Assessment of genotoxic potential of Cr(VI) in the mouse duodenum: An in
silico comparison with mutagenic and nonmutagenic carcinogens across tissues
Chad M. Thompson
a,⇑, J. Gregory Hixon
b, Deborah M. Proctor
c, Laurie C. Haws
b, Mina Suh
c,
Jonathan D. Urban
b, Mark A. Harris
a aToxStrategies, Inc., Katy, TX 77494, USA
bToxStrategies, Inc., Austin, TX 78759, USA c
ToxStrategies, Inc., Orange County, CA 92688, USA
a r t i c l e
i n f o
Article history:
Received 6 January 2012 Available online 15 June 2012
Keywords:
Hexavalent chromium (Cr(VI)) Toxicogenomics
Mutagens Nonmutagens
Principal components analysis (PCA) Logistic regression classification
a b s t r a c t
In vitrostudies on hexavalent chromium [Cr(VI)] indicate that reduced forms of this metal can interact with DNA and cause mutations. Recently, Cr(VI) was shown to induce intestinal tumors in mice; however, Cr(VI) elicited redox changes, cytotoxicity and hyperplasia – suggesting involvement of tissue injury rather than direct mutagenesis. Moreover, toxicogenomic analyses indicated limited evidence for DNA damage responses. Herein, we extend these toxicogenomic analyses by comparing the gene expression patterns elicited by Cr(VI) with those of four mutagenic and four nonmutagenic carcinogens. To date, tox-icogenomic profiles for mutagenic and nonmutagenic duodenal carcinogens do not exist, thus duodenal gene changes in mice were compared to those elicited by hepatocarcinogens. Specifically, duodenal gene changes in mice following exposure to Cr(VI) in drinking water were compared to hepatic gene changes previously identified as potentially discriminating mutagenic and nonmutagenic hepatocarcinogens. Using multivariate statistical analyses (including logistic regression classification), the Cr(VI) gene responses clustered apart from mutagenic carcinogens and closely with nonmutagenic carcinogens. These findings are consistent with other intestinal data supporting a nonmutagenic mode of action (MOA). These findings may be useful as part of a full weight of evidence MOA evaluation for Cr(VI)-induced intestinal carcinogenesis. Limitations to this analysis will also be discussed.
Ó2012 Elsevier Inc.
1. Introduction
Inhalation of hexavalent chromium [Cr(VI)] has long been rec-ognized to pose a carcinogenic risk to the lung (IARC, 1990). Oral exposure to Cr(VI) at environmentally relevant levels has been widely considered not to pose a cancer risk due to reduction of Cr(VI) to poorly absorbed Cr(III) by bodily fluids and cellular con-stituents (De Flora et al., 1997; Proctor et al., 2002; US EPA, 1991). However, chronic exposure to very high levels of Cr(VI) in drinking water resulted in intestinal neoplasms in mice in a recent 2-year bioassay (NTP, 2008). Notably, the carcinogenic Cr(VI) concentrations in the drinking water were bright yellow and were associated with reduced water intake due to unpalatability (NTP, 2008; Thompson et al., 2011b). Considering that the intestinal carcinogenesis in mice occurred at unusually high Cr(VI)
concentrations, it is critical to understand the mode of action (MOA) of the intestinal tumors in mice because it informs the rel-evance of these tumors to humans as well as the low-dose extrap-olation methods employed for the derivation of Cr(VI) toxicity criteria in various media (e.g. drinking water). To this end, a com-prehensive research program was conducted to gather critical data needed to inform the MOA underlying Cr(VI)-induced intestinal carcinogenesis (Kopec et al., 2012a; Thompson et al., 2011a,b).
An important consideration in these studies is whether Cr(VI) acts via a mutagenic or nonmutagenic MOA in the small intestine. Mutagens interact directly with DNA and are generally thought to exhibit a non-thresholded dose–response,1whereas nonmutagenic
(i.e. indirect) genotoxic carcinogens and nongenotoxic carcinogens exhibit thresholded behavior (Bolt et al., 2004; Eastmond, 2008). As part of our research into the MOA of Cr(VI)-induced intestinal car-cinogenesis, in vivo micronucleus formation and k-ras mutations
0273-2300Ó2012 Elsevier Inc.
http://dx.doi.org/10.1016/j.yrtph.2012.05.019
⇑Corresponding author. Address: ToxStrategies, Inc., 23501 Cinco Ranch Blvd., Suite G265, Katy, TX, USA. Fax: +1 832 218 2756.
E-mail addresses: cthompson@toxstrategies.com (C.M. Thompson), ghixon@
toxstrategies.com(J. Gregory Hixon),dproctor@toxstrategies.com(D.M. Proctor),
lhaws@toxstrategies.com(L.C. Haws),msuh@toxstrategies.com(M. Suh),jurban@ toxstrategies.com(J.D. Urban),mharris@toxstrategies.com(M.A. Harris).
1
However, according toUS EPA (2005a), ‘‘Special attention is important when the data support a nonlinear mode of action but there is also a suggestion of mutagenicity. Depending on the strength of the suggestion of mutagenicity, the assessment may justify a conclusion that mutagenicity is not operative at low doses and focus on a nonlinear approach. . .’’.
Contents lists available atSciVerse ScienceDirect
Regulatory Toxicology and Pharmacology
j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / y r t p hOpen access under CC BY-NC-ND license.
were assessed in duodenal tissue sections of mice exposed to Cr(VI) up to 90 days, and were found to be negative (Harris et al., 2012; O’Brien et al., in preparation). Toxicogenomic evaluation of re-sponses to Cr(VI) in the mouse small intestine indicated activation of Nrf2 signaling at relatively low exposure concentrations (Kopec et al., 2012a), consistent with clear alteration in cellular redox status in similarly treated mice (Thompson et al., 2011b). Several genes in-volved in DNA repair were elevated by day 8 of exposure to carcin-ogenic concentrations of Cr(VI), and functional analysis indicated enrichment of DNA repair pathways at the highest Cr(VI) concentra-tions (Kopec et al., 2012a). Beyond transcriptional and functional analyses, it is of interest to scientists and risk assessors to examine whether the genomic signature/profile of Cr(VI) is similar to that of known mutagens. Due in part to the low incidence of cancer in the small intestine (Greaves, 2007), there are insufficient toxicoge-nomic profile data from the rodent small intestine with which to compare the duodenal toxicogenomic data of mice treated with Cr(VI). There are, however, toxicogenomic comparisons for muta-genic and nonmutamuta-genic carcinogens in rodent liver.
Ellinger-Ziegelbauer et al. (2005)exposed rats to carcinogenic concentrations of eight hepatocarcinogens (four mutagenic and four nonmutagenic) for up to 2 weeks, and identified a subset of genes that were useful for distinguishing between mutagenic and nonmutagenic hepatic carcinogens. In the absence of compa-rable intestinal data, we compared the differential expression of genes in the mouse duodenum following Cr(VI) treatment with the differential expression reported by Ellinger-Ziegelbauer and colleagues. To facilitate the comparison of the differential expres-sion of genes across nine chemicals, we used data reduction tech-niques (e.g. principal components analysis, PCA) and multivariate analyses to make unbiased evaluation of whether Cr(VI) was sim-ilar to the mutagenic or nonmutagenic carcinogens. The results indicate that the expression pattern induced by Cr(VI) more clo-sely follows that latter. These findings, notwithstanding the limi-tations discussed herein, may be useful as part of a weight of evidence to evaluate the MOA for Cr(VI)-induced intestinal carcinogenesis.
2. Material and methods
2.1. Animal treatments and tissue preparation
Test substance, animal husbandry, and study design have been described in detail elsewhere (Thompson et al., 2011b). Briefly, fe-male B6C3F1 mice (Charles Rivers Laboratories International, Inc.) were providedad libitumaccess to Cr(VI), as sodium dichromate dihydrate (SDD), in drinking water at concentrations ranging from 0.3–520 mg/L. After 7 and 90 days of exposure (referred to herein respectively as day 8 and 91), animals were euthanized using CO2. For toxicogenomic analyses, duodenal samples were scraped
and processed as described previously (Kopec et al., 2012a).
2.2. Microarray analysis of Cr(VI) data
Details on mouse 4x44 K Agilent whole-genome oligonucleo-tide microarrays and data analysis for SDD-elicited duodenal gene expression at day 8 and 91 are described inKopec et al. (2012a). In brief, total RNA was isolated according to the manufacturer’s pro-tocol with an additional acid phenol:chloroform extraction, resus-pended in RNA storage solution (Ambion Inc., Austin, TX), quantified (A260), and quality was assessed by evaluation of the A260/A280 ratio and by visual inspection of 1
l
g total RNA on a denaturing gel. Dose-dependent changes in gene expression were examined using mouse 444 K Agilent whole-genome oligonu-cleotide microarrays (Agilent Technologies, Inc., Santa Clara, CA).All experiments were performed with three biological replicates and independent labeling of each sample (Cy3 and Cy5, and dye swap) for every dose group at each time point. Microarray data were normalized using a semi-parametric approach (Eckel et al., 2004, 2005).
2.3. Gene expression data selection for comparisons
These Cr(VI) gene expression data were compared to previously published gene expression data for four genotoxic and four non-genotoxic hepatic carcinogens (Ellinger-Ziegelbauer et al., 2005). The genotoxic carcinogens were 2-nitrofluorene (2-NF), dimethyl-nitrosamine (DMN), 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanone (NNK), and aflatoxin B1 (AB1); the nongenotoxic carcinogens were methapyrilene (MPy), diethylstilbestrol (DES), Wy-14643 (WY), and piperonylbutoxide (PBO). Importantly, the four ‘‘genotoxic’’ hepatic carcinogens were characterized as induc-ing DNA modification and causinduc-ing mutations and physical distor-tion of DNA (Ellinger-Ziegelbauer et al., 2004, 2005), and have been characterized as mutagens by the U.S. EPA (US EPA, 2005b; US EPA, 2007). Thus, we will use the terminology mutagenic/non-mutagenic as opposed to the genotoxic/nongenotoxic terminology used by Ellinger-Ziegelbauer and colleagues. Moreover, the term genotoxic is somewhat ambiguous. For example, WY can induce oxidative DNA damage (i.e. genotoxicity), yet many scientists con-sider carcinogens that work primarily through oxidative stress and oxidative damage as nongenotoxic (Ellinger-Ziegelbauer et al., 2005; Klaunig et al., 1998).
Treatment of rats with these eight carcinogens resulted in sig-nificant expression of 651 probe sets, corresponding to 477 non-redundant genes, as measured by Affymetrix RG U34A arrays, which were further grouped into 23 toxicological categories ( Ellin-ger-Ziegelbauer et al., 2005). Seven categories (comprised of 139 genes) were discussed in greater detail by Ellinger-Ziegelbauer and colleagues:oxidative stress/DNA response(13),oxidative stress/ protein damage response(25), oxidative stress response(13),stress response (9), regeneration (34), cell survival and/or proliferation
(25), and cell cycle progression (20). Several (but not all) of the genes in these categories were differentially expression by the two classes of carcinogens. Gene expression data for all 477 genes were obtained from the Supplementary Material in Ellinger-Zieg-elbauer et al. (2005). Data were available for 1, 3, 7 and 14 days of exposure; however, only the 7 day exposure data were averaged (3 replicates for DMN, NNK, AB1, MPy, DES, Wy, PBO and 2 repli-cates for 2-NF) in order to obtain a single day 8 expression value for each of the 139 genes for each of the 8 carcinogens.
The treatment doses employed in Ellinger-Ziegelbauer et al. (2005) were concentrations known to be carcinogenic to rats in longer-term bioassays; moreover, the doses also caused histopa-thological changes in the course of their short-term study. For con-sistency, we therefore selected the 520 mg/L SDD (182 mg/L Cr(VI)) treatment group because it is carcinogenic in a 2-year bioassay (NTP, 2008), and elicited histopathological lesions in the mouse duodenum at day 8 (Thompson et al., 2011b).
To allow for direct comparison between our data and those of
Ellinger-Ziegelbauer et al. (2005), 651 significant Affymetrix rat probes from Ellinger-Ziegelbauer et al. were converted to unique HomoloGeneIDs using Database for Annotation, Visualization, and Integrated Discovery and mapped to the mouse whole-genome 444 K Agilent array. Of the 477 unique genes, orthologous mapping identified 395 (82%) mouse genes for which Cr(VI)-elic-ited gene expression changes at 520 mg/L SDD were available. Of the 139 unique genes in the 7 aforementioned categories, orthol-ogous mapping identified 116 (83%) mouse genes for which Cr(VI)-elicited gene expression changes at 520 mg/L SDD were available.
2.4. Toxicogenomic data analysis
Expression data from Supplemental material in Ellinger-Ziegelbauer and the Cr(VI) study were uploaded for transcription factor analysis using Ingenuity Pathway Analysis (IPA) with filter-ing criteria of 1.5-fold change in expression. Hierarchical clusterfilter-ing was performed using MultiExperiment Viewer (MeV v. 4.6.0) of the TM4 microarray software suite (Saeed et al., 2003).
2.5. Data reduction via principal components analysis (PCA)
The objective of this analysis is to use differential gene expres-sion data from 116 orthologs to investigate the relationship (if any) between those changes and genotoxicity for eight carcinogens with MOAs characterized as mutagenic or nonmutagenic, and then use that relationship as the basis for classifying a chemical for which the genotoxicity is uncertain, Cr(VI). To reduce the dimension of the data from the 116 inputs to a more manageable number, unro-tated orthogonal PCA was employed – which is a standard tech-nique for data reduction (Hair et al., 1998; Tabachnick and Fidell, 2007). The gene expression data, provided as ratios, were normal-ized by log10transformation, which is the natural transform for
ra-tio data (Keene, 1995). The Pearson correlation matrix amongst the transformed variables was used as the input to the PCA. Horn’s par-allel analysis as modified by Glorfeld (Glorfeld, 1995; Horn, 1965) – a simulation-based test in which eigenvalues for the first, second, third, etc. components from a series of simulated random datasets of the same size and number of variables as the actual data are compared with the eigenvalues from the actual data – was used to determine the number of components to retain. Only those com-ponents from the actual data whose eigenvalues exceed the 95th percentile of the values for the corresponding component from the simulations are retained. The PCA allows the bulk of the infor-mation from the entire set of 116 genes to be captured by a much smaller and more manageable number of retained components, which are themselves linear combinations of the original 116 gene expression changes. These retained component scores served as the inputs to the logistic regression and cluster analyses described below. The PCA and the subsequent analyses using the retained component scores were conducted in R (R Development Core Team).
2.6. Logistic regression classification analysis
The retained component scores of the four mutagenic and four nonmutagenic carcinogens were used as inputs to a logistic regres-sion classification analysis. A logistic regresregres-sion analysis was used because the research question involves a dichotomous outcome (mutagenic or nonmutagenic) and the logistic function is a bounded function ideally suited for problems involving dichoto-mous outcomes (Montgomery et al., 2006). The logistic regression analysis maps the retained component scores to the genotoxicity outcome (mutagenic vs. nonmutagenic) for the 8 carcinogens. The parameters from this analysis were then used to assess the re-tained component scores for Cr(VI) to determine its most probable classification.2
2.7. Cluster analysis
To further explore the question of whether the gene changes associated with exposures to Cr(VI) in drinking water more closely
resembled those associated with known mutagenic or known non-mutagenic carcinogens, a cluster analysis was performed using the retained component scores for the four mutagenic and four non-mutagenic carcinogens. The general method involved maximum likelihood estimation via expectation maximization, with the qual-ity of the candidate cluster solutions indexed by the Bayesian Information Criterion (BIC).3Models where the clusters could have
a variable size, variable shape (within the ellipse family), and vari-able orientation were considered. With only four mutagenic and four nonmutagenic carcinogens, the cluster analysis was limited to deter-mining if the data comprised one cluster or two, as there were insuf-ficient data to consider models with greater numbers of clusters. This was not a limitation because we were only interested in two outcomes (mutagenic and nonmutagenic). The cluster analysis was conducted using the mclust package for R (Fraley and Raftery, 2002, 2006). As with the logistic regression classifier analysis, after assessing the four mutagenic and four nonmutagenic carcinogens, the cluster analysis solution was then used to classify Cr(VI).
3. Results
3.1. Comparison of gene expression
Following treatment of rats with eight carcinogens, Ellinger-Ziegelbauer et al. (2005)identified 477 non-redundant genes that were significantly altered by at least one carcinogenP1.7-fold. Of these 477 genes, expression values for 139 genes comprising seven toxicological categories were provided inTable 2of Ellin-ger-Ziegelbauer et al. (2005). Two categories,Oxidative stress/DNA damage responseand Cell cycle progressionwere especially noted as differentially affected by the two classes of carcinogens.Table 1
lists the expression values for genes in these two categories,viz. the maximal change in gene expression after either 1, 3, 7, or 14 days of exposure. Broadly, the genes listed forCell cycle progres-sionas well asApex1were up-regulated by the nonmutagenic car-cinogens, whereas the genes listed in their Oxidative stress/DNA damage responsecategory were up-regulated by mutagenic carcin-ogens (Ellinger-Ziegelbauer et al., 2009, 2005). Listed in the center column ofTable 1are the fold-change values after 7 days of expo-sure to 520 mg/L SDD (182 mg/L Cr(VI)). Generally, the magnitude and direction of change for each gene following Cr(VI) exposure is more similar to nonmutagenic compounds.
To further compare the Cr(VI) gene changes with those of the eight carcinogens inEllinger-Ziegelbauer et al. (2005), we obtained the expression values for the 7-day treatment groups from the Supplemental material in Ellinger-Ziegelbauer et al. (2005). Of the 477 non-redundant rat genes, 395 (82%) orthologous mouse genes were identified (see Section2.3). Expression data for these 395 orthologs were uploaded into IPA and subject to transcription factor analysis. As shown inTable 2, p53 activity was predicted to be activated in three of the four mutagenic carcinogens, but only one of four nonmutagenic carcinogens. Notably, p53 was not pre-dicted to be activated using this set of 395 orthologs (Table 2) or when using the entire day 8 520 mg/L SDD microarray dataset (data not shown).
During IPA analysis, we noted that Mdm2 was increased
P1.5-fold by all mutagenic carcinogens but was unchanged by the nonmutagenic carcinogens at day 8. Other notable gene changes fromTable 1areCdkn1a,Ccng1andMgmt, which are all regulated by p53 and purported to discriminate between muta-genic and nonmutamuta-genic carcinogens (Boverhof and Gollapudi, 2
For a general discussion of the use of a logistic-based classifier for this purpose, seeHastie et al. (2001). For a discussion of the mathematical superiority of a logistic-based classifier relative to another classical alternative, discriminant function analysis, seePress and Wilson (1978).
3The BIC is a commonly used index of model quality that optimizes the trade-off
between complexity and fit – in essence finding the most parsimonious model that fits the data well.
2010; Ellinger-Ziegelbauer et al., 2009, 2005; Waters et al., 2010). The fold-change expression levels of these four genes in the mouse duodenum after 7 and 90 days of exposure to Cr(VI) indicate that onlyCcng1was elevatedP1.5-fold, and only at day 8 (Fig. 1A– B). In comparison to the eight carcinogens from Ellinger-Zieg-elbauer et al. (2005), the gene changes for Cr(VI) were on par with those of the four nonmutagenic carcinogens (Fig. 1C).
We expanded the gene comparisons for the nine carcinogens to the 139 genes discussed inEllinger-Ziegelbauer et al. (2005). Of the 139 rat genes, 116 (83%) orthologous mouse genes were identified, and subjected to hierarchical clustering in order to better visualize the expression data, as well as address the question of whether the Cr(VI)-induced gene changes were more similar to mutagenic or nonmutagenic carcinogens. The overall gene expression pattern in-duced by Cr(VI) was more similar to the nonmutagenic carcino-gens, however, Cr(VI) also clustered separately from the four
nonmutagenic compounds as indicated by the dendrogram (Fig. 2). The most noticeable similarities with the nonmutagenic compounds can be seen in the categories ofOxidative Stress/Protein Damage ResponseandRegeneration, whereas the most noticeable dissimilarities with the mutagenic compounds can be seen in the category ofOxidative Stress/DNA Response(Fig. 2).
3.2. Principal components analysis
As explained in the Material and methods, the gene expression data following seven days of exposure to the eight carcinogens in
Ellinger-Ziegelbauer et al. (2005)and for Cr(VI) were analyzed by PCA. Horn’s parallel analysis (Glorfeld, 1995; Horn, 1965) of the PCA results showed that only the first three components were sig-nificant (i.e. they had eigenvalues exceeding 95% of the eigenvalues for the corresponding component in equivalently sized and
Table 2
Transcription factor analysis.a
2-NF DMN NNK AB1 SDDb
Mpy DES WY PBO
TP53 STAT3 TP53 RELA MYC MYCN MYCN PPARA NFE2L2
MYCN EZH2 MYCN STAT3 STAT5B MYC RELA PPARG PPARA
CDKN2A ;PPARD MED1 TP53 FOXM1 PPARA YY1 RXRA ;NFYA
RELA ;CEBPD RELA ;CEBPB MYCN ;SREBF1 NFE2L2 MYCN ;MYOD1
EZH2 TP73 GFI1 ;SMARCB1 TP53 ESRRA ;SMAD3
TP73 ;FOXO3 KEAP1 ;MYOD1 NFATC2 TFAM ;CREBBP
TRIM24 ;HNF4A ATF4 PPARGC1B ;EP300
;SMARCA4 ;MLXIPL ;PPARGC1B TRIM24
;PPARD ;TCF3 ;MLX MYC
;TCF3 ;CREBBP ;MLXIPL HSF1
;PXR ;RXRA Estrogen receptor
;CEBPA ;SREBF1 ;SPI1
;PML ;SREBF2 ;TP53 ;IRF7 ;CREBBP ;Rb ;STAT1 ;SMARCB1 ;IRF7 ;MED1 ;CTNNB1 ;PPARA ;NR1I3 ;PPARGC1A ;HNF4A PXR = PXR ligand-PXR-Retinoic acid-RXRa.
a Based on 395 orthologs, withP1.5-fold filter, day 8 data only. b 520 mg/L SDD.
Table 1
Select Comparisons of Genes Involved in DNA Damage Response and Cell Cycle Progression. Gene symbol Mutagensa
Nonmutagensa
2-NF DMN NNK ABI SDDb
MPy DES WY PBO
Cell cycle progression
Ccnb1 2.0 2.2 1.1 1.2 3.0 4.8 2.1 5.3 2.7 Top2a 1.8 2.2 1.3 1.1 3.4 2.4 1.8 3.9 3.9 Mcm6 1.7 1.7 1.1 1.2 1.4 1.8 2.2 3.4 7.8 Pcna 1.7 1.6 1.1 1.4 2.2 1.9 1.4 2.4 3.4 Cdc20 1.4 1.0 1.1 1.3 5.8 3.5 2.1 6.4 2.9 Pttg1 1.1 1.1 1.2 1.1 1.5 3.1 1.9 3.8 1.5 Tubb5 1.3 1.7 1.1 1.5 4.6 2.4 2.1 3.0 3.0
Oxidative stress/DNA damage
Mdm2 5.7 3.3 5.5 8.8 1.3 2.0 1.4 1.3 2.5 Mgmt 3.2 2.7 2.7 4.9 1.2 1.5 1.3 1.4 1.1 Cdkn1a 17.0 23.1 14.5 13.0 1.1 1.4 1.5 1.1 1.3 Ccng1 11.2 6.1 7.8 12.8 1.9 2.1 1.9 2.1 1.4 Apex1 2.6 1.2 1.2 1.1 4.9 2.1 2.3 2.3 2.0 a
Gene list and fold-change expression values were obtained fromFig. 2andTable 2ofEllinger-Zigelbauer et al. (2005). Mutagenic and nonmutagenic carcinogens were administered at known carcinogenic doses (based on longer-term studies) daily for up to 14 days. Italicized values represent maximum change from either 1 and/or 3 days of exposure, whereas non-italicized values represent a maximum change after 7 or 14 days of exposure as reported inTable 2ofEllinger-Ziegelbauer et al. (2005). All nonmutagenic compounds elicited weak to moderate hyperplasia, whereas the mutagenic compounds produced variably hypertrophy, necrosis, and mitosis.
b
Values for Cr(VI) represent changes in the duodenum after 7 days of exposure to 520 mg/L SDD (182 mg/L Cr(VI)), which induced histopathological lesions such as hyperplasia, atrophy, and cytoplasmic vacuolization (Thompson et al., 2011b).
structured random datasets). Thus, the set of 116 ortholog expres-sion scores was reduced to three component scores. Those three components respectively accounted for 33.86%, 29.02%, 19.44% of the overall variance and had standard deviations of 1.52, 1.41, and 1.15. The fourth component had a standard deviation below
1 (0.68), which is another indication (beyond the Horn’s parallel analysis) that the fourth and all subsequent components were
Fig. 1.Fold change of select genes involved in DNA damage response. Fold change expression ofMgmt, Cdkn1a, Ccng1,andMdm2after 7 (A) and 90 (B) days of exposure to varying concentrations of Cr(VI) in drinking water (Kopec et al., 2012a). (C) Expression of these four genes after 7 days of exposure to mutagenic (DMN, 2-NF, NNK, AB1) and nonmutagenic (WY, PBO, MPy, DES) carcinogens ( Ellinger-Ziegelbauer et al., 2005). Also shown are expression values for Cr(VI) in the 520 mg/ L treatment groups after 7 and 90 days of exposure. Note: dotted line indicates 1.5-fold increase in expression.
Cell cycle
progression
Cell survival/
proliferation
Oxidative stress/
protein damage
response
Oxidative stress/
DNA response
Stress response
Regeneration
Oxidative stress
response
Ddpy d Xdh Cc nd1 Cdk 16 Cdk 4 Cdk 1 Cc nb1 Pttg1 Cdc 20 Hmgb2 Top2a Pc na Mc m6 Stmn1 Tuba1c Tubb5 Nap1l1 H2afz H2afx Cs tb Rb1 Igfbp1 Igfbp2 Gas 6 Cx c l10 Vegfa Gdf15 Polr2i Ts c 22d1 Ddah1 Nupr1 Nrg1 Hdc Serpinb9 Fam162a Ty ms Lgals 3 Igfbp3 Dus p6 Timp1 Bax Cdk n1a Cgrrf1 Btg2 Apex 1 Mgmt Fn1 Gas s 45a Mdm2 Cc ng1 Cgref1 Zmat3 Tmem47 Hs pb1 Hs pd1 Cc t3 Hs pe1 Hs ph1 Hs pa9 Grpel1 Ide Cts d Ps ma1 Ps ma2 Ps ma4 Ps ma5 Ps mb1 Ps mb3 Ps mb4 Ps mc 1 Ps mc 2 Ps mc 5 Ps md1 Ps mc 4 Ps mb5 Ps ma7 Ac o1 Sepw1 Hs f1 Mt1 Mt2 Gs s Gc lc Nqo1 Hmox 1 Ephx 1 Atf3 Fx c 1 Tat Rps 27 Rps 3 Rps 5 Odc 1 Eif4ebp Ddx 21 Ps at1 Eef1g Fk bp4 Rps 19 Phgdh Rpl23 Npm1 Rps 4x Eef2 Rpl8 Rps 24 Rps 7 Rpl17 Rpl10a Rps 15a Klf6 Tmem120 Nfk bia My c Egr1 Rhob Fold change-2 1 2
Fig. 2.Hierarchical clustering. Hierarchical clustering of 116 orthologs in seven categories previously reported to be differentially expressed by mutagenic (DMN, 2-NF, NNK, AB1) and nonmutagenic (WY, PBO, MPy, DES) carcinogens. The dendrogram indicates that Cr(VI) (in the form of SDD) more closely clusters with the nonmutagenic carcinogens.
inconsequential. In aggregate, the three retained component scores represented over 80% of the variance contained in the entire set of 116 values.
3.3. Logistic regression classification analysis
Using the three component scores identified based on the PCA as independent variables and the known classification (i.e., muta-genic or nonmutamuta-genic) of the eight carcinogens from Ellinger-Ziegelbauer et al. (2005)as the dependent variable, the logistic regression analysis showed that the first component was signifi-cantly related to the classification,X^2(1) = 11.09, p< 0.001. The second and third components were not significant. The overall analysis provided an excellent fit to the data, as a comparison of the predicted values from the analysis to the actual classification of the eight carcinogens as either mutagenic or nonmutagenic re-vealed a perfect correspondence. When the parameters from this analysis were used to assess the retained component scores for Cr(VI) to determine its most probable classification, Cr(VI) was classified as a nonmutagenic carcinogen with a probability of nearly one (deviating in the fifteenth decimal place).
3.4. Cluster analysis
Because the logistic regression classification analysis indicated that the third principal component was not important in relation to the distinction between mutagenic and nonmutagenic carcino-gens, the cluster analysis focused only on the first two component scores. (The third component was ‘‘significant’’ as per Horn’s paral-lel analysis, but that significant variance is not related to the issue of classification as either mutagenic or nonmutagenic.) The cluster analysis showed that the BIC was maximized for a two cluster solu-tion, indicating that the first two component scores for the four mutagenic and four nonmutagenic carcinogens formed two dis-tinct clusters (Fig. 3). The inner ellipse is the standard variance el-lipse, the outer ellipse is the 95% confidence boundary, and the differing symbols depict the groupings assigned by the cluster analysis. It should be noted that the cluster analysis routine was not supplied with information as to the classification of each car-cinogen as either mutagenic or nonmutagenic, but only the first two component scores based on the PCA of the ortholog expression data. On that basis, the cluster analysis segregated the carcinogens into two visibly distinct groups. The group on the left is, in fact, the
four mutagenic carcinogens, and the group on the right is the four nonmutagenic carcinogens. With regard to Cr(VI), the cluster anal-ysis assigns a probability of nearly one (deviating in the ninth dec-imal place) that Cr(VI) belongs to the nonmutagenic group (Fig. 3).
3.5. Validation of logistic classifier and cluster analysis performance
To validate the overall performance of the logistic regression classifier and cluster analysis techniques used to classify Cr(VI), we first used these techniques to classify each of the eight known carcinogens. Specifically, we left out one of the carcinogens as a test case, used the remaining seven cases as inputs to the logistic regression and cluster analysis, and then used the results to classify the eighth case. We repeated this analysis leaving out each of the eight known cases in turn, and compared the classifications with that ascribed by Ellinger-Ziegelbauer et al. (2005). The logistic regression classifier correctly classified all eight of the left-out cases on the basis of the other seven, and the cluster analysis cor-rectly classified six of the eight left-out cases. The overall perfor-mance across these two techniques, 14 correct classifications out of 16, is significantly better than chance,X^2(1) = 9,p< 0 .01.
4. Discussion
Kopec et al. (2012a)recently summarized the effects of Cr(VI) on the duodenal transcriptome in mice following 7 or 90 days of exposure in drinking water. Therein, Cr(VI) was shown to increase the expression of several genes involved in base excision repair re-lated to oxidative DNA damage (Sedelnikova et al., 2010) including
Ape1,Parp1,Pcna,Fen1andLig1. In addition, functional enrichment analysis indicated enrichment of DNA repair pathways related to mismatch repair (520 mg/L SDD), BRCA1 signaling (520 mg/L SDD), and nucleotide excision repair (170 and 520 mg/L SDD) at day 8 but not day 91. To further investigate genotoxic responses at the transcript level, we compared the gene expression changes induced by Cr(VI) in the duodenum with gene changes induced by four mutagenic and four nonmutagenic hepatocarcinogens. Duodenal transcripts were compared to liver transcripts because the rarity of small intestinal cancer means transcript data for small intestinal carcinogens are not readily available. Therefore, gene expression patterns elicited by Cr(VI) in the duodenum were com-pared to genes differentially effected by mutagenic and nonmuta-genic hepatocarcinogens (Ellinger-Ziegelbauer et al., 2005).
Fig. 3.Cluster analysis. Cluster analysis of 116 orthologs results in the 4 mutagenic carcinogens (triangles) clustering together and apart from the 4 nonmutagenic carcinogens (squares). The inner ellipse is the standard variance ellipse, and the outer ellipse is the 95% confidence boundary. The square with crosshair represents Cr(VI).
The comparison of a select set of genes inTable 1andFig. 1 sug-gested that the transcript changes elicited by Cr(VI) in the mouse duodenum were not similar to the transcript changes induced by mutagenic carcinogens in the liver. Furthermore, transcription fac-tor analysis (Table 2) suggested that p53 signaling was activated by 3 of 4 mutagens and only 1 of 4 nonmutagens. As p53 signaling is well known to be activated by DNA damage, it is somewhat sur-prising that p53 was not activated by all 4 mutagens. Nevertheless, the observation that p53 signaling was not predicted to be activated by Cr(VI) at 520 mg/L SDD, suggests that Cr(VI) was not acting like a mutagenic carcinogen. The multivariate analytical approaches (Figs. 2 and 3) further suggest that the toxicogenomic expression profile in the duodenal epithelium following 7 days of exposure to Cr(VI) in drinking water more closely resembled the expression pro-files of nonmutagenic hepatic carcinogens. In total, these analyses lend support to the hypothesis that the carcinogenic MOA of Cr(VI) in the small intestine does not involve a mutagenic MOA.
It should be stressed that these analyses do not prove that Cr(VI) acts via a nonmutagenic MOA; however the results herein are consistent with published evidence that Cr(VI) induces redox changes at lower concentrations than those that caused cancer in a 2-year bioassay (Kopec et al., 2012a; Thompson et al., 2011b), as well as evidence that neitherk-rasmutations or crypt micronu-cleus formation could be detected in duodenal tissues of mice ex-posed to Cr(VI) for up to 90 days (Harris et al., 2012; O’Brien et al., in preparation). Although oxidative DNA damage can lead to muta-tion, this does not mean that compounds that induce oxidative stress (e.g. WY) have a mutagenic MOA (Ellinger-Ziegelbauer et al., 2005; Klaunig et al., 1998). It should also be recognized that the term nonmutagenic is not synonymous with nongenotoxic. As discussed in several recent reviews on chromium (Holmes et al., 2008; Nickens et al., 2010; Zhitkovich, 2011), Cr(VI) may work through epigenetic mechanisms such as altered DNA methylation. For example, lung biopsies from workers occupationally exposed to chromate exhibit fewer p53 point mutations than expected, but instead show signs of chromosomal instability including re-duced expression and hypermethylation of MutL homolog 1 (MLH1) (Hirose et al., 2002; Kondo et al., 1997). More recently, Cr(VI) has been shown to increase methylation of histone H3 lysine 9 (H3K9) and produced time- and dose-dependent decreases in MLH1in vitro(Sun et al., 2009).
According to NTP, only two other chemicals tested by the NTP program have caused increased incidences of neoplasms in the mouse intestine,viz. captan ando-nitrotoluene (NTP, 2008). The latter caused tumors of the large intestine (NTP, 2002), and mech-anistic studies suggest thato-nitrotoluene (2-nitrotoluene) carcin-ogenicity may involve oxidative DNA damage (Watanabe et al., 2010). Captan and the structurally similar folpet induce similar histopathological changes in the mouse duodenum as Cr(VI), are reactive toward thiols, and have been determined to have a non-mutagenic MOA (Cohen et al., 2010; US EPA, 2004). More recently, it was reported that indium phosphide caused small intestinal tu-mors in mice (Chandra et al., 2010). However, NTP characterized the increased incidences of neoplasms of the small intestines of male mice (from inhalationstudies) as marginal and ‘‘mayhave been related to exposure to indium phosphide’’ (emphases added) (NTP, 2001). Nevertheless, lung tumors induced by indium phos-phide are thought to involve oxidative stress (Gottschling et al., 2001; Upham and Wagner, 2001). Thus, the chemicals inducing intestinal tumors in mice investigated to date generally appear to have oxidant properties that may be responsible for tumorigenesis. Importantly, we have recently shown that Cr(VI) exposure signifi-cantly decreases the GSH/GSSG ratio in the duodenal epithelium of mice and induces Nrf2 signaling (Kopec et al., 2012a; Thompson et al., 2011b). Moreover, Nrf2 activation at day 91 was recently confirmed by IPA transcription factor analysis (Kopec et al., 2012b).
Despite the aforementioned consistency with other findings from our studies into the MOA for Cr(VI)-induced intestinal car-cinogenesis (Harris et al., 2012; Kopec et al., 2012a; Thompson et al., 2011b; O’Brien et al., in preparation), the genomic profile comparisons described herein have several limitations. First, the microarray platforms (Affymetric vs. Agilent), specific gene probes (60 vs. 25 nucleotide oligomers), signal normalization, and statistical analyses differed between Kopec et al. (2012a)
andEllinger-Ziegelbauer et al. (2005). Notably, a basic premise/ assumption in our analysis is that, for example, a 2-fold change in gene expression in the Ellinger-Ziegelbauer et al. dataset is ‘‘equivalent’’ to a 2-fold change in the Kopec et al. dataset. Second, the gene changes were compared across species (mouse vs. rat). Notably, there is a federal effort to compare genomic re-sponses to chemicals in yeast,C.elegans, zebrafish embryos, and mice in the pursuit of identifying ‘‘expression signature profiles’’ that can be used to predict toxicological responses in humans (NIEHS, 2011). Thus, it is anticipated that cellular responses to DNA damage, oxidative stress, and proliferation are likely to be, in part, similar across species. Recent studies indicate that basal gene expression is quite similar across species (Chan et al., 2009; Zheng-Bradley et al., 2010). Whether this species concordance holds for response to xenobiotics is less clear; how-ever, our own studies conducted in mice and rats generally indi-cate similar gene expression changes in the intestines of both species in response to Cr(VI) (Kopec et al., 2012b). The difference in tumor outcome in the intestine of rats and mice may arise from overall differences in dosimetry (pharmacokinetics) rather than pharmacodynamics (Proctor et al., in press).
While gene expression patterns appear to be similar across species at the tissue level, the basal gene expression patterns across tissues appear to be different – even within the same spe-cies (Chan et al., 2009; Jonker et al., 2009; Zheng-Bradley et al., 2010). Thus, a third uncertainty in the analyses herein is whether differential gene responses in the liver should be com-pared to differential gene responses in the duodenum. Jonker et al. (2009)exposed mice to two genotoxic and nongenotoxic carcinogens and reported that there was little overlap in the genomic responses to the carcinogens in tissues examined (liver, spleen, bladder, blood, and lymph nodes) using a microarray of 8205 oligonucleotides. Nevertheless, it is well accepted that stress and DNA repair mechanisms are highly conserved across species, and even phyla, on a structural and pathway level ( Rob-ertson et al., 2009; Taylor and Lehmann, 1998). Again, the effort to compare genomic responses to chemicals across phyla (NIEHS, 2011) suggests that many scientists believe that some of the cel-lular responses governing oxidative stress and DNA damage are likely to be shared across cell types. Interestingly, analyses by
Zheng-Bradley et al. (2010) indicate that gene expression pat-terns from different cell lines were more homogeneous than their respective tissues of origin; the authors speculated that this might be due to immortalization or the lower variability in cell culture samples. Regardless of the reason(s), the fact that gene expression profiles differ between cultured cells and their tissues of origin suggest that extrapolation of gene changes from
in vitrocell models to in vivomodels (i.e. body organs) may be no less uncertain than extrapolatingin vivo gene changes across tissues.
Finally, it is worth noting that the Cr(VI) exposures associated with lung cancer are primarily of particulate form and are associ-ated with inflammation (Nickens et al., 2010). In contrast we ob-served relatively little histological, biochemical, or genomic evidence for inflammation in our studies (Kopec et al., 2012; Thompson et al., 2011b). Therefore, we draw no conclusions about the applicability of the findings herein pertaining to soluble Cr(VI) in the duodenum to particulate Cr(VI) in the lung.
5. Conclusions
This study does not attempt to identify gene changes that dis-tinguish between mutagenic and nonmutagenic carcinogens, but rather compare the gene changes elicited by Cr(VI) in the duode-num to a relatively small set of genes previously shown to be dif-ferentially expressed following exposure to mutagenic and nonmutagenic carcinogens. Data reduction and multivariate statis-tical analyses were used in order to remove subjectivity from these comparisons. Notwithstanding the limitations discussed previ-ously, the observation that the gene changes elicited by Cr(VI) are more similar to nonmutagenic than mutagenic carcinogens is consistent with evidence that Cr(VI) induced redox changes in the intestine at both carcinogenic and noncarcinogenic concentra-tions, as well as lack of evidence for micronucleus formation and
k-rasmutation after 90 days of exposure to Cr(VI) concentrations in drinking water as high as 182 mg/L. Thus, while there is substan-tial data (mainly from in vitro studies) that indicate Cr(VI) can interact directly with DNA, target tissue data do not support a mutagenic MOA for Cr(VI) in the small intestine. As part of a weight of evidence evaluation of the MOA for Cr(VI)-induced intes-tinal carcinogenesis, the analyses in this report add further support to the hypothesis that the intestinal tumors in the mouse duode-num are the result of a nonmutagenic MOA.
Conflict of interest
None.
Acknowledgments
The authors would like to thank Dr. Anna Kopec for assistance with toxicogenomic data. We also thank Drs. Michael Dourson, Da-vid Gaylor, Lucy Anderson, Rebecca Fry and Timothy R. Zacharew-ski for critical review of an earlier version of portions of this manuscript. This work was funded by Cr(VI) Panel of the American Chemistry Council.
References
Bolt, H.M. et al., 2004. Carcinogenicity categorization of chemicals-new aspects to be considered in a European perspective. Toxicology Letters 151, 29–41. Boverhof, D.R., Gollapudi, B.B., 2010. Application of Toxicogenomics in Safety
Evaluation and Risk Assessment. John Wiley & Sons, Hoboken, NJ.
Chan, E.T. et al., 2009. Conservation of core gene expression in vertebrate tissues. Journal of Biology 8, 33.
Chandra, S.A. et al., 2010. Chemical carcinogenesis of the gastrointestinal tract in rodents: an overview with emphasis on NTP carcinogenesis bioassays. Toxicologic Pathology 38, 188–197.
Cohen, S.M. et al., 2010. Carcinogenic mode of action of folpet in mice and evaluation of its relevance to humans. Critical Reviews Toxicology 40, 531–545. De Flora, S. et al., 1997. Estimates of the chromium(VI) reducing capacity in human body compartments as a mechanism for attenuating its potential toxicity and carcinogenicity. Carcinogenesis 18, 531–537.
Eastmond, D.A., 2008. Evaluating genotoxicity data to identify a mode of action and its application in estimating cancer risk at low doses: A case study involving carbon tetrachloride. Environmental and Molecular Mutagenesis 49, 132–141. Eckel, J.E. et al., 2004. Empirical bayes gene screening tool for time-course or dose-response microarray data. Journal of Biopharmaceutical Statistics 14, 647–670. Eckel, J.E. et al., 2005. Normalization of two-channel microarray experiments: a
semiparametric approach. Bioinformatics 21, 1078–1083.
Ellinger-Ziegelbauer, H. et al., 2009. Application of toxicogenomics to study mechanisms of genotoxicity and carcinogenicity. Toxicology Letters 186, 36–44. Ellinger-Ziegelbauer, H. et al., 2004. Characteristic expression profiles induced by
genotoxic carcinogens in rat liver. Toxicological Sciences 77, 19–34. Ellinger-Ziegelbauer, H. et al., 2005. Comparison of the expression profiles induced
by genotoxic and nongenotoxic carcinogens in rat liver. Mutation Research 575, 61–84.
Fraley, C., Raftery, A.E., 2002. Extending the Linear Model with R. Chapman and Hall/ CRC, London.
Fraley, C., Raftery, A.E., 2006. MCLUST Version 3 for R: Normal Mixture Modeling and Model-based Clustering. Chapman and Hall/CRC, London.
Glorfeld, L.W., 1995. An improvement on Horn’s parallel analysis methodology for selecting the correct number of factors to retain. Educational and Psychological Measurement 55, 377–393.
Gottschling, B.C. et al., 2001. The role of oxidative stress in indium phosphide-induced lung carcinogenesis in rats. Toxicological Sciences: An Official Journal of the Society of Toxicology. 64, 28–40.
Greaves, P., 2007. Histopathology of Preclinical Toxicity Studies. Elsevier-Academic Press, London.
Hair, J.F. et al., 1998. Multivariate Data Analysis. Prentice-Hall, Upper Saddle River. Harris, M. A., et al., 2012. Assessment of genotoxic potential of Cr(VI) in the intestine via in vivo intestinal micronucleus assay and in vitro high content analysis in differentiated and undifferentiated Caco-2. 51st Annual Meeting of the Society of Toxicology, San Francisco, CA.
Hastie, T. et al., 2001. The Elements of Statistical Learning: Data Mining, Inference, and Prediction. Springer-Verlag, New York.
Hirose, T. et al., 2002. Frequent microsatellite instability in lung cancer from chromate-exposed workers. Molecular Carcinogenesis 33, 172–180.
Holmes, A.L. et al., 2008. Carcinogenicity of hexavalent chromium. Indian Journal of Medical Research 128, 353–372.
Horn, J.L., 1965. A rationale and test for the number of factors in factor analysis. Psychometrika 30, 179–185.
IARC, 1990. Chromium, nickel and welding. IARC Monographs on the Evaluation of Carcinogenic Risks to Humans 49, 1–648.
Jonker, M.J. et al., 2009. Finding transcriptomics biomarkers for in vivo identification of (non-)genotoxic carcinogens using wild-type and Xpa/p53 mutant mouse models. Carcinogenesis 30, 1805–1812.
Keene, O.N., 1995. The log transformation is special. Statistics in Medicine 14, 811–819. Klaunig, J.E. et al., 1998. The role of oxidative stress in chemical carcinogenesis.
Environmental Health Perspectives 106 (Suppl 1), 289–295.
Kondo, K. et al., 1997. Mutations of the p53 gene in human lung cancer from chromate-exposed workers. Biochemical and Biophysical Research Communications 239, 95–100.
Kopec, A.K. et al., 2012a. Genome-wide gene expression effects in B6C3F1 mouse intestinal epithelia following 7 and 90 days of exposure to hexavalent chromium in drinking water. Toxicology and Applied Pharmacology 259, 13–26. Kopec, A.K. et al., 2012b. Comparative toxicogenomic analysis of oral Cr(VI) exposure effects in rat and mouse small intestinal epithelia. Toxicology and applied pharmacology 262, 124–138.
Montgomery, D.C. et al., 2006. Introduction to Linear Regression Analysis. Wiley, New York.
Nickens, K.P. et al., 2010. Chromium genotoxicity: a double-edged sword. Chemico-Biological Interactions 188, 276–288.
NIEHS, Comparative Genomic Responses to Environmental Toxicants. Principal Investigator: Jonathan H. Freedman. In support of the NIEHS Toxicogenomics Research Consortium Cooperative Research Program. 2011.
NTP, 2001. NTP technical report on the toxicology and carcinogenesis studies of indium phosphide (CAS No. 22398–80-7) in F344/N rats and B6C3F1 mice (inhalation studies), NTP TR 499. NIH Publication No. 01-4433.
NTP, 2002. NTP technical report on the toxicology and carcinogenesis studies of o-nitrotoluene (CAS No. 88-72-2) in F344/N rats and B6C3F1 mice (feed studies), NTP TR 504. NIH Publication No. 02-4438.
NTP, 2008. NTP technical report on the toxicology and carcinogenesis studies of sodium dichromate dihydrate (CAS No. 7789–12-0) in F344/N rats and B6C3F1 mice (drinking water studies), NTP TR 546. NIH Publication No. 08-5887. Press, S.J., Wilson, S., 1978. Choosing between logistic regression and discriminant
analysis. Journal of the American Statistical Association 73, 699–705. Proctor, D.M. et al., 2002. Is hexavalent chromium carcinogenic via ingestion? A
weight-of-evidence review. Journal of Toxicology and Environmental Health A 65, 701–746.
Proctor, D. M., et al., in press. Hexavalent chromium reduction kinetics in rodent stomach contents. Chemosphere.
Robertson, A.B. et al., 2009. DNA repair in mammalian cells: base excision repair: the long and short of it. Cellular and Molecular Life Sciences 66, 981–993. Saeed, A.I. et al., 2003. TM4: a free, open-source system for microarray data
management and analysis. BioTechniques 34, 374–378.
Sedelnikova, O.A. et al., 2010. Role of oxidatively induced DNA lesions in human pathogenesis. Mutation Research 704, 152–159.
Sun, H. et al., 2009. Modulation of histone methylation and MLH1 gene silencing by hexavalent chromium. Toxicology and Applied Pharmacology 237, 258–266. Tabachnick, B. G., Fidell, L. S., 2007. Using Multivariate Statistics. Pearson, Boston. Taylor, E.M., Lehmann, A.R., 1998. Conservation of eukaryotic DNA repair
mechanisms. International Journal of Radiation Biology 74, 277–286. Thompson, C.M. et al., 2011a. Application of the U.S. EPA mode of action Framework
for purposes of guiding future research: a case study involving the oral carcinogenicity of hexavalent chromium. Toxicological Sciences 119, 20–40. Thompson, C.M. et al., 2011b. Investigation of the mode of action underlying the
tumorigenic response induced in B6C3F1 mice exposed orally to hexavalent chromium. Toxicological Sciences 123, 58–70.
US EPA, 1991. National primary drinking water regulations-synthetic organic chemicals and inorganic chemicals; monitoring for unregulated contaminants; national primary drinking water regulations implementation; national secondary drinking water regulations. Final rule. Federal Register. 56, 3526– 3597.
US EPA, 2004. Captan; cancer reclassification; amendment of reregistrationeligibility decision; notice of availability. Federal Register 69, 68357–68360.
US EPA, 2005a. Guidelines for carcinogen risk assessment, EPA/630/P-03/001F. Risk Assessment Forum: US Environmental Protection Agency, Washington, DC. US EPA, 2005b. Supplemental guidance for assessing susceptibility from early-life
exposure to carcinogens, EPA/630/R-03/003F. Risk Assessment Forum: US Environmental Protection Agency, Washington, DC.
US EPA, 2007. Framework for Determining a Mutagenic Mode of Action for Carcinogenicity: Using EPA’s 2005 Cancer Guidelines and Supplemental Guidance for Assessing Susceptibility from Early-Life Exposure to Carcinogens. US Environmental Protection Agency, Washinton, DC.
Upham, B.L., Wagner, J.G., 2001. Toxicant-induced oxidative stress in cancer. Toxicological Sciences: An Official Journal of the Society of Toxicology 64, 1–3.
Watanabe, C. et al., 2010. DNA damage and estrogenic activity induced by the environmental pollutant 2-nitrotoluene and its metabolite. Environmental Health and Preventive Medicine 15, 319–326.
Waters, M.D. et al., 2010. Characterizing and predicting carcinogenicity and mode of action using conventional and toxicogenomics methods. Mutation Research 705, 184–200.
Zheng-Bradley, X. et al., 2010. Large scale comparison of global gene expression patterns in human and mouse. Genome Biology 11, R124.
Zhitkovich, A., 2011. Chromium in drinking water: sources, metabolism, and cancer risks. Chemical Research in Toxicology 24, 1617–1629.
