TOXICOLOGICAL SCIENCES 99(1), 326–337 (2007)

doi:10.1093/toxsci/kfm150

Advance Access publication June 11, 2007

Multicenter Study of Acetaminophen Hepatotoxicity Reveals theImportance of Biological Endpoints in Genomic Analyses

Richard P. Beyer,*,1 Rebecca C. Fry,†,1 Michael R. Lasarev,‡,1 Lisa A. McConnachie,*,1 Lisiane B. Meira,†,1

Valerie S. Palmer,‡,1 Christine L. Powell,§,1 Pamela K. Ross,§,1 Theo K. Bammler,* Blair U. Bradford,§ Alex B. Cranson,‡Michael L. Cunningham,{ Rickie D. Fannin,{ Gregory M. Higgins,‡ Patrick Hurban,k Robert J. Kayton,‡

Kathleen F. Kerr,* Oksana Kosyk,§ Edward K. Lobenhofer,k Stella O. Sieber,{ Portia A. Vliet,* Brenda K. Weis,{

Russel Wolfinger,kj Courtney G. Woods,§ Jonathan H. Freedman,kk,2 Elwood Linney,kk,2 William K. Kaufmann,§,2

Terrance J. Kavanagh,*,2 Richard S. Paules,{,2 Ivan Rusyn,§,2,3 Leona D. Samson,†,2 Peter S. Spencer,‡,2 William Suk,{,2

Raymond J. Tennant,{,2 Helmut Zarbl,#,2 and Members of the Toxicogenomics Research Consortium

*University of Washington, Seattle, Washington 98195; †Massachusetts Institute of Technology, Cambridge, Massachusetts 02139;

‡Oregon Health & Science University, Portland, Oregon 97239; §University of North Carolina, Chapel Hill, North Carolina 27599;{National Institute of Environmental Health Sciences, Research Triangle Park, North Carolina 27709; kCogenics, a Division of Clinical Data, Inc.,

Morrisville, North Carolina 27560; kjSAS Institute, Inc., Cary, North Carolina 27513; kkDuke University Medical Center, Durham, North Carolina 27710;

and #Fred Hutchinson Cancer Research Center, Seattle, Washington 98109

Received April 19, 2007; accepted June 5, 2007

Gene expression profiling is a widely used technique with data

from the majority of published microarray studies being publicly

available. These data are being used for meta-analyses and

in silico discovery; however, the comparability of toxicogenomic

data generated in multiple laboratories has not been critically

evaluated. Using the power of prospective multilaboratory

investigations, seven centers individually conducted a common

toxicogenomics experiment designed to advance understanding of

molecular pathways perturbed in liver by an acute toxic dose of

N-acetyl-p-aminophenol (APAP) and to uncover reproducible

genomic signatures of APAP-induced toxicity. The nonhepato-

toxic APAP isomer N-acetyl-m-aminophenol was used to identify

gene expression changes unique to APAP. Our data show that

c-Myc is induced by APAP and that c-Myc–centered interactomes

are the most significant networks of proteins associated with liver

injury. Furthermore, sources of error and data variability among

Centers and methods to accommodate this variability were

identified by coupling gene expression with extensive toxicological

evaluation of the toxic responses. We show that phenotypic

anchoring of gene expression data is required for biologically

meaningful analysis of toxicogenomic experiments.

Key Words: liver injury; toxicogenomics; phenotypic anchoring.

The biology of liver and other tissues in normal and disease

states increasingly is being probed using global approaches such as

microarray transcriptional profiling. Acceptance of this technology

is based principally on a satisfactory level of reproducibility of

data among laboratories and across platforms (Bammler et al.,2005; Shi et al., 2006). Achieving reasonable concordance in

expression data among laboratories requires standardized

protocols for RNA labeling, hybridization, microarray process-

ing, data acquisition, and normalization (Bammler et al., 2005).

The majority of standardization efforts have focused on

(1) comparing gene expression data from the analysis of

reference samples distributed among several laboratories

(Bammler et al., 2005; Pennie et al., 2004), or (2) developing

guidelines for documenting experimental details and results

when publishing microarray data (Ball et al., 2002). The issue

of reproducibility and reliability of genomics data obtained

from similar (standardized) biological experiments performed

in different laboratories is crucial to the generation and utility

of large databases of microarray results (Becker, 2001; Miles,

2001). While several recent studies uncovered important lim-

itations of expression profiling of chemical injury to cells and

tissues (Baker et al., 2004; Beekman et al., 2006; Ulrich et al.,2004), determining the effects of intralaboratory variables on

the reproducibility, validity, and general applicability of the re-

sults that are generated by different laboratories and deposited

into publicly available databases remains a gap.

The National Institute of Environmental Health Sciences

(NIEHS) established the Toxicogenomics Research Consor-

tium to apply the collective and specialized expertise from

academic institutions to address issues in integrating gene

expression profiling, bioinformatics, and general toxicology.

Key elements include developing standardized practices for

1 Equally contributing first author.2 Equally contributing senior author.3 To whom correspondence should be addressed at Department of

Environmental Sciences and Engineering, School of Public Health,

University of North Carolina at Chapel Hill, CB #7431, Chapel Hill, NC

27599. Fax: (919) 843-2596. E-mail: iir@unc.edu.

� The Author 2007. Published by Oxford University Press on behalf of the Society of Toxicology. All rights reserved.For Permissions, please email: journals.permissions@oxfordjournals.org

by guest on July 3, 2014http://toxsci.oxfordjournals.org/

Dow

nloaded from

gene expression studies, and conducting systematic assess-

ments of the reproducibility of traditional toxicity endpoints

and microarray data within and among laboratories. To this

end, the consortium selected the classical hepatotoxicant acet-

aminophen (APAP) (James et al., 2003; Kaplowitz, 2005), for

its proof-of-concept experiments. Despite more than 30 years

of research on APAP, we are far from a complete under-

standing of the mechanisms of liver injury, risk factors, and

molecular markers that predict clinical outcome after poisoning

(Watkins et al., 2006).

To evaluate both mechanisms and interlaboratory variation in

injury response, the toxicogenomic analysis of APAP-induced

hepatotoxicity was performed at seven geographically dispersed

Centers. Parallel studies with N-acetyl-m-aminophenol (AMAP),

the nonhepatotoxic isomer of APAP, provided a method to

isolate transcripts associated with hepatotoxicity. This study

identified potential sources of interlaboratory variability when

microarray analyses are conducted on samples generated in

different laboratories. The results also show that phenotypic

anchoring of gene expression data is required for biologically

meaningful meta-analysis of genomic experiments. This study

reports an extensive temporal pathway map of APAP- and

AMAP-specific, as well as aminophenol-nonspecific responses

in mouse liver, and uncovers a central role of Myc-regulated

signaling networks in APAP-induced acute liver injury.

Furthermore, we demonstrate that transcriptional responses in

livers of mice and rats are similar when equitoxic doses of APAP

are considered.

MATERIALS AND METHODS

Animals and treatments. To minimize the potential effects of environ-

mental and genetic variables, all participating laboratories received animals,

feed, bedding, and test article preparations from centralized sources. Each

participating laboratory used 45 10- to 11-week-old male isogenic C57BL/6J

mice (Jackson Labs, Bar Harbor, ME, or Sacramento, CA). The mice were

housed individually in an environmentally controlled area with temperature

(69–75�F), humidity control (35–65%), and a 12-h light/dark cycle under

pathogen-free conditions. The bedding and animal chow used by all

laboratories were provided by Battelle (Columbus, OH). Mice were maintained

on irradiated NTP-2000 chow (Ziegler Brothers, Inc., East Berlin, PA) and

bedding was on Sani-Chips (P.J. Murphy, Montville, NJ).

All participating laboratories were required to adhere to a standard protocol

for treatments, treatment termination, specimen collection, and processing, and

all endpoints were measured at centralized locations. After acclimation (12–15

days), animals were fasted for 18-h overnight prior to treatment and for 3 h after

dosing; water was given ad libitum. Mice, randomly assigned to treatment

groups (five per treatment, per time point, per center), were dosed (between

8 and 10 A.M. local time) by oral gavage with 10 ml/kg body weight of vehicle

(methylcellulose, 0.5% wt/vol), AMAP (300 mg/kg), or N-acetyl-p-amino-

phenol (APAP; 300 mg/kg). The experimental conditions were selected based

on the dose (300–600 mg/kg) and time (up to 48 h) finding study conducted at

one of the participating laboratories prior to the experiment (Supplemental

Table 1). The formulations of vehicle and test compounds used in this study

were prepared at Battelle and supplied to all laboratories to assure consistency

in drug concentrations. Before treatment, solutions were stirred continuously to

assure that proper mixing occurred while dosing. Mice were terminated under

Nembutal anesthesia (100 mg/kg, i.p.) at 6, 12, and 24 h after treatment.

At termination, whole blood was collected from the vena cava or by cardiac

puncture, transferred to 1.1-ml microcentrifuge tubes with gel additive carrier/

clot activator for serum separation (Sarstedt, Aktlengesellschaft & Co.,

Numbrecht, Germany), and centrifuged at 16,000 3 g for 20 min at room tem-

perature. Serum was collected from the supernatant above the gel, transferred to

a new tube, frozen in liquid nitrogen, and then stored at �80�C until analyses.

Liver samples were sectioned according to the standard operating procedure to

assure consistency of the collections. The left lateral lobe was used for

histopathological evaluation and RNA isolation, the median lobe was collected

for histopathological evaluation, and the remainder of the median lobe and right

anterior lobe was collected for glutathione measurements. Liver samples for

histopathology (3- to 5-mm-thick tissue slice) were fixed in 10% neutral

buffered formalin. All other liver samples were immediately frozen in liquid

nitrogen and stored at �80�C until analyses. Leftover solutions of the vehicle

and test chemicals, serum samples, as well as liver samples designated for

histopathology and glutathione measurements, were shipped on dry ice to

Battelle for further processing. RNA was extracted from each liver sample at

the treatment center, flash frozen, and stored at �80�C for shipment to

Paradigm Array Labs (Now Cogenics, a Division of Clinical Data, Inc.,

Morrisville, NC) where microarray experiments were performed. All animal

studies for this project were approved by each Institution’s respective Animal

Care and Use Committee prior to the start of the study.

Clinical chemistry. All serum samples were analyzed with a centrifugal

analyzer (Hitachi 911, Chula Vista, CA) at Battelle using procedures supplied

by the manufacturer. Variables analyzed included alanine aminotransferase

(ALT, historical control values from male B6C3F1 mice: 51 ± 25 U/l,

minimum/maximum: 21/112 U/l, n ¼ 34), aspartate aminotransferase (no

historical control available), sorbitol dehydrogenase (30 ± 9 U/l, min/max: 20/

63 U/l, n ¼ 36), and bile salts/acids (23 ± 5lM/l, min/max: 15/35lM/l, n ¼ 35).

The complete data set is provided in Supplemental Table 2. Serum enzyme

measurements were not performed on a small number of animals (15 in total)

from which blood samples were not available.

Histological methods. Liver tissue samples were embedded in paraffin,

sectioned, stained with hematoxylin and eosin (H&E), and examined for

pathological changes at Battelle where the diagnoses were subject to Quality

assurance and pathology peer review following standard National Toxicology

Program (NTP) procedure (Boorman et al., 2002). In addition to standard

qualitative pathology reads of the liver sections, a semiquantitative method of

estimating a fraction of necrotic tissue in histopathological sections of the liver

was applied (Mauton, 2002). For this purpose, a grid with 100 evenly spaced

points was overlaid on printed images of H&E-stained liver sections taken at

3100 magnification. All of the tissues available on each slide were evaluated.

The percent liver necrosis score was determined as the number of points lying

within necrotic areas divided by a total number of points within the tissue section.

Histopathological sections from all Centers were assessed for percent liver

necrosis score in a blinded fashion by one trained individual. These results were

independently verified by a board certified veterinary pathologist. These data are

provided in Supplemental Table 2.

Hepatic glutathione analysis. The method was based on the determination

of thiols by high-performance liquid chromatography with fluorescence

detection (Toyo’oka et al., 2001). Liver samples (median and right anterior

lobes) from individual animals were analyzed for reduced and total glutathione

at Battelle. Briefly, 1 mg of liver was placed in a test tube with 1 ml of 5%

trichloroacetic acid containing 5mM ethylenediaminetetraacetic acid (EDTA),

homogenized, vortexed, and centrifuged. Next, 50 ll of homogenate was

transferred to a test tube with 950 ll of 5mM EDTA containing 0.1M borax

(pH 9.3). Then, 100-ll aliquots were placed in duplicate tubes for each liver

sample. Fluorobenzofurazan-4-sulfonic acid ammonium salt was added for

reduced glutathione and tri-n-butylphosphine in acetonitrile was added to the

other tube for total glutathione. High-performance liquid chromatography with

fluorescence detection was used to determine total and reduced glutathione and

compared against calibration standards that were processed with each assay.

These data are provided in Supplemental Table 2.

MULTICENTER STUDY OF ACETAMINOPHEN HEPATOTOXICITY 327

by guest on July 3, 2014http://toxsci.oxfordjournals.org/

Dow

nloaded from

RNA isolation. Total RNA was isolated at each Center according to the

standardized protocol. Frozen liver samples (~30 mg from the left lateral lobe)

of four randomly selected mice from each study group were homogenized in

the extraction buffer supplied in the RNeasy mini kit (Qiagen, Valencia, CA).

Then, RNA was isolated using the spin column technology as detailed by the

manufacturer. RNA quality and quantity were determined spectrophotometri-

cally from the absorbance at 260 and 280 nm. Aliquots of 25 lg RNA were

frozen at �70�C and shipped to Cogenics for microarray analysis.

Microarray hybridizations. All gene expression profiling was performed

at Cogenics. One microgram of total RNA from individual mouse liver samples

was amplified and labeled with a fluorescent dye (Cy3), and a common

reference—pooled mouse liver messenger RNA (mRNA) isolated from the

livers of 100 male C57BL6/J mice (see Bammler et al., 2005 for details)—was

amplified and labeled with Cy5 using Agilent Technologies (Palo Alto, CA)

Low RNA Input Linear Amplification labeling kit following the manufacturer’s

protocol. The quantity of the resulting fluorescently labeled cRNA was

evaluated using a Nanodrop ND-100 spectrophometer (Nanodrop Technolo-

gies, Wilmington, DE) and its integrity measured using an Agilent Bioanalyzer.

Equal amounts of Cy3 and Cy5-labeled cRNA (750 ng) from the individual

animals and from the pooled control, respectively, were hybridized to an

Agilent Mouse Oligo Microarray (~21,000 features, catalogue# G4121) for

17 h at 65�C. The hybridized microarrays were washed and scanned using an

Agilent G2565BA scanner. Data were extracted from the scanned image using

Agilent Feature Extraction software version 7.5. There were two noted study

protocol deviations due to technical issues: Center #6 supplied RNA samples

for only one time point (6 h), and two samples from AMAP-treated animals

(one each for 12- and 24-h time points) from Center #3 were not included in

follow-up analyses due to poor microarray data quality as determined by

Feature Extraction software (excessive background signal intensity and large

number of saturated features).

Data analysis. Raw microarray data were processed and analyzed with

tools in the Bioconductor (Gentleman et al., 2004) software package.

Following the recommendations of the Toxicogenomics Research Con-

sortium’s standardization experiment (Bammler et al., 2005; Qin and Kerr,

2004), median Cy3 and Cy5 fluorescence intensities for each spot on the array

were used as the signal and no background adjustment was performed. Two

microarrays (Center 3) were excluded from further processing and analysis due

to poor quality. Intensities flagged as saturated signal by the Feature Extraction

software (Agilent Technologies) were treated as missing values. This resulted

in less than 1% missing data per array (median 0.1%, maximum 0.9%). Cy3/

Cy5 ratios were generated for each spot and these values were normalized for

intensity and spatially dependent systematic variation by regional lowess smooth-

ing (Cui et al., 2003). This data set (raw and normalized data files) is publicly

available from cebs.niehs.nih.gov (accession number 009-00001-0010-000-1).

Using the normalized data, genes with significant evidence for differential ex-

pression were identified using the Significance Analysis of Microarrays (SAM)

methodology (Tusher et al., 2001), as implemented in the Bioconductor package

‘‘samr.’’ The k-nearest-neighbor method was used to impute missing values

(Troyanskaya et al., 2001). For each of the three time points (6, 12, and 24 h) we

compared each pair of the three treatment groups for a total of nine pairwise

group comparisons. For each pairwise comparison, we treated ‘‘Center’’ as a

blocking factor and used 500 permutations (permuting the treatment labels

within Center) to estimate the reference null distributions for the significance

tests.

Correlations between liver toxicity and gene expression. The Spearman

correlation coefficient between gene expression and percent liver necrosis was

computed separately for each Center. A weighted median (Birkes and Dodge,

1993) of these individual correlation coefficients was computed with weights

proportional to the sample size from each Center (giving greater weight to

correlation coefficients from Centers with more data). Significance for this

(weighted) median correlation was determined by an approximate permutation

test (199 random permutations plus the original configuration). The p value

summarizes the proportion of random permutations that produced a (weighted)

median correlation coefficient at least as extreme as the one observed.

The number of permutations considered led to a resolution for the p value of

1/200 ¼ 0.005; a 95% confidence interval (Wilson’s interval; Agresti and

Coull, 1998) for p was computed to provide limits on the actual p value.

A (weighted) median correlation was deemed significant if p < 0.01 together

with an upper confidence limit for p strictly less than 0.05, a condition that

provides some assurance that a higher number of permutations would still pro-

duce small p values. The dashed-gray lines on Figure 4a show a robust linear fit

(Frees, 1991) applied to each Center’s data.

Variability analysis. An analysis of the variability of the microarray data

was conducted by fitting a random effects analysis of variance (ANOVA)

model on a feature-by-feature basis to the normalized log ratios (Jin et al.,

2001; Wolfinger et al., 2001). Eight variance components in the model

correspond to treatment, time, Center, their two- and three-way interactions,

and the residual variance. Since each animal was observed only once, the latter

is a confounded effect of animal-to-animal and technical variability. All effects

are considered random in this analysis in order to obtain a balanced partitioning

of estimated total variability. All effects are assumed to be normally distributed,

and variance components were estimated using restricted maximum likelihood

via JMP Microarray software (SAS Institute Inc., Cary, NC).

Analysis of biological networks. Network analysis was performed on lists

of genes generated from the earlier analyses described above as being

differentially expressed. Ingenuity Pathways Knowledge Base software (In-

genuity Systems, Inc., Redwood City, CA) was used for the analysis of significant

interactions. Ingenuity Pathways Knowledge Base is a database that contains data

of individually curated relationships between gene objects (e.g., genes, mRNAs,

and proteins) to generate significant biological networks and pathways.

A significance score ( p value) for each network and/or canonical pathway is

calculated according to the fit of the test gene set versus genes in the particular

pathway. The score of each identified network is calculated as the negative log

p value which indicates the likelihood that the genes in a test gene list can be

found together in a network due to random chance. In this study, a score

corresponding to p < 10�20 or lower was used to select highly significant

biological networks.

Gene ontology analysis. Lists of genes generated by various statistical

analyses of the microarray data were submitted to the GoMiner web-based

(http://discover.nci.nih.gov/gominer/) tool (Zeeberg et al., 2003) for identifi-

cation of the biological processes that are overrepresented. Statistical

significance of the biological process enrichment was determined using

Fisher’s exact p value based on the relative overrepresentation of genes within

a given gene ontology (GO) category as compared to a random sample of genes

probed on the array that was used in these experiments (see above). GO

pathways with a p < 0.05 and a minimum of five gene transcripts represented

from that category were considered for further analysis.

Species comparison of gene expression and toxicity data. Publicly

available (ntp.niehs.nih.gov/go/15716) gene expression and liver toxicity data

generated from F344/N rats that were exposed to either subtoxic (50 or 150

mg/kg) or toxic (1500 or 2000 mg/kg) doses of APAP for 6, 18, 24, or 48 h

were used. In order to identify APAP-responsive genes in both species, each of

the probes present on the Rat Oligonucleotide Array (Agilent G4130A) used in

the above study, and the Mouse Oligonucleotide Array used in this work, were

mapped to Entrez Gene IDs (where possible). Homologous genes present on

both microarray platforms were identified by using National Center for

Biotechnology Information HomoloGene tool by requiring that both Entrez

Gene IDs provided reciprocal BLAST hits. This resulted in the identification in

3076 candidate homologs that were measured on both microarrays. These

results were further restricted to 962 candidates by selecting genes with

a twofold or higher APAP-induced expression difference (as compared to time-

matched vehicle control animals separately for each species/time/dose group)

and a log ratio p value less than 0.001 in at least six treated animals (either rat

or mouse, see Supplemental Table 7 for a complete list of transcripts).

328 BEYER ET AL.

by guest on July 3, 2014http://toxsci.oxfordjournals.org/

Dow

nloaded from

RESULTS

Reducing Variability and Assuring Confidencein Gene Expression Studies

This study was designed to (1) advance understanding of the

molecular pathways perturbed in the liver by a toxic dose of

APAP by contrasting toxic compound-specific responses with

those of a nontoxic structural analog (AMAP), and (2) apply

the power of a multiple laboratory investigation to uncover

a stereotypic genomic signature of APAP-induced toxicity. The

experiment achieved the overall goal of inducing liver injury

with APAP (300 mg/kg, ig), while both vehicle and AMAP

showed no adverse effects (Figs. 1a and 1b, Supplemental

Table 2). Greater than 20% necrosis was observed in ~55% of

the APAP-treated animals at each time point and correlated

well (rPearson ¼ 0.846) with serum ALT (Fig. 1c). However,

toxic injury induced by APAP showed variability across

Centers and animals. Animal-to-animal variability within in-

dividual Centers was less than variation among centers, and

one Center (#2) did not induce liver injury in 12 out of 15 mice

treated with APAP. Furthermore, serum ALT levels in vehicle-

treated animals from Center 3 were elevated, about threefold

higher than in other Centers. Other clinical chemistry data

largely followed the trends of changes in ALT (Supplemental

Table 2). A decrease in reduced glutathione content at 6 h

following APAP treatment was observed (Fig. 1d, samples

from Center #2 were not provided). However, the magnitude of

glutathione depletion varied considerably from< 5% (Center #5)

FIG. 1. Toxicological assessment of APAP-induced injury in mouse liver. Severity and time-course (6, 12, and 24 h) of liver necrosis (a) and elevations in

serum ALT activity (b) associated with APAP treatment across Centers that performed the in vivo portion of the study. Mean values for up to five animals per

treatment group are shown for APAP-treated animals. An average of up to 15 measurements (all time points) is shown for the vehicle category. (c) Correlation plot

between liver necrosis and serum ALT measurements at the level of the individual animal. Data from vehicle- and APAP-treated animals (all centers combined) are

plotted according to the time point (n ¼ 6 h, þ ¼ 12 h, 3 ¼ 24 h). (d) Fraction (mean values for up to five animals) of reduced glutathione (GSH, reported as % of

glutathione content in vehicle-treated mice at each corresponding time point) in livers of APAP-treated animals. See Supplemental Table 2 for a complete data table

on the clinical chemistry and histopathology measurements collected in this study.

MULTICENTER STUDY OF ACETAMINOPHEN HEPATOTOXICITY 329

by guest on July 3, 2014http://toxsci.oxfordjournals.org/

Dow

nloaded from

to> 60% (Center #7). A weak negative correlation was observed

between the fraction of the reduced glutathione and necrosis

(rPearson: �0.537 [6 h], �0.413 [12 h], and �0.377 [24 h]);

however, it should be noted that glutathione depletion in mouse

liver after a hepatotoxic dose of APAP is known to occur within

60 min of dosing (Mitchell et al., 1973) and the time points in this

study may be less informative.

The statistical significance of changes in gene expression

was determined in pairwise comparisons between treatments

for each time point and a Venn diagram of the overlap was

constructed (Fig. 2a, Table 1, Supplemental Table 3). Genes in

the m5 sector, those differentially expressed between APAP-

and vehicle-treated mice and also between the APAP- and

AMAP-treated mice, were considered to represent an APAP-

selective transcriptional response in mouse liver. Thus, a non-

redundant list of genes (7806 transcripts) whose expression

changed significantly and specifically due to APAP treatment

at any of the time points was compiled and the transcriptional

response was visualized (Fig. 2b, top). When measurements of

liver necrosis and serum ALT in the same animals are also

displayed in a yellow–blue heat map (Fig. 2b), a pronounced

concordance between APAP hepatotoxicity and transcriptional

response is evident.

To assess the relative contributions of the different sources

of biological, technical, and experimental variability, we fitted

a random effects ANOVA model separately to data from each

feature in the normalized microarray data. Through this

statistical test, overall observed variability for each feature on

the microarray was partitioned into components corresponding

to the effects of treatment, time, Center, two- and three-way

interactions, and the residual variability (a combination of

biological and technical variability). Figure 3 shows the

distribution of R2 (the proportion of total variability explained

for each feature), as well as box plots for each of the eight

sources of variability detailed above. Results are shown for

(1) all 21,318 features on the Agilent array (Figs. 3a and 3b),

and (2) the 7806 transcripts (see Fig. 2) representing the

APAP-specific expression profile in liver (Figs. 3a inset and

3c). When all transcripts on the array are considered, R2

distribution is bimodal and centered on 20% and 70% of

explained variability. On average, less than half of the overall

variability observed in this standardization experiment can be

attributed to the effects of time, treatments, and Center, or their

combinations. However, genes in the selectively APAP-

hepatotoxic signature fall clearly in the upper R2 mode where

considerably more variation is due to the effects of time,

FIG. 2. Toxicogenomic assessment of APAP-, AMAP-selective, and

‘‘aminophenol-like’’ responses in mouse liver. (a) Venn diagram illustrating

the distribution of gene expression alterations associated with APAP, AMAP or

vehicle treatment in liver. Gene lists were generated using SAM (false

discovery rate < 0.01) for each comparison between treatments. m sectors

designate groups of genes selected in each pairwise comparison (see Table 1 for

the total number of transcripts in each sector; sector assignment for each gene

on the microarray is provided in supplemental materials). (b) Heat maps of

toxicological phenotypes (yellow–blue, liver necrosis, serum ALT activity) and

gene expression (red–green) for genes identified in sectors m5 (APAP-specific,

top), m3 (AMAP-specific, middle), and m6 (aminophenol-like nonspecific,

bottom). Samples are ordered by Center (1–7), treatment (Vehicle, AMAP, and

APAP), and time (6, 12, 24 h). Genes are clustered (average linkage) according

to correlation (centered).

330 BEYER ET AL.

by guest on July 3, 2014http://toxsci.oxfordjournals.org/

Dow

nloaded from

treatments, and Center, or their combinations. Furthermore, the

residual variability in these APAP hepatotoxicity–associated

genes is reduced to 40% and treatment- and Center-specific

effects are more pronounced than those of time and the two-

and three-way interactions.

Probing Mechanisms by Linking Gene Expressionwith Phenotypic Endpoints

A Spearman correlation analysis of all features on the array

identified 3505 (6 h), 6085 (12 h), and 5988 (24 h) transcripts

with a significant association between gene expression and

percent liver necrosis (Table 1, Fig. 4a, and Supplemental

Table 4). The majority of these transcripts are found in sector m5

(75% [6 h], 57% [12 h], and 72% [24 h]). Figure 4a graphically

illustrates the Spearman correlation-based analysis by display-

ing two representative examples of the positive and negative

association between gene expression and the ‘‘phenotypic

anchor’’—percent liver necrosis—for two genes known to be

involved in the mechanism of liver toxicity by APAP. The

dashed-gray lines show a robust linear fit applied to each

Center’s data and provide a visual guide of the consistency of the

association.

Data mining of this signature using GO annotations was

used to assign functional classes and to determine the

biological processes defining the APAP-selective transcrip-

tional response associated with necrosis (Table 1, p < 0.01

columns). These lists were segregated by time point and into

positively and negatively correlating groups. Temporal changes

in biological processes identified by this analysis are shown in

Figures 4b and 4c and Supplemental Table 5. To further

characterize the APAP-selective hepatotoxic response, we used

the Ingenuity protein interactome analysis (Supplemental

Table 6). In all time points, Myc-regulated interactomes were

identified as the most significant (p < 10�37, Fig. 4d).

Furthermore, a majority of the members of the Myc-centric

interactomes were shared between time points with most

pronounced overlap at 12 and 24 h after APAP treatment.

Determining Toxicant-Specific Gene Expressionthrough Structural Analogues

Since it has been suggested that gene expression profiling

may mis-classify transcriptional changes occurring in response

to xenobiotic exposure as signs of toxicity (Hayes and Bradfield,

2005), this experiment also included a treatment with the

nontoxic stereo-isomer of APAP, AMAP. AMAP caused no

observable adverse health outcomes in any of the seven Centers;

however, it did evoke an AMAP-selective transcriptional re-

sponse (sector m3, Fig. 2b, middle and Table 1). While the total

number of transcripts modulated by AMAP is much smaller than

those changed by APAP, GO-based analysis of the biological

processes identified several significant categories at 12 and 24 h

(Fig. 4e). Further analysis identified genes that respond to both

APAP and AMAP similarly. These 397 transcripts (sector m6,

Fig. 2b, bottom and Table 1) comprise a nonspecific response to

aminophenol-like xenobiotics and thus should not be included in

the analysis of the drug-specific response. These genes lack a

consistent pattern in expression changes, nonetheless catabolism

and protein localization appear to be small but significant

biological processes (Fig. 4f).

Validating Biological Responses across Rodent Models

To compare the transcriptional responses to hepatotoxic

doses of APAP between mouse and rat we combined the gene

expression data obtained in this study to previously published

data (http://ntp.niehs.nih.gov/go/15716) on expression profiling

of rat liver samples from time-course (2–48 h) and dose–

response (50, 150, 1500, and 2000 mg/kg) analysis of APAP.

TABLE 1

Spearman Correlation Analysis of Gene Expression and Necrosis Score

6 h 12 h 24 h

Significance of correlation

Total # of

transcripts

Significance of correlation

Total # of

transcripts

Significance of correlation

Total # of

transcriptsp* < 0.01 p > 0.01 p* < 0.01 p > 0.01 p* < 0.01 p > 0.01

m0 314 14,658 14,972 777 12,771 13,548 452 11,519 11,971

m1 312 1088 1400 449 922 1371 403 1050 1453

m2 1 18 19 4 120 124 13 145 158

m3 11 21 32 36 148 184 73 146 219

m4 209 867 1076 698 666 1364 373 1193 1566

m5 2619 1116 3735 3884 515 4399 4331 1055 5386

m6 3 34 37 90 85 175 48 203 251

m7 36 11 47 147 6 153 295 19 314

Total 3505 17,813 21,318 6085 15,233 21,318 5988 15,330 21,318

Note. The number of genes that are significantly associated ( p < 0.01 and *upper confidence interval � 0.05, see Fig. 3 for examples) between gene expression

and percent necrosis in the liver at each time point (6, 12, or 24 h) and within a particular m sector (see Fig. 2a) is shown.

MULTICENTER STUDY OF ACETAMINOPHEN HEPATOTOXICITY 331

by guest on July 3, 2014http://toxsci.oxfordjournals.org/

Dow

nloaded from

Even though the data sets were generated using two different

microarrays, albeit of the same platform and at the same

reference laboratory, it was possible to compare and contrast

gene expression results at the level of individual genes (Fig. 5).

A yellow–blue heat map in Figure 5 displays measurements of

necrosis and serum ALT activity from the same animals in both

data sets (Supplemental Table 8). Similar to the data that were

obtained with mice (Fig. 2b), a high degree of concordance

between APAP-induced toxicity and transcriptional response is

evident in individual animals, both mice and rats. Seventy-one

transcripts (r¼ 0.83) constitute a species-independent universal

transcriptional response to APAP (Fig. 5, red box). The

transcripts for both Myc and Jun oncogenes are located within

this signature. Several GO processes were downregulated

following treatment with APAP as exemplified by a set of

52 transcripts (r ¼ 0.76, Fig. 5, green box).

DISCUSSION

The use of genomics has created both opportunities and

challenges. The concept of gene expression profiles as sig-

natures of toxicant classes, disease subtypes, or other biological

endpoints (Hamadeh et al., 2002a,b), together with public

availability of expression data from multiple studies, has

spurred the development of multiple databases and meta-

analysis for comparative study of toxicity based on reference

compounds (Waters et al., 2003). Disparate data streams may

include gene expression, body and organ-weight measure-

ments, clinical chemistry, and histopathology findings for

several tissues and require careful collection, management, and

integration (Mattes et al., 2004). While it has been recognized

that validated biological endpoints, phenotypic anchors, are

essential for interpreting toxicological outcomes (Paules, 2003;

Powell et al., 2006), no study has systematically approached

the issue of the variability in expression and toxicity data

through multilaboratory comparisons.

The effects of early and/or low-dose effects of APAP in the

rodent liver were considered recently and it was shown that

gene expression is a sensitive means of identifying indicators

of adverse effects that precede overt toxicity (Heinloth et al.,2004). The design of the current study was weighted toward

uncovering robust genomic signatures of APAP-induced tox-

icity that were reproducible in studies conducted across

multiple laboratories by focusing on doses and time points

that were selected to result in a clear phenotype (i.e., liver

injury) that would facilitate anchoring pathobiologicalFIG. 3. Variability analysis of microarray data. (a) Histogram showing

a distribution of R2 statistic (proportion of total variability explained by the

statistical model) across transcripts considered in this experiment. A three-way

variance components model was fit separately to all 21,318 (light green

histogram) transcripts on the microarray, or only 7806 transcripts from Venn

sector m5 (dark green inset). Corresponding box plot is also shown—top and

bottom of the box are the 25th and 75th percentiles, inside the box is a mean

bar, red bracket indicates shortest half of the data. (b) Box plots of variance

component estimates, standardized (sum ¼ 1) within each transcript (21,318

transcripts). (c) Box plots of standardized variance component estimates for the

7806 m5 transcripts. Animal/Tech—the residual variance, which is a combina-

tion of animal-to-animal and technical variability. Other variance components

correspond to the effects of Treatment, Time, Center, and their two- and three-

way interactions.

332 BEYER ET AL.

by guest on July 3, 2014http://toxsci.oxfordjournals.org/

Dow

nloaded from

endpoints with microarray data. Furthermore, this study firmly

establishes that careful consideration of the traditional toxicity

measurements, so-called ‘‘phenotypic anchors,’’ adds confi-

dence to the interpretation of data from novel technologies,

such as gene expression profiling. However, while ‘‘pheno-

typic anchoring’’ provides insights regarding mechanism of

FIG. 4. Phenotypic anchoring of gene expression and liver pathology reveals common biological pathways across Centers. (a) Plots show examples of the

positive (top, glutamate-cysteine ligase, catalytic subunit, Gclc) and negative (bottom, cytochrome P450 2e1, Cyp2e1) association between gene expression and

necrosis in the liver at 6 h after treatment. Points show individual animal’s response to APAP (red), AMAP (blue), or vehicle (green) treatment; gray dotted lines

show a robust linear fit to data from each of seven Centers. The (weighted) median Spearman correlation value (rs) is indicated together with a p value determined

from a permutation test (95% confidence interval for p is provided in parentheses). (b, c) Gene ontology analysis of transcripts in m5 sector that significantly

correlate with liver necrosis. The number of genes associated with each process is given in parentheses. Biological pathways affected in all time points are

highlighted by a box. (d) Ingenuity pathway analysis of transcripts in m5 sector that significantly correlate (positively, red, or negatively, green) with liver necrosis

identifies Myc-centered interactome as the top scoring network. Significant network interactions were identified for gene sets (same as in c and d) representing 6-h

post-APAP treatment (inner blue circle), 12-h post-APAP treatment (middle circle), and 24-h post-APAP treatment (outer circle). Temporality of the response is

displayed by line connections. Proteins connected to Myc by lines were found to be common between two and three time points. Solid lines represent physical

interactions (i.e., binding), whereas arrows represent any other type of indirect cellular interactions. Proteins that are not connected to Myc were found in the

interactomes specific for each particular time point. Different shapes of the nodes represent functional classification of the proteins that are shown: square,

cytokines; vertical rhombus, enzymes; horizontal rhombus, peptidases; triangle, phosphatases; inverted triangle, kinases; vertical oval, transmembrane receptors;

horizontal oval, transcription regulators; trapezoid, transporters; circle, other. (e, f) Gene ontology analysis of transcripts significantly correlating with AMAP (m3)

or aminophenol-like compound response (m6), respectively.

MULTICENTER STUDY OF ACETAMINOPHEN HEPATOTOXICITY 333

by guest on July 3, 2014http://toxsci.oxfordjournals.org/

Dow

nloaded from

toxicity, it may not provide additional value in terms of

prediction.

Despite our best efforts to standardize the study a consider-

able degree of animal-to-animal variability in the extent of liver

injury was observed. While the overwhelming majority of

studies in the literature report data on toxic effects in the form of

mean ± standard error, we chose to consider and display clinical

chemistry and histopathology data for individual animals,

which clearly demonstrates that a treatment-specific gene

expression signature is only present in subjects that exhibited

the pathologic effect of treatment. Furthermore, our study used

a correlation-based analytical approach to associate the extent of

liver toxicity caused by APAP (as % liver necrosis) with gene

expression. We argue that such correlation-based analytical

approach to associate the extent of liver toxicity with gene

expression is especially powerful for the selection of genes that

change in expression at a certain statistical significance and that

respond to treatment and/or pathophysiological endpoints.

Thus, despite the lack of evident toxicity in some APAP-

treated animals, a signature could be determined and used with

confidence to map gene expression profiles to pathways.

Previous studies indicated that comparability of gene

expression experiments among laboratories can be affected by

numerous technical variables (Bammler et al., 2005). An ad-

ditional layer of complexity arises when entire biological

experiments are performed in parallel, introducing additional

sources of variability stemming from technical, genetic, epi-

genetic, or stochastic factors. While this study did not consider

effects of genetics (since all mice were from an inbred line) or

microarray platform (since all experiments were performed on

the same commercial platform at a centralized location), we

found that the largest source of variability in a real-life biological

experiment was technical/animal whereby animal-to-animal

technical or biological deviations affected the outcome. Possible

contributors to this variability include differences in animal

husbandry, duration, and dose of anesthesia, methods, and

timing of blood draws, timing of tissue excision, storage, and

preparation, etc. Overall, this result underscores the convolution

of biological responses in complex systems and the importance

of taking each step of the experiment into consideration.

Inevitably, the high sensitivity of new assays creates additional

noise in the measurements and makes it more challenging to

extract meaningful knowledge with confidence. However, when

only those genes that constitute a phenotypically anchored

signature are considered, as shown in this work, the relative

proportion of uncontrolled variability diminishes. Thus, our

results argue strongly that the biological response–based filters

increase confidence and reduce variability in the microarray data

even if animal-to-animal variability is considerable.

This work, with the added confidence of a multilaboratory-

based investigation, reveals a number of biological pathways

that are responsive to APAP in mouse liver; however, current

study design does not allow one to conclude that other

centrilobular hepatotoxicants might produce a similar or

overlapping pattern of gene expression as seen for APAP

(i.e., the specificity of the response). Furthermore, our analysis

identifies genes that are not only consistently up-/down-

regulated, but also are significantly correlated with the toxicity

FIG. 5. Biological pathways affected by APAP in mouse and rat liver. Heat map of toxicological phenotypes (liver necrosis, serum ALT activity) and gene

expression from integrated mouse (this study) and rat (National Center for Toxicogenomics, NIEHS) data sets. See ‘‘Experimental Procedures’’ for the description

of the process of gene selection. A complete list of genes used in this analysis is available in Supplementary Table 7. Samples are ordered by Center (1–7, mouse

data only), treatment (AMAP or APAP, where applicable), and time (mouse: 6, 12, 24 h; rat: 2, 18, 24, and 48 h). Genes are clustered (average linkage) according

to correlation (centered).

334 BEYER ET AL.

by guest on July 3, 2014http://toxsci.oxfordjournals.org/

Dow

nloaded from

phenotype. While this work considered transcriptional

responses to APAP-induced liver injury and did not include

proteomic analysis, we argue that gene expression changes are

key secondary indicators of the cell’s response to stress and

thus can be used successfully to identify important biological

pathways involved in causation of, or response to, tissue

damage. It has been postulated that loss of mitochondrial

function (Donnelly et al., 1994) and concomitant generation of

oxidative stress (Jaeschke, 1990), as well as activation of

inflammatory cytokine and chemokine cascades, are all in-

volved in APAP-induced hepatotoxicity (James et al., 2003;

Kaplowitz, 2005; Watkins et al., 2006). We show that APAP-

related induction of programmed cell death accompanied by

inflammatory response and the parallel upregulation of the cell

cycle, activation of kinases, and metabolism are prominent at

all time points considered in this study. A number of in-

flammatory cells and mediator molecules are thought to be

involved in parenchymal cell damage. Neutrophils are an

essential component of the innate immune response at both the

initiation and progression levels. Recently, direct evidence that

neutrophil accumulation in the liver contributes to the

progression and severity of APAP-induced liver injury was

presented (Liu et al., 2006). Our data confirm that neutrophil

chemotaxis (i.e., Fcgr3, Itgb2, and Fcer1g) is one of the

significant biological processes that is induced by APAP

treatment and correlates significantly with liver necrosis.

The c-Jun N-terminal kinase (Jnk) signaling pathway is

primarily activated by proinflammatory cytokines and stress

stimuli. In vitro and in vivo studies have shown that APAP

induces a sustained activation of Jnk that contributes to

hepatocellular necrosis (Gunawan et al., 2006). The propensity

for Jnk to mediate cell death has been postulated to be

mediated by the antagonistic cross-talk between Jnk and nu-

clear factor kappa B (NF-jB). APAP has been shown to inhibit

NF-jB transactivation in the liver and in turn amplify Jnk

tipping the scale from cell survival to cell death (Matsumaru

et al., 2003). Our data show that MAP kinases, including

members of the Jnk signaling pathway (Traf2, Mapk8ip3, and

Dusp10), are positively associated with APAP-induced liver

necrosis. As expected, this pathway is most active at 6 h when

oxidative stress and the production of proinflammatory cyto-

kines from the influx of inflammatory cells are at their peak.

Liver regeneration after damage depends on the ability of

cells to progress through the cell cycle and proliferate (Fausto

et al., 1995). In vitro, it has been shown that APAP interferes

with the ability of cells to respond to growth factor stimuli since

APAP prevented degradation of inhibitor of jB, thus blocking

activation of NF-jB, phosphorylation of Raf1, and upregulation

ofMyc (Boulares et al., 1999). While in vitro data should be con-

sidered with caution, data from this in vivo study also show that

c-Myc is induced by APAP and that c-Myc–centered inter-

actomes are the most significant networks of proteins associated

with liver injury. Jun and Fos are also induced in liver by

APAP treatment, and networks that encompass these two key

regulators of AP-1–specific transcriptional response are second-

most significant at all three time points. Indeed, the relationship

between chemical stress and the synthesis of glutathione,

including the role of these genes, transcription factors, and

glutathione-synthesizing enzymes (e.g., glutamate-cysteine

ligase) has been documented (Kitteringham et al., 2000).

Most of the significantly downregulated biological processes

were found as early as 6 h after treatment with APAP. Lipid

metabolism, electron transport, and cofactor metabolism are

repressed at all time points. This observation likely reflects

a temporary suspension of physiological functioning of the

liver that occurs under conditions of massive necrosis. Indeed,

it was previously suggested that transcript levels of enzymes

involved in major energy-consuming biochemical pathways are

downregulated after APAP treatment (Heinloth et al., 2004),

and the mitochondrial dysfunction in liver after APAP

treatment is well known (Hinson et al., 2004; Knight and

Jaeschke, 2002; Ruepp et al., 2002).

Use of toxic and nontoxic structural analogues is a helpful

method to isolate gene expression changes that are most likely

to be associated with the phenotype of interest—in this case,

liver damage. AMAP and APAP are structurally similar com-

pounds and both are metabolized to reactive intermediates

(Rashed and Nelson, 1989). AMAP is believed to be nontoxic

(Nelson, 1980), even though it is known to produce protein

adducts, glutathione depletion, and other effects (Rashed et al.,1990) postulated as contributors to APAP-mediated liver in-

jury. The difference in the response to these isomers is thought

to reflect lower and shorter depletion of glutathione by AMAP

(Rashed et al., 1990). AMAP was also shown to have an effect

on the liver proteome qualitatively similar to APAP (Fountou-

lakis et al., 2000). Even though the response to AMAP was less

pronounced, it was most extensive at 12 and 24 h after treat-

ment and many of the biological processes are shared between

the two chemicals.

Rat and mouse models have been used extensively to

investigate organ-specific effects and mechanisms of toxicity.

The availability of detailed genetic information in these model

systems and humans provides greater utility of the toxicity data

obtained in animals (Cohen, 2004). Since regulatory elements

of evolutionarily related species are conserved, gene expression

signatures reflecting similar phenotypes in the species are also

likely to be conserved. Indeed, the cross-species comparison of

gene expression patterns revealed great similarity in the effects

of toxic doses of APAP in both species. When the extent of liver

injury was compared to the transcriptional response, it was clear

that similar expression changes occur in rat and mouse liver

under conditions of acute organ damage by APAP. Apoptosis,

cell cycle, stress, and transcriptional response, coupled with

downregulation of metabolism and transport, are common

biological processes between mouse and rat in response to

APAP and are likely to be in humans as well. For example, Mycand Jun signify a representative cross-species APAP-induced

response. Although it is yet to be determined whether these

MULTICENTER STUDY OF ACETAMINOPHEN HEPATOTOXICITY 335

by guest on July 3, 2014http://toxsci.oxfordjournals.org/

Dow

nloaded from

molecular events drive liver injury in humans as well, the

relative similarity of the sequence of liver injury events, its

dependency on glutathione content and inflammatory response,

and the overall phenotype shared among rodents and man

indicates that the signaling pathways driven by the mobilization

of energy resources and activation of tissue repair may have

a role in prognosis for human acute drug-induced liver injury.

The clear gain realized from this comparative genomics

approach is connecting molecular and histopathological fea-

tures of human liver injury to mouse and rat models with greater

confidence. Establishing this molecular relationship between

species should provide new opportunities to explore research

avenues into molecular pathogenesis, treatment, and prevention

of adverse drug-induced liver toxicity in humans.

SUPPLEMENTARY DATA

Supplementary data are available online at http://toxsci.

oxfordjournals.org/.

FUNDING

National Institutes of Health (U19-ES011384, U19-

ES011375, U19-ES011387, U19-ES011391, U19-ES011399,

and N01-ES25497).

ACKNOWLEDGMENTS

The authors wish to thank David Schwartz (NIEHS) for contributing to

the study design. The members of the Toxicogenomics Research Consortium

also acknowledge the efforts of David Balshaw and Michael Humble of the

NIEHS, and support of Molly Vallant of the NTP and the technical assistance

of Laurene Fomby, Tammy Wheat, Daphne Vasconcelos, Michael Ryan, Brian

Burback, and Milton Hejtacik (all from Battelle), and the Investigation

Genomics Group of Cogenics. This work was supported (in part) by the

Intramural Research Program of the National Institutes of Health and NIEHS.

REFERENCES

Agresti, A., and Coull, B. A. (1998). ‘‘Approximate’’ is better than ‘‘exact’’ for

interval estimation of binomial proportions. Am. Stat. 52, 119–126.

Baker, V. A., Harries, H. M., Waring, J. F., Duggan, C. M., Ni, H. A.,

Jolly, R. A., Yoon, L. W., De Souza, A. T., Schmid, J. E., Brown, R. H.,

et al. (2004). Clofibrate-induced gene expression changes in rat liver:

A cross-laboratory analysis using membrane cDNA arrays. Environ. Health

Perspect. 112, 428–438.

Ball, C. A., Sherlock, G., Parkinson, H., Rocca-Sera, P., Brooksbank, C.,

Causton, H. C., Cavalieri, D., Gaasterland, T., Hingamp, P., Holstege, F.,

et al. (2002). Standards for microarray data. Science 298, 539.

Bammler, T., Beyer, R. P., Bhattacharya, S., Boorman, G. A., Boyles, A.,

Bradford, B. U., Bumgarner, R. E., Bushel, P. R., Chaturvedi, K., Choi, D.,

et al. (2005). Standardizing global gene expression analysis between

laboratories and across platforms. Nat. Methods. 2, 351–356.

Becker, K. G. (2001). The sharing of cDNA microarray data. Nat. Rev.

Neurosci. 2, 438–440.

Beekman, J. M., Boess, F., Hildebrand, H., Kalkuhl, A., and Suter, L. (2006).

Gene expression analysis of the hepatotoxicant methapyrilene in primary rat

hepatocytes: An interlaboratory study. Environ. Health Perspect. 114,

92–99.

Birkes, D., and Dodge, Y. (1993). Alternative Methods of Regression. John

Wiley, New York, NY.

Boorman, G. A., Haseman, J. K., Waters, M. D., Hardisty, J. F., and Sills, R. C.

(2002). Quality review procedures necessary for rodent pathology databases

and toxicogenomic studies: The National Toxicology Program experience.

Toxicol. Pathol. 30, 88–92.

Boulares, H. A., Giardina, C., Navarro, C. L., Khairallah, E. A., and

Cohen, S. D. (1999). Modulation of serum growth factor signal transduction

in Hepa 1-6 cells by acetaminophen: An inhibition of c-myc expression,

NF-kappaB activation, and Raf-1 kinase activity. Toxicol. Sci. 48, 264–274.

Cohen, S. M. (2004). Risk assessment in the genomic era. Toxicol. Pathol.

32(Suppl. 1), 3–8.

Cui, X., Kerr, M. K., and Churchill, G. A. (2003). Transformations for cDNA

microarray data. Stat. Appl. Genet. Mol. Biol.2, Article4

Donnelly, P. J., Walker, R. M., and Racz, W. J. (1994). Inhibition of

mitochondrial respiration in vivo is an early event in acetaminophen-induced

hepatotoxicity. Arch. Toxicol. 68, 110–118.

Fausto, N., Laird, A. D., and Webber, E. M. (1995). Liver regeneration. 2. Role

of growth factors and cytokines in hepatic regeneration. FASEB J. 9,

1527–1536.

Fountoulakis, M., Berndt, P., Boelsterli, U. A., Crameri, F., Winter, M.,

Albertini, S., and Suter, L. (2000). Two-dimensional database of mouse liver

proteins: Changes in hepatic protein levels following treatment with

acetaminophen or its nontoxic regioisomer 3-acetamidophenol. Electropho-

resis 21, 2148–2161.

Frees, E. W. (1991). Trimmed slope estimates for simple linear regression.

J. Stat. Plann. Infer. 27, 203–221.

Gentleman, R. C., Carey, V. J., Bates, D. M., Bolstad, B., Dettling, M.,

Dudoit, S., Ellis, B., Gautier, L., Ge, Y., Gentry, J., et al. (2004).

Bioconductor: Open software development for computational biology and

bioinformatics. Genome Biol. 5, R80.

Gunawan, B. K., Liu, Z. X., Han, D., Hanawa, N., Gaarde, W. A., and

Kaplowitz, N. (2006). c-Jun N-terminal kinase plays a major role in murine

acetaminophen hepatotoxicity. Gastroenterology 131, 165–178.

Hamadeh, H. K., Bushel, P. R., Jayadev, S., DiSorbo, O., Bennett, L., Li, L.,

Tennant, R., Stoll, R., Barrett, J. C., Paules, R. S., et al. (2002a). Prediction

of compound signature using high density gene expression profiling. Toxicol.

Sci. 67, 232–240.

Hamadeh, H. K., Bushel, P. R., Jayadev, S., Martin, K., DiSorbo, O., Sieber, S.,

Bennett, L., Tennant, R., Stoll, R., Barrett, J. C., et al. (2002b). Gene

expression analysis reveals chemical-specific profiles. Toxicol. Sci. 67,

219–231.

Hayes, K. R., and Bradfield, C. A. (2005). Advances in toxicogenomics. Chem.

Res. Toxicol. 18, 403–414.

Heinloth, A. N., Irwin, R. D., Boorman, G. A., Nettesheim, P., Fannin, R. D.,

Sieber, S. O., Snell, M. L., Tucker, C. J., Li, L., Travlos, G. S., et al. (2004).

Gene expression profiling of rat livers reveals indicators of potential adverse

effects. Toxicol. Sci. 80, 193–202.

Hinson, J. A., Reid, A. B., McCullough, S. S., and James, L. P. (2004).

Acetaminophen-induced hepatotoxicity: Role of metabolic activation,

reactive oxygen/nitrogen species, and mitochondrial permeability transition.

Drug Metab. Rev. 36, 805–822.

Jaeschke, H. (1990). Glutathione disulfide formation and oxidant stress during

acetaminophen-induced hepatotoxicity in mice in vivo: The protective effect

of allopurinol. J Pharmacol. Exp. Ther. 255, 935–941.

336 BEYER ET AL.

by guest on July 3, 2014http://toxsci.oxfordjournals.org/

Dow

nloaded from

James, L. P., Mayeux, P. R., and Hinson, J. A. (2003). Acetaminophen-induced

hepatotoxicity. Drug Metab. Dispos. 31, 1499–1506.

Jin, W., Riley, R. M., Wolfinger, R. D., White, K. P., Passador-Gurgel, G., and

Gibson, G. (2001). The contributions of sex, genotype and age to tran-

scriptional variance in Drosophila melanogaster. Nat. Genet. 29, 389–395.

Kaplowitz, N. (2005). Idiosyncratic drug hepatotoxicity. Nat. Rev. DrugDiscov. 4, 489–499.

Kitteringham, N. R., Powell, H., Clement, Y. N., Dodd, C. C., Tettey, J. N.,

Pirmohamed, M., Smith, D. A., McLellan, L. I., and Kevin, P. B. (2000).

Hepatocellular response to chemical stress in CD-1 mice: Induction of early

genes and gamma-glutamylcysteine synthetase. Hepatology 32, 321–333.

Knight, T. R., and Jaeschke, H. (2002). Acetaminophen-induced inhibition of

Fas receptor-mediated liver cell apoptosis: Mitochondrial dysfunction versus

glutathione depletion. Toxicol. Appl. Pharmacol. 181, 133–141.

Liu, Z. X., Han, D., Gunawan, B., and Kaplowitz, N. (2006). Neutrophil

depletion protects against murine acetaminophen hepatotoxicity. Hepatology

43, 1220–1230.

Matsumaru, K., Ji, C., and Kaplowitz, N. (2003). Mechanisms for sensitization

to TNF-induced apoptosis by acute glutathione depletion in murine

hepatocytes. Hepatology 37, 1425–1434.

Mattes, W. B., Pettit, S. D., Sansone, S. A., Bushel, P. R., and Waters, M. D.

(2004). Database development in toxicogenomics: Issues and efforts.

Environ. Health Perspect. 112, 495–505.

Mauton, P.R. (2002). Principles and Practices of Unbiased Stereology: AnIntroduction for Bioscientists. The Johns Hopkins University Press,

Baltimore, MD.

Miles, M. F. (2001). Microarrays: Lost in a storm of data? Nat. Rev. Neurosci. 2,

441–443.

Mitchell, J. R., Jollow, D. J., Potter, W. Z., Gillette, J. R., and Brodie, B. B.

(1973). Acetaminophen-induced hepatic necrosis. IV. Protective role of

glutathione. J. Pharmacol. Exp. Ther. 187, 211–217.

Nelson, E. B. (1980). The pharmacology and toxicology of meta-substituted

acetanilide I: Acute toxicity of 3-hydroxyacetanilide in mice. Res. Commun.Chem. Pathol. Pharmacol. 28, 447–456.

Paules, R. (2003). Phenotypic anchoring: Linking cause and effect. Environ.

Health Perspect. 111, A338–A339.

Pennie, W., Pettit, S. D., and Lord, P. G. (2004). Toxicogenomics in risk

assessment: An overview of an HESI collaborative research program.

Environ. Health Perspect. 112, 417–419.

Powell, C. L., Kosyk, O., Ross, P. K., Schoonhoven, R., Boysen, G.,

Swenberg, J. A., Heinloth, A. N., Boorman, G. A., Cunningham, M. L.,

Paules, R. S., et al. (2006). Phenotypic anchoring of acetaminophen-induced

oxidative stress with gene expression profiles in rat liver. Toxicol. Sci. 93,

213–222.

Qin, L. X., and Kerr, K. F. (2004). Empirical evaluation of data transformations

and ranking statistics for microarray analysis. Nucleic Acids Res. 32,

5471–5479.

Rashed, M. S., Myers, T. G., and Nelson, S. D. (1990). Hepatic protein

arylation, glutathione depletion, and metabolite profiles of acetaminophen

and a non-hepatotoxic regioisomer, 3#-hydroxyacetanilide, in the mouse.

Drug Metab. Dispos. 18, 765–770.

Rashed, M. S., and Nelson, S. D. (1989). Characterization of glutathione

conjugates of reactive metabolites of 3#-hydroxyacetanilide, a non-

hepatotoxic positional isomer of acetaminophen. Chem. Res. Toxicol. 2,

41–45.

Ruepp, S. U., Tonge, R. P., Shaw, J., Wallis, N., and Pognan, F. (2002).

Genomics and proteomics analysis of acetaminophen toxicity in mouse liver.

Toxicol. Sci. 65, 135–150.

Shi, L., Reid, L. H., Jones, W. D., Shippy, R., Warrington, J. A., Baker, S. C.,

Collins, P. J., de Longueville, F., Kawasaki, E. S., Lee, K. Y., et al. (2006).

The MicroArray Quality Control (MAQC) project shows inter- and intra-

platform reproducibility of gene expression measurements. Nat. Biotechnol.

24, 1151–1161.

Toyo’oka, T., Tanabe, J., and Jinno, H. (2001). Determination of rat

hepatocellular glutathione by reversed-phase liquid chromatography with

fluorescence detection and cytotoxicity evaluation of environmental pol-

lutants based on the concentration change. Biomed. Chromatogr. 15,

240–247.

Troyanskaya, O., Cantor, M., Sherlock, G., Brown, P., Hastie, T.,

Tibshirani, R., Botstein, D., and Altman, R. B. (2001). Missing value

estimation methods for DNA microarrays. Bioinformatics 17, 520–525.

Tusher, V. G., Tibshirani, R., and Chu, G. (2001). Significance analysis of

microarrays applied to the ionizing radiation response. Proc. Natl. Acad. Sci.

U.S.A. 98, 5116–5121.

Ulrich, R. G., Rockett, J. C., Gibson, G. G., and Pettit, S. D. (2004). Overview

of an interlaboratory collaboration on evaluating the effects of model

hepatotoxicants on hepatic gene expression. Environ. Health Perspect. 112,

423–427.

Waters, M., Boorman, G., Bushel, P., Cunningham, M., Irwin, R., Merrick, A.,

Olden, K., Paules, R., Selkirk, J., Stasiewicz, S., et al. (2003). Systems

toxicology and the Chemical Effects in Biological Systems (CEBS)

knowledge base. EHP Toxicogenomics 111, 15–28.

Watkins, P. B., Kaplowitz, N., Slattery, J. T., Colonese, C. R., Colucci, S. V.,

Stewart, P. W., and Harris, S. C. (2006). Aminotransferase elevations in

healthy adults receiving 4 grams of acetaminophen daily: A randomized

controlled trial. J. Am. Med. Assoc. 296, 87–93.

Wolfinger, R. D., Gibson, G., Wolfinger, E. D., Bennett, L., Hamadeh, H.,

Bushel, P., Afshari, C., and Paules, R. S. (2001). Assessing gene significance

from cDNA microarray expression data via mixed models. J. Comput. Biol.

8, 625–637.

Zeeberg, B. R., Feng, W., Wang, G., Wang, M. D., Fojo, A. T., Sunshine, M.,

Narasimhan, S., Kane, D. W., Reinhold, W. C., Lababidi, S., et al. (2003).

GoMiner: A resource for biological interpretation of genomic and proteomic

data. Genome Biol. 4, R28.

MULTICENTER STUDY OF ACETAMINOPHEN HEPATOTOXICITY 337

by guest on July 3, 2014http://toxsci.oxfordjournals.org/

Dow

nloaded from