Importance of dosage standardization for interpreting transcriptomal signature profiles: Evidence from studies of xenoestrogens

Abstract

To obtain insights into similarities and differences in the biological actions of related drugs or toxic agents, their transcriptomal signature profiles (TSPs) have been examined in a large number of studies. However, many such reports did not provide proper justification for the dosage criteria of each agent. Using a well characterized cell culture model of estrogen-dependent proliferation of MCF7 human breast cancer cells, we demonstrate how different approaches to dosage standardization exert critical influences on TSPs, leading to different and even conflicting conclusions. Using quantitative cellular response (QCR)-based dosage criteria, TSPs were determined by Affymetrix microarray when cells were proliferating at comparable rates in the presence of various estrogens. We observed that TSPs of the xenoestrogens (e.g., genistein or bisphenol A) were clearly different from the TSP of 17β-estradiol; namely, the former strongly enhanced expression of genes involved in mitochondrial oxidative phosphorylation, whereas the latter showed minimal effects. In contrast, TSPs for genistein and 17β-estradiol were indistinguishable by using the marker gene expression-based dosage criteria, conditions in which there was comparable expression of the mRNA transcripts for the estrogen-inducible WISP2 gene. Our findings indicate that determination and interpretation of TSPs in pharmacogenomic and toxicogenomic studies that examine the transcriptomal actions of related agents by microarray require a clear rationale for the dosage standardization method to be used. We suggest that future studies involving TSP analyses use quantitative and objective dosage standardization methods, such as those with quantitative cellular response or marker gene expression-based dosage criteria.

• breast cancer
• estrogen
• pharmacogenomics
• toxicogenomics
• transcriptome

Full text

Importance of dosage standardization for interpreting
transcriptomal signature profiles: Evidence from
studies of xenoestrogens
Toshi Shioda*
†
, Jessica Chesnes*, Kathryn R. Coser*, Lihua Zou
‡
, Jingyung Hur*, Kathleen L. Dean*,
Carlos Sonnenschein
§
, Ana M. Soto
§
, and Kurt J. Isselbacher*
†
*Department of Tumor Biology and Molecular Profiling Laboratory, Massachusetts General Hospital Center for Cancer Research, Charlestown, MA 02129;
‡
Division of Computational Biology, Harvard Bauer Center for Genomics Research, Cambridge, MA 02138; and
§
Department of Anatomy and Cell Biology,
Tufts University School of Medicine, Boston, MA 02111
Contributed by Kurt J. Isselbacher, June 26, 2006
To obtain insights into similarities and differences in the biological
actions of related drugs or toxic agents, their transcriptomal
signature profiles (TSPs) have been examined in a large number of
studies. However, many such reports did not provide proper
justification for the dosage criteria of each agent. Using a well
characterized cell culture model of estrogen-dependent prolifera-
tion of MCF7 human breast cancer cells, we demonstrate how
different approaches to dosage standardization exert critical in-
fluences on TSPs, leading to different and even conflicting conclu-
sions. Using quantitative cellular response (QCR)-based dosage
criteria, TSPs were determined by Affymetrix microarray when cells
were proliferating at comparable rates in the presence of various
estrogens. We observed that TSPs of the xenoestrogens (e.g.,
genistein or bisphenol A) were clearly different from the TSP of
17
␤
-estradiol; namely, the former strongly enhanced expression of
genes involved in mitochondrial oxidative phosphorylation,
whereas the latter showed minimal effects. In contrast, TSPs for
genistein and 17
␤
-estradiol were indistinguishable by using the
marker gene expression-based dosage criteria, conditions in which
there was comparable expression of the mRNA transcripts for the
estrogen-inducible WISP2 gene. Our findings indicate that deter-
mination and interpretation of TSPs in pharmacogenomic and
toxicogenomic studies that examine the transcriptomal actions of
related agents by microarray require a clear rationale for the
dosage standardization method to be used. We suggest that future
studies involving TSP analyses use quantitative and objective
dosage standardization methods, such as those with quantitative
cellular response or marker gene expression-based dosage criteria.
breast cancer 兩 estrogen 兩 pharmacogenomics 兩 toxicogenomics 兩
transcriptome
M
icroarray determinations of transcriptomal changes in-
duced by biologically active substances have bec ome in-
creasingly popular in pharmacogenomics research. For example,
a large number of studies have reported time- and dose-
dependent aspects of transcriptomal changes induced by estro-
gen ic and antiestrogenic agents in cell culture and an imal models
(1–17). To obtain insights into similarities and differences in the
ef fects of multiple hor monal agents, it has been common
practice to determine their transcriptomal signature profiles
(TSPs; ref. 18). TSP analyses of antiestrogens such as tamoxifen,
raloxifen, or fulvestrant have suggested a molecular basis for
their partial estrogen-like activities (1–6). Studies of TSPs for
17
␤
-estradiol (E
2
) and xenoestrogens have provided insights into
a genomics-based classification of estrogens and their mecha-
n isms of action (6–9).
To determine TSPs of hormonally active agents, estrogen
t arget cells and tissues such as estrogen-dependent human
breast cancer cell cultures (1–7, 12–14) and the rodent uterus
(8, 9, 16, 17) have been used as standards in in vitro and in vivo
model systems, respectively. For both systems, the time-
dependent aspects of the agent-induced transcriptomal
changes have been well characterized (3, 5, 9, 12–14, 16, 17).
In contrast, relatively few studies have systematically examined
the dosage-dependent aspects of the hor mone-induced tran-
scriptomal changes (6, 19). In many studies, TSPs of estrogens
and antiestrogens were deter mined by using a relatively high
single dose of each agent. For example, whereas the prolifer-
ative effect of E
2
on estrogen-t arget MCF7 human breast
cancer cells is saturated at ⬇0.1 nM (12, 19), w ith few
exceptions (12, 19) most studies have determined TSPs of this
model using E
2
at 1–10 nM (1–7, 10, 13, 14). However, we
observed that the transcriptomal ef fects of E
2
on MCF7 cells
showed a strong dosage dependency, and that high concen-
trations of E
2
(⬎0.1 nM) induced ex pression of mRNA
transcripts for TGF-
␣
and stromal cell-derived factor-1 (SDF-
1), which c ould not be induced by lower c oncentrations of E
2
,
yet these lower concentrations induced maximal cell prolifer-
ation (19). Thus, interpret ation of TSP dat a using a single
saturating c oncentration of a hor monal agent may have sig-
n ificant limit ations.
In the present study, by characterizing the transcriptomal
ef fects of E
2
and xenoestrogens in MCF7 cells, we demonstrate
the importance of using objective and quantitative criteria to
st andardize the dosage of hormonal agents in TSP-based phar-
mac ogenomics. We propose two approaches to dosage st andard-
ization and discuss their advantages and limitations.
Results
To characterize the dose-dependent aspects of estrogen effects
on MCF7 cell proliferation, we measured the increase in cell
yield during 120 h of exposure to varying c oncentrations of E
2
and of xenoestrogens (Fig. 1A). E
2
, the natural an imal estrogen,
strongly enhanced MCF7 cell proliferation at 3–100 pM.
Bisphenol A and p-nonylphenol, representative plastic-related
xenoestrogens, enhanced MCF7 cell proliferation at 3–100 nM.
Gen istein and daidzein, representative isoflavone phytoestro-
gens that bind preferably to estrogen receptor (ER)
␤
(20), were
weak xenoestrogens that stimulated MCF7 cell proliferation at
c oncentrations of ⱖ0.1
␮
M. Tris (4-hydroxyphenyl)-4-propyl-1-
pyrazole (PPT), a synthetic selective agonist for ER
␣
, increased
Conflict of interest: no conflicts declared.
Freely available online through the PNAS open access option.
Abbreviations: E
2
,17
␤
-estradiol; MGE, marker gene expression; PAR, pairwise angle ratio;
PPT, 1,3,5-tris(4-hydroxyphenyl)-4-propyl-1-pyrazole; QCR, quantitative cellular response;
TSP, transcriptomal signature profile.
Data deposition: The microarray data presented in this study have been deposited in the
Gene Expression Omnibus (GEO) database, www.ncbi.nlm.nih.gov兾geo (accession no.
GSE5200).
†
To whom correspondence may be addressed. E-mail: tshioda@partners.org or kisselbacher@
partners.org.
© 2006 by The National Academy of Sciences of the USA
www.pnas.org兾cgi兾doi兾10.1073兾pnas.0605341103 PNAS
兩
August 8, 2006
兩
vol. 103
兩
no. 32
兩
12033–12038
GENETICS

MCF7 cell proliferation relatively strongly at 0.1–10 nM. Our
120-h assay clearly demonstrated a quantitative dif ference be-
t ween 30 pM E
2
and 3
␮
M genistein on MCF7 cell proliferation
ef fects. However, there was no observed difference in the ef fects
of 30 pM E
2
and 10
␮
M genistein. Assays of cell proliferation at
shorter time periods of estrogen exposure (e.g., 48 h) were not
quantit atively dif ferent in distinguishing the effects of 30 pM E
2
and 3
␮
M genistein (data not shown).
Our preliminary ex periments on the time-dependent aspects
of the transcriptomal effects of E
2
in MCF7 cells revealed that
many known estrogen-inducible genes reached their maximum
levels of expression only after 48 h of exposure. A previous study
by us showed consistent reproducibility of the transcriptomal
profiling of MCF7 cells when exposed to varying concentrations
of E
2
for 48 h (19). However, MCF7 cell transcriptomes deter-
mined after 120 h of exposure to E
2
showed poorer reproduc-
ibilit y, presumably reflecting that cells were no longer synchro-
n ized in their transcriptomal responses to E
2
(dat a not shown).
Therefore, although the estrogen effects on cell proliferation
were assayed after 120 h of exposure, transcriptomal effects were
deter mined at 48 h.
To determine TSPs of xenoe strogens using DNA microarrays, we
defined two criteria for dosage standardization. The first, quanti-
tative cellular response (QCR)-based dosage criteria, identified
doses that induce an equal QCR. Using the QCR criteria, estrogen
concentrations that supported MCF7 cell proliferation with a
strength equal to 30 pM E
2
were identified, namely, 3 nM PPT, 10
nM bisphenol A, 10 nM p-nonylphenol, 1
␮
M daidzein, and 10
␮
M
genistein (Fig. 1 A, red circles). Interestingly, even though 30 pM E
2
and 10
␮
M genistein equally enhanced MCF7 cell proliferation, the
latter induced expre ssion of the mRNA transcripts for WISP2
⬇3-fold more strongly than the former (Fig. 4B, which is published
as supporting information on the PNAS web site). Because WISP2
is a representative E
2
-inducible gene whose expression is strongly
enhanced from 1 to 60 pM E
2
in MCF7 cells (19), our observation
demonstrated a significant discrepancy between the proliferative
and the transcriptional effects of estrogens. By exposing MCF7 cells
to varying concentrations of genistein, we observed that 30 pM E
2
and 3
␮
M genistein equally induced WISP2 mRNA (48 h of
exposure; Fig. 1B). However, at this concentration, genistein stim-
ulated MCF7 cell proliferation only half as much as 30 pM E
2
(120
h of exposure; Fig. 1 A). Based on these observations, we defined
the second criteria of estrogen dosage standardization, namely,
marker gene expression (MGE)-based dosage criteria that identi-
fied concentrations of E
2
(30 pM) and genistein (3
␮
M) that equally
induced mRNA for WISP2 (Fig. 1A, blue circles).
The xenoestrogen TSPs were determined by using Af fymetrix
(Sant a Clara, CA) GeneChip DNA microarrays by exposing
MCF7 cells to these agents for 48 h at the dosage chosen by the
QCR and MGE criteria. Initial characterization of the xe-
noestrogen TSPs with 20 represent ative estrogen-inducible
genes (19) revealed that the transcriptomal effects of E
2
and
gen istein at the dosage deter mined by the MGE criteria were
remark ably similar (Fig. 4A). In c ontrast, transcriptomal effects
of E
2
and the xenoestrogens (including genistein) on these
representative marker genes were clearly different when their
dosage was determined by the QCR criteria (Fig. 4 B–D). The
ex pression profiles of these 20 marker genes were similar be-
t ween genistein and daidzein (Fig. 4B ). Profiles for bisphenol A
and p-nonylphenol also showed apparent similarit y (Fig. 4C).
The profile for PPT fit between the phytoestrogen pattern and
the plastic estrogen pattern (Fig. 4D). The remarkable differ-
ences observed among the TSPs using the QCR criteria sug-
gested that different estrogens have discrete transcriptomal
actions that can be used to discriminate the E
2
, soybean isofla-
vones, and man-made chemical xenoestrogens even when their
phar macological or toxicological effects are comparable. In
c ontrast, the TSPs of E
2
and genistein, which were clearly
dif ferent using the QCR criteria, were strikingly similar using the
MGE criteria, suggesting that the transcriptomal actions of E
2
and genistein were essentially identical. These conflicting ob-
servations exemplify the significant but often-overlooked risk of
interpreting the transcriptomal profiling data beyond the limi-
t ations of the experimental conditions.
The TSPs of estrogens were further characterized by hierar-
chical clustering analysis involving 1,675 informative genes
whose expression was strongly affected by at least one of the
estrogens (Fig. 2A). This analysis involved 24 TSP sets repre-
senting seven dif ferent estrogens and control vehicle. The TSPs
of each agent were determined by three independently per-
for med cell culture experiments and were grouped into five
major classes. Consistent with the in itial characterization involv-
ing the 20 selected genes, E
2
and gen istein at the concentrations
deter mined by MGE criteria [shown as genistein(L) in Fig. 2 A]
for med a single class (TSP names shown in red). The t wo
phy toestrogens (gen istein and daidzein) formed a separate class
(TSP names in blue), and the t wo plastic-related chemicals
(bisphenol A and p-nonylphenol) formed another class (TSP
names in green). PPT and vehicle (0.1% ethanol) showed
patterns different from those observed for the above-mentioned
estrogens (TSP names in black or purple, respectively).
A
B
Fig. 1. Determination of xenoestrogen concentrations based on QCR and
MGE-based dosage criteria. (A) Estrogen-dependent proliferation of MCF7
cells. Effects of xenoestrogens on cell proliferation were determined by the
120-h E-SCREEN assay, as described (19). The number of cells in the wells
supplemented with 100 pM E
2
was determined by a Coulter (Fullerton, CA) cell
counter and defined as 100% relative cell number, which corresponded to
⬇50% confluence. Numbers of cells cultured in the presence of the xenoestro-
gens were expressed relative to this condition. Each point represents mean ⫾
SEM (n ⫽ 5). Xenoestrogen concentrations determined by QCR criteria are
indicated by red circles; E
2
and genistein concentrations determined by MGE
criteria are indicated by blue circles. (B) Effects of 48 h of exposure to E
2
(30 pM)
and genistein (3
␮
M) on expression of the WISP2 mRNA transcripts in MCF7
cells. Real-time quantitative PCR determination of amount of WISP2 mRNA by
three independent sets of experiments is shown.
12034
兩
www.pnas.org兾cgi兾doi兾10.1073兾pnas.0605341103 Shioda et al.

Fig. 2. Clustering analysis of xenoestrogen effects on MCF7 cell transcriptome. (A) Heat-map representation of 2D hierarchical clustering analysis of the TSPs
of xenoestrogens. Concentrations of the estrogenic agents were determined with QCR criteria except genistein(L) (low concentration), which was determined
by using MGE criteria. Bars below the heat map indicate seven subclusters of genes. (B) Statistical significance of similarities between the TSPs of xenoestrogens.
Histogram shows the joint null distribution of the PAR
ij
. Dots indicate observed PAR
ij
between the indicated pairs of groups of xenoestrogen TSPs. The P values
indicate how unlikely the observed TSP similarity would be to happen by chance. (C) Details of xenoestrogen effects on expression of representative
estrogen-discriminating marker genes. Each column represents a marker gene, and the points indicate fold changes in mRNA expression over vehicle control in
the log scale. The pattern of the representative unchanged genes is also shown as control. BPA, bisphenol A; PNP, p-nonylphenol.
Shioda et al. PNAS
兩
August 8, 2006
兩
vol. 103
兩
no. 32
兩
12035
GENETICS

To objectively examine the similarities between the TSP classes,
pairwise angle ratios (PARs) were calculated as measurements for
distance s between two TSP groups. To obtain the P values for the
TSP similarities, the null distribution of PARs was calculated by
using randomly generated PARs, and the locations of the measured
PARs were determined (Fig. 2B). As expected from the heat-map
pattern (Fig. 2 A), a strong similarity between TSPs for E
2
and the
low-concentration genistein (3
␮
M; MGE criteria) was confirmed
by this approach (P ⬍ 0.0002). A strong similarity between TSPs for
daidzein and high-concentration genistein (10
␮
M, QCR criteria)
was also shown (P ⬍ 0.0002). A marginally significant similarity
between TSPs for bisphenol A and p-nonylphenol was demon-
strated by the PAR location at the foot of the null distribution curve
(P ⬍ 0.002).
The clustering analysis identified seven subclusters of class-
discriminating marker genes (Fig. 2 A, bottom bars, and Table 1,
which is published as supporting information on the PNAS web
site). Subclusters 1a, 1b, and 1c were similar in their expression
patterns, together forming cluster 1; subclusters 3a and 3b formed
cluster 3; and subclusters 2 and 4 were unique. Expression profiles
of 20 genes representing each subcluster are shown in Fig. 2C
(subclusters 1a, 2, 3a, and 4) and Fig. 5, which is published as
supporting information on the PNAS web site (all subclusters).
Profiles of unchanged genes are also shown as control. Cluster 1
consisted of genes induced by the xenoestrogens but not by E
2
. The
low MGE-criteria concentration of genistein [3
␮
M, shown as
genistein(L)] also did not induce the cluster 1 genes. Cluster 2
involved genes strongly down-regulated by E
2
as well as by low and
high (10
␮
M, QCR criteria) concentrations of genistein and daid-
zein. Expre ssion of these cluster 2 genes was only modestly affected
by bisphenol A, p-nonylphenol, or PPT. In contrast, expression of
cluster 3 genes was strongly up-regulated by E
2
, by low and high
concentrations of genistein and daidzein, but not by bisphenol A or
PPT. Interestingly, the cluster 3 genes were induced by p-
nonylphenol, demonstrating the differential effects of the two
plastic-related xenoestrogens. Cluster 4 involved genes most
strongly induced by daidzein, high-concentration genistein, and
p-nonylphenol; they were induced relatively weakly by E
2
, low-
concentration genistein, and bisphenol A. Taken together, these
results provide further support for the following concepts: (i) TSPs
determined by using the QCR-criteria dosage reflect the divergent
transcriptomal effects of the estrogenic agents when their cell
proliferation responses are set to be equal; and (ii) the TSPs
determined by using the MGE-criteria dosage reflect the maximum
degree of similarity in their transcriptomal effects.
To predict the biological outcomes of the differential transcrip-
tomal effects of the xenoestrogens, we performed gene ontology兾
pathway analyses using the Kyoto Encyclopedia of Genes and
Genomes (KEGG) PATHWAY database (21). As expected from
their cell proliferation effects, large numbers of genes involved in
cell cycle progression were up-regulated by both E
2
and the
xenoestrogens (Table 2, which is published as supporting informa-
tion on the PNAS web site). Genes involved in purine and pyrim-
idine metabolism that are required for DNA synthesis as well as
genes involved in steroid biosynthesis were also significantly acti-
vated by the e strogens. Interestingly, of the 103 gene s involved in
mitochondrial oxidative phosphorylation, 46–83% were up-
regulated by the xenoestrogens. Fig. 6, which is published as
supporting information on the PNAS web site, shows that the
p-nonylphenol up-regulated genes were involved in all five oxidative
phosphorylation c omplexes. However, E
2
, PPT, and low-
concentration genistein did not significantly affect the expre ssion of
these genes. Analyses using the Gene Ontology database (22)
revealed re sults similar to the KEGG database (data not shown).
To compare the effects of each xenoestrogen on the expression
of oxidative phosphorylation-related genes, amounts of these genes
(82 genes, p-nonylphenol-induced) were plotted in Fig. 3A.Al-
though expre ssion of most of these genes was not significantly
enhanced by E
2
or by low-concentration genistein, they were
increased 2-fold or greater by high-concentration genistein and
other xenoestrogens. PPT also significantly enhanced expression of
these genes but more weakly than other xenoestrogens. Expre ssion
of the mRNA transcripts for the uncoupling proteins (UCP-1, -2,
and -3) that reflect mitochondrial biogenesis (23) was not affected
by E
2
or by low-concentration genistein; interestingly, UCP-1
expre ssion was moderately enhanced by genistein (Fig. 3B). In
contrast, expression of the UCP genes was strongly enhanced by the
xenoestrogens, but expre ssion of GAPDH and
␤
-actin (ACTB) was
unaffected by any estrogen. Thus, when the QCR criteria were used
for dosage standardization, the xenoestrogens (including genistein)
significantly increased expre ssion of the oxidative phosphorylation-
related genes, but E
2
did not. In contrast, when the MGE criteria
were used to determine concentrations of E
2
and genistein, neither
affected expression of these genes. These findings indicate that the
use of different criteria for dosage standardization for pharmacog-
enomics studie s may lead to different conclusions.
Fig. 3. Xenoestrogen induction of genes involved in mitochondrial oxidative
phosphorylation. (A) Xenoestrogen induction of 82 genes involved in oxida-
tive phosphorylation. The low [genistein(L), 3
␮
M] and high concentrations
[genistein, 10
␮
M] of genistein were determined by using MGE and QCR
criteria, respectively. Each point indicates fold increase in mRNA amounts in
estrogen-exposed MCF7 cells over vehicle-exposed control (average of three
microarray data, log scale). Averages of fold increase of all points for each
estrogen are shown on the top of the graph (mean ⫾ SEM, n ⫽ 82). (B)
Estrogen-induced expression of mRNA for mitochondrial uncoupling protein
(UCP1–3) and two housekeeping genes, GAPDH and
␤
-actin (ACTB). Each bar
represents fold increase in mRNA amounts in estrogen-exposed MCF7 cells
over vehicle-exposed control (average of three microarray data).
12036
兩
www.pnas.org兾cgi兾doi兾10.1073兾pnas.0605341103 Shioda et al.

Discussion
Microarray technology is increasingly being used in pharmacolog-
ical and toxicological studies. Several guidelines such as Minimum
Information About a Microarray Experiment (MIAME) and its
toxicology-adapted version MIAME兾Tox (24), as well as the
Minimum Information Needed for a Toxicology experiment (MIN-
Tox; ref. 25), have been proposed for standardized description of
experiments for pharmaco- and toxicogenomic studies. Unfortu-
nately, however, these guidelines do not specify any method for
dosage standardization in studying biologically active agents. Al-
though we have emphasized the critical importance of the dosage-
dependent aspects of the experimental determination of TSPs by
DNA microarray (19), this concern is often overlooked. For exam-
ple, microarray experiments comparing the TSPs of estrogen-
related agents are often performed with doses selected either
arbitrarily or empirically, namely, without apparent objective or
quantitative criteria for dosage standardization (1–7, 10, 13, 14).
To provide some guidance for dosage standardization of phar-
maco- and toxicogenomic experiments, in the present study, we
describe two approaches: the QCR criteria, that provide the TSPs
representing different agents known to induce an equal QCR; and
the MGE criteria, that provide the TSPs including reference marker
genes that are expre ssed equally in the presence of different agents
(Fig. 1). The TSPs determined using these two criteria were
remarkably different (Fig. 2) and thus led to different and con-
flicting conclusions (Fig. 3). Experiments using the QCR criteria
indicate that, in MCF7 cells, the TSPs of xenoestrogens differed
significantly from the TSP of E
2
(Fig. 2). TSP pathway analysis
showed that the xenoestrogens (including genistein) significantly
increased expre ssion of the genes involved in mitochondrial oxida-
tive phosphorylation, whereas E
2
showed no effects on expre ssion
of these genes (Fig. 3). However, it must be noted that these
observations were valid only at estrogen concentrations that en-
hanced MCF7 cell proliferation to comparable levels (Fig. 1). In
contrast, experiments performed using the MGE criteria convinc-
ingly demonstrated that the transcriptomal effects of genistein and
E
2
were practically identical (Fig. 2). In other words, the transcrip-
tomal effects of genistein and E
2
in MCF7 cells became indistin-
guishable at certain concentrations. Thus, the MGE criteria may be
more appropriate than the QCR criteria when the objective of a
study is to determine the maximum TSP similarities of different
agents. In contrast, the QCR criteria may be useful to characterize
difference s in mechanisms of action as well as side effects of agents
showing quantitatively similar biological actions. Table 3, which is
published as supporting information on the PNAS web site, sum-
marizes the features of different dosage criteria for the determi-
nation of TSPs.
The concept of the QCR and MGE dosage standardization
criteria should be applicable not only to studies of estrogenic
agents but more generally for other types of phar maco兾
toxic ogenomic studies involving both in vitro and in vivo models.
It is important to mention that the QCR and MGE criteria may
be applicable to studies involv ing human subjects or wildlife
an imals as long as a specific and quantitative agent-induced
phenot ype or marker gene is available (see Table 3). For
example, applying the MGE criteria, a clinical study may be able
to select patient-derived specimens whose drug-inducible MGEs
are comparable and ask whether the behavior of other genes is
similar when different drugs are used. Similarly, using the QCR
criteria, wildlife animals exposed to a group of environmental
c ontaminants showing similar and specific effects on a common
t arget tissue may be stratified before transcriptomal profiling. To
facilit ate dat abase-based bioinfor matics research on genomics
dat a, it would be important to have a description of the dosage
st andardization criteria (e.g., QCR, MGE, or empirical) when
TSPs and information on the doses of agents are incorporated
into databases. For this purpose, it may be helpful to extend the
MI AME兾Tox (Minimum Information About a Microarray Ex-
periment兾Tox) and MIN-Tox (Minimum Information Needed
for a Toxicolog y Ex periment) guidelines accordingly.
In summary, we have pre sented evidence that different criteria
for dosage standardization of biologically active agents may result
in conflicting and兾or misleading conclusions of a pharmaco兾
toxicogenomic study. We have also called attention to the risk of
interpreting TSP data where hormonal doses were selected on
either an empirical or an arbitrary basis. We have also described
applications of two dosage-standardizing criteria, namely, QCR and
MGE criteria. To correctly interpret genomics studies generating
TSPs, it is critical to specify the methods and criteria used for dosage
standardization and to provide the rationale for their use.
Materials and Methods
Cell Culture. Culture c onditions and the 120-h E-SCREEN pro-
liferation assay of MCF7 cells (BUS stock) were previously
described (19). To deter mine the transcriptomal effects of
estrogens, cells were washed three times with phenol red-free
DMEM and incubated at 37°C fo r1hinthefinal wash. The
medium was then changed to phenol red-free DMEM supple-
mented with 5% charcoal兾dextran-stripped FCS (HyClone, Lo-
gan, UT), and cells were cultured for 48 h in the presence of
estrogens. All plasticware was carefully selected to minimize
xenoestrogen contamination. As reported (26), contamination
of labware-derived xenoestrogens exerted critical effects on both
MCF7 cell proliferation and gene expression.
Microarray Experiment and Data Analysis. Total RNA extraction and
Affymetrix U133A GeneChip DNA microarray experiments were
performed as described in our previous study (19). Twenty-four
microarray data sets, which represented eight different estrogen
treatments, were obtained from three independently performed
cell culture experiments for each treatment. To perform 2D hier-
archical clustering analysis, 22,283 genes on the array were filtered
by using the following criteria: (i) the scaled intensity value was
⬎600 for at least three microarray data sets; (ii) the difference
between the maximum and minimum intensity value was ⬎300; and
(iii) the standard deviation of the gene vector was ⬎300. Expression
intensities of 1,675 informative genes were log-transformed and
analyzed by using Cluster software and visualized with TreeView
software (27). Complete linkage clustering analysis was performed
with median center for both genes and array data sets using
centered correlation as a similarity metric. Gene ontology analysis
was performed by using the GeneSifter.net on-line service (vizX-
labs, Seattle, WA), which served as a communication tool to the
Kyoto Encyclopedia of Genes and Genomes database (28) and the
Gene Ontology Consortium Database (22).
To objectively evaluate similarities of transcriptomal effects
bet ween t wo estrogens, we defined the PAR
ij
as
PAR
ij
⫽
cos
⫺1
共
␳
ij
兲
min
k
僆
兵
i, j
其
lⰻ
兵
i, j
其
cos
⫺1
共
␳
kl
兲
共1 ⱕ i ⬍ j ⱕ 8兲, [1]
where
␳
ij
is the sample correlation of gene expression under
treatments i and j (1 ⱕ i ⬍ j ⱕ 8). The inverse cosine of the
sample correlation measures the angle between the two gene
ex pression vectors, with cos
⫺1
(1) ⫽ 0 when gene ex pression is
perfectly positively correlated under the two treatments and
cos
⫺1
(⫺1) ⫽
␲
for perfect negative correlation. cos
⫺1
(
␳
ij
) can
therefore be interpreted as the gene-expression distance be-
t ween treatments i and j. When PAR
ij
, which is the ratio of these
dist ances, is significantly small, the transcriptomal effects of
treatments i and j are significantly close to each other. To assess
the significance of the observed PAR
ij
, the joint null distribution
of PAR
ij
was computed for a randomly sampled set of 1,675 genes
f rom the list of 22,283 genes. The observed PAR
ij
was c ompared
Shioda et al. PNAS
兩
August 8, 2006
兩
vol. 103
兩
no. 32
兩
12037
GENETICS

with the distribution of min
i,j
PAR
ij
across 5,000 samples from
this null distribution to obtain a P value as:
p
ij
⫽
再
f raction of simulated min
l,m
PAR
lm
smaller than observed PAR
ij
冎
. [2]
Real-Time Quantitative PCR (RTQ-PCR). RTQ-PCR was performed
using TaqMan PCR Master Mix and ABI Prism 7700 thermal
c ycler (Applied Biosystems, Framingham, MA). PCR primers
for human WISP2 were 5⬘-CATGAGAGGCACACCGAAGA
(sense) and 5⬘- GCACCTTTGAGAGGAGGCAG (antisense),
and a TaqMan probe for this mRNA transcript was 5⬘-VIC-
CCACCTCCTGGCCTTCTCCCTCC-TAMRA.
We thank Ben Wittner, Paul Grosu, and Reddy Gali for useful discus-
sions and Kremena Star for in itial contributions to microarray data
interpretation. We also thank James Signorovitch for his contributions
to the statistical evaluation of similarities between transcriptomal pro-
files. This study was funded in part by Susan Komen Breast Cancer
Foundation Grant BCTR0503620 (to T.S.), National Institutes of Health
R01 Grants CA82230 (to T.S.) and ES012301 (to A.M.S.), and the
Foundation for Research in Cell Biology and Cancer (T.S.).
1. Vendrell, J. A., Magnino, F., Danis, E., Duchesne, M. J., Pinloche, S., Pons, M.,
Birnbaum, D., Nguyen, C., Theillet, C. & Cohen, P. A. (2004) J. Mol.
Endocrinol. 32, 397–414.
2. Inoue, A., Yoshida, N., Omoto, Y., Oguchi, S., Yamori, T., Kiyama, R. &
Hayashi, S. (2002) J. Mol. Endocrinol. 29, 175–192.
3. Wang, D. Y., Fulthorpe, R., Liss, S. N. & Edwards, E. A. (2004) Mol.
Endocrinol. 18, 402–411.
4. Gadal, F., Starzec, A., Bozic, C., Pillot-Brochet, C., Malinge, S., Ozanne, V.,
Vicenzi, J., Buffat, L., Perret, G., Iris, F. & Crepin, M. (2005) J. Mol .
Endocrinol. 34, 61–75.
5. Frasor, J., Stossi, F., Danes, J. M., Komm, B., Ly ttle, C. R. & Katzenellenbogen,
B. S. (2004) Cancer Res. 64, 1522–1533.
6. Jorgensen, M., Vendelbo, B., Skakkebaek, N. E. & Leffers, H. (2000) Environ.
Health Perspect. 108, 403–412.
7. Terasaka, S., Aita, Y., Inoue, A., Hayashi, S., Nishigaki, M., Aoyagi, K., Sasak i,
H., Wada-Kiyama, Y., Sakuma, Y., Akaba, S., et al. (2004) Environ. Health
Perspect. 112, 773–781.
8. Watanabe, H., Suzuki, A., Goto, M., Lubahn, D. B., Handa, H. & Iguchi, T.
(2004) J. Mol. Endocrinol. 33, 243–252.
9. Moggs, J. G., Ashby, J., Tinwell, H., Lim, F. L., Moore, D. J., Kimber, I. &
Orphanides, G. (2004) Environ. Health Perspect. 112, 1137–1142.
10. Harvell, D. M., Richer, J. K., Allred, D. C., Sartorius, C. A. & Horwitz, K. B.
(2006) Endocrinology 147, 700–713.
11. Stossi, F., Barnett, D. H., Frasor, J., Komm, B., Lyttle, C. R. & Katzenellen-
bogen, B. S. (2004) Endocrinology 145, 3473–3486.
12. Lobenhofer, E. K., Bennett, L., Cable, P. L., Li, L., Bushel, P. R. & Afshari,
C. A. (2002) Mol. Endocrinol. 16, 1215–1229.
13. Soulez, M. & Parker, M. G. (2001) J. Mol. Endocr inol. 27, 259–274.
14. Frasor, J., Danes, J. M., Komm, B., Chang, K. C., Ly ttle, C. R. & Katzenel-
lenbogen, B. S. (2003) Endocrinology 144, 4562–4574.
15. Ma, X. J., Salunga, R., Tuggle, J. T., Gaudet, J., Enright, E., McQuary, P.,
Payette, T., Pistone, M., Stecker, K., Zhang, B. M., et al. (2003) Proc. Natl.
Acad. Sci . USA 100, 5974–5979.
16. Moggs, J. G., Tinwell, H., Spur way, T., Chang, H. S., Pate, I., Lim, F. L., Moore,
D. J., Soames, A., Stuckey, R., Currie, R., et al. (2004) Environ. Health Perspect.
112, 1589–1606.
17. Watanabe, H., Suzuki, A., Kobayashi, M., Takahashi, E., It amoto, M., Lubahn,
D. B., Handa, H. & Iguchi, T. (2003) J. Mol. Endocrinol. 30, 347–358.
18. Merrick, B. A. & Bruno, M. E. (2004) Curr. Opin. Mol. Ther. 6, 600–607.
19. Coser, K. R., Chesnes, J., Hur, J., Ray, S., Isselbacher, K. J. & Shioda, T. (2003)
Proc. Natl . Acad . Sci . USA 100, 13994–13999.
20. Hall, J. M. & Korach, K. S. (2002) J. Biol. Chem. 277, 44455–44461.
21. Kanehisa, M., Goto, S., Kawashima, S., Okuno, Y. & Hattori, M. (2004) Nucleic
Acids Res. 32, D277–D280.
22. Ashburner, M., Ball, C. A., Blake, J. A., Botstein, D., Butler, H., Cherry, J. M.,
Dav is, A. P., Dolinski, K., Dwight, S. S., Eppig, J. T., et al . (2000) Nat. Genet.
25, 25–29.
23. Wu, Z., Puigserver, P., Andersson, U., Zhang, C., Adelmant, G., Mootha, V.,
Troy, A., Cinti, S., L owell, B., Scarpulla, R. C., et al. (1999) Cell 98, 115–124.
24. Bao, W., Schmid, J. E., Goetz, A. K., Ren, H. & Dix, D. J. (2005) Reprod.
Toxicol. 19, 411–419.
25. Leighton, J. K., Brown, P., Ellis, A., Harlow, P., Harrouk, W., Pine, P. S., Robison,
T., Rosario, L. & Thompson, K. (2006) Environ. Health Perspect. 114, 573–578.
26. Ishikawa, T., Tak ano, K., Yasufuku-Tak ano, J., Fujita, T., Igarashi, T., Miura,
M. & Hata, K. (2001) Nat. Biotechnol . 19, 812.
27. Eisen, M. B., Spellman, P. T., Brown, P. O. & Botstein, D. (1998) P roc. Natl.
Acad. Sci . USA 95, 14863–14868.
28. Kanehisa, M., Goto, S., Hattori, M., Aoki-Kinoshit a, K. F., Itoh, M., Ka-
washima, S., Katayama, T., Araki, M. & Hirakawa, M. (2006) Nucleic Acids Res.
34, D354–D357.
12038
兩
www.pnas.org兾cgi兾doi兾10.1073兾pnas.0605341103 Shioda et al.