Environmental Toxicology and Chemistry, Vol. 29, No. 3, pp. 660–668, 2010
# 2009 SETAC
Printed in the USA
DOI: 10.1002/etc.70
MOLECULAR DOCKING AND COMPARATIVE MOLECULAR SIMILARITY INDICES
ANALYSIS OF ESTROGENICITY OF POLYBROMINATED DIPHENYL ETHERS
AND THEIR ANALOGUES
WEIHUA YANG,yz XIAOHUA LIU,y HONGLING LIU,y YANG WU,y JOHN P. GIESY,y§k# and HONGXIA YU*y
yState Key Laboratory of Pollution Control and Resource Reuse, School of the Environment, Nanjing University, Nanjing 210093, China
zSchool of Chemistry and Chemical Engineering, Xuzhou Normal University, Xuzhou 221116, China
§Department of Biomedical and Veterinary Biosciences and Toxicology Centre, University of Saskatchewan, Saskatoon, Saskatchewan S7N 5A2, Canada
kZoology Department, National Food Safety and Toxicology Center, and Center for Integrative Toxicology, Michigan State University, East Lansing,
Michigan 48824, USA
#Biology and Chemistry Department, City University of Hong Kong, Kowloon, Hong Kong Special Administrative Region, China
(Submitted 9 July 2009; Returned for Revision 4 September 2009; Accepted 22 September 2009)
Abstract— Molecular docking and three-dimensional quantitative structure–activity relationships (3D-QSAR) were used to develop
models to predict estrogenicity of polybrominated diphenyl ethers (PBDEs), para-hydroxylated polybrominated diphenyl ethers (paraHO-PBDEs), and brominated bisphenol A compounds to the human estrogen receptor a (hERa). Based on the molecular conformations
developed from the molecular docking, predictive comparative molecular similarity indices analysis (CoMSIA) models were developed.
The results of CoMSIA modeling with region focusing included were: leave-one-out (LOO) cross-validated coefficient
q2(LOO) ¼ 0.722 (all 26 compounds), q2(LOO) ¼ 0.633 (the training set, 20 compounds), q2(LMO, two groups) ¼ 0.520 0.155
(26 compounds), q2(LMO, five groups) ¼ 0.665 0.068 (26 compounds), predictive r2, r2pred ¼ 0.686 (the test set, 6 compounds), and
Q2EXT ¼ 0.678. The 3D-QSAR can be used to infer the activities of compounds with similar structural characteristics. The interaction
mechanism between compounds and the hERa was explored. Hydrogen bonding of the compound with Glu353 in the hERa is an
important determinant of the estrogenic activity of para-HO-PBDEs and brominated bisphenol A. Environ. Toxicol. Chem.
2010;29:660–668. # 2009 SETAC
Keywords—Endocrine-disrupting activity
Surflex dock
Receptor-based
Ligand–receptor interaction mechanism
Three-dimensional quantitative structure–activity relationships
been previously developed for interactions of PBDEs with the
aryl hydrocarbon receptor (AhR) [9], but the 3D-QSAR study
for AhR does not involve the interaction analysis between
PBDEs and amino acid residues in the active site of AhR,
because the crystal structure of AhR is not available.
The present study was conducted to develop predictive 3DQSAR models of the activity of PBDEs and PBDE analogues to
the human estrogen receptor a (hERa) and to study the mechanisms of ligand–receptor interaction. To explore further the
probable binding site of the hERa, binding conformations were
determined by use of the Surflex dock. Docking conformations
were assumed to be the actual bioactive binding conformations.
Based on the alignment derived from the docking conformations, comparative molecular similarity indices analysis
(CoMSIA) models were developed to predict the estrogenicity
of PBDEs and their analogues to the hERa.
INTRODUCTION
Polybrominated diphenyl ethers (PBDEs) are present in air,
water, soil, fish, birds, marine mammals, and even human blood,
adipose tissue, and breast milk [1–4]. Humans can be directly
exposed to hydroxylated polybrominated diphenyl ethers
(HO-PBDEs) [5]. Polybrominated diphenyl ether concentrations in the environment are increasing exponentially over time,
as indicated by a marked increase in human breast milk [6].
Polybrominated diphenyl ethers are agonists and/or antagonists of the androgen, progesterone, and estrogen receptors
[7,8]. This, combined with the ubiquitous distribution and
increasing concentrations of PBDEs, has led to concern over
their possible toxic effects. To assess further the potential for
PBDEs, HO-PBDEs, and brominated bisphenol A, quantitative
structure–activity relationships (QSAR) models were developed. Because the interaction between endocrine disrupting
compounds and receptors is a three-dimensional (3D) process,
3D-QSAR is more likely to produce accurate predictions of
activity of compounds to the estrogen receptor (ER). To date
there have been no reports of 3D-QSAR models to predict the
estrogenicity of HO-PBDEs to the ER. Similar models have
MATERIALS AND METHODS
Data sets and biological activity
The data sets used in these studies covered a series of
PBDEs, para-HO-PBDEs, and (brominated) bisphenols A compounds [7], which have been shown to possess or not to possess
estrogenic activity. The estrogenic potency to the hERa was
estimated by use of the ER-CALUX (chemically activated
luciferase gene expression) transactivation assay in which the
All Supplemental Data may be found in the online version of this article.
* To whom correspondence may be addressed
(yuhx@nju.edu.cn).
Published online 2 December 2009 in Wiley InterScience
(www.interscience.wiley.com).
660
3D-QSAR study on estrogenicity of PBDE analogues
gene for luciferase is under transcriptional control of the
response elements for activated ER receptors. Median effective
concentration (EC50) values are expressed as micromolar. For
modeling purposes, compounds for which activity was observed
but for which EC50 values were greater than 10 mM were
assigned a value of 15 mM. Inactive compounds within tested
concentrations were given an EC50 value one log unit greater
(100 mM) than the greatest tested concentration 10 mM [10]
(Table 1). These assumed values were added to the data set to be
able to model the full range of activity (i.e., from very active to
moderately active to nonactive compounds). The EC50 values
were converted to pEC50 (logEC50, M) values, which were
used as dependent variables in QSAR analysis. The structures of
the compounds used in the present study are given in Table 1.
Molecular docking and alignment
Estimates of initial conformations of compounds were
developed from coordinates of X-ray crystallographic data
(Cambridge Structure Database) for 11 target compounds
(i.e., BDE15, BDE28, BDE30, BDE32, BDE47, BDE51,
BDE138, BDE166, BPA, TriBBPA, TBBPA). Conformations
of compounds not contained in the Cambridge Database were
inferred from structurally analogous compounds [11]. The
geometries of these compounds were subsequently optimized
using the Minimize module in SYBYL7.3 (Tripos) by calculating the energy minimized conformation by use of the Powell
method. The Tripos force field (distance-dependent dielectric)
was used to reach a final energy convergence gradient value of
0.001 kcal/(mol Å). Gasteiger-Hückel charges were assigned
to each compound. The minimized structures were used as
initial conformations for docking calculations. The Surflex dock
program interfaced with SYBYL 7.3 was used to dock the
compounds to the active site of the human ERa. The crystal
structure of the human ERa ligand-binding domain (LBD)
when bound to 17b-estradiol (E2; PDB code 1ERE) was
obtained from the Protein Data Bank (http://www.rcsb.org/
pdb/). Prior to making docking calculations, E2 was extracted
from the crystal structure, the structural water molecules were
removed, and hydrogen atoms were added in standard geometry
by using Biopolymer modulators. Kollman-all atom charges
were assigned to protein atoms. Automated molecular docking
with default parameters sets produced 10 options of binding
conformation for each ligand. The top TotalScore conformation
was selected as the most likely bioactive conformation. The 3D
alignment approach used was based on the top-ranked molecular conformations obtained from the Surflex dock analysis.
Bioactive conformations were assigned Gasteiger-Hückel
charges and imported into a SYBYL molecular database for
CoMSIA studies without further energy minimization.
3D-QSAR computer modeling
The CoMSIA method incorporates five physicochemical
properties: steric, electrostatic, hydrophobic, hydrogen bond
donor, and hydrogen bond acceptor. The CoMSIA descriptors
were derived according to the methods described by Klebe et al.
[12]. To derive the CoMSIA descriptor fields for aligned
molecules, a 3D cubic lattice with grid spacing of 2 Å in x,
y, and z directions was created. A default value of 0.3 was used
as the attenuation factor a.
Environ. Toxicol. Chem. 29, 2010
661
The method of partial least-squares (PLS) regression was
used to correlate variations in the biological activities with
variations in the respective descriptors. The predictive values of
PLS models were evaluated by using the leave-one-out (LOO)
cross-validation method. To improve the signal-noise ratio for
CoMSIA, a minimum d value (column filtering) of 2.0 kcal/mol
was used. A cross-validated coefficient, q2, characterized the
predictive ability of the models. To establish the model for
predicting the activity of the compounds in the training set and
the test set, a noncross-validation was performed with the
optimal number of components. To refine the models, region
focusing [13,14] was performed on conventional CoMSIA
models. Discriminant power values were used as weights with
different weighing factors applied in addition to grid spacing to
obtain more predictive models. Because the LOO procedure
itself does not necessarily guarantee the maximum predictive
power of the models [15,16], a more powerful statistical
evaluation, the leave-many-out (LMO) procedure, was also
applied. The LMO cross-validation, which was divided automatically by the program into training and test sets, was
performed for the entire set of compounds. The LMO crossvalidation was conducted based on either two or five groups.
Two groups mean that the model was built by using 50% of the
available data, and the model obtained was tested on the other
50% of compounds. Five groups mean that the model was built
by using 80% of the available data and the model obtained was
tested on the other 20% of compounds. External validation is
the only way to establish a reliable QSAR model [17]. To test
the utility of the model as a predictive tool, an external set of
compounds with known activities but not used in model generation (test set) was predicted. The predictive r2 value (r2pred)
and the external validation parameter (Q2EXT) based on
molecules in the test set were used to evaluate the predictive
power of the CoMSIA models.
RESULTS
CoMSIA statistical results
A receptor-based alignment procedure was used. All the
molecules were aligned based on their docking conformations
without a separation of training and test sets. For all of the 26
compounds, the CoMSIA model yielded q2LOO ¼ 0.466, which
is less satisfactory, and r2 ¼ 0.966. The next step was to generate statistically significant 3D QSAR models in terms of
cross-validated correlation coefficients with the least standard
deviation using 3D QSAR tools (region focusing) that are
available in SYBYL. Region focusing was weighted by a
discriminant power value of 1.5 and a grid spacing of 1.0 Å.
Use of region focusing on the 26 compounds model yielded
values of q2LOO ¼ 0.722, r2 ¼ 0.949. All statistical data are
given in Table 2.
To assess further the robustness and statistical confidence of
the derived models, bootstrapping analysis for 100 runs was
performed (Table 2). The q2boot and standard error of estimate of
bootstrapping (100 runs) are 0.978 and 0.151, respectively,
indicating that the model is stable and has high internal predictive ability.
Next, a cross-validation analysis was applied to the entire set
of compounds (without separating training and test sets) to
investigate the stability of the CoMSIA model. The model was
Table 1. Structure and activity of the target compoundsa
Empirical measurementb
EC50 (mM)c
pEC50 (M)
Predicted
pEC50 (M)
Residual
pEC50 (M)
100
15
3.4
5.1
15
3.1
7.3
2.9
100
15
15
2.5
3.9
100
100
100
100
4.00
4.82
5.47
5.29
4.82
5.51
5.14
5.54
4.00
4.82
4.82
5.60
5.41
4.00
4.00
4.00
4.00
4.74
5.23
5.67
5.21
4.73
4.84
5.12
5.45
3.73
4.91
4.57
4.92
5.25
4.13
4.06
4.29
4.02
0.74
0.41
0.20
0.08
0.09
0.67
0.02
0.09
0.27
0.09
0.25
0.68
0.16
0.13
0.06
0.29
0.02
Bisphenol A
0.3
6.52
6.44
0.08
MBBPA
0.5
6.30
6.44
0.14
diBBPA
0.4
6.40
6.26
0.14
triBBPAd
15
4.82
5.13
0.31
TBBPA
100
4.00
3.94
0.06
4-Phenoxyphenol
1.7
5.77
5.90
0.13
T2-like HO-BDE
0.1
7.00
6.96
0.04
T3-like HO-BDEd
0.5
6.30
6.61
0.31
T4-like HO-BDE
100
4.00
4.26
0.26
Panel A
PBDEs
BDE-15d
BDE-28
BDE-30
BDE-32
BDE-47
BDE-51d
BDE-71
BDE-75
BDE-77
BDE-85
BDE-99d
BDE-100
BDE-119
BDE-138
BDE-153
BDE-166
BDE-190d
Br substituted site
4,40 2,4,40 2,4,62,40 ,62,20 ,4,40 2,20 ,4,60 2,30 ,40 ,62,4,40 ,63,30 ,4,40 2,20 ,3,4,40 2,20 ,4,40 ,52,20 ,4,40 ,62,30 ,4,40 ,62,20 ,3,4,40 ,50 2,20 ,4,40 ,5,50 2,3,4,40 ,5,62,3,30 ,4,40 ,5,6-
Panel B
Panel C
a
PBDEs ¼ polybrominated diphenyl ethers; BDE ¼ brominated diphenyl ether; MBBPA ¼ monobromobisphenol A; diBBPA ¼ 3,30 -dibromobisphenol A;
triBBPA ¼ 3,30 ,5-tribromobisphenol A; TBBPA ¼ 3,30 ,5,50 -tetrabromobisphenol A; T2 ¼ 3,5-diiodothyronine; T3 ¼ 3,30 ,5-triiodothyronine; T4 ¼ 3,30 ,5,50 tetraiodothyronine; HO-BDE ¼ hydroxylated brominated diphenyl ether; EC50 ¼ median effective concentration.
b
Experimental EC50 values cited from biological activity data reported by Meerts et al. [7].
c
Compounds for which activity was observed but with EC50 values greater than 10 mM were given a response of 15 mM. Inactive compounds within tested
concentration were given an EC50 value one log unit greater (100 mM) than the greatest tested concentration 10 mM.
d
The compounds in the test set for model validation.
3D-QSAR study on estrogenicity of PBDE analogues
Environ. Toxicol. Chem. 29, 2010
Table 2. Statistical data for comparative molecular similarity indices
analysis (CoMSIA) models with region focusing included
Statistical parameters
q2LOOa
PLS componentsb
SEPc
r2ncvd
SEEe
Ftestf
r2predg
Q2EXTh
Contribution
Steric
Electrostatic
Hydrophobic
H-bond donor
H-bond acceptor
2
q booti
SEEbootj
q22cvk
SD2cvl
q25cvm
SD5cvn
26-Compound model
20-Compound model
0.722
6
0.562
0.949
0.241
58.953
—
—
0.633
5
0.675
0.940
0.272
44.068
0.686
0.678
0.187
0.186
0.229
0.194
0.203
0.978
0.151
0.520
0.155
0.665
0.068
0.109
0.254
0.249
0.176
0.213
—
—
—
—
—
—
a
Cross-validated correlation coefficient after the leave-one-out procedure.
Optimal number of principal components.
c
Standard error of prediction.
d
Noncross-validated correlation coefficient.
e
Standard error of estimate.
f
Ratio of r2ncv explained to unexplained ¼ r2ncv/(1 r2ncv).
g
Predicted correlation coefficient for the test set of compounds.
h
External validation parameter.
i
Average of correlation coefficient for 100 samplings using the bootstrappin
procedure.
j
Average standard error of estimate for 100 samplings using the bootstrapping procedure.
k
Average cross-validated correlation coefficient for 100 runs using the two
cross-validation group.
l
Standard deviation of average cross-validated correlation coefficient for
100 runs.
m
Average cross-validated correlation coefficient for 100 runs using the five
cross-validation group.
n
Standard deviation of average cross-validated correlation coefficient for
100 runs.
b
cross-validated using two (leave-half-out) and five (leave 20%
out) cross-validation groups. The LMO cross validation was
performed 100 times with the same region-focusing setting
(Table 2).
663
experimental (y) values of relative potencies for the test set,
y þ b or y~r ¼ a0 y þ b0 ) as represented by the coefficient of
(yr ¼ a~
determination R2, was determined [17]. Regressions of y against
y and y~r0 ¼ k0 y) were
y~ or y~ against y through the origin, (yr0 ¼ k~
2
02
characterized by R 0 and R 0, respectively (Supplemental
Data, Fig. S1). The values of the coefficients of determination
were R2 ¼ 0.7146, R20 ¼ 0.6994, R0 20 ¼ 0.6813, respectively,
whereas the values for k and k0 were 0.9813 and 1.0109,
respectively. The models were considered to be acceptable
predictive QSAR models because they satisfied the conditions
suggested [17]: q2 > 0.5, R2 > 0.6, R20 or R0 20 close to R2, and
the corresponding k, k0 values are between 0.85 and 1.15. The
predicted pEC50 values for the training set and the test set,
based on CoMSIA model with region focusing included, are
shown in Table 1. The graphic results for the experimental
versus predicted activities of both training set and test set are
given in Figure 1.
CoMSIA contour maps
In the CoMSIA steric field, the green (sterically favorable)
and yellow (sterically unfavorable) contours represent 80% and
20% level contributions, respectively (Fig. 2A). To aid in the
visualization, the most potent compound BDE100 among
PBDEs is overlaid on the map. A great green contour around
the 20 ,40 Br substituents indicates that a sterically bulky group
(e.g., Br) is favored in this region. This is in line with the
experimental estrogenic activity measurements, e.g., the estrogenic activity order is 2,4,6-tri-BDE30 (EC50 ¼ 3.4 mM) <
2,4,40 ,6-tetra-BDE75 (EC50 ¼ 2.9 mM) < 2,20 ,4,40 ,6-pentaBDE100 (EC50 ¼ 2.5 mM).
The electrostatic contour map of the CoMSIA model is
shown in Figure 2B. Similarly, in the CoMSIA electrostatic
field, the red (electronegative charge favorable) and blue
(electropositive charge favorable) contours represent 80 and
20% level contributions, respectively. The electrostatic contour
shows only blue polyhedra favoring electropositive moieties, so
the hydroxyl group (a strong electropositive substituent) at the
benzene ring is a very important feature for an estrogenic
activity.
Validation of the CoMSIA models
In addition to the statistical evaluation, the assessment of its
predictive ability is also an essential requirement for a 3DQSAR model. The CoMSIA calculations were performed on the
training set, and the model obtained was tested on the test set.
Selection of the training set (20 compounds) and the test set (six
compounds) was made such that the test set comprised structurally diverse compounds with a range of biological activities
similar to that of the training set. A great q2 (0.633) was
obtained when region focusing was included for the training
set. In addition, the r2pred was 0.686 and Q2EXT was 0.678 for the
test set (Table 2). The predictive power of the model was
deemed to be acceptable (q2 > 0.5, r2pred > 0.4). The relatively
great value of q2 obtained in the LOO appeared to be necessary
but not sufficient for the model to have great predictive power
[17]. The degree of concordance between the predicted (~
y) and
Fig. 1. Predicted versus experimental estrogenic activities (pEC50) for
polybrominated diphenyl ethers and their analogues in the training set (solid
circles) and the test set (open circles), with region focusing included.
664
Environ. Toxicol. Chem. 29, 2010
W. Yang et al.
Fig. 2. Comparative molecular similarity indices analysis (CoMSIA) standard deviation/coefficients (stdev.coeff) contour map. (A) CoMSIA steric contour map;
the compound BDE100 is displayed as a reference. (B) CoMSIA electrostatic contour map; the compound BDE100 is displayed as a reference. (C) CoMSIA
hydrophobic contour maps; the compound BDE100 is displayed as a reference. (D) CoMSIA hydrogen bond donor and hydrogen bond acceptor contour maps; the
compound 3,5-diiodothyronine-like hydroxylated brominated diphenyl ether (T2-like HO-BDE) is displayed as a reference. [Color figure can be seen in the online
version of this article, available at www.interscience.wiley.com.]
In the CoMSIA hydrophobic field, the yellow (hydrophobic
favorable) and white (hydrophobic unfavorable or hydrophilic
favorable) contours also represent 80 and 20% level contributions, respectively (Fig. 2C). The great yellow contour near the
20 ,60 Br substituents indicates that hydrophobic Br group is
favored in these regions. This also demonstrated that the estrogenic activity of BDE100 is more than that of BDE30.
The hydrogen bond donor and receptor contour map of the
CoMSIA model in the presence of the most potent compound,
3,5-diiodothyronine-like hydroxylated brominated diphenyl
ether (T2-like HO-BDE), is shown in Figure 2D. The cyan
(hydrogen bond donor favorable) and purple (hydrogen bond
donor unfavorable) contours represent 65% and 35% level
contributions, respectively, in the hydrogen bond donor fields.
In the CoMSIA hydrogen bond acceptor field, the magenta
(hydrogen bond acceptor favorable) and red (hydrogen bond
acceptor unfavorable) contours represent 65% and 35% level
contributions, respectively. A magenta contour around the O
atom of OH on the phenolic ring represents the higher activity
of compounds having a hydrogen bond acceptor group at this
position, such as the estrogenic activity of BDE30, which,
lacking OH as a hydrogen bond acceptor group at this
position, is less than that of T2-like HO-BDE (i.e., 40 -HOBDE30).
3D-QSAR study on estrogenicity of PBDE analogues
Environ. Toxicol. Chem. 29, 2010
665
Docking study
Brominated bisphenol A compounds
It has been long understood that the phenolic ring is normally
associated with binding to the hERa. One hundred eight of one
hundred thirty hERa-active chemicals (83%) in the NCTR data
set contain a phenolic ring. The contribution of the phenolic ring
to binding is more significant than any other structural feature
[18]. Among the target compounds, brominated bisphenol A
compounds and para-HO-PBDEs all contain phenolic ring
moieties, so a docking study was conducted to explore
how —OH on two target compounds interacts with the hERa.
In addition, the conformations of PBDEs having no OH group
were also analyzed.
The docking conformations of brominated bisphenol A
compounds suggest that the presence of the Br and hydroxyl
moieties are important in binding to the hERa (Fig. 3). For
monobromobisphenol A (MBBPA; Fig. 3A), the hydroxyl
group with an ortho-substituted Br has hydrogen bond interaction with the carboxylate of Glu353 and the guanidinium
group of Arg394, whereas the other hydroxyl group acts as a
donor to form a hydrogen bond with the oxygen of the hydroxyl
group of Thr347. For 3,30 -dibromobisphenol A (diBBPA;
Fig. 3B), one hydroxyl group makes a hydrogen bond to the
carboxylate group of Glu353, whereas another hydroxyl group
Fig. 3. Ligand binding in the human estrogen receptor a ligand-binding domain. (A) Monobromobisphenol A. (B) 3,30 -Dibromobisphenol A. (C) 3,30 ,5Tribromobisphenol A. (D) 3,30 ,5,50 -Tetrabromobisphenol A. Residues important for interactions between the human estrogen receptor a ligand-binding domain
and the ligands are displayed in all panels. [Color figure can be seen in the online version of this article, available at www.interscience.wiley.com.]
666
Environ. Toxicol. Chem. 29, 2010
forms a hydrogen bond with His524. 3,30 ,5-Tribromobisphenol
A (triBBPA; Fig. 3C) forms hydrogen bonds with Glu353,
Gly521, and His524. However, 3,30 ,5,50 -tetrabromobisphenol
A (TBBPA; Fig. 3D), which showed no estrogenic potency
within the tested concentrations, exhibits completely different
conformation with other brominated bisphenols A compounds.
The carbon atom connecting the two phenyl rings of TBBPA
is located near the hydrophilic Thr347, whereas the carbon
atom connecting two phenyl ring is located in a hydrophobic
region around Leu391 and Ile424 for MBBPA, diBBPA, and
triBBPA. 3,30 ,5,50 -Tetrabromobisphenol A forms hydrogen
bonds with His524 and Gly521 but does not form a hydrogen
bond with Glu353. All of these observations demonstrate that
the difference in number and position of the Br among brominated bisphenol A compounds is important in different conformations and furthermore forms hydrogen bonds with
W. Yang et al.
different amide residues that result in different activities for
the hERa.
Hydroxylated polybrominated diphenyl ethers
The para-HO-PBDEs (Fig. 4) that were investigated in the
present study exhibited different docking conformations. The
hydroxyl groups moieties of 4-(2,4,6-tribromophenoxy)phenol
(T2-like HO-BDE) and 2-bromo-4-(2,4,6-tribromophenoxy)phenol (T3-like HO-BDE) are all positioned in Glu353,
Arg394 end of the binding site. 3,5-Diiodothyronine(T2)-like
HO-BDE (Fig. 4A) forms a hydrogen bond with Glu353. The
hydroxyl group of 3,30 ,5-triiodothyronine (T3)-like HO-BDE
(Fig. 4B) interacts with Glu353 and Arg394. 3,30 ,5,50 -Tetraiodothyronine (T4)-like HO-BDE (Fig. 4C), which was demonstrated no estrogenic effect within the tested concentration
range, exhibits a docking conformation different from that of
Fig. 4. Ligand binding in the human estrogen receptor a ligand-binding domain. (A) 4-(2,4,6-Tribromophenoxy)phenol (i.e., T2-like HO-BDE). (B) 2-Bromo-4(2,4,6-tribromophenoxy)phenol (i.e., T3-like HO-BDE). (C) 2,6-Dibromo-4-(2,4,6-tribromophenoxy)phenol (i.e., T4-like HO-BDE). Residues important for
interaction between the human estrogen receptor a ligand-binding domain and the ligands are displayed in all panels. [Color figure can be seen in the online version
of this article, available at www.interscience.wiley.com.]
3D-QSAR study on estrogenicity of PBDE analogues
T2-like HO-BDE; the hydroxyl group on T4-like HO-BDE is
located in the His524 end of hERa LBD. 3,30 ,5,50 -tetraiodothyronine(T4)-like HO-BDE forms hydrogen bonds with
His524 and Gly521; this is similar to TBBPA.
In comparing para-HO-PBDEs with brominated bisphenol
A compounds, it can be seen that both TBBPA and T4-like
HO-BDE, which did not show significant potency relative to
estrogen within the tested concentration range, do not form a
hydrogen bond with Glu353 but form hydrogen bonds with
His524 and Gly521. Thus, it was concluded that formation of a
hydrogen bond with Glu353 may play a critical role in discriminating the estrogenic potency of para-HO-PBDEs and
brominated bisphenol A compounds.
In addition, three para-HO-PBDEs examined in the literature [19] were docked to the hERa, 40 -OH-BDE17, which was
a more potent estrogen agonist, formed hydrogen bonds with
Glu353 and Arg394 (Supplemental Data, Fig. S2), whereas the
other para-HO-PBDEs, 40 -HO-BDE49 and 4-HO-BDE42,
which were less potent, both form hydrogen bonds with
Gly521 and His524 (Supplemental Data, Fig. S2). These results
demonstrate that forming hydrogen bonds with Glu353 in the
LBD of the hERa is important in determining estrogenicity.
Polybrominated diphenyl ethers
The structure of hERa can accept ligands containing an
aromatic ring and a number of different hydrophobic groups
[20]. Docking conformations of several PBDEs, which did not
cause any expression in the transactivation assay and were thus
not estrogenic, demonstrate that their Br substituents are close
to the hydrophilic residues Glu353 and Arg394; i.e., their
hydrophobic Br substituents did not match well with the hERa
LBD. This may partially explain why they had no estrogenic
potential within the tested concentration range. For example,
the para-Br substituents of BDE-15 are near hydrophilic residues His524, Glu353, and Arg394; i.e., hydrophobic Br substituents on BDE-15 are unfavorable for binding to the hERa.
On the other hand, the Br substituents of BDE-100 are located
near the hydrophobic residues Ile424, Leu525, and Leu540.
Thus, hydrophobic Br substituents on BDE-100 match well with
the hydrophobic cavity (Supplemental Data, Fig. S3).
DISCUSSION
Biological activity data for compounds were from ERCALUX bioassay, which is an hER-mediated luciferase
reporter gene assay in MCF-7 cells. Reporter gene assays
may act as a mechanistic tool to characterize receptor-mediated
endocrine activity [21]. Binding of PBDE analogues to ER in
target cells results in the initiation of specific transcription
activation events. The ER binding is a major determinant or
rate-determining step in the reporter gene assay [22]. Fang et al.
[22] obtained a good linear relationship (r2 ¼ 0.78) between the
ER binding and yeast assays across investigated chemical
classes; i.e., the two assays correlate very well for estrogenic
agonists. In the present study, activity was partially demonstrated through binding (e.g., docking study section).
The docking study showed that formation of a hydrogen
bond with Glu353 may play a critical role in discriminating
estrogenic potency of para-HO-PBDEs and brominated bisphenol A compounds. The crystal structure of E2 bound to the ER
Environ. Toxicol. Chem. 29, 2010
667
reveals that the 3-OH of E2 forms hydrogen bonds with Glu353,
Arg394, and a water molecule, whereas the 17b-OH forms only
one hydrogen bond with His524. Elimination or modification of
either of these two OH groups significantly reduces binding
affinity for the receptor, but the effect is greater at the 3-position
than at the 17b-position; namely, the 3-OH moiety is more
important than the 17b-OH in ER binding [18]. This partially
supports the assumption that forming a hydrogen bond with
residue Glu353 is more important in hERa binding for paraHO-PBDEs and brominated bisphenol A compounds.
It can be seen that hydroxylated PBDEs would be expected
to have greater estrogenic activity than the nonhydroxylated
analogues from empirical measurements of the relative estrogenic potency of these compounds [7]. The results of another
study [19] suggested that the weak estrogenic effects of DE-71
are due to hydroxylated metabolites of individual congeners.
These results are similar to those observed for hydroxylated
PCBs, which are more potent estrogen agonists than are nonhydroxylated PCBs [23]. The relative potency of individual
PCB congeners ranged from 25- to 650-fold less than the
corresponding hydroxylated PCB congeners [24], so it can
be inferred that the —OH group, which can form a hydrogen
bond with hERa LBD, plays a critical role in PBDE analogues’
estrogenic activity.
For the successful use of QSARs to predict toxicological and
fate effects, especially in a regulatory setting, there is a growing
realization that the applicability domain of the model must be
defined [25]. The definition of the applicability domain includes
the chemical properties, structural features, and biological
effects of the training set of compounds and is most reliable
when based on an understanding of the underlying chemical
mechanisms of the toxic endpoint. The structural classes
include PBDEs, para-HO-PBDEs, and brominated bisphenol
A compounds. The models built could predict the estrogenic
activity of these PBDE analogues. The 3D contour maps
generated from these models were analyzed individually and
provide the regions in space where interactive fields may
influence the activity. It should be noted that it is not appropriate
to infer the activity only from the contour of the reference
molecule, because the alignment in the present study is based on
docking conformations, not on common substructure. It is well
known that compounds having similar structure may exhibit
very different docking conformations; this is also concluded
from this docking study. Of course, the activity difference
between compounds having similar docking conformations
can be explained by use of the contour map, such as BDE30,
BDE75, and BDE100 in the present study. Thus, the contour
map and docking conformations must be combined to predict
estrogenic activity of PBDE analogues.
CONCLUSIONS
Understanding intermolecular interactions of PBDE congeners, para-HO-PBDEs, and brominated bisphenol A compounds with hERa was achieved by development of
molecular docking analysis and 3D-QSAR models. The docked
conformation of target compounds to the hERa LBD showed
that the ability to form a hydrogen bond with the Glu353 residue
was critical in discriminating among the relative potencies of
para-HO-PBDEs and brominated bisphenols A compounds.
668
Environ. Toxicol. Chem. 29, 2010
The PBDEs that lack hydroxyl groups to anchor them in the
binding site probably exhibit weak estrogenic activity through
hydrophobic interactions.
SUPPLEMENTAL DATA
Figure S1. A regression between observed and predicted (A)
and between predicted and observed (B) activities for PBDEs
and PBDE analogues from the test set.
Figure S2. 40 -HO-BDE17, 40 -HO-BDE49, and 4-HOBDE42 binding to the hERa LBD. Residues important for
interaction between the hERa LBD and the ligands are displayed in all panels.
Figure S3. COMSIA contour plots for hydrophobic fields
with BDE-15 as reference molecule (left) and with BDE-100 as
reference molecule (right). The yellow/white polyhedra depict
favorable sites for hydrophobic/hydrophilic groups. (504 KB
PDF)
Acknowledgement—This work was supported by the National Natural
Science Foundation of China (grant 20737001) and the 863 Program of China
(grant 2006AA06Z424), jointly funded by the Research Grants Council of
Hong Kong and NSFC (grant 20518002/N_CityU110/05). Portions of the
research were supported by a Discovery Grant from the National Science and
Engineering Research Council of Canada (project 6807) and a grant from the
Western Economic Diversification Canada (project 6971 and 6807). J.P.
Giesy’s participation in the project was supported as an at-large Chair
Professorship from City University of Hong Kong and by an Area of
Excellence grant (AoE P-04/04).
REFERENCES
1. Darnerud PO, Gunnar ES, Johannesson T, Larsen PB, Viluksela M. 2001.
Polybrominated diphenyl ethers: occurrence, dietary exposure and
toxicology. Environ Health Perspect 109:49–68.
2. Hites RA. 2004. Polybrominated diphenyl ethers in the environment and
in people: A meta-analysis of concentrations. Environ Sci Technol
38:945–956.
3. He JZ, Robrock KR, Alvarez-Cohen L. 2006. Microbial reductive
debromination of polybrominated diphenyl ethers (PBDEs). Environ Sci
Technol 40:4429–4434.
4. Mai BX, Chen SJ, Luo XJ, Chen LG, Yang QS, Sheng GY, Peng PA, Fu
JM, Zeng EY. 2005. Distribution of polybrominated diphenyl ethers in
sediments of the Pearl River Delta and adjacent South China Sea.
Environ Sci Technol 39:3521–3527.
5. Qiu XH, Bigsby RM, Hites RA. 2009. Hydroxylated metabolites of
polybrominated diphenyl ethers in human blood samples from the United
States. Environ Health Perspect 117:93–98.
6. Toms LM, Harden FA, Symons RK, Burniston D, Fürst P, Müller JF.
2007. Polybrominated diphenyl ethers (PBDEs) in human milk from
Australia. Chemosphere 68:797–803.
7. Meerts IA, Letcher RJ, Hoving S, Marsh G, Bergman A, Lemmen JG, van
der Burg B, Brouwer A. 2001. In vitro estrogenicity of polybrominated
diphenyl ethers, hydroxylated PDBEs, and polybrominated bisphenol A
compounds. Environ Health Perspect 109:399–407.
8. Hamers T, Kamstra JH, Sonneveld E, Murk AJ, Kester MHA, Andersson
PL, Legler J, Brouwer A. 2006. In vitro profiling of the endocrinedisrupting potency of brominated flame retardants. Toxicol Sci 92:157–
173.
W. Yang et al.
9. Wang YW, Liu HX, Zhao CY, Liu HX, Cai ZW, Jiang GB. 2005.
Quantitative structure–activity relationship models for prediction of the
toxicity of polybrominated diphenyl ether congeners. Environ Sci
Technol 39:4961–4966.
10. Harju M, Hamers T, Kamstra JH, Sonneveld E, Boon JP, Tysklind M,
Andersson PL. 2007. Quantitative structure–activity relationship
modeling on in vitro endocrine effects and metabolic stability involving
26 selected brominated flame retardants. Environ Toxicol Chem 26:816–
826.
11. Tamura H, Ishimoto Y, Fujikawa T, Aoyama H, Yoshikawa H,
Akamatsu M. 2006. Structural basis for androgen receptor agonists and
antagonists: Interaction of SPEED 98-listed chemicals and related
compounds with the androgen receptor based on an in vitro reporter gene
assay and 3D-QSAR. Bioorg Med Chem 14:7160–7174.
12. Klebe G, Abraham U, Mietzner T. 1994. Molecular similarity indexes in
a comparative analysis (CoMSIA) of drug molecules to correlate and
predict their biological activity. J Med Chem 37:4130–4146.
13. Jójárt B, Martinek TA, Márki Á. 2005. The 3D structure of the binding
pocket of the human oxytocin receptor for benzoxazine antagonists,
determined by molecular docking, scoring functions and 3D-QSAR
methods. J Comput Aided Mol Des 19:341–356.
14. Gilbert KM, Boos TL, Dersch CM, Greiner E, Jacobson AE, Lewis D,
Mateck D, Prisinzano TE, Zhang Y, Rothman RB, Rice KC, Venanzia
CA. 2007. DAT/SERT selectivity of flexible GBR 12909 analogs
modeled using 3D-QSAR methods. Bioorg Med Chem 15:1146–1159.
15. Novellino E, Fattorusso C, Greco G. 1995. Use of comparative molecular
field analysis and cluster analysis in series design. Pharm Acta Helv
70:149–154.
16. Kubinyi H, Hamprecht FA, Mietzner T. 1998. Three-dimensional
quantitative similarity–activity relationships (3D QSAR) from SEAL
similarity matrices. J Med Chem 41:2553–2564.
17. Golbraikh A, Tropsha A. 2002. Beware of q2! J Mol Graph Model
20:269–276.
18. Fang H, Tong WD, Shi LM, Blair R, Perkins R, Branham W, Hass BS,
Xie Q, Dial SL, Moland CL, Sheehan DM. 2001. Structure–activity
relationships for a large diverse set of natural, synthetic, and environmental estrogens. Chem Res Toxicol 14:280–294.
19. Mercado-Feliciano M, Bigsby RM. 2008. Hydroxylated metabolites
of the polybrominated diphenyl ether mixture DE-71 are weak
estrogen receptor alpha ligands. Environ Health Perspect 116:1315–
1321.
20. Anstead GM, Carlson KE, Katzenellenbogen JA. 1997. The estradiol
pharmacophore: Ligand structure—estrogen receptor binding affinity
relationships and a model for the receptor binding site. Steroids 62:268–
303.
21. Sun H, Xu XL, Qu JH, Hong X, Wang YB, Xu LC, Wang XR. 2008.
4-Alkylphenols and related chemicals show similar effect on the function
of human and rat estrogen receptor a in reporter gene assay.
Chemosphere 71:582–588.
22. Fang H, Tong WD, Perkins R, Soto AM, Prechtl NV, Sheehan DM. 2000.
Quantitative comparisons of in vitro assays for estrogenic activities.
Environ Health Perspect 108:723–729.
23. Celik L, Lund JDD, Schiøtt B. 2008. Exploring Interactions of
endocrine-disrupting compounds with different conformations of the
human estrogen receptor a ligand binding domain: A molecular docking
study. Chem Res Toxicol 21:2195–2206.
24. Layton AC, Sanseverino J, Gregory BW, Easter JP, Sayler GS, Schultz
TW. 2002. In vitro estrogen receptor binding of PCBs: Measured activity
and detection of hydroxylated metabolites in a recombinant yeast assay.
Toxicol Appl Pharmacol 180:157–163.
25. Aptula AO, Roberts DW, Cronin MTD, Schultz TW. 2005. Chemistry–
toxicity relationships for the effects of diand trihydroxybenzenes to
Tetrahymena pyriformis. Chem Res Toxicol 18:844–854.