (2006) 7:80.[CrossRef][Medline]
Dombrowski F, Flaschka C, Klotz L, von Netzer B, Schulz C, Lehnert H, Evert M. Hepatocellular neoplasms after intrahepatic transplantation of ovarian fragments into ovariectomized rats. Hepatology (2006) 43:857–867.[CrossRef][Web of Science][Medline]
Eadon G, Kaminsky L, Silkworth J, Aldous K, Hilker D, O'Keefe P, Smith R, Gierthy J, Hawley J, Kim N, et al. Calculation of 2,3,7,8-TCDD equivalent concentrations of complex environmental contaminant mixtures. Environ. Health Perspect. (1986) 70:221–227.[Web of Science][Medline]
Fletcher N, Wahlstrom D, Lundberg R, Nilsson CB, Nilsson KC, Stockling K, Hellmold H, Hakansson H. 2,3,7,8-Tetrachlorodibenzo-p-dioxin (TCDD) alters the mRNA expression of critical genes associated with cholesterol metabolism, bile acid biosynthesis, and bile transport in rat liver: A microarray study. Toxicol. Appl. Pharmacol. (2005) 207:1–24.[Web of Science][Medline]
Flores-Morales A, Stahlberg N, Tollet-Egnell P, Lundeberg J, Malek RL, Quackenbush J, Lee NH, Norstedt G. Microarray analysis of the in vivo effects of hypophysectomy and growth hormone treatment on gene expression in the rat. Endocrinology (2001) 142:3163–3176.
Frueh FW, Hayashibara KC, Brown PO, Whitlock JP Jr. Use of cDNA microarrays to analyze dioxin-induced changes in human liver gene expression. Toxicol. Lett. (2001) 122:189–203.[CrossRef][Web of Science][Medline]
Gentleman RC, Carey VJ, Bates DM, Bolstad B, Dettling M, Dudoit S, Ellis B, Gautier L, Ge Y, Gentry J, et al. Bioconductor: Open software development for computational biology and bioinformatics. Genome Biol. (2004) 5:R80.[CrossRef][Medline]
Hanlon PR, Zheng W, Ko AY, Jefcoate CR. Identification of novel TCDD-regulated genes by microarray analysis. Toxicol. Appl. Pharmacol. (2005) 202:215–228.[CrossRef][Web of Science][Medline]
Hayes KR, Zastrow GM, Nukaya M, Pande K, Glover E, Maufort JP, Liss AL, Liu Y, Moran SM, Vollrath AL, et al. Hepatic Transcriptional Networks Induced by Exposure to 2,3,7,8-Tetrachlorodibenzo-p-dioxin. Chem. Res. Toxicol. (2007) 20:1573–1581.[CrossRef][Web of Science][Medline]
Irizarry RA, Bolstad BM, Collin F, Cope LM, Hobbs B, Speed TP. Summaries of Affymetrix GeneChip probe level data. Nucleic Acids Res. (2003) 31:e15.
Klaunig JE, Kamendulis LM. The role of oxidative stress in carcinogenesis. Annu. Rev. Pharmacol. Toxicol. (2004) 44:239–267.[CrossRef][Web of Science][Medline]
Klebanov L, Yakovlev A. How high is the level of technical noise in microarray data? Biol. Direct (2007) 2:9.[CrossRef][Medline]
Knerr S, Schrenk D. Carcinogenicity of 2,3,7,8-tetrachlorodibenzo-p-dioxin in experimental models. Mol. Nutr. Food Res. (2006) 50:897–907.[CrossRef][Web of Science][Medline]
Kociba RJ, Keyes DG, Beyer JE, Carreon RM, Gehring PJ. Long-term toxicologic studies of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) in laboratory animals. Ann. N. Y. Acad. Sci. (1979) 320:397–404.[CrossRef][Medline]
Lemieux S. Probe-level linear model fitting and mixture modeling results in high accuracy detection of differential gene expression. BMC Bioinformatics (2006) 7:391.[CrossRef][Medline]
Li C, Wong WH. Model-based analysis of oligonucleotide arrays: Expression index computation and outlier detection. Proc. Natl. Acad. Sci. U. S. A. (2001) 98:31–36.
Liu X, Milo M, Lawrence ND, Rattray M. Probe-level measurement error improves accuracy in detecting differential gene expression. Bioinformatics (2006) 22:2107–2113.
Lucier GW, Tritscher A, Goldsworthy T, Foley J, Clark G, Goldstein J, Maronpot R. Ovarian hormones enhance 2,3,7,8-tetrachlorodibenzo-p-dioxin-mediated increases in cell proliferation and preneoplastic foci in a two-stage model for rat hepatocarcinogenesis. Cancer Res. (1991) 51:1391–1397.
Maglich JM, Stoltz CM, Goodwin B, Hawkins-Brown D, Moore JT, Kliewer SA. Nuclear pregnane X receptor and constitutive androstane receptor regulate overlapping but distinct sets of genes involved in xenobiotic detoxification. Mol. Pharmacol. (2002) 62:638–646.
Martinez JM, Afshari CA, Bushel PR, Masuda A, Takahashi T, Walker NJ. Differential toxicogenomic responses to 2,3,7,8-tetrachlorodibenzo-p-dioxin in malignant and nonmalignant human airway epithelial cells. Toxicol. Sci. (2002) 69:409–423.
Mayes BA, McConnell EE, Neal BH, Brunner MJ, Hamilton SB, Sullivan TM, Peters AC, Ryan MJ, Toft JD, Singer AW, et al. Comparative carcinogenicity in Sprague-Dawley rats of the polychlorinated biphenyl mixtures Aroclors 1016, 1242, 1254, and 1260. Toxicol. Sci. (1998) 41:62–76.
Mode A, Gustafsson JA. Sex and the liver—A journey through five decades. Drug Metab. Rev. (2006) 38:197–207.[CrossRef][Web of Science][Medline]
Morrissey RE, Harris MW, Diliberto JJ, Birnbaum LS. Limited PCB antagonism of TCDD-induced malformations in mice. Toxicol. Lett. (1992) 60:19–25.[CrossRef][Web of Science][Medline]
National Toxicology Program. NTP toxicology and carcinogenesis studies of 2,2',4,4',5,5'-hexachlorobiphenyl (PCB 153) (CAS No. 35065-27-1) in female Harlan Sprague-Dawley rats (gavage studies). Natl. Toxicol. Program. Tech. Rep. Ser. (NTP TR 529) (2006a) 1–168.
National Toxicology Program. NTP toxicology and carcinogenesis studies of 2,3,4,7,8-pentachlorodibenzofuran (PeCDF) (CAS No. 57117-31-4) in female Harlan Sprague-Dawley rats (gavage studies). Natl. Toxicol. Program. Tech. Rep. Ser. (NTP TR 525) (2006b) 1–198.
National Toxicology Program. NTP toxicology and carcinogenesis studies of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) (CAS No. 1746-01-6) in female Harlan Sprague-Dawley rats (gavage studies). Natl. Toxicol. Program. Tech. Rep. Ser. (NTP TR 521) (2006c) 1–232.
National Toxicology Program. NTP toxicology and carcinogenesis studies of 3,3',4,4',5-pentachlorobiphenyl (PCB 126) (CAS No. 57465-28-8) in female Harlan Sprague-Dawley rats (gavage studies). Natl. Toxicol. Program. Tech. Rep. Ser. (NTP TR 520) (2006d) 1–246.
National Toxicology Program. NTP toxicology and carcinogenesis studies of a binary mixture of 3,3',4,4',5-pentachlorobiphenyl (PCB 126) (CAS No. 57465-28-8) and 2,2',4,4',5,5'-hexachlorobiphenyl (PCB 153) (CAS No. 35065-27-1) in female Harlan Sprague-Dawley rats (gavage studies). Natl. Toxicol. Program. Tech. Rep. Ser. (NTP TR 530) (2006e) 1–258.
Nebert DW, Roe AL, Dieter MZ, Solis WA, Yang Y, Dalton TP. Role of the aromatic hydrocarbon receptor and [Ah] gene battery in the oxidative stress response, cell cycle control, and apoptosis. Biochem. Pharmacol. (2000) 59:65–85.[CrossRef][Web of Science][Medline]
Nukaya M, Takahashi Y, Gonzalez FJ, Kamataki T. Aryl hydrocarbon receptor-mediated suppression of GH receptor and Janus kinase 2 expression in mice. FEBS Lett. (2004) 558:96–100.[CrossRef][Web of Science][Medline]
Ovando BJ, Vezina CM, McGarrigle BP, Olson JR. Hepatic gene downregulation following acute and subchronic exposure to 2,3,7,8-tetrachlorodibenzo-p-dioxin. Toxicol. Sci. (2006) 94:428–438.
Pastorelli R, Carpi D, Campagna R, Airoldi L, Pohjanvirta R, Viluksela M, Hakansson H, Boutros PC, Moffat ID, Okey AB, et al. Differential expression profiling of the hepatic proteome in a rat model of dioxin resistance: Correlation with genomic and transcriptomic analyses. Mol. Cell. Proteomics (2006) 5:882–894.
Peters AK, van Londen K, Bergman A, Bohonowych J, Denison MS, Van den BM, Sanderson JT. Effects of polybrominated diphenyl ethers on basal and TCDD-induced ethoxyresorufin activity and cytochrome P450-1A1 expression in MCF-7, HepG2, and H4IIE cells. Toxicol. Sci. (2004) 82:488–496.
Pinheiro JC, Bates DM. Mixed-effects models in S and S-Plus. In: Series: Statistics and Computing—Chambers J, Eddy W, Härdle W, Sheather S, Tierney L, eds. (2004) 1st ed. New York: Springer. 1–528.
Puga A, Barnes SJ, Chang C, Zhu H, Nephew KP, Khan SA, Shertzer HG. Activation of transcription factors activator protein-1 and nuclear factor-kappaB by 2,3,7,8-tetrachlorodibenzo-p-dioxin. Biochem. Pharmacol. (2000) 59:997–1005.[CrossRef][Web of Science][Medline]
R Development Core Team. R: A Language and Environment for Statistical Computing (2006) R Foundation for Statistical Computing, Vienna, Austria. Available at: http://www.R-project.org.
Safe S. Limitations of the toxic equivalency factor approach for risk assessment of TCDD and related compounds. Teratog. Carcinog. Mutagen. (1997) 17:285–304.[CrossRef][Web of Science][Medline]
Schuetz EG, Brimer C, Schuetz JD. Environmental xenobiotics and the antihormones cyproterone acetate and spironolactone use the nuclear hormone pregnenolone X receptor to activate the CYP3A23 hormone response element. Mol. Pharmacol. (1998) 54:1113–1117.
Silkworth JB, Koganti A, Illouz K, Possolo A, Zhao M, Hamilton SB. Comparison of TCDD and PCB CYP1A induction sensitivities in fresh hepatocytes from human donors, Sprague-Dawley rats, and rhesus monkeys and HepG2 cells. Toxicol. Sci. (2005) 87:508–519.
Slatter JG, Cheng O, Cornwell PD, de Souza A, Rockett J, Rushmore T, Hartley D, Evers R, He Y, Dai X, et al. Microarray-based compendium of hepatic gene expression profiles for prototypical ADME gene-inducing compounds in rats and mice in vivo. Xenobiotica (2006) 36:902–937.[CrossRef][Web of Science][Medline]
Smyth GK. Limma: Linear models for microarray data. In: Bioinformatics and Computational Biology Solutions using R and Bioconductor—Gentleman R, Casey V, Dudoit S, Irizarry R, Huber W, eds. (2005) New York: Springer. 397–420.
Stahlberg N, Merino R, Hernandez LH, Fernandez-Perez L, Sandelin A, Engstrom P, Tollet-Egnell P, Lenhard B, Flores-Morales A. Exploring hepatic hormone actions using a compilation of gene expression profiles. BMC Physiol. (2005) 5:8.[CrossRef][Medline]
Starr TB, Greenlee WF, Neal RA, Poland A, Sutter TR. The trouble with TEFs. Environ. Health Perspect. (2001) 107:A492–A493.[CrossRef]
Storey JD. The positive false discovery rate: A Bayesian interpretation and the q-value. Ann. Stat. (2003) 31:2013–2035.[CrossRef]
Suh J, Kang JS, Yang KH, Kaminski NE. Antagonism of aryl hydrocarbon receptor-dependent induction of CYP1A1 and inhibition of IgM expression by di-ortho-substituted polychlorinated biphenyls. Toxicol. Appl. Pharmacol. (2003) 187:11–21.[CrossRef][Web of Science][Medline]
Tabb MM, Kholodovych V, Grun F, Zhou C, Welsh WJ, Blumberg B. Highly chlorinated PCBs inhibit the human xenobiotic response mediated by the steroid and xenobiotic receptor (SXR). Environ. Health Perspect. (2004) 112:163–169.[Web of Science][Medline]
Thackaberry EA, Jiang Z, Johnson CD, Ramos KS, Walker MK. Toxicogenomic profile of 2,3,7,8-tetrachlorodibenzo-p-dioxin in the murine fetal heart: Modulation of cell cycle and extracellular matrix genes. Toxicol. Sci. (2005) 88:231–241.
Tijet N, Boutros PC, Moffat ID, Okey AB, Tuomisto J, Pohjanvirta R. Aryl hydrocarbon receptor regulates distinct dioxin-dependent and dioxin-independent gene batteries. Mol. Pharmacol. (2006) 69:140–153.
Vamvakas A, Keller J, Dufresne M. In vitro induction of CYP 1A1-associated activities in human and rodent cell lines by commercial and tissue-extracted halogenated aromatic hydrocarbons. Environ. Toxicol. Chem. (1996) 15:814–823.[CrossRef][Web of Science]
Van den Berg M, Birnbaum L, Bosveld AT, Brunstrom B, Cook P, Feeley M, Giesy JP, Hanberg A, Hasegawa R, Kennedy SW, et al. Toxic equivalency factors (TEFs) for PCBs, PCDDs, PCDFs for humans and wildlife. Environ. Health Perspect. (1998) 106:775–792.[Web of Science][Medline]
Van den Berg M, Birnbaum LS, Denison M, De Vito M, Farland W, Feeley M, Fiedler H, Hakansson H, Hanberg A, Haws L, et al. The 2005 World Health Organization re-evaluation of human and mammalian toxic equivalency factors for dioxins and dioxin-like compounds. Toxicol. Sci. (2006) 93:223–241.
van den Huevel ER, Geeven G, Bauerschmidt S, Polman JEM. Evaluation of an Affymetrix high-density oligonucleotide microarray platform as a measurement system. Qual. Reliability Eng. Int. (2005) 21:491–508.[CrossRef]
van der Laan MJ, Pollard KS. A new algorithm for hybrid hierarchical clustering with visualization and the bootstrap. J. Stat. Plann. Inf. (2003) 117:275–303.[CrossRef]
Venables WN, Ripley BD. Modern Applied Statistics with S (2002) LLC, New York: Springer Science & Business Media.
Vezina CM, Walker NJ, Olson JR. Subchronic exposure to TCDD, PeCDF, PCB126, and PCB153: Effect on hepatic gene expression. Environ. Health Perspect. (2004) 112:1636–1644.[Web of Science][Medline]
Walker NJ, Gastel JA, Costa LT, Clark GC, Lucier GW, Sutter TR. Rat CYP1B1: An adrenal cytochrome P450 that exhibits sex- dependent expression in livers and kidneys of TCDD-treated animals. Carcinogenesis (1995) 16:1319–1328.
Walker NJ, Wyde ME, Fischer LJ, Nyska A, Bucher JR. Comparison of chronic toxicity and carcinogenicity of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) in 2-year bioassays in female Sprague-Dawley rats. Mol. Nutr. Food Res. (2006) 50:934–944.[CrossRef][Web of Science][Medline]
Waxman DJ, O'Connor C. Growth hormone regulation of sex-dependent liver gene expression. Mol. Endocrinol. (2006) 20:2613–2629.
Wiebel FJ, Wegenke M, Kiefer F. Bioassay for determining 2,3,7,8-tetrachlorodibenzo-p-dioxin equivalents (TEs) in human hepatoma HepG2 cells. Toxicol. Lett. (1996) 88:335–338.[CrossRef][Web of Science][Medline]
Wu Z, Irizarry RA, Gentleman R, Martinez-Murillo F, Spencer F. A model-based background adjustment for oligonucleotide expression arrays. J. Am. Stat. Assoc. (2004) 99:909–917.[CrossRef][Web of Science]
Wyde ME, Cambre T, Lebetkin M, Eldridge SR, Walker NJ. Promotion of altered hepatic foci by 2,3,7,8-tetrachlorodibenzo-p-dioxin and 17beta-estradiol in male Sprague-Dawley rats. Toxicol. Sci. (2002) 68:295–303.
Wyde ME, Eldridge SR, Lucier GW, Walker NJ. Regulation of 2,3,7,8-tetrachlorodibenzo-p-dioxin-induced tumor promotion by 17 beta-estradiol in female Sprague–Dawley rats. Toxicol. Appl. Pharmacol. (2001) 173:7–17.[CrossRef][Web of Science][Medline]
Xu L, Li AP, Kaminski DL, Ruh MF. 2,3,7,8 tetrachlorodibenzo-p-dioxin induction of cytochrome P4501A in cultured rat and human hepatocytes. Chem. Biol. Interact. (2000) 124:173–189.[CrossRef][Web of Science][Medline]
Yager JD, Liehr JG. Molecular mechanisms of estrogen carcinogenesis. Annu. Rev. Pharmacol. Toxicol. (1996) 36:203–232.[CrossRef][Web of Science][Medline]
Yang Y, Abel SJ, Ciurlionis R, Waring JF. Development of a toxicogenomics in vitro assay for the efficient characterization of compounds. Pharmacogenomics (2006) 7:177–186.[CrossRef][Web of Science][Medline]
Zakharkin SO, Kim K, Mehta T, Chen L, Barnes S, Scheirer KE, Parrish RS, Allison DB, Page GP. Sources of variation in Affymetrix microarray experiments. BMC Bioinformatics (2005) 6:214.[CrossRef][Medline]
Zeiger M, Haag R, Hockel J, Schrenk D, Schmitz H-J. Inducing effects of dioxin-like polychlorinated biphenyls on CYP1A in the human hepatoblastoma cell line HepG2, the rat hepatoma cell line H4IIE, and rat primary hepatocytes: Comparison of relative potencies. Toxicol. Sci. (2001) 63:65–73.
CiteULike 
Connotea 
Del.icio.us What's this?
This article has been cited by other articles:
|
|
 |
|
  E. A. Carlson, C. McCulloch, A. Koganti, S. B. Goodwin, T. R. Sutter, and J. B. Silkworth Divergent Transcriptomic Responses to Aryl Hydrocarbon Receptor Agonists between Rat and Human Primary Hepatocytes Toxicol. Sci., November 1, 2009; 112(1): 257 - 272. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 M. S. Yates, Q. T. Tran, P. M. Dolan, W. O. Osburn, S. Shin, C. C. McCulloch, J. B. Silkworth, K. Taguchi, M. Yamamoto, C. R. Williams, et al. Genetic versus chemoprotective activation of Nrf2 signaling: overlapping yet distinct gene expression profiles between Keap1 knockout and triterpenoid-treated mice Carcinogenesis, June 1, 2009; 30(6): 1024 - 1031. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||