A Decade of Toxicogenomic Research and Its Contribution to Toxicological Science

Division of Bioinformatics and Biostatistics, National Center for Toxicological Research, US Food and Drug Administration, 3900 NCTR Road, Jefferson, AR 72079, USA.
Toxicological Sciences (Impact Factor: 3.85). 07/2012; 130(2). DOI: 10.1093/toxsci/kfs223
Source: PubMed


Toxicogenomics enjoyed considerable attention as a ground-breaking addition to conventional toxicology assays at its inception.
However, the pace at which toxicogenomics was expected to perform has been tempered in recent years. Next to cost, the lack
of advanced knowledge discovery and data mining tools significantly hampered progress in this new field of toxicological sciences.
Recently, two of the largest toxicogenomics databases were made freely available to the public. These comprehensive studies
are expected to stimulate knowledge discovery and development of novel data mining tools, which are essential to advance this
field. In this review, we provide a concise summary of each of these two databases with a brief discussion on the commonalities
and differences between them. We place our emphasis on some key questions in toxicogenomics and how these questions can be
appropriately addressed with the two databases. Finally, we provide a perspective on the future direction of toxicogenomics
and how new technologies such as RNA-Seq may impact this field.


Available from: Juergen Borlak, Dec 31, 2015
A Decade of Toxicogenomic Research and Its Contribution to
Minjun Chen,*
Min Zhang,* Jürgen Borlak, and Weida Tong*
*Division of Bioinformatics and Biostatistics, National Center for Toxicological Research, US Food and Drug Administration, Jefferson, Arkansas 72079; and
Center of Pharmacology and Toxicology, Hannover Medical School, D-30625 Hannover, Germany
These authors contributed equally to thisstudy.
To whom correspondence should be addressed at Division of Bioinformatics and Biostatistics, National Center for Toxicological Research, US Food and
Drug Administration, 3900 NCTR Road, Jefferson, AR 72079. Fax: (870) 543-7854. E-mail:
Received May 9, 2012; accepted July 6, 2012
Toxicogenomics enjoyed considerable attention as a
ground-breaking addition to conventional toxicology assays at
its inception. However, the pace at which toxicogenomics was
expected to perform has been tempered in recent years. Next to
cost, the lack of advanced knowledge discovery and data min-
ing tools signicantly hampered progress in this new eld of
toxicological sciences. Recently, two of the largest toxicogenomics
databases were made freely available to the public. These com-
prehensive studies are expected to stimulate knowledge discovery
and development of novel data mining tools, which are essential to
advance this eld. In this review, we provide a concise summary
of each of these two databases with a brief discussion on the com-
monalities and differences between them. We place our emphasis
on some key questions in toxicogenomics and how these questions
can be appropriately addressed with the two databases. Finally,
we provide a perspective on the future direction of toxicogenomics
and how new technologies such as RNA-Seq may impact this eld.
Key Words: bioinformatics; toxicogenomics; liver; systems
Toxicology has traditionally relied on animal testing to deter-
mine the risk of a chemical compound to humans based on
well-established cytological, physiologic, metabolic, and mor-
phologic endpoints (Suter etal., 2004). The rodent model is com-
monly used to identify toxic substances such as carcinogens and
reproductive toxins. Toxicological studies require a large num-
ber of animals to allow statistically signicant conclusions to be
drawn, and the 2-year carcinogenicity bioassays are routinely
conducted to assess the potential of tumorigenicity in animals
and relevant risk in humans. Consequently, the current toxico-
logical testing is of high cost in terms of time, labor, compound
Disclaimer: The views presented in this article do not necessarily reect
those of the U.S. Food and Drug Administration.
synthesis, and animals used. It becomes a substantial investment
in drug development and environmental health evaluation (Ulrich
and Friend, 2002; Waters and Fostel, 2004). Furthermore, animal
testing is not a fail-safe paradigm (Olson etal., 2000). There is a
need to constantly improve safety testing strategies.
The value of microarray technology was quickly real-
ized in the toxicology community soon after its introduction
in the mid-1990s (DeRisi et al., 1996; Schena et al., 1995;
Wodicka et al., 1997), which has led to a new scientic
subdiscipline termed toxicogenomics (Nuwaysir etal., 1999).
Through integrating genomic technology with bioinformatics,
toxicogenomics has enjoyed widespread attention as an alter-
native means to study the underlying molecular mechanisms
of toxicity and address challenges that are difcult to over-
come by conventional toxicology methods (Nuwaysir et al.,
1999). The broader concept of toxicogenomics encompasses
transcriptomics, proteomics, and metabolomics (Afshari etal.,
2011; Cui and Paules, 2010). In practice, microarray-based
toxicogenomics is still the main application. With microar-
rays, expression levels of tens of thousands of genes can be
simultaneously monitored, permitting the assessment of
alterations in gene expression proles induced by different
compounds or associated with different physiological condi-
tions. Importantly, the large number of genes tested together
provides opportunities to identify gene patterns and signatures
that provide unique insight into a drug’s toxicity that are dif-
cult to recognize by conventional technologies. Therefore,
toxicogenomics was highly expected to revolutionize the
traditional approaches for assessing toxicity (Boverhof and
Zacharewski, 2006) and has been considered as a paradigm
shift in toxicology.
Many studies have demonstrated the value of toxicogenomics
(Ellinger-Ziegelbauer etal., 2008; Fielden etal., 2007; Gerecke
etal., 2009; Huang etal., 2010; Low etal., 2011; Suter etal.,
toxicological sciences 130(2), 217–228 (2012)
Advance Access publication July 12, 2012
Published by Oxford University Press 2012.
by guest on December 30, 2015 from
Page 1
2003; Yang etal., 2006; Zidek etal., 2007). For example, it has
been suggested that toxicogenomics biomarkers can discriminate
drug candidates that have a greater potential to cause toxicity in
susceptible patient populations despite no conventional indicators
of toxicity being observed in preclinical studies (McBurney
et al., 2009, 2012). Equally, more sensitive biomarkers for
early toxicity detection can be derived from a “subtoxic dose”
in which the injury occurs at the molecular but not at the
phenotypic level or in clinical chemistry measures (Lühe etal.,
2005). Ellinger-Ziegelbauer et al. (2008) applied a 14-day
toxicogenomics design to develop a gene signature to distinguish
genotoxic carcinogens from non-genotoxic carcinogens, an
endeavor that usually requires the 2-year bioassay.
In this review, the publication trend in PubMed for toxicog-
enomics was analyzed, which exhibited the plateauing of the
eld. It is likely that the recent release of two large toxicogenom-
ics databases in the public domain will stimulate knowledge
discovery and data mining tool development. Subsequently, a
concise summary of these databases is provided with a brief
discussion on the commonality and differences among them.
The focus is then shifted to key questions in toxicogenomics
and how to appropriately address these questions with these
databases. Finally, a perspective on the future direction of toxi-
cogenomics is provided with a discussion on how RNA-Seq
will impact this eld.
The early success of toxicogenomics stimulated its wide-
spread application. By surveying the publications indexed
by PubMed using the “toxicogenomics” query (bar chart in
Fig.1A), a rapid growth of publications was found during the
period from 2000 to 2006. However, the growth has atted
since then. A decreased trend was observed in the Voluntary
eXploratory Data Submission (VXDS) program of the US
Food and Drug Administration (FDA) in Supplementary g-
ure 1 (Goodsaid et al., 2010), although the specic reasons
are unknown. Some factors have been speculated to be respon-
sible for the fading optimism, including the lack of uniform
designs, multiplicity of normalization and analysis strategies,
questionable reproducibility of microarray data across plat-
forms, absence of data quality control measures and standards,
and lack of effective data sharing and reporting (Boverhof and
Zacharewski, 2006). However, most of these issues are not
toxicogenomics specic and thus are unlikely to be the key
contributors to the slowing progress in toxicogenomics. For
example, the use of microarrays in the genomics area as a whole
has continually ourished (bar chart in Fig. 1B). Moreover,
some factors such as data sharing (Brazma etal., 2001), data
quality control (Shi etal., 2006), genomic biomarker develop-
ment (Shi etal., 2010a), and the need for careful phenotypic
anchoring of genomics data in toxicology (Bammler et al.,
2005; Beyer etal., 2007) have been adequately addressed in
community-wide efforts. Thus, the factors contributing to the
decline should be beyond the technologyitself.
In the period of 2000–2010, by reading abstracts and associ-
ated Medical Subject Heading terms of the published papers
containing the keyword “toxicogenomics,” four categories can
be dened: (1) “Review,” (2) “Mechanism,” (3) “Biomarker,
and (4) “Others.As shown in Figure 2, “Biomarker” and
“Review” notably decreased since 2006, whereas both
“Mechanism” and “Others” continued to increase. The analysis
implies that although toxicogenomics remains a valuable tool
for mechanistic study, its application for biomarker discovery
and development has diminished.
Developing reliable and robust toxicogenomics biomarkers
requires a large number of tested compounds (Ulrich and Friend,
2002), and thus requires signicant funding sources to con-
duct such studies. The FDA-led community wide MicroArray
Quality Control Phase-II project selected three toxicogenom-
ics datasets and three cancer microarray datasets (Shi et al.,
2010b). The sample size of the cancer datasets on average was
signicantly larger than those of the toxicogenomics datasets.
More samples often mean better chances at nding more robust
and better biomarkers, but also indicate a higher cost particu-
larly for animal studies. By comparing the percentage of the
projects funded by the National Institute of Health (NIH) using
the query of “toxicogenomics” against “genomics,” the per-
centage of NIH-funded projects was found to correlate with
the percentage of publications in both elds during the period
of 2000–2010 (solid line in Fig.1). Obviously, the proportion
of NIH-funded projects for toxicogenomics slightly decreased
after 2007 compared with still increased but with a smaller
slop for genomics. The trend difference in NIH-funded pro-
jects between toxicogenomics and genomics is consistent with
that in publications in PubMed, suggesting the lack of funding
plays some role in the fading optimum on toxicogenomics.
Considering the signicant cost involved in a large-scale
toxicogenomics experiment, most reported studies (particu-
larly those from the academic institutes) are small in terms of
the number of compounds tested. This has limited the applica-
tion of extensive and comprehensive bioinformatics approaches
and thus reduced the essential knowledge discovery opportu-
nity to enable interrogation of data beyond ontology and other
“guilt-by-association” considerations in biomarker discovery.
In 2011, two large toxicogenomics databases were made
freely available to the public: the Japanese Toxicogenomics
Project (TGP or TG-GATEs) and DrugMatrix. The TGP
was performed by the Japanese National Institute of Health
Science, the National Institute Biomedical Innovation, and
15 pharmaceutical companies (
by guest on December 30, 2015 from
Page 2
open-tggates/search.html; Uehara etal., 2010). DrugMatrix was
generated by Iconix Pharmaceuticals (Ganter etal., 2005)and
was purchased and publicly released by the U.S. National
Institute of Environmental Health Sciences recently (https://ntp. Both databases place an
emphasis on marketed drugs and liver gene expression data.
Some drugs were tested equally by the TGP and DrugMatrix
effort in multiple doses and treatment durations as well as in
vitro and in vivo study designs. The biggest advantages of these
two databases over the existing publicly available databases
such as GEO (Barrett etal., 2005), ArrayExpress (Brazma etal.,
2003), ArrayTrack (Tong et al., 2003), CEBS (Waters et al.,
2003), and others (Burgoon etal., 2006; Hayes etal., 2005) are:
(1) the uniform experimental design makes the comparative
study between chemical treatments more straightforward and
relevant; (2) the large number of marketed drugs proled
provides an unprecedented opportunity to comprehensively
assess the utility of the microarray-based preclinical models for
predicting human specic toxicity; and (3) both in vitro and in
vivo studies for the same set of chemicals allow determination
of the similarity and difference between two systems in
predictive toxicology.
The TGP (Uehara etal., 2010) tested 170 compounds, mainly
medicinal drugs. Supplemented with some kidney studies, their
main target organ is the liver. Data from some 20,000 arrays
were generated in both in vitro and in vivo experiments. As
summarized in Table 1, the in vivo experiments used male
Sprague Dawley rats with two different experimental designs,
single- and repeated-dose study. For both designs, a 1-week
dose range nding study was performed rst to determine the
FIG. 1. The yearly publications and NIH-funded projects related to toxicogenomics (A) or genomics (B) compared with all publications/projects in a specic
year. Publications and projects were queried from PubMed and Research Portfolio Online Reporting Tools (RePORT) of NIH (
reporter.cfm), respectively. The “genomics” query used keywords of (microarray OR “gene expression proling” OR genomics). The “toxicogenomics” query
used keywords of (toxicogenomics OR ((microarray OR “gene expression proling” OR genomics) AND (toxicology OR carcinogenicity OR carcinogenesis OR
tumorigenicity OR genotoxicity OR non-genotoxicity OR hepatotoxicity OR “liver toxicity” OR nephrotoxicity OR “kidney toxicity” OR “cardiovascular toxic-
ity” OR cardiotoxicity OR immunotoxicity OR “reproductive toxicity” OR “skin sensitization” OR “cutaneous toxicity” OR “endocrine disruption” OR neurotox-
icity OR “hematologic toxicity” OR “pulmonary toxicity” OR “gastrointestinal toxicity” OR “musculoskeletal toxicity” OR “urinary toxicity”)).
by guest on December 30, 2015 from
Page 3
maximum tolerated dose (MTD) of each compound. Based on
these data (e.g., body weight and food changes, organ weight,
and gross pathology), the MTD was set as the highest dose in
the repeated-dose study. In the single-dose study, almost all
compounds were treated at the same dose as their repeated-
dose study to facilitate comparison of gene expression between
single- and repeated-dose studies. Specically, in the single-
dose study, rats were treated in one of the three dose levels (low,
medium, and high at a 1:3:10 ratio) with concurrent controls
and sacriced at 3, 6, 9, or 24 h after a single administration. In
the repeated-dose study, rats were also treated at the three dose
levels (low, medium, and high at a 1:3:10 ratio) with concurrent
controls, but sacriced at 24 h after the last dose of repeated
administration for 3, 7, 14, and 28 days. For each dose/time
group, the gene expression data were analyzed with three
animals per group. Other data obtained include histological
examination, blood chemistry, hematology, body weight, organ
weight, and general symptoms. In addition, two types of in
vitro studies, primary hepatocytes from Sprague Dawley male
rats and human donors, were used. They were treated with
three dose levels (low, medium, and high at ratio 1:5:25) with
concurrent controls and harvested for gene expression analysis
at 2, 8, and 24 h after treatment.
In the case of DrugMatrix (Ganter et al., 2005), over 600
drugs were tested in vivo and in vitro using male Sprague Dawley
rats. Two dose levels were applied in the in vivo experiments;
the fully effective dose, dened as the dose used for treating
disease (converted from human), and the MTD dened as 50%
reduction in weight gain over control after 5days of daily dos-
ing. A maximum of 13 tissues including liver, kidney, heart,
and bone marrow were harvested for gene expression analy-
sis. Microarray studies were done after 0.25, 1, 3, and 5 (some
were replaced with 7/14/30/90) days of rst administration
with 3 biological replicates in each group. The GE CodeLink
RU1 10,000 rat array was applied to all the samples, whereas
Affymetrix RG230-2.0 arrays were only used for ~5000 sam-
ples (about half are liver samples). Overall data from about
10,000 arrays are available from liver tissues. Histopathology
analysis, serum chemistry test, hematology, organ weight, and
gross observation were also performed. The in vitro experi-
ments were done using primary hepatocytes harvested from
male Sprague Dawley rats. Microarray analysis was performed
at 16 and 24 h after the treatment.
Table1 summarizes key information about the two datasets.
They both (1) are drug-centric; (2) primarily focus on the
rat liver; (3) include substantial data for the kidney; (4) use
the same Affymetrix array chip (Rat RG230-2.0) for gene
expression proling (the majority of samples in DrugMatrix
were proled with the CodeLink chip); and importantly,
(5) include conventional toxicological data for predictive
modeling and phenotypic anchoring. The high degree of
commonality between two databases provides an opportunity
for the cross-lab comparison and meta-analysis, including
assessment of technical performance, transferability of
genomic markers, statistically validating each others ndings,
The highest commonalities between TGP and DrugMatrix
databases are that (1) both assayed a large number of marketed
drugs, of which 73 drugs are in common; (2) both used the same
male Sprague Dawley strain; and (3) similar in vitro and in vivo
experimental designs were implemented in both studies. The
high commonality between two databases provides a unique
opportunity for cross-database comparison and meta-analysis.
However, there are two key differences between two databases
that need to be taken into account in the comparative analysis
and meta-analysis. These are: (1) TGP used Affymetrix, whereas
DrugMatrix predominately applied CodeLink, although some
5,000 arrays were also performed on the Affymetrix platform
and (2) the determination of MTD is different.
It is worthwhile to mention another toxicogenomics dataset
that was also released recently from a European effort. The
InnoMed PredTox project (
browse_studies.seam) was carried out among 14 pharmaceu-
tical companies, 2 small-to-mid-sized enterprises, and 3 uni-
versities with a goal of discovering molecular mechanisms of
liver and kidney toxicity (Suter etal., 2011). Drug candidates
from the participating companies that were discontinued at the
preclinical phase due to toxicological ndings in the liver and/
or kidney were selected as test compounds, including 14 pro-
prietary drug candidates and 2 reference toxic compounds (i.e.,
gentamicin and troglitazone). The male Wistar rat was the test
species. The animals were divided into three groups (ve rats
per group and per time point), treated at two dose levels (low
and high) with a concurrent vehicle controls. Dosages were
selected based on pre-existing information with the aim of
achieving target organ toxicity after 2 weeks of exposure with
the high dose. The rats were sacriced after 1, 3, or 14days of
repeated dosing following an overnight fasting period. Serum,
plasma, blood, as well as liver and kidney tissues were col-
lected at necropsy for further investigations. Conventional toxi-
cological endpoints were collected alongside transcriptomics,
proteomics, and metabolomics proles for all animals. Thus,
this is an interesting dataset for systems biology. However, we
decide for not including it in this review due to the fact that it
involves only 16 compounds, and their chemical identity is not
disclosed. In addition, it uses a different rat strain (i.e., Wistar),
which will complicate the cross-database analysis with TGP
and DrugMatrix.
The two publically available databases together provide an
unprecedented opportunity to revisit some of the important
promises in toxicogenomics, paving the way to further advance
toxicogenomics and its role in risk assessment and regulatory
decision-making. Table2 summarizes key questions that have
been postulated in toxicogenomics which could be potentially
by guest on December 30, 2015 from
Page 4
Summary of Two Published Toxicogenomics Databases, i.e., TGP and DrugMatrix
Dataset DrugMatrix (Iconix) TGP (Japan)
Species Male Sprague Dawley rat Male Sprague Dawley rat Human donor
Study type In vivo In vitro In vivo In vitro In vitro
Dose type Repeat dose Single dose Single dose Repeat dose Single dose Single dose
Dose level Control
, low, and high Control
and high Control, low, middle, and high
Sample collection after
0.25, 1, 3, 5days plus some 7+ days 16 and 24 h 3, 6, 9, and 24 h 24 h after the treatment
of 3, 7, 14, and 28days
2, 8, and 24 h 2, 8, and 24 h
Biological replicates Triplicate Duplicate Triplicate Duplicate Duplicate
Microarray Platform Affymetrix RG230-2.0 and CodeLink RU1 arrays Affymetrix RG230-2.0 array Affymetrix human U133
plus 2.0 array
No. compounds tested in liver CodeLink: 343
Affymetrix: 201
CodeLink: 120
Affymetrix: 126
Phase I: 131
Phase II: 27
Phase I: 131
Phase II: 12
Phase I: 131
Phase II: 14
Phase I: 119
Phase II: 39
No. arrays tested in liver CodeLink: 5264
Affymetrix: 2218
CodeLink: 780
Affymetrix: 939
7378 6765 3370 2610
Microarray data available for
other tissues
Kidney, heart, thigh muscle, bone
marrow, spleen, brain, and intestine
Not available Kidney Not available Not available
Items examined Histopathology, body/organ weight,
food consumption, hematology, and
blood chemistry
Not available Histopathology, body/organ weight, food con-
sumption, hematology, and blood chemistry
Cell viability (lactate dehydrogenase release and
DNA contents)
Controls were shared by a group of compounds.
Phase II compounds were released in February, 2012; some compounds tested in phase II missed some doses or time points compared with those in phase I.
by guest on December 30, 2015 from
Page 5
addressed by both databases. Some of these questions are dis-
cussed in more depth in the following subsections.
This review is not meant to exhaustively discuss all the
applications ever proposed in toxicogenomics. We do realize
that some of these toxicogenomics applications are extremely
important, such as toxicogenomics for the mechanistic study.
The toxicogenomics-based mechanistic research has become a
commonly accepted approach to study the molecular mecha-
nisms underlying toxicity (Cui and Paules, 2010), as illustrated
in Figure 2. Many reviews have summarized the mechanistic
studies using toxicogenomics data (Blomme etal., 2009; Foster
et al., 2007; Ulrich and Friend, 2002). Nonetheless, the two
databases provide a repository of ~30,000 liver gene expression
proles associated with ~500 drugs from a wide range of phar-
maceutical indications, thus offering a signicant advantage
over any other datasets in the literature to study mechanisms of
toxicity. We are also not going to discuss the concept and util-
ity of the “reference” toxicogenomics database, an active eld
in the earliest days of toxicogenomics, because both TGP and
DrugMatrix were developed at very beginning to serve that pur-
pose by interrogating the potential toxicity liability of a com-
pound based on its gene signature against these precalculated
signatures using the entire set of compounds in the reference
databases. Another important application in toxicogenomics
which will not be discussed here is to develop microarray-based
predictive models using the short-term toxicogenomics design
for supplementing the 2-year carcinogenicity bioassay for such
as non-genotoxic carcinogenicity, which were also investigated
using both DrugMatrix (Fielden etal., 2007) and TGP (Uehara
etal., 2011).
Replacing Animal Models With In Vitro Assays Coupled
With Toxicogenomics
In addition to ethical considerations, animal studies are
resource-intensive regarding time and cost. Even short-term in
vivo experiments still need a large quantity of compounds (g),
and compound availability is frequently a limiting factor at the
early stage of drug discovery. Meanwhile, animal welfare pre-
sents a strong incentive to reduce testing in animals. Efforts
have been made to explore in vitro systems to supplement or
even replace animal models for safety assessment of food addi-
tives, cosmetic chemicals, etc. Anotable effort in Europe is to
develop alternatives to animal testing by encouraging methods to
“reduce, rene and replace” (3Rs) animal uses under the REACH
(Registration, Evaluation and Authorization of Chemicals) initia-
tive (Abbott, 2005). In the United States, the FDA in its new initi-
atives in advancing regulatory science has promoted the effort of
animal-free approaches with emphasis on testing methods based
on in silico and in vitro approaches (Hamburg, 2011).
Traditional in vitro cytotoxicity assays can reduce the com-
pound requirement to the necessary level, but at that measure,
endpoints such as cell lysis are of limited predictive value. Studies
have demonstrated the potential of toxicogenomics-based in
vitro systems for toxicity assessment. The early endeavor by
Waring etal. (2001) was encouraging, followed by further dem-
onstration using primary cultured rat hepatocytes to identify two
FIG. 2. The trend of yearly publications related to (1) “Review,” (2) “Biomarker,” (3) “Mechanism,” and (4) “Others” among those queried with the single
keyword of “toxicogenomics.” The publications were annotated based on Medical Subject Heading (MeSH) terms or reading abstracts. (1) “Review”—these
publications are classied by MeSH as “review”; (2) “Biomarker”—it relates to development of classiers or signature genes that separate different modes of
action, chemical toxicity, or toxicity types; (3) “Mechanism”—it mainly involves identication of differentially expressed genes between two or more conditions
(e.g., treated vs. control) that are subsequently used to identify altered pathways, gene/protein functions, and/or regulatory networks to understand the underlying
mechanisms of toxicity; and (4) “Others”—the publications do not fall into any of the three aforementioned categories (most are commentary, database related,
or development of novel algorithms).
by guest on December 30, 2015 from
Page 6
toxicological classes (Yang etal., 2006). However, it is not an
unknown fact that in vitro systems are different from in vivo
systems, particularly in metabolic functions that are specically
important for drug toxicity. By comparing various animal-free
liver models against liver tissue in terms of their gene expression
proles, Boess etal. (2003) demonstrated that the gene expres-
sion of cultured primary hepatocytes has moderate similarity
to the liver, whereas two selected cell lines are quite different
from whole liver. The authors suggested that although in vitro
experiments are an indispensable research tool, the limitations
of the experimental design of cell culture studies should be kept
in mind (Jaeschke, 2003).
One aspect of toxicogenomics-based safety assessment is to
determine the “guilt-by-association” between compounds and
then relate such association to toxicity endpoints. Consequently,
whether an in vitro system holds a potential to replace an in
vivo system can be viewed as whether these two systems have
the same “association” patterns. Areliable conclusion of such
analysis is more pronounced if it is performed on a large num-
ber of compounds, for which the current literature evidence is
limited. Fortunately, such an opportunity resides in both TGP
and DrugMatrix where both in vitro and in vivo toxicogenom-
ics studies were carried out for the same sets of compounds.
Specically, the TGP dataset provided 145 compounds tested
in rat primary hepatocytes and 158 compounds tested in human
primary hepatocytes, and the rat in vivo liver toxicity data
for most of these compounds are available. DrugMatrix also
provided gene expression data on rat primary hepatocytes for
about 120 compounds, and about 60 of them also have the rat
in vivo liver gene expression data.
Key Issues/Questions in Toxicogenomics That Could Be Potentially Addressed by TGP and/or DrugMatrix Databases
Predictive toxicology
1. Replacing animal models with in vitro assays coupled with toxicogenomics
This is one of the major efforts, mainly under the REACH initiative in Europe, to develop alternative testing systems
with the aim to improve the screening throughput and to reduce animal use and compounds tested.
2. Predicting toxicity prior to conventional endpoints
A short-term toxicogenomics study based on exposure time within a few days could provide the similar accuracy in
toxicity assessment as this is commonly conducted in, e.g., the 28-day conventional animal study. This application
allows integration of toxicity evaluation into an early stage of drug discovery, which could result in savings in both
time and resources required for the conventional way of toxicity assessment.
3. Supplementing the 2-year bioassay with toxicogenomics
The 2-year bioassay is a norm to assess, e.g., non-genotoxic carcinogenicity. The successful application of
toxicogenomics in this eld will tremendously reduce cost, animal use, and time required for this type of safety
1. Reliable biomarker
It has been hypothesized and demonstrated in some cases that toxicogenomics improves specicity and/or sensitivity
of the current toxicity biomarkers. For example, liver function tests such as elevated serum transaminases (e.g., AST,
ALT) lack sensitivity for drug-induced liver injury. Toxicogenomics could generate more reliable biomarkers for liver
2. “Subtox” biomarker
The “subtox” biomarker announces the existence of gene expression signals at a dose level where the conventional indi-
cators of toxicity (e.g., clinical chemistry or histopathology) are not observed but occurs in a higher dose level. This
has an important clinical implication in early diagnosis for, e.g., acetaminophen-induced liver injury.
3. Translational biomarker
It is hypothesized but with limited demonstration that toxicogenomics biomarkers can discriminate drug candidates that
have the potential to cause toxicity in susceptible patients from drugs that do not have this potential despite no conven-
tional indicators being observed in preclinical studies.
Mechanism and mode of action
1. Reference database for categorizing chemicals
Given that a large number compounds has been proled with microarray, different gene signatures corresponding specic
mechanisms can be predetermined based on the group of compounds sharing the same mechanisms. Subsequently, the
gene expression prole of an unknown agent (e.g., a single compound or a mixture) can be compared against these
signatures, and putative mechanisms or mode of actions could be postulated for the unknown agent.
2. Drug-pair approach to study toxicity mechanisms
It is hypothesized that if two compounds are pharmacologically closely related with similar chemical structure but one
is toxic whereas the other is not, the difference between two should be related to off-target events. With toxicogenom-
ics, the off-target-related pathways can be examined via the side-by-side comparison of the pair to study underlying
mechanisms of toxicity.
3. Phenotypic anchoring
This relates specic alterations in gene expression proles to specic adverse effects dened by conventional parameters
of toxicity. Consequently, the underlying mechanisms eliciting the toxicity can be understood to assist drug candidate
selection and drug development decision making and can also be used to guide the design of follow-up experiments.
by guest on December 30, 2015 from
Page 7
Predict Toxicity Prior to Conventional Endpoints
The concept of molecular events preceding pathological end-
points in time has driven the development of short-term toxi-
cogenomics assays for toxicity that are commonly assayed in
a long-term animal study. For pharmaceutical companies, this
means better prioritization of compounds in drug development
and earlier identication of potential “show-stopping” toxici-
ties (Ruepp etal., 2005). In other words, toxicogenomics can
remove candidates with unacceptable safety margins in the
early drug discovery process, which would result in tremen-
dous savings in both time and resources (Yang etal., 2004).
Since Afshari et al. (1999) proposed that gene expression
changes could occur prior to the onset of pathological symp-
toms, many supportive ndings have been published. Ruepp
et al. (2005) claimed to nd several examples in their study
indicating an earlier detection of toxicological events by tran-
script proling, which precedes pathology. For example, in
the study of tacrine hepatotoxicity, the 6h transcript proles
agged 4 out of 5 animals to have liver damage (cholestasis),
but histological analysis did not nd changes until 24 h. Roth
et al. (2011) reported gene expressions changes in rat liver
treated with a histamine-3 receptor agonist after a single acute
administration. Their analysis of gene expression changes
strongly suggested the development of toxicity while histo-
pathology did not identify a clear liver toxicity; the toxicog-
enomics ndings were conrmed in a 2-week repeated-dose rat
study where prominent liver pathology occurred. However, the
ndings from other studies indicated otherwise. Foster et al.
(2007) argued that in their 3-year research with an analysis of
33 compounds, they found toxicogenomics is no more or even
less sensitive than the traditional endpoints of histopathology
in some cases, particularly when histopathological changes are
focal and multifocal innature.
The controversial ndings in the literature could stem from
the small number of tested compounds studied, which limits
the ability to draw conclusive results. The two large datasets
contain detailed histological and clinical chemistry data, offer-
ing several ways to verify whether toxicogenomics is a more
sensitive tool to detect toxicity compared with the conventional
approach. For example, the TGP repeated-dose study was car-
ried out at 3, 7, 14, and 28days. Thus, the expression patterns
and gene signatures from the shorter treatment duration (i.e., 3,
7, or 14days) can be used against to the histopathological nd-
ings in the 28-day treatment to assess whether the molecular
events in the early time points correlate with the pathological
Translational Biomarkers
The safety assessment in humans relies heavily on ani-
mal studies, which requires a reliable translational biomarker
from the tested animals that predict toxicity in humans. Of the
compounds that cleared preclinical testing, a study showed
that only 71% of all human toxicities can be reasonably pre-
dicted with animal models, and that the accuracy varied among
different organs (Olson et al., 2000). Toxicological changes
occurring in preclinical species are not necessarily relevant to
humans partly because of species differences in cell biology,
physiology, or responses to changes induced by compounds.
Moreover, despite the vigorous safety testing during the drug
development process, rare adverse drug reactions (ADRs) of
new medicinal products cannot be predicted at the time of mar-
ket introduction. Thus, rare ADRs are a leading cause for the
withdrawal of drugs from the market. There are two basic sce-
narios in which cross-species comparisons may be a challenge:
(1) false positive, in which toxicities observed in animal models
are not relevant to humans and (2) false negative, in which tox-
icities observed in human cannot be detected by animal models.
An understanding of the molecular mechanisms of toxicologi-
cal changes can help establish the relevance to humans (Yang
etal., 2004) and thus develop better translational biomarkers to
minimize false positives and false negatives.
The majority of toxicogenomics-based studies, particularly
in the study of drug-induced liver injury (DILI), have focused
on nding signals in animals for drugs that are missed with the
conventional approach (scenario 2: false negative). Drugs (e.g.,
troglitazone) did not cause obvious liver toxicity in preclini-
cal studies or during human clinical trials, but were found to
cause severe liver injury in humans, albeit at a relatively low
incidence, once the drug reached the market and was exposed
to a large number of patients (Kaplowitz, 2005). Whether toxi-
cogenomics could help in identifying the liability of these com-
pounds in animal models is still not clear, but some promising
cases were reported. Lühe etal. (2005) reported a short-term
rat study investigating three different antidiabetic compounds,
in which histopathology ndings indicated that two compounds
caused steatosis, but the third one did not cause any visible
signs of hepatotoxicity. However, rat gene expression prol-
ing classied all three compounds as causing steatosis, and
follow-up studies truly identied the third compound as caus-
ing steatosis in dogs with histopathology showing liver necrosis
and microsteatosis. We also examined a toxicogenomics design
using rats to predict human specic DILI but the results were
disappointing (Zhang etal., 2012).
Although the mechanistic investigation could lead to relevant
translational biomarkers, proling a large number of drugs for
a comparative analysis is another way forward for translational
biomarker discovery. This approach requires a large number of
proled drugs, some with human specic toxicity and others
with no such potential. This requirement can be realized in both
TGP and DrugMatrix but with a signicant challenge. Both TGP
and DrugMatrix contain gene expression data primarily for mar-
keted drugs. Classifying a marketed drug as a toxic agent such
as DILI is a challenge. For example, the distinction between
DILI drugs and non-DILI ones could be a false dichotomy
because the causality of some drugs and liver injury is difcult
to establish. Thus, the utility of both TGP and DrugMatrix for
translational biomarkers largely lies in how the marketed drugs
are accurately annotated for certain toxicity such as DILI. Some
by guest on December 30, 2015 from
Page 8
of this information is available from the FDA drug labeling sys-
tem. For example, we annotated the DILI potential in humans
for 289 drugs of 389 tested compounds in the DrugMatrix data-
set and for 128 drugs of 170 tested compounds in the TGP data-
set (Chen etal., 2011). The human toxicity data coupled with in
vivo/in vitro toxicogenomics data in the TGP and DrugMatrix
databases will benet translational biomarker development for
predicting human adverse events and mechanism studies for
cross-species extrapolation.
Drug-Pair Approach to Study Toxicity Mechanisms
Some drugs are more prone to elicit toxicity than others despite
sharing a similar chemical structure or being in the same thera-
peutic category with the same mode of action. Drug pair repre-
sents a pair of drugs that possess similar chemical structure and
act on the same therapeutic target but display discordant toxic risk
in humans (McBurney etal., 2009, 2012). One of the most strik-
ing examples is ibuprofen versus ibufenac. Both are non-steroidal
anti-inammatory drugs with only one methyl group setting them
apart. Ibuprofen is an over-the-counter drug and it has been on
the market more than 30years with limited case reports for DILI.
In contrast, ibufenac was marketed in 1966 and quickly with-
drawn in February, 1968 due to severe hepatotoxicity. Late clini-
cal studies demonstrated that ibufenac caused elevated alanine
aminotransferase (ALT) in 12/36 patients and jaundice in 5/400
cases. The small difference in chemical structure but huge differ-
ence in human toxicity between a pair of drugs such as ibuprofen/
ibufenac makes the drug-pair approach specically attractive to
identify the mechanisms associating with one but not with the
other in the pair. It has been suspected that if two compounds
are pharmacologically closely related with similar chemical
structures but one is toxic whereas the other is not, the difference
between the two could be related to off-target events. With toxi-
cogenomics, the off-target-related pathways can be exploited via
a side-by-side comparison of a drug pair to delineate underlying
mechanisms differentiating thepair.
We conducted a systematic analysis for both databases to
determine the potential drug pairs for the DILI research. We
dene that two drugs are considered as a pair if they share the
similar chemical structure with structure similarity > 0.5, and
they share the same therapeutic effect that is dened in the
fourth level of the Anatomical Therapeutic Chemical (ATC)
system of the World Health Organization Collaborating Centre
(, and they have a distinct DILI annota-
tion based on the FDA-approved drug label. First, we assessed
the chemical structure similarity between any pair of the com-
pounds using the Tanimoto metric based on the extended con-
nectivity ngerprints (ECFP-4) descriptors calculated using
Pipeline Pilot (version 8.0, Accelrys Inc., San Diego, CA).
As a result, 284 pairs with their Tanimoto similarity scores of
0.5 were identied. Secondly, therapeutic categories of each
pair were compared based on the ATC system where drugs
were classied in groups at ve different levels according to
the organ or system on which they act and their therapeutic,
pharmacological, and chemical properties. Two drugs were con-
sidered as a potential pair if they belong to the same classica-
tion in the fourth level of ATC system (i.e., chemical subgroup).
Finally, the difference in DILI risk between two drugs in a pair
was assessed with their DILI classication based on the DILI
scoring system using the FDA-approved drug labels (Chen
etal., 2011). Two scenarios were considered for a pair with dif-
ferent DILI risk: (1) one was most-DILI-concern, whereas the
other is either less-DILI-concern or no-DILI-concern; and (2)
both drugs are most-DILI-concern but one is a withdrawn drug
and the other is still on the market. The analysis yielded 16 drug
pairs as summarized in Table3, and their chemical structures
were provided in Supplementary gure2. Most identied drug
pairs were from DrugMatrix, which might relate to the fact that
the development of drug classes in DrugMatrix is based on
structure-activity relationship. The availability of gene expres-
sion data for these 16 pairs in TGP and DrugMatrix offers a
distinct opportunity to study DILI-specic mechanisms.
Toxicogenomics enjoyed its widespread attention as a revo-
lutionary alternative to conventional toxicology studies from
its inception. However, the pace at which toxicogenomics was
expected to have signicant impact has atted in recent years.
Several areas of focus related to these two databases can be
foreseen to advance toxicogenomics.
First, the current toxicogenomics studies are still liver-dominant
followed by kidney as evident in both discussed datasets. The
organ complexity at the cellular level is a challenge. DrugMatrix
contains minimum datasets for multi-organ gene expression pro-
les, whereas TGP only contains data for liver and kidney. Thus,
toxicogenomics should progress into integration between organs
and into other organs beyond the liver and kidney. Secondly, turn-
ing data into knowledge remains the challenge but with the devel-
opment of advanced knowledge discovery and data mining tools
using large datasets such as the opportunities offered by these two
large databases, the next innovation cycle in toxicogenomics can
be foreseen with bioinformatics.
Besides the need in developing effective knowledge discovery
tools and entering inter-organ investigation, technology innova-
tions will also impact the science of toxicogenomics. Notably,
whole-transcriptome sequencing using next-generation sequenc-
ing technologies, i.e., RNA-Seq, is a newly emerging technology
for both mapping and quantifying transcriptomes. Compared
with DNA microarrays, RNA-Seq provides a more sensitive
and precise measurement of transcript levels. RNA-Seq has a
very low background signal and does not have an upper limit
for quantication. Consequently, it has a large dynamic range
of expression levels over which transcripts can be detected,
estimated to be greater than 9000-fold (Nagalakshmi et al.,
2008). By contrast, DNA microarrays lack sensitivity for genes
expressed either at low or very high levels and therefore have a
by guest on December 30, 2015 from
Page 9
much smaller dynamic range (hundreds fold level). Additionally,
RNA-Seq has also been shown to be highly accurate for quanti-
fying expression levels and has high levels of reproducibility for
both technical and biological replicates (Marioni et al., 2008).
These improvements over DNA microarrays could provide new
opportunities for toxicogenomics to identify more sensitive bio-
markers, the biggest promise toxicogenomics offered. The most
signicant advantage associated with RNA-Seq is its ability to
identify new types of biomarkers (alternative splicing, mutation,
isoform-specic expression, non-coding RNA, etc.) which is dif-
cult to realize with microarrays. The cost of RNA-Seq currently
is comparable to microarrays and thus may eventually replaced
microarrays and move toxicogenomics forwards toward fulll-
ing its old promises. However, critical assessment of RNA-Seq
to toxicogenomics needs to be carefully conducted to understand
whether the technology really adds the value to enhance our
understanding of the underlying mechanism of toxicity, whether
the technology delivers a better prediction system and whether
the biomarkers from the large microarray databases such as these
discussed in this review can be directly applied to RNA-Seq data.
Currently, the FDA-led community-wide MAQC consortium is
entering the third phase to address these issues. The results and
conclusions from this SEquencing Quality Control (SEQC)
project will shed the light on perspective use of RNA-Seq in
Supplementary data are available online at http://toxsci.
German Federal Ministry for Education and Research as
part of the Virtual Liver Network initiative (Grant number 031
6154to J.B.). J.B. is also the recipient of an ORISE Stipend of
the FDA, which is gratefully acknowledged.
Abbott, A. (2005). Animal testing: More than a cosmetic change. Nature 438,
Sixteen Drug Pairs Identied from the Tested Drugs in TGP and DrugMatrix
Drug pair Tanimoto similarity Therapeutic category Hepatotoxicity Database source
Epirubicin 0.99 Anthracyclines and related substances More toxic DrugMatrix
Doxorubicin Less toxic
Epirubicin 0.92 Anthracyclines and related substances More toxic DrugMatrix
Daunorubicin Less toxic
Erythromycin 0.88 Macrolides More toxic DrugMatrix
Clarithromycin Less toxic
Sulfathiazole 0.88 Sulfonamides More toxic DrugMatrix
Sulfadiazine Less toxic
Clomipramine 0.81 Non-selective monoamine reuptake inhibitors More toxic TGP
Imipramine Less toxic
Erythromycin 0.80 Macrolides More toxic DrugMatrix
Azithromycin Less toxic
Epirubicin 0.79 Anthracyclines and related substances More toxic DrugMatrix
Idarubicin Less toxic
Ibufenac 0.73 Non-steroids anti-inammatory and anti-rheumatic products More toxic DrugMatrix
Ibuprofen Less toxic
Iproniazid 0.71 Hydrazides for treatment of tuberculosis More toxic TGP
Isoniazid Less toxic
Mercaptopurine 0.70 Purine analogs More toxic DrugMatrix
Tioguanine Less toxic
Rifampin 0.64 Antibiotics for treatment of tuberculosis More toxic DrugMatrix
Rifabutin Less toxic
Ciprooxacin 0.62 Fluoroquinolones More toxic DrugMatrix
Lomeoxacin Less toxic
Bromfenac 0.62 Non-steroids anti-inammatory and anti-rheumatic products More toxic DrugMatrix
Diclofenac Less toxic
Rosiglitazone 0.58 Thiazolidinediones More toxic DrugMatrix
Pioglitazone Less toxic
Nortriptyline 0.52 Non-selective monoamine reuptake inhibitors More toxic DrugMatrix
Amitriptyline Less toxic
Troglitazone 0.50 Thiazolidinediones More toxic DrugMatrix
Pioglitazone Less toxic
by guest on December 30, 2015 from
Page 10
Afshari, C. A., Hamadeh, H. K., and Bushel, P. R. (2011). The evolution of
bioinformatics in toxicology: Advancing toxicogenomics. Toxicol. Sci.
120(Suppl.1), S225–S237.
Afshari, C. A., Nuwaysir, E. F., and Barrett, J. C. (1999). Application of com-
plementary DNA microarray technology to carcinogen identication, toxi-
cology, and drug safety evaluation. Cancer Res. 59, 4759–4760.
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 labora-
tories and across platforms. Nat. Methods 2, 351–356.
Barrett, T., Suzek, T. O., Troup, D. B., Wilhite, S. E., Ngau, W. C., Ledoux, P.,
Rudnev, D., Lash, A. E., Fujibuchi, W., and Edgar, R. (2005). NCBI GEO:
Mining millions of expression proles–database and tools. Nucleic Acids
Res. 33(Database issue), D562–D566.
Beyer, R. P., Fry, R. C., Lasarev, M. R., McConnachie, L. A., Meira, L. B.,
Palmer, V. S., Powell, C. L., Ross, P. K., Bammler, T. K., Bradford, B. U.,
etal. (2007). Multicenter study of acetaminophen hepatotoxicity reveals the
importance of biological endpoints in genomic analyses. Toxicol. Sci. 99,
Blomme, E. A., Yang, Y., and Waring, J. F. (2009). Use of toxicogenomics to
understand mechanisms of drug-induced hepatotoxicity during drug discov-
ery and development. Toxicol. Lett. 186, 22–31.
Boess, F., Kamber, M., Romer, S., Gasser, R., Muller, D., Albertini, S., and
Suter, L. (2003). Gene expression in two hepatic cell lines, cultured primary
hepatocytes, and liver slices compared to the in vivo liver gene expression
in rats: Possible implications for toxicogenomics use of in vitro systems.
Toxicol. Sci. 73, 386–402.
Boverhof, D. R., and Zacharewski, T. R. (2006). Toxicogenomics in risk assess-
ment: Applications and needs. Toxicol. Sci. 89, 352–360.
Brazma, A., Hingamp, P., Quackenbush, J., Sherlock, G., Spellman, P.,
Stoeckert, C., Aach, J., Ansorge, W., Ball, C. A., Causton, H. C., etal. (2001).
Minimum information about a microarray experiment (MIAME)-toward
standards for microarray data. Nat. Genet. 29, 365–371.
Brazma, A., Parkinson, H., Sarkans, U., Shojatalab, M., Vilo, J.,
Abeygunawardena, N., Holloway, E., Kapushesky, M., Kemmeren, P., Lara,
G. G., etal. (2003). ArrayExpress—A public repository for microarray gene
expression data at the EBI. Nucleic Acids Res. 31, 68–71.
Burgoon, L. D., Boutros, P. C., Dere, E., and Zacharewski, T. R. (2006).
dbZach: A MIAME-compliant toxicogenomic supportive relational data-
base. Toxicol. Sci. 90, 558–568.
Chen, M., Vijay, V., Shi, Q., Liu, Z., Fang, H., and Tong, W. (2011).
FDA-approved drug labeling for the study of drug-induced liver injury. Drug
Discov. Today 16, 697–703.
Cui, Y., and Paules, R. S. (2010). Use of transcriptomics in understanding
mechanisms of drug-induced toxicity. Pharmacogenomics 11, 573–585.
DeRisi, J., Penland, L., Brown, P. O., Bittner, M. L., Meltzer, P. S., Ray, M.,
Chen, Y., Su, Y. A., and Trent, J. M. (1996). Use of a cDNA microarray to
analyse gene expression patterns in human cancer. Nat. Genet. 14, 457–460.
Ellinger-Ziegelbauer, H., Gmuender, H., Bandenburg, A., and Ahr, H. J. (2008).
Prediction of a carcinogenic potential of rat hepatocarcinogens using toxi-
cogenomics analysis of short-term in vivo studies. Mutat. Res. 637, 23–39.
Fielden, M. R., Brennan, R., and Gollub, J. (2007). A gene expression bio-
marker provides early prediction and mechanistic assessment of hepatic
tumor induction by nongenotoxic chemicals. Toxicol. Sci. 99, 90–100.
Foster, W. R., Chen, S. J., He, A., Truong, A., Bhaskaran, V., Nelson, D. M.,
Dambach, D. M., Lehman-McKeeman, L. D., and Car, B. D. (2007). A ret-
rospective analysis of toxicogenomics in the safety assessment of drug can-
didates. Toxicol. Pathol. 35, 621–635.
Ganter, B., Tugendreich, S., Pearson, C. I., Ayanoglu, E., Baumhueter, S.,
Bostian, K. A., Brady, L., Browne, L. J., Calvin, J. T., Day, G. J., et al.
(2005). Development of a large-scale chemogenomics database to improve
drug candidate selection and to understand mechanisms of chemical toxicity
and action. J. Biotechnol. 119, 219–244.
Gerecke, D. R., Chen, M., Isukapalli, S. S., Gordon, M. K., Chang, Y. C.,
Tong, W., Androulakis, I. P., and Georgopoulos, P. G. (2009). Differential
gene expression proling of mouse skin after sulfur mustard exposure:
Extended time response and inhibitor effect. Toxicol. Appl. Pharmacol.
234, 156–165.
Goodsaid, F. M., Amur, S., Aubrecht, J., Burczynski, M. E., Carl, K., Catalano,
J., Charlab, R., Close, S., Cornu-Artis, C., Essioux, L., etal. (2010). Voluntary
exploratory data submissions to the US FDA and the EMA: Experience and
impact. Nat. Rev. Drug Discov. 9, 435–445.
Hamburg, M. A. (2011). Advancing regulatory science. Science 331, 987.
Hayes, K. R., Vollrath, A. L., Zastrow, G. M., McMillan, B. J., Craven, M.,
Jovanovich, S., Rank, D. R., Penn, S., Walisser, J. A., Reddy, J. K., et al.
(2005). EDGE: Acentralized resource for the comparison, analysis, and dis-
tribution of toxicogenomic information. Mol. Pharmacol. 67, 1360–1368.
Huang, J., Shi, W., Zhang, J., Chou, J. W., Paules, R. S., Gerrish, K., Li, J., Luo,
J., Wolnger, R. D., Bao, W., etal. (2010). Genomic indicators in the blood
predict drug-induced liver injury. Pharmacogenomics J. 10, 267–277.
Jaeschke, H. (2003). Are cultured liver cells the right tool to investigate mecha-
nisms of liver disease or hepatotoxicity? Hepatology 38, 1053–1055.
Kaplowitz, N. (2005). Idiosyncratic drug hepatotoxicity. Nat. Rev. Drug
Discov. 4, 489–499.
Low, Y., Uehara, T., Minowa, Y., Yamada, H., Ohno, Y., Urushidani, T.,
Sedykh, A., Muratov, E., Kuz’min, V., Fourches, D., etal. (2011). Predicting
drug-induced hepatotoxicity using QSAR and toxicogenomics approaches.
Chem. Res. Toxicol. 24, 1251–1262.
Lühe, A., Suter, L., Ruepp, S., Singer, T., Weiser, T., and Albertini, S. (2005).
Toxicogenomics in the pharmaceutical industry: Hollow promises or real
benet? Mutat. Res. 575, 102–115.
Marioni, J. C., Mason, C. E., Mane, S. M., Stephens, M., and Gilad, Y. (2008).
RNA-seq: An assessment of technical reproducibility and comparison with
gene expression arrays. Genome Res. 18, 1509–1517.
McBurney, R. N., Hines, W. M., Von Tungeln, L. S., Schnackenberg, L. K.,
Beger, R. D., Moland, C. L., Han, T., Fuscoe, J. C., Chang, C. W., Chen,
J. J., etal. (2009). The liver toxicity biomarker study: Phase Idesign and
preliminary results. Toxicol. Pathol. 37, 52–64.
McBurney, R. N., Hines, W. M., Von Tungeln, L. S., Schnackenberg, L. K., Beger,
R. D., Moland, C. L., Han, T., Fuscoe, J. C., Chang, C. W., Chen, J. J., etal.
(2012). The liver toxicity biomarker study phase I: Markers for the effects of
tolcapone or entacapone. Toxicol. Pathol, doi: 10.1177/0192623312444026.
Nagalakshmi, U., Wang, Z., Waern, K., Shou, C., Raha, D., Gerstein, M.,
and Snyder, M. (2008). The transcriptional landscape of the yeast genome
dened by RNA sequencing. Science 320, 1344–1349.
Nuwaysir, E. F., Bittner, M., Trent, J., Barrett, J. C., and Afshari, C. A. (1999).
Microarrays and toxicology: The advent of toxicogenomics. Mol. Carcinog.
24, 153–159.
Olson, H., Betton, G., Robinson, D., Thomas, K., Monro, A., Kolaja, G.,
Lilly, P., Sanders, J., Sipes, G., Bracken, W., etal. (2000). Concordance of
the toxicity of pharmaceuticals in humans and in animals. Regul. Toxicol.
Pharmacol. 32, 56–67.
Roth, A., Boess, F., Landes, C., Steiner, G., Freichel, C., Plancher, J. M., Raab,
S., de Vera Mudry, C., Weiser, T., and Suter, L. (2011). Gene expression-based
in vivo and in vitro prediction of liver toxicity allows compound selection at
an early stage of drug development. J. Biochem. Mol. Toxicol. 25, 183–194.
Ruepp, S., Boess, F., Suter, L., de Vera, M. C., Steiner, G., Steele, T., Weiser, T.,
and Albertini, S. (2005). Assessment of hepatotoxic liabilities by transcript
proling. Toxicol. Appl. Pharmacol. 207(Suppl.2), 161–170.
Schena, M., Shalon, D., Davis, R. W., and Brown, P. O. (1995). Quantitative
monitoring of gene expression patterns with a complementary DNA micro-
array. Science 270, 467–470.
by guest on December 30, 2015 from
Page 11
Shi, L., Campbell, G., Jones, W. D., Campagne, F., Wen, Z., Walker, S. J., Su,
Z., Chu, T. M., Goodsaid, F. M., Pusztai, L., etal. (2010b). The MicroArray
Quality Control (MAQC)-II study of common practices for the development
and validation of microarray-based predictive models. Nat. Biotechnol. 28,
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.
Shi, Q., Hong, H., Senior, J., and Tong, W. (2010a). Biomarkers for drug-induced
liver injury. Expert Rev. Gastroenterol. Hepatol. 4, 225–234.
Suter, L., Babiss, L. E., and Wheeldon, E. B. (2004). Toxicogenomics in predic-
tive toxicology in drug development. Chem. Biol. 11, 161–171.
Suter, L., Haiker, M., De Vera, M. C., and Albertini, S. (2003). Effect of
two 5-HT6 receptor antagonists on the rat liver: A molecular approach.
Pharmacogenomics J. 3, 320–334.
Suter, L., Schroeder, S., Meyer, K., Gautier, J. C., Amberg, A., Wendt, M.,
Gmuender, H., Mally, A., Boitier, E., Ellinger-Ziegelbauer, H., etal. (2011).
EU framework 6 project: Predictive toxicology (PredTox)—Overview and
outcome. Toxicol. Appl. Pharmacol. 252, 73–84.
Tong, W., Cao, X., Harris, S., Sun, H., Fang, H., Fuscoe, J., Harris, A.,
Hong, H., Xie, Q., Perkins, R., et al. (2003). ArrayTrack–support-
ing toxicogenomic research at the U.S. Food and Drug Administration
National Center for Toxicological Research. Environ. Health Perspect.
111, 1819–1826.
Uehara, T., Minowa, Y., Morikawa, Y., Kondo, C., Maruyama, T., Kato, I.,
Nakatsu, N., Igarashi, Y., Ono, A., Hayashi, H., et al. (2011). Prediction
model of potential hepatocarcinogenicity of rat hepatocarcinogens using
a large-scale toxicogenomics database. Toxicol. Appl. Pharmacol. 255,
Uehara, T., Ono, A., Maruyama, T., Kato, I., Yamada, H., Ohno, Y., and
Urushidani, T. (2010). The Japanese toxicogenomics project: Application of
toxicogenomics. Mol. Nutr. Food Res. 54, 218–227.
Ulrich, R., and Friend, S. H. (2002). Toxicogenomics and drug discovery: Will
new technologies help us produce better drugs? Nat. Rev. Drug Discov. 1,
Waring, J. F., Ciurlionis, R., Jolly, R. A., Heindel, M., and Ulrich, R. G. (2001).
Microarray analysis of hepatotoxins in vitro reveals a correlation between
gene expression proles and mechanisms of toxicity. Toxicol. Lett. 120,
Waters, M., Boorman, G., Bushel, P., Cunningham, M., Irwin, R., Merrick, A.,
Olden, K., Paules, R., Selkirk, J., Stasiewicz, S., etal. (2003). Systems toxi-
cology and the Chemical Effects in Biological Systems (CEBS) knowledge
base. EHP Toxicogenom. 111(1T), 15–28.
Waters, M. D., and Fostel, J. M. (2004). Toxicogenomics and systems toxicol-
ogy: Aims and prospects. Nat. Rev. Genet. 5, 936–948.
Wodicka, L., Dong, H., Mittmann, M., Ho, M. H., and Lockhart, D. J. (1997).
Genome-wide expression monitoring in Saccharomyces cerevisiae. Nat.
Biotechnol. 15, 1359–1367.
Yang, Y., Abel, S. J., Ciurlionis, R., and Waring, J. F. (2006). Development of
a toxicogenomics in vitro assay for the efcient characterization of com-
pounds. Pharmacogenomics 7, 177–186.
Yang, Y., Blomme, E. A., and Waring, J. F. (2004). Toxicogenomics in drug
discovery: From preclinical studies to clinical trials. Chem. Biol. Interact.
150, 71–85.
Zhang, M., Chen, M., and Tong, W. (2012). Is toxicogenomics a more reliable
and sensitive biomarker than conventional indicators from rats to predict
drug-induced liver injury in humans? Chem. Res. Toxicol. 25, 122–129.
Zidek, N., Hellmann, J., Kramer, P. J., and Hewitt, P. G. (2007). Acute hepato-
toxicity: Apredictive model based on focused illumina microarrays. Toxicol.
Sci. 99, 289–302.
by guest on December 30, 2015 from
Page 12
  • Source
    • "Acute, sublethal effects were also investigated using genes involved in the cellular stress response and endocrine regulation as endpoints . Toxicogenomics is gaining momentum because information concerning genes respond to different chemical compounds is highly powerful for identifying molecular mechanisms and cellular pathways specific to the mode of action of a number of toxicants and drugs (Chen et al., 2012). "
    [Show abstract] [Hide abstract] ABSTRACT: Several organic UV filters have hormonal activity in vertebrates, as demonstrated in fishes, rodents and human cells. Despite the accumulation of filter contaminants in aquatic systems, research on their effects on the endocrine systems of freshwaters invertebrates is scarce. In this work, the effects of five frequently used UV filters were investigated in embryos and larvae of Chironomus riparius, which is a reference organism in ecotoxicology. LC50 values for larvae as well as the percentage of eclosion of eggs were determined following exposures to: octyl-p-methoxycinnamate (OMC) also known as 2-ethylhexyl-4-methoxycinnamate (EHMC); 4-methylbenzylidene camphor (4MBC); 4-hydroxybenzophenone (4HB); octocrylene (OC); and octyldimethyl-p-aminobenzoate (OD-PABA). To assess sublethal effects, expression levels of the genes coding for the ecdysone receptor (EcR) and heat shock protein HSP70 were investigated as biomarkers for endocrine and stress effects at the cellular level. Life-stage-dependent sensitivity was found. In embryos, all of the UV filters provoked a significant overexpression of EcR at 24h after exposure. OC, 4MBC and OD-PABA also triggered transcriptional activation of the hsp70 stress gene in embryos. In contrast, in larvae, only 4MBC and OMC/EHMC increased EcR and hsp70 mRNA levels and OD-PABA upregulated only the EcR gene. These results revealed that embryos are particularly sensitive to UV filters, which affect endocrine regulation during development. Most UV filters also triggered the cellular stress response, and thus exhibit proteotoxic effects. The differences observed between embryos and larvae and the higher sensitivity of embryos highlight the importance of considering different life stages when evaluating the environmental risks of pollutants, particularly when analyzing endocrine effects.
    Full-text · Article · Jul 2016 · Science of The Total Environment
    • "To this end, so-called omics methods are being deployed ; omics approaches are sometimes viewed as " high-throughput " , but it can be argued that even though vast amounts of data are generated , this is not necessarily done in a high-throughput manner, as the data analysis can be demanding. In this context, toxicogenomics is a generic term commonly referring to molecular approaches to screen for alterations in gene expression and products of protein function in living systems subjected to toxicological challenge (Chen et al., 2012). The term comprises transcriptomics, proteomics, and other more recent approaches such as metabolomics and epigenomics, which are, in essence, related to different steps along the complex chain of events of gene expression and its consequences. "
    [Show abstract] [Hide abstract] ABSTRACT: Engineered nanomaterials are being developed for a variety of technological applications. However, the increasing use of nanomaterials in society has led to concerns about their potential adverse effects on human health and the environment. During the first decade of nanotoxicological research, the realization has emerged that effective risk assessment of the multitudes of new nanomaterials would benefit from a comprehensive understanding of their toxicological mechanisms, which is difficult to achieve with traditional, low-throughput, single end-point oriented approaches. Therefore, systems biology approaches are being progressively applied within the nano(eco)toxicological sciences. This novel paradigm implies that the study of biological systems should be integrative resulting in quantitative and predictive models of nanomaterial behaviour in a biological system. To this end, global 'omics' approaches with which to assess changes in genes, proteins, metabolites, etc are deployed allowing for computational modelling of the biological effects of nanomaterials. Here, we highlight omics and systems biology studies in nanotoxicology, aiming towards the implementation of a systems nanotoxicology and mechanism-based risk assessment of nanomaterials.
    No preview · Article · Dec 2015 · Toxicology and Applied Pharmacology
    • "Toxicogenomics combines toxicology with omics technologies to investigate the mechanisms underlying a toxicological response (Waters and Fostel, 2004). Microarray-based gene expression profiling still remains the core technological platform in toxicogenomic research (Chen et al., 2012). It is a well-established technique and provides genome-wide information on transcriptomic changes (Shi et al., 2006) and is used to obtain better insight in the molecular mechanisms underlying drug-induced liver toxicity (Cheng et al., 2011; Cui and Paules, 2010; Nuwaysir et al., 1999). "
    [Show abstract] [Hide abstract] ABSTRACT: In order to improve attrition rates of candidate-drugs there is a need for a better understanding of the mechanisms underlying drug-induced hepatotoxicity. We aim to further unravel the toxicological response of hepatocytes to a prototypical cholestatic compound by integrating transcriptomic and metabonomic profiling of HepG2 cells exposed to Cyclosporin A. Cyclosporin A exposure induced intracellular cholesterol accumulation and diminished intracellular bile acid levels. Performing pathway analyses of significant mRNAs and metabolites separately and integrated, resulted in more relevant pathways for the latter. Integrated analyses showed pathways involved in cell cycle and cellular metabolism to be significantly changed. Moreover, pathways involved in protein processing of the endoplasmic reticulum, bile acid biosynthesis and cholesterol metabolism were significantly affected. Our findings indicate that an integrated approach combining metabonomics and transcriptomics data derived from representative in vitro models, with bioinformatics can improve our understanding of the mechanisms of action underlying drug-induced hepatotoxicity. Furthermore, we showed that integrating multiple omics and thereby analyzing genes, microRNAs and metabolites of the opposed model for drug-induced cholestasis can give valuable information about mechanisms of drug-induced cholestasis in vitro and therefore could be used in toxicity screening of new drug candidates at an early stage of drug discovery. Copyright © 2015. Published by Elsevier Ltd.
    No preview · Article · Apr 2015 · Toxicology in Vitro
Show more