An efficient steroid pharmacophore-based strategy to identify new aromatase inhibitors

Article abstract of DOI:10.1016/j.ejmech.2009.05.003

Aromatase, an enzyme involved in the conversion of androgens into estrogens, is an important target for the endocrine treatment of breast cancer. Aromatase inhibition is usually achieved with steroids structurally related to the substrate of catalysis or,

Full text of DOI:10.1016/j.ejmech.2009.05.003

European Journal of Medicinal Chemistry 44 (2009) 4121–4127
Contents lists available at ScienceDirect
European Journal of Medicinal Chemistry
journal homepage: http://www.elsevier.com/locate/ejmech
Original article
An efficient steroid pharmacophore-based strategy to identify
new aromatase inhibitors
,
a,
**
Marco A.C. Neves a, Teresa C.P. Dinis b, Giorgio Colombo c , M. Luisa Sa e Melo
*
´
a
ˆ
´
ˆ
´
Centro de Estudos Farmaceuticos, Lab. Quımica Farmaceutica, Faculdade de Farmacia, Universidade de Coimbra, 3000-548 Coimbra, Portugal
Centro de Neurociencias, Lab. Bioquımica, Faculdade de Farmacia, Universidade de Coimbra, 3000-548 Coimbra, Portugal
c Istituto di Chimica del Riconoscimento Molecolare, CNR, 20131 Milano, Italy
b
ˆ
´
´
a r t i c l e i n f o
a b s t r a c t
Article history:
Aromatase, an enzyme involved in the conversion of androgens into estrogens, is an important target for
the endocrine treatment of breast cancer. Aromatase inhibition is usually achieved with steroids struc-
turally related to the substrate of catalysis or, alternatively, with azole non-steroid compounds.
Received 20 November 2008
Accepted 7 May 2009
Available online 19 May 2009
Substituted androstenedione derivatives with D1 D6 and D1,6 unsaturations and 6-alkyl/6-phenyl
,
aliphatic substitutions, are among the most potent steroid aromatase inhibitors known to date. In this
paper we have combined the common pharmacophoric and shape features of these molecules into a new
pharmacophore model, useful for virtual screening of large compound databases. Small subsets of the
best fitting anti-aromatase candidates were extracted from the NCI database and experimentally tested
on an in vitro assay with human placental aromatase. New potent aromatase inhibitors were identified
such as compounds 8 and 14. Considering the lack of a crystal structure for the aromatase enzyme, this
ligand-based method is a valuable tool for the virtual screening of new aromatase inhibitors.
Ó 2009 Elsevier Masson SAS. All rights reserved.
Keywords:
Pharmacophore modeling
Virtual screening
Aromatase inhibitors
Breast cancer
1. Introduction
site cavity as the natural substrate [4]. Formestane and exemestane,
second and third generation aromatase inhibitors, are successful
Estrogen deprivation is an effective approach for the endocrine
treatment of hormone sensitive breast cancer in postmenopausal
women. Aromatase, the enzyme responsible for the conversion of
androgens into estrogens, is therefore an important pharmacolog-
ical target [1]. The aromatization reaction is a three-step trans-
formation involving two hydroxylations at the 19-methyl group of
androstenedione and testosterone, and a final oxidative decar-
bonylation. Each reaction consumes a single mole of molecular
oxygen and NADPH [2].
Since androstenedione is the preferred substrate for the enzyme
[3], the initial development of aromatase inhibitors was focused on
this basic scaffold, substituted at several positions. Most of these
molecules are competitive inhibitors and bind to the same active
examples of steroid aromatase inhibitors developed with this
approach. Indeed, both compounds have been approved for clinical
use [5,6]. Other examples of potent inhibitors include androstene-
dione derivatives with D1 6 and 1,6 unsaturations and 6-n-alkyl or
, D D
6-n-phenyl aliphatic substitutions [7–10]. These compounds high-
lighted the presence of a hydrophobic pocket close to the C6 posi-
tion of the steroid nucleus. The length and shape of this substitution
were found to be critically important to the activity [11].
Besides competitive aromatase binding, several steroid aroma-
tase inhibitors are converted by the enzyme into reactive inter-
mediates, which are able to cause time-dependent inactivation.
These compounds are known as mechanism-based inactivators.
The activation step is triggered during a normal catalytic process
and depends on the presence of NADPH. Typically, a reactive
electrophilic intermediate is formed and immediately reacts with
a nucleophilic residue within the active site [12].
Although a crystallographic 3D structure of the aromatase
enzyme is still not available, several X-ray structures of homologous
mammalian and human cytochrome P450 enzymes [13–16] have
been used as templates to build homology models [17–19]. Key
atomic details that ultimately determine molecular interactions were
identified in structure-based studies and used in the rational design
of new aromatase inhibitors [20–22]. Despite clear advances in the
Abbreviations: AG, aminoglutethimide; ESP, electrostatic surface potential; HBA,
hydrogen bond acceptor; HOMO, highest occupied molecular orbital; HYD,
hydrophobic group; LUMO, lowest unoccupied molecular orbital; NCI, National
Cancer Institute; PSA, polar surface area; RMSD, root mean square deviation; VS,
virtual screening.
* Corresponding author. Tel.: þ39 0228500031; fax: þ39 0228901239.
** Corresponding author. Tel.: þ351 239488475; fax: þ351 239488471.
E-mail addresses: g.colombo@icrm.cnr.it (G. Colombo), samelo@ci.uc.pt
(M.L. Sa´ e Melo).
0223-5234/$ – see front matter Ó 2009 Elsevier Masson SAS. All rights reserved.
doi:10.1016/j.ejmech.2009.05.003

4122
M.A.C. Neves et al. / European Journal of Medicinal Chemistry 44 (2009) 4121–4127
Fig. 1. A) Training set of C6-substituted steroid aromatase inhibitors used for the common-features pharmacophore model generation [7–10]. B) Common-features pharmacophore
model of C6-substituted steroid aromatase inhibitors. The STR-HYP pharmacophoric query had five features: two hydrogen bond acceptors (HBA1 and HBA2, green) and three
hydrophobic groups (HYD1, HYD2 and HYD3, cyan). The training set inhibitors (1, cyan, 2, brown, 3, yellow, 4, blue, 5, green, 6, purple) are represented at the best fit alignment to
the model. (For interpretation of the references to colour in figure legends, the reader is refered to the web version of this article).
overall quality of aromatase homology models, their usefulness in
ligand–protein high throughput docking experiments of large
compound databases remains to be demonstrated. On the other
hand, ligand-based pharmacophore models such as 3D-QSAR CoMFA
developed for several classes of aromatase inhibitors, led to the
design of very potent molecules [23–26], and a recent study by
Langer and coworkers highlighted the value of pharmacophore
models in virtual screening of large electronic compound databases,
using a model derived from potent non-steroid aromatase inhibitors
[27].
In this work, we have summarized information about C6-
substituted androstenedione derivatives, potent steroid aromatase
inhibitors, into a ligand-based strategy to identify new anti-
aromatase hits. A pharmacophore model recapitulating the most
essential structural and functional features linked to activity was
built and used to screen the NCI database. The most fitted hits were
evaluated experimentally and new potent aromatase inhibitors
were identified.
Each point accounts for an important chemical feature, such as
hydrogen bond donors/acceptors, hydrophobic groups, negative/
positive ionizable groups and aromatics. Due to the basic structures
of the compounds used, hydrogen bond acceptors (HBA) and
hydrophobic groups (HYD) were selected for the common-features
alignment procedure. Besides its strong potency, this training set of
molecules was chosen in order to account for the effect of different
lengths, shapes and volumes of the hydrophobic moiety at C6, as
well as its stereochemistry in relation to the steroid framework.
Ten different pharmacophore hypotheses were automatically
generated by the Catalyst software, with four or five pharmaco-
phore features and alignment scores ranging from 54 to 66. The
HipHop algorithm used in this study scores each alignment based
on the degree of superimposition of all training set compounds, as
well as its estimated rarity [30]. The top ranked solutions had two
HBA groups and three HYD moieties, whereas less ranked solutions
had only two acceptors and two hydrophobic groups. Visual
inspection of the training set molecules aligned to the top ranked
solution, STR-HYP (Fig. 1B), revealed that the two hydrogen bond
acceptors matched the 3-oxo (HBA1) and the 17-oxo (HBA2)
groups. One of the hydrophobic groups, HYD1, superimposes the
19-methyl and the A–B ring junction, HYD2 matches ring C and the
18-methyl, and the third apolar feature, HYD3, is related to the
hydrophobic moiety linked at C6. Due to the rigid nature of the
steroid scaffold, most of the common-features solutions were very
similar to the top ranked pharmacophore model, with slight
differences in the projection vectors of HBA1 and HBA2 (the loca-
tion of hypothetical hydrogen bond donors) and the positioning of
HYD3. The top ranked solution was therefore selected for the
following steps.
2. Results and discussion
2.1. Pharmacophore modeling
Androstenedione derivatives are among the most potent steroid
aromatase inhibitors found to date. In particular, it has been
postulated that n-alkyl and n-phenyl aliphatic groups linked at
position C6 increase the anti-aromatase activity due to the pres-
ence of a hydrophobic cavity at the enzyme binding site. In this
sense, we have used a training set of potent C6-substituted
androstenedione derivatives reported in the literature [7–10]
(Fig. 1A), namely, 6
b-ethylandrosta-1,4-diene-3,17-dione (1), 6-
ethylandrosta-1,4,6-triene-3,17-dione (2), 6-n-propylandrosta-
1,4,6-triene-3,17-dione (3), 6-benzylandrosta-4,6-diene-3,17-dione
Table 1
NCI database screening results.
(4), 6a-phenethylandrost-4-ene-3,17-dione (5) and 6-phenethy-
Hypothesis
Hits
landrosta-1,4,6-triene-3,17-dione (6), to derive a common-features
pharmacophore model using the HipHop [28] algorithm of the
Catalyst software [29]. Briefly, the program identifies chemical
features common to a training set of active compounds and
generates hypotheses for their activity. These hypotheses are
spatial dispositions of pharmacophoric points providing the
compounds’ relative alignment in the binding site of the enzyme.
STR-HYP
16 212
2189
1462
19
STR-HYP þ Shapea
STR-HYP þ Shape þ Filteringb
Visual inspection
a
Compound 5 was converted into a shape query and combined with the initial
hypothesis. The minimum similarity tolerance was set to 0.5.
b
Filters applied: Lipinski Rule of Five, rotatable bonds ꢀ 8, PSA < 150.

M.A.C. Neves et al. / European Journal of Medicinal Chemistry 44 (2009) 4121–4127
4123
Fig. 2. NCI database hits selected based on the STR-HYP þ Shape pharmacophore model.
Cytochrome P450 enzymes have an inner binding cavity
accessible from the outside through one or more ligand channels
[31]. Therefore, in order to bind to the active site, aromatase
inhibitors must have appropriate shape and volume. A set of
inclusion volumes based on the shape of compound 5 were applied
to STR-HYP, and the steric tolerance was adjusted to allow good
shape complementarity with the training set molecules. A new
pharmacophore model combining pharmacophoric and shape
features was obtained (STR-HYP þ Shape).
expected to be false positives due to the presence of protruding
groups that might clash with aromatase binding site residues.
Furthermore, in order to increase the ‘‘drug-likeness’’ of the new
anti-aromatase candidates, several filters were applied, namely
a Lipinski Rule of Five filter [34], and filters based on the maximum
number of rotatable bonds (not more than 8) and polar surface area
(PSA < 150). This procedure reduced the number of compounds to
1462 (0.5% of the database, Table 1). The molecules were then
superimposed with the pharmacophore model and visually inspec-
ted, 19 of them being selected based on a good root mean square
deviation(RMSD)fittothemodel. Of these,10 wereavailablefromthe
NCI database (Fig. 2) and were obtained for experimental evaluation.
Before performing the biochemical evaluation, the hits were
inspected on a large electronic collection of organic chemistry
(CrossFire Beilstein) using the MDL CrossFire Commander [35].
Additional searches were performed using the PubChem
Compound database [36], a publicly available resource with
chemical and biological information of small molecules, including
results from NCI anti-cancer drug screenings. It was found that
none of them had been previously tested experimentally as aro-
matase inhibitor.
2.2. Virtual screening
The NCI open chemical repository collection is a large library of
synthetic and natural compounds, with more than 260 000
different structures [32]. This library has been used to screen, both
in vitro and in vivo, for new anti-cancer and anti-viral agents, with
the goal of identifying and evaluating novel chemical leads and
their underlying biological mechanisms of action. The electronic
version of the NCI repository (NCI database) was downloaded from
the ZINC website [33] and converted into a multiconformer data-
base using the catDB utility program of the Catalyst software [29].
An initial virtual screening (VS) run with the STR-HYP pharma-
cophore model identified 16 212 hits (5.6% of the database, Table 1),
and the modified pharmacophore hypothesis, STR-HYP þ Shape,
yielded 2189 hits (0.8% of total number of compounds, Table 1).
Most of the molecules excluded, based on shape features, are
2.3. Biochemical evaluation
The compounds selected using our ligand-based VS strategy
were biochemically evaluated for the ability to inhibit the enzyme
aromatase. The molecules were initially screened at 10
mM and
100 M concentrations, followed by a full concentration–response
m
Table 2
study, allowing the determination of the half-maximal inhibitory
concentration (IC50) reported in Table 2. Formestane and amino-
glutethimide (AG), second and first generation aromatase inhibi-
tors, were also tested, in the same assay conditions, as reference
compounds.
Aromatase inhibition activity of NCI hits selected based on the STR-HYP þ Shape
pharmacophore model. Aminoglutethimide and formestane were tested as refer-
ence aromatase inhibitors.
Compound id
NCI code
IC50
–
0.274 ꢁ 0.004
9.8 ꢁ 0.2
15.5 ꢁ 0.1
–
(
m
M)a
Inhibition at 100 mM
21%b
–
–
7
8
9
10
11
12
13
14
NSC76982
NSC93358
NSC94891
NSC122427
NSC136718
NSC302379
NSC383467
NSC613604
NSC688803
NSC692587
–
–
Table 3
N.O.c
Enzyme kinetic parameters for compounds 8 and 14, and type of aromatase inhi-
bition, based on a kinetic and time-dependent inactivation procedure.
–
N.O.c
126 ꢁ 15
0.678 ꢁ 0.007
176 ꢁ 3
–
–
–
26%
–
b
Compound id
Ki (
m
M)a
Type of inhibitiona
KI (
m
M)b
kinact (minꢂ1
)
15
16
8
14
0.266 ꢁ 0.002
0.385 ꢁ 0.008
Competitive
Competitive
N.O.c
21.4 ꢁ 1.0
N.O.c
0.608 ꢁ 0.044
–
Aminoglutethimide
Formestane
10.0 ꢁ 0.1
a
Apparent inhibition constants (Ki) were calculated by a nonlinear regression
analysis using the Michaelis–Menten equation, and the type of inhibition was
determined by a Lineweaver–Burk plot.
–
0.092 ꢁ 0.004
–
a
Results are shown as the mean ꢁ SEM of three independent experiments.
b
c
b
Inhibition at 30
m
M.
KI and kinact were obtained by a Kitz–Wilson plot.
Inactivation was not observed at concentrations ꢀ2
c
Inhibition was not observed at concentrations ꢀ100
m
M.
mM.

4124
M.A.C. Neves et al. / European Journal of Medicinal Chemistry 44 (2009) 4121–4127
more potent than AG. However, these two molecules are less active
than formestane (IC50 ¼ 0.092
mM), the second generation aroma-
tase inhibitor tested.
Kinetic analysis of the enzyme activity was also performed. The
M) and
kinetic constants, Michaelis–Menten constant (Km ¼ 0.094
m
maximum velocity of catalysis (Vmax ¼ 163.7 pmol of sub-
strate minꢂ1 mgꢂ1 of protein) were calculated under initial velocity
conditions. The type of inhibition was characterized using a Line-
weaver–Burk plot. As expected, the most potent aromatase inhib-
itors (compounds 8, Ki ¼ 0.266
m
M, and 14, Ki ¼ 0.385
mM),
inhibited the enzyme in a competitive manner (Table 3).
Compounds 8 and 14 were further tested for their ability to
cause time-dependent inactivation of aromatase. Compound 14,
but not compound 8, was able to inactivate the enzyme in the
presence of NADPH, with a pseudo first order kinetics during
the first 12 min of incubation (Fig. 3). Kitz–Wilson analysis [37] of
the results obtained, gave a kinact of 0.608 minꢂ1 and KI of 21.4
mM.
Since the Ki observed from the competition kinetics is lower than
the KI obtained from the inactivation experiments, this suggests
that the covalent binding of the inhibitor to the active site of the
enzyme is the rate-limiting step of the inactivation. On the other
hand, addition of substrate androstenedione in excess prevented
inactivation (Fig. 4A), as well as not including NADPH in the
medium (Fig. 4B). This suggests that the inhibitor acts at or near
the active site of aromatase, and, since NADPH was essential for the
time-dependent aromatase activity loss by compound 14, that the
inhibitor transformation into a reactive intermediate depends on
Fig. 3. Time- and concentration-dependent inactivation of human placental aromatase
by compound 14 in the presence of NADPH. The concentrations of inhibitor used were
0 mM (-), 2.5 mM (:), 5 mM (;) and 7.5 mM (A). A Kitz–Wilson plot of the same data
is shown in the inset. Each point represents the mean of three independent assays and
the vertical bars, the standard error of the mean (SEM).
enzyme catalysis. Furthermore, the nucleophilic trapping agent L-
cysteine did not prevent enzyme inactivation to a significant extent
(Fig. 4C), suggesting a covalent bond formation at the active site,
between aromatase and the reactive electrophilic intermediate,
therefore preventing diffusion of the activated inhibitor to the
surrounding media.
Most of the compounds selected showed anti-aromatase activity
in the assay conditions used (Table 2). Compounds 7, 13, 15 and 16
are weak aromatase inhibitors with IC50 higher than 100
Compounds 9 (IC50 ¼ 9.8 M) and 10 (IC50 ¼ 15.5 M), have anti-
aromatase potencies comparable to the first generation aromatase
inhibitor tested, aminoglutethimide (IC50 ¼ 10.0 M), and, more
interestingly, compounds 8 and 14 have an IC50 in the nanomolar
range. Compound 8 (IC50 ¼ 0.274 M) is 36 times more active than
aminoglutethimide, and compound 14 (IC50 ¼ 0.678 M) 15 times
mM.
m
m
2.4. Stereoelectronic characterization
m
The strong anti-aromatase activity of compound 8, a B-nor
steroid with similar hydrophobic core compared to the substrate
androstenedione, prompted us to further evaluate the molecular
geometry and electronic properties of these structures using high
m
m
Fig. 4. Effect of androstenedione (A), NADPH (B) and L-cysteine (C) on the time-dependent inactivation of human placental aromatase by compound 14. A) 7.5
nedione was incubated with (-) or without (:) 7.5 M of inhibitor. Incubations of 7.5 M of inhibitor without androstenedione (;) were also performed. B) 7.5 M of inhibitor was
incubated with (:) or without (-) NADPH. C) 0.5 mM of L-cysteine was incubated with (-) or without (:) 7.5 M of inhibitor. Incubations of 7.5 M of inhibitor without L-cysteine
(;) were also performed. Each point represents the mean of three independent assays and the vertical bars, the standard error of the mean.
mM of androste-
m
m
m
m
m

M.A.C. Neves et al. / European Journal of Medicinal Chemistry 44 (2009) 4121–4127
4125
Fig. 5. A) Superimposition of the minimized structures of androstenedione (grey) and B-nor-androstenedione (white) at the ab initio HF/6-31G** level. The molecules were
superimposed based on RMSD of carbon atoms at the A, C and D rings, and represented on a side (right) and top view (left). B) Electrostatic surface potential (ESP), HOMO and LUMO
valence orbitals derived for androstenedione (top) and B-nor-androstenedione (bottom). The ESP was mapped on the 0.02 e/Å3 electron density isocontour derived from ab initio
HF/6-31G** calculations (V ¼ 0.1 eV, blue; V ¼ ꢂ0.1 eV, red). The HOMO and LUMO are represented at an orbital amplitude of 0.1 (blue) and ꢂ0.1 (red). (For interpretation of the
references to colour in figure legends, the reader is refered to the web version of this article).
level ab initio quantum chemistry methods. It was found that the
geometries of androstenedione and its B-nor derivative are very
superimposable, with an RMSD of 0.31 Å based on pairwise align-
ment of the A, C and D ring carbons (Fig. 5A). Furthermore, the
distances between the hydrogen bond acceptor groups, an impor-
tant pharmacophoric feature in our model, are very similar (10.44 Å
in androstenedione and 10.38 Å in the B-nor derivative). Slight
differences were however identified, namely the shape and size of
the B ring, and the location of the acceptor linked to the ring A. The
B ring of the nor-steroid, a cyclopentane, adopts an envelope
conformation which is less bulky than the cyclohexane chair in
androstenedione. The 3-oxo groups are located 0.63 Å apart based
on our superimposition.
Electronic properties of the molecules were also calculated,
namely the electrostatic surface potential (ESP) and the valence
orbitals, i.e. the highest occupied molecular orbital (HOMO) and the
lowest unoccupied molecular orbital (LUMO). These properties are
similar in both compounds (Fig. 5B). Negative potential was found
in both carbonyls and along the O]C3–C4]C5 conjugation due to
pharmacophoric features with steric restrictions and ‘‘drug-like-
ness’’ filters allowed the isolation of a small subset, enriched in
strong aromatase inhibitors, from the large NCI database.
The screening methodology was validated experimentally by
testing some of the most promising VS hits on an in vitro assay, and
new potent aromatase inhibitors were found. 6-Methyl-B-nor-
androstenedione (8) was one of the most interesting compounds
identified, with a low nanomolar IC50 and a competitive mechanism
of inhibition. The strong anti-aromatase potency was rationalized
based on structural and physicochemical similarities between the
B-nor-androstenedione scaffold and the natural substrate of the
enzyme. To the best of our knowledge, this is the first report of
B-nor-androgens as aromatase inhibitors. These compounds repre-
sent an important new structural class of anti-aromatase agents and
should be further optimized. Compound 14 was another interesting
molecule identified, combining strong competitive inhibition
properties with mechanism-based inactivation of the enzyme.
Compounds 9 and 10 had anti-aromatase potencies comparable to
aminoglutethimide.
p
electron delocalization. The HOMO and the LUMO are located at
The value of experimentally validated virtual screening approa-
ches of large compound databases relies on fast and affordable
identification of new hit compounds for particular targets of interest.
However, hits identified with such approaches are usually non-opti-
mized structures. Therefore, it is not surprising that the new aroma-
tase inhibitors reported in this study are less potent than formestane.
Nonetheless, starting with a training set of compounds from a single
class, we were able to increase the chemical diversity of aromatase
inhibitors, identifying interesting new scaffolds which can be further
explored by lead optimization.
In conclusion, we have described and validated a new ligand-
based VS methodology for new aromatase inhibitors based on
a pharmacophore model with common-features of steroid inhibi-
tors. The screening of a large compound database was very fast and
new potent and chemically diverse aromatase inhibitors could be
the A-ring, on the delocalized system. Therefore, both compounds
are expected to share a similar aromatase recognition mechanism
and reactivity.
3. Conclusions
In this paper we have built a new pharmacophore model for an
important class of aromatase inhibitors and used it in a virtual
screening study for new anti-aromatase hits. Previous knowledge
on the binding determinants of C6-substituted androstenedione
derivatives to the aromatase active site was essential to this ligand-
based approach. A hydrophobic pocket close to the C6 of steroid
inhibitors was explored to improve the binding affinity of the new
anti-aromatase candidates. Therefore, the combination of essential

4126
M.A.C. Neves et al. / European Journal of Medicinal Chemistry 44 (2009) 4121–4127
identified. Moreover, this methodology has a broader application
for a large variety of compound databases.
Microsomal protein content was determined by the biuret
method using bovine serum albumin as standard.
4.5. Concentration–response and kinetic studies
4. Computational and experimental methods
Aromatase activity was evaluated according to the method
described by Siiteri and Thompson [40]. The concentration–
response and initial velocity experiments were performed as
4.1. Materials and general methods
The NCI selected compounds were obtained from the Drug
Synthesis and Chemistry Branch, Developmental Therapeutics
Program, Division of Cancer Treatment and Diagnosis of the National
Cancer Institute. DL-Aminoglutethimide, androstenedione, for-
mestane and NADPH were purchased from Sigma–Aldrich (St. Louis,
previously described [20,41]. Briefly, microsomal protein (30
m
g),
[1b
-3H] androstenedione (6.6 ꢄ 105 dpm) and NADPH (270
mM)
were used for the concentration–response experiment with an
incubation time of 20 min. The molecules in study were initially
MO, U.S.A.). The [1b
-3H] androstenedione (specific activity of 25.3 Ci/
tested at 10
concentration–response study with at least
ranging from 0.01 M to 160 M. For the initial velocity study the
concentration of [1
-3H] androstenedione was varied from 7.5 to
mM and 100
mM concentrations, followed by a full
8
concentrations
mmol) and the liquid scintillation cocktail Optiphase Hisafe 2 were
purchased from PerkinElmer (Boston, MA, U.S.A.). Radioactive
samples were counted on a Packard Tri-Carb 2900 TR Liquid Scin-
tillation Analyzer. All the other reagents were of adequate grade for
biochemical analysis.
m
m
b
100 nM and the incubation time was set to 5 min. Three different
concentrations of each inhibitor were tested. The tritiated water
formed during the conversion of the tritiated substrate, [1b
-3H]
androstenedione, to estrone was quantified by liquid scintillation
counting. Each assay was performed three times in duplicate and
the results were treated by nonlinear regression analysis.
4.2. Pharmacophore modeling
Pharmacophore design was performed using the Catalyst soft-
ware [29]. The aromatase inhibitors were initially submitted to the
catDB utility program and a conformational search with internal
energy minimization was performed using the best quality gener-
ation type. A maximum of 250 conformers were saved within an
energy window of 20 kcal/mol above the global minimum. The
HipHop algorithm [28] of Catalyst was used for the common-
features pharmacophore model, using hydrophobic and hydrogen
bond acceptor functions. The ‘‘Principal’’ value was set to 2 for
molecule 3 (all features in the molecule were considered to build
the pharmacophore model) and 1 for the other compounds (at least
one mapping for each generated hypothesis was found). The
‘‘maximum omitted features’’ value was set to 1 for all the mole-
cules (all but one feature was forced to map) and default settings
were used for the other options. A similarity tolerance of 0.5 was
used in the shape query. This value was chosen in order to match all
training set molecules.
4.6. Time-dependent inactivation assay
Several concentrations of compounds 8 and 14 (up to ca.10 times
the IC50) were incubated at 37 ꢃC in a medium containing sodium
phosphate buffer (67 mM), pH 7.5, microsomal protein (300 g) and
m
NADPH (900 M), in a final volume of 500 L. Aliquots (50 mL) were
m
m
removed in duplicate at several times (0, 4, 8 and 12 min), and
immediately diluted in sodium phosphate buffer (67 mM), pH 7.5,
[1
b
-3H] androstenedione (6.6 ꢄ 105 dpm) and NADPH (270
mM) in
a final volume of 500 m
L. The mixturewas then incubated at 37 ꢃC for
20 min, and the extent of the aromatization reaction was deter-
mined by liquid scintillation counting as described previously. Each
assay was performed three times. First order inactivation constants
(kobs), at each inactivator concentration, were obtained from the
slope of linear regressions of log aromatase activity remaining
versus incubation time plots, multiplied by 2.303. The KI and kinact
were determined from the slope and y intercept of a Kitz–Wilson
plot [37], respectively. Inactivation studies in the absence of NADPH
were performed in the same manner, but NADPH was omitted in the
initial incubation. For the same studies in the presence of andros-
4.3. Virtual screening
The NCI database was downloaded from the 2007 release of the
ZINC database [33] and converted into a multiconformer Catalyst
database using the ‘‘FAST’’ conformational analysis model of the
catDB utility program. A maximum of 100 conformations were
generated and saved for each molecule. Pharmacophore searches
were performed with ‘‘fast flexible database search’’ settings.
Instant JChem [38] was used for management, search and
prediction of molecular descriptors for the NCI hits. A Lipinski Rule
of Five [34] filter was applied (molecular weight under 500 g/mol,
not more than 5 hydrogen bond donors, not more than 10 hydrogen
bond acceptors and calculated partition coefficient, c log P, less than
5), as well as a filter based on the maximum number of rotatable
bonds (not more than 8) and the maximum PSA (less than 150).
tenedione or
L-cysteine, the substrate (7.5 mM) or L-cysteine
(0.5 mM) was included in the initial incubation.
4.7. Ab initio calculation details
The minimum energy conformations and electronic properties
were determined by ab initio quantum chemistry calculations.
Building blocks from the standard libraries of MAESTRO [42] were
used to generate the initial geometry for the molecules in study,
followed by
a conformational search with the Systematic
Unbounded Multiple Minimum (SUMM) [43] routine implemented
in MACROMODEL v8.1 [44], using the Merck Molecular Force Field
(MMFF) [45] and the Generalized Born equation/Surface Area (GB/
SA) continuum solvation model [46] with parameters for water
4.4. Human placental isolation
(dielectric constant
3 of 78). The molecular mechanics geometries
Human term placental microsomes were obtained by differen-
tial centrifugation, according to the method described by Ryan [39],
and were used as a source of aromatase. The microsomes were
obtained by differential centrifugation and were resuspended in
a buffer containing sodium phosphate (0.1 M), sucrose (0.25 M),
glycerol (20%) and dithiothreitol (0.5 mM), pH 7.4, and stored in
aliquots at ꢂ80 ꢃC until needed.
were further optimized with Gaussian 98 [47] using a split-valence
basis set with polarization d-orbitals added to heavy atoms and
polarization p-orbitals added to hydrogens (HF/6-31G**).
The optimized geometries were used to calculate electronic
properties, namely the total density, ESP, HOMO and LUMO.
Contour surfaces were represented using Molden v4.6 software
[48].

M.A.C. Neves et al. / European Journal of Medicinal Chemistry 44 (2009) 4121–4127
4127
[21] M. Cepa, G. Correia-Da-Silva, E.J.T. da Silva, F.M.F. Roleira, Y.Y. Hong, S. Chen,
N.A. Teixeira, Biol. Chem. 389 (2008) 1183–1191.
Acknowledgements
[22] T. Jackson, L.W.L. Woo, M.N. Trusselle, A. Purohit, M.J. Reed, B.V.L. Potter,
ChemMedChem 3 (2008) 603–618.
[23] S. Gobbi, A. Cavalli, A. Rampa, F. Belluti, L. Piazzi, A. Paluszcak, R.W. Hartmann,
M. Recanatini, A. Bisi, J. Med. Chem. 49 (2006) 4777–4780.
[24] F. Leonetti, A. Favia, A. Rao, R. Aliano, A. Paluszcak, R.W. Hartmann, A. Carotti, J.
Med. Chem. 47 (2004) 6792–6803.
[25] M. Recanatini, A. Cavalli, Bioorg. Med. Chem. 6 (1998) 377–388.
[26] A. Cavalli, G. Greco, E. Novellino, M. Recanatini, Bioorg. Med. Chem. 8 (2000)
2771–2780.
[27] D. Schuster, C. Laggner, T.M. Steindl, A. Palusczak, R.W. Hartmann, T. Langer, J.
Chem. Inf. Model. 46 (2006) 1301–1311.
ˆ
We thank Fundaça˜o para a Ciencia e a Tecnologia (FCT) through
POCI for financial support and the Drug Synthesis and Chemistry
Branch, Developmental Therapeutics Program, Division of Cancer
Treatment and Diagnosis of the National Cancer Institute for kindly
providing the compounds screened in this study. We also acknowl-
edge the Maternity Daniel de Matos, Coimbra, Portugal for providing
the human placenta. M. A. C. Neves thanks FCT for a Ph.D. grant
(SFRH/BD/17624/2004).
[28] O.O. Clement, A.T. Mehl, in: O.F. Gu¨ner (Ed.), Pharmacophore Perception,
Development, and Use in Drug Design, International University Line, La Jolla,
CA, 2000, pp. 69–84.
[29] Catalyst, Release Version 4.11, Accelrys Software Inc, 2007, www.accelrys.com.
[30] D. Barnum, J. Greene, A. Smellie, P. Sprague, J. Chem. Inf. Comput. Sci. 36
(1996) 563–571.
[31] K. Schleinkofer, Sudarko, P.J. Winn, S.K. Ludemann, R.C. Wade, EMBO Rep. 6
(2005) 584–589.
[32] Developmental Therapeutics Program, NCI/NIH, 2008, http://dtp.nci.nih.gov.
[33] J.J. Irwin, B.K. Shoichet, J. Chem. Inf. Model. 45 (2005) 177–182.
[34] C.A. Lipinski, F. Lombardo, B.W. Dominy, P.J. Feeney, Adv. Drug Deliv. Rev. 23
(1997) 3–25.
[35] MDL CrossFire Commander, Release Version 7.0SP2, Elsevier MDL, 2005,
www.mdli.com.
[36] PubChem Compound database, 2008, available at: http://www.ncbi.nlm.nih.
gov/sites/entrez/.
[37] R. Kitz, I.B. Wilson, J. Biol. Chem. 237 (1962) 3245–3249.
[38] Instant JChem, Release Version 2.0.0, ChemAxon, 2008, www.chemaxon.com.
[39] K.J. Ryan, J. Biol. Chem. 234 (1959) 268–272.
[40] P.K. Siiteri, E.A. Thompson, J. Steroid Biochem. 6 (1975) 317–322.
[41] M.A.C. Neves, T.C.P. Dinis, G. Colombo, M.L.S. Melo, J. Steroid Biochem. Mol.
Biol. 110 (2008) 10–17.
[42] MAESTRO, Release Version 8.0.314, Schrodinger, 2008, www.schrodinger.com.
[43] J.M. Goodman, W.C. Still, J. Comput. Chem. 12 (1991) 1110–1117.
[44] F. Mohamadi, N.G.J. Richards, W.C. Guida, R. Liskamp, M. Lipton, C. Caufield,
G. Chang, T. Hendrickson, W.C. Still, J. Comput. Chem. 11 (1990) 440–467.
[45] T.A. Halgren, J. Comput. Chem. 17 (1996) 490–519.
References
[1] I.E. Smith, M. Dowsett, N. Engl. J. Med. 348 (2003) 2431–2442.
[2] B. Meunier, S.P. de Visser, S. Shaik, Chem. Rev. 104 (2004) 3947–3980.
[3] Y. Shimizu, C. Yarborough, Y. Osawa, J. Steroid Biochem. Mol. Biol. 44 (1993)
651–656.
[4] R.W. Brueggemeier, Breast Cancer Res. Treat. 30 (1994) 31–42.
[5] R.C. Coombes, E. Hall, L.J. Gibson, R. Paridaens, J. Jassem, T. Delozier, S.E. Jones,
I. Alvarez, G. Bertelli, O. Ortmann, A.S. Coates, E. Bajetta, D. Dodwell,
R.E. Coleman, L.J. Fallowfield, E. Mickiewicz, J. Andersen, P.E. Lonning,
G. Cocconi, A. Stewart, N. Stuart, C.F. Snowdon, M. Carpentieri, G. Massimini,
J.M. Bliss, N. Engl. J. Med. 350 (2004) 1081–1092.
[6] P.E. Goss, T.J. Powles, M. Dowsett, G. Hutchison, A.M.H. Brodie, J.C. Gazet,
R.C. Coombes, Cancer Res. 46 (1986) 4823–4826.
[7] M. Numazawa, M. Oshibe, S. Yamaguchi, M. Tachibana, J. Med. Chem. 39 (1996)
1033–1038.
[8] M. Numazawa, M. Oshibe, S. Yamaguchi, Steroids 62 (1997) 595–602.
[9] M. Numazawa, S. Yamaguchi, J. Steroid Biochem. Mol. Biol. 67 (1998) 41–48.
[10] M. Numazawa, S. Yamaguchi, Steroids 64 (1999) 187–196.
[11] M. Numazawa, M. Shelangouski, M. Nagasaka, Steroids 65 (2000) 871–882.
[12] R.W. Brueggemeier, J.C. Hackett, E.S. Diaz-Cruz, Endocr. Rev. 26 (2005)
331–345.
[13] P. Rowland, F.E. Blaney, M.G. Smyth, J.J. Jones, V.R. Leydon, A.K. Oxbrow,
C.J. Lewis, M.G. Tennant, S. Modi, D.S. Eggleston, R.J. Chenery, A.M. Bridges, J.
Biol. Chem. 281 (2006) 7614–7622.
[14] P.A. Williams, J. Cosme, V. Sridhar, E.F. Johnson, D.E. Mcree, J. Inorg. Biochem.
81 (2000) 183–190.
[15] P.A. Williams, J. Cosme, A. Ward, H.C. Angova, D.M. Vinkovic, H. Jhoti, Nature
424 (2003) 464–468.
[16] P.A. Williams, J. Cosme, D.M. Vinkovic, A. Ward, H.C. Angove, P.J. Day,
C. Vonrhein, I.J. Tickle, H. Jhoti, Science 305 (2004) 683–686.
[17] A.D. Favia, A. Cavalli, M. Masetti, A. Carotti, M. Recanatini, Proteins 62 (2006)
1074–1087.
[18] S. Karkola, H.D. Holtje, K. Wahala, J. Steroid Biochem. Mol. Biol.105 (2007) 63–70.
[19] C. Loge, M. Le Borgne, P. Marchand, J.M. Robert, G. Le Baut, M. Palzer,
R.W. Hartmann, J. Enzyme Inhib. Med. Chem. 20 (2005) 581–585.
[20] M.A.C. Neves, T.C.P. Dinis, G. Colombo, M.L.S. Melo, ChemMedChem 2 (2007)
1750–1762.
[46] W. Clark Still, Anna Tempczyk, Ronald C. Hawley, Thomas Hendrickson, J. Am.
Chem. Soc. 112 (1990) 6127–6129.
[47] M.J. Frisch, G.W. Trucks, H.B. Schlegel, G.E. Scuseria, M.A. Robb, J.R. Cheeseman,
V.G. Zakrzewski, J.A. Montgomery Jr., R.E. Stratmann, J.C. Burant, S. Dapprich,
J.M. Millam, A.D. Daniels, K.N. Kudin, M.C. Strain, O. Farkas, J. Tomasi, V. Bar-
one, M. Cossi, R. Cammi, B. Mennucci, C. Pomelli, C. Adamo, S. Clifford, J.
Ochterski, G.A. Petersson, P.Y. Ayala, Q. Cui, K. Morokuma, D.K. Malick, A.D.
Rabuck, K. Raghavachari, J.B. Foresman, J. Cioslowski, J.V. Ortiz, A.G. Baboul,
B.B. Stefanov, G. Liu, A. Liashenko, P. Piskorz, I. Komaromi, R. Gomperts, R.L.
Martin, D.J. Fox, T. Keith, M.A. Al-Laham, C.Y. Peng, A. Nanayakkara, C. Gon-
zalez, M. Challacombe, P.M.W. Gill, B.G. Johnson, W. Chen, M.W. Wong, J.L.
Andres, M. Head-Gordon, E.S. Replogle, J.A. Pople, Gaussian 98, Revision A.7,
Gaussian, Inc., Pittsburgh, PA, USA, 1998, www.gaussian.com.
[48] G. Schaftenaar, J.H. Noordik, J. Comput.-Aided Mol. Des. 14 (2000) 123–134.