Pharmacophore based receptor modeling: The case of adenosine A3 receptor antagonists. An approach to the optimization of protein models

Article abstract of DOI:10.1021/jm051112+

To design and synthesize new potent and selective antagonists of the human A₃ adenosine receptor, pharmacophoric hypotheses were generated with the software Catalyst for a comprehensive set of compounds retrieved from previous literature. Three of these pharmacophores were used to drive the optimization of a molecular model of the receptor built by homology modeling. The alignment of the ligands proposed by Catalyst was then used to manually dock a set of known A₃ antagonists into the binding site, and as a result, the model was able to explain the different binding mode of very active compounds with respect to less active ones and to reproduce, with good accuracy, free energies of binding. The docking highlighted that the nonconserved residue Tyr254 could play an important role for A₃ selectivity, suggesting that a mutagenesis study on this residue could be of interest in this respect. The reliability of the whole approach was successfully tested by rational design and synthesis of new compounds.

Full text of DOI:10.1021/jm051112+

J. Med. Chem. 2006, 49, 4085-4097
4085
Pharmacophore Based Receptor Modeling: The Case of Adenosine A3 Receptor Antagonists.
An Approach to the Optimization of Protein Models
Andrea Tafi,† Cesare Bernardini,† Maurizio Botta,*,† Federico Corelli,† Matteo Andreini,‡ Adriano Martinelli,*,‡
Gabriella Ortore,‡ Pier Giovanni Baraldi,*,§ Francesca Fruttarolo,§ Pier Andrea Borea,# and Tiziano Tuccinardi‡
Dipartimento Farmaco Chimico Tecnologico, UniVersita` degli Studi di Siena, Via A. Moro, 53100 Siena, Italy, Dipartimento di
Scienze Farmaceutiche, UniVersita` di Pisa, Via Bonanno 6, 56126 Pisa, Italy, Dipartimento di Scienze Farmaceutiche, UniVersita` degli
Studi di Ferrara, Via Fossato di Mortara 17-19, 44100 Ferrara, Italy, and Dipartimento di Medicina Clinica e Sperimentale-Sezione di
Farmacologia, UniVersita` degli Studi di Ferrara, Via Fossato di Mortara 17-19, 44100 Ferrara, Italy
ReceiVed NoVember 4, 2005
To design and synthesize new potent and selective antagonists of the human A3 adenosine receptor,
pharmacophoric hypotheses were generated with the software Catalyst for a comprehensive set of compounds
retrieved from previous literature. Three of these pharmacophores were used to drive the optimization of a
molecular model of the receptor built by homology modeling. The alignment of the ligands proposed by
Catalyst was then used to manually dock a set of known A3 antagonists into the binding site, and as a result,
the model was able to explain the different binding mode of very active compounds with respect to less
active ones and to reproduce, with good accuracy, free energies of binding. The docking highlighted that
the nonconserved residue Tyr254 could play an important role for A3 selectivity, suggesting that a mutagenesis
study on this residue could be of interest in this respect. The reliability of the whole approach was successfully
tested by rational design and synthesis of new compounds.
Introduction
2,4-dialkylpyridine17-19 and pyridine derivatives,20,21 have been
synthesized and tested as hA3AR antagonists.
Adenosine is an ubiquitous neuromodulator that acts by
stimulating four cell surface receptors (A1, A2A, A2B, A3), all
being part of the huge family of the G-protein-coupled receptors
(GPCRs). In particular, A3 receptors, discovered in 1992 by
Zhuo et al.,1 have been reported to be activated by a higher
concentration of their natural ligand with respect to other
subtypes, suggesting a pathophysiological role during hypoxic
stress and other cellular damage.2 While A3 activation results
in general hypotension and mast-cells degranulation,3,4 selective
antagonists of these receptors may be used in clinical practice
as anti-inflammatory5 as well as cerebroprotective6,7 and anti-
asthmatic agents.8
To design and synthesize a new generation of potent and
selective antagonists with improved absorption, distribution,
metabolism, and excretion (ADME) profiles as well as increased
water solubility compared to previous analogues, we set up a
drug design approach, in which ligand-based information and
homology modeling were combined throughout to increase the
success possibilities. Many studies have been published in which
the docking procedure was used as an alignment tool for the
development of 3D-QSAR models.22-24 Furthermore, some
recent studies have demonstrated that GPCR models of higher
accuracy can be produced if homology modeling, based on the
rhodopsin X-ray template, is supported by experimental struc-
tural constraints appropriate for active or inactive receptor
conformations25 or if crude models are optimized by including
ligand-based information.26 Regarding this last approach, Klebe
and co-workers have recently developed the MOBILE method,
where homology models are refined by including information
about bioactive ligands as spatial restraints,26 and have applied
it to the discovery of neurokinin-1 receptor antagonists.27
In the past decade, because of the importance of this new
biological target, a great effort has been made to design and
synthesize new potent and selective agonists and antagonists
of the human A3 adenosine receptor (hA3AR). Because of the
lack of experimental 3D structural data about the hA3AR binding
site, the rational design of hA3AR antagonists has been
commonly pursued through the construction of receptor models
or, alternatively, by QSAR/3D-QSAR approaches (such as
CoMFA) using experimental data obtained by previously
synthesized ligands. As a result of all this reseach, several classes
of compounds, including pyrazolo[4,3-e]-1,2,4-triazolo[1,5-c]-
pyrimidine derivatives,9-12 triazoloquinazoline derivatives,13
isoquinoline and quinazoline analogues,14-16 and 3,5-diacyl-
In this study we have actually merged the capabilities of two
generally complementary computational methodologies, that is,
pharmacophore28 and homology modeling,29 to build a predictive
three-dimensional model of the human A3 receptor. The aim of
our approach was to gain information from a set of ligands by
means of a 3D-QSAR methodology to drive the refinement of
the receptor model and to increase the probability of identifying
the most important features for ligand recognition.
* To whom correspondence should be addressed. For M.B.: phone, +39-
577-234306; fax, +39-577-234333; e-mail, botta@unisi.it. For A.M.: phone,
+39-50-2219556; fax, +39-50-2219605; e-mail, marti@farm.unipi.it. For
P.G.B.: phone, +39-532-291293; fax, +39-532-291296; e-mail, baraldi@
unife.it.
Accordingly, we first generated several common feature
hypotheses for human A3 receptor antagonists and screened the
three most promising pharmacophores. These pharmacophores
were all characterized by good statistical parameters and
prediction ability; moreover, they were very similar in terms
of feature composition and disposition in three-dimensional
space.
† Universita` degli Studi di Siena.
‡ Universita` di Pisa.
§ Dipartimento di Scienze Farmaceutiche, Universita` degli Studi di
Ferrara.
# Dipartimento di Medicina Clinica e Sperimentale-Sezione di Farma-
cologia, Universita` degli Studi di Ferrara.
10.1021/jm051112+ CCC: $33.50 © 2006 American Chemical Society
Published on Web 06/21/2006

4086 Journal of Medicinal Chemistry, 2006, Vol. 49, No. 14
Tafi et al.
Table 1. hA3AR Antagonists Considered for Pharmacophore Generation and Validation
compd
R1
R2
R3
R4
R5
X
1a
1b
1c
1d
1e
NHCONH-(4-SO3H-Ph)
NH2
ethyl
ethyl
methyl
ethyl
ethyl
ethyl
ethyl
ethyl
methyl
ethyl
methyl
propyl
propyl
H
methyl
ethyl
NHCONH-(3-Cl-Ph)
NHCONH-4-CH3-Ph
NHCONH-4-F-Ph
NHCONH-(2-OCH3-Ph)
NHCONH-(2-Cl-Ph)
NHCONH-(3-Cl-Ph)
NHCONH-Ph
NHCONH-(4-OCH3-Ph)
NHCONH-4-pyridyl
NH2
1f
1g
1h
1i
1j
1k
1la
1ma
1na
1oa
1pa
1qa
1ra
2a
2b
2c
2d
2ea
3a
3b
3c
3d
3e
3f
3g
3h
3i
3j
3k
3l
NHCONH-(4-OCH3-Ph)
NH2
NHCONH-(4-OCH3-Ph)
NHCONH-(4-NO2-Ph)
NHCONH-Ph
NHCONH-(4-SO3H-Ph)
NH2
propyl
propyl
NHCO-Ph
NHCOCH2-Ph
NHCOCH2CH3
NHCOCH3
NHCO-(4-OCH3-Ph)
NHCO-Ph
2-pyridyl
2-pyridyl
2-pyridyl
2-pyridyl
2-pyridyl
2-pyridyl
2-pyridyl
2-pyridyl
H
2-pyridyl
3-pyridyl
3-CH3-2-pyridyl
N,N-diethylamino
1-pyrrolidinyl
2-pyridyl
2-pyridyl
2-pyridyl
O-ethyl
CH
CH
CH
CH
CH
CH
CH
CH
N
N
N
N
N
NHCO-(4-CH3-Ph)
NHCO-(3,4-CH3-Ph)
NHCO-(3,4-OCH3-Ph)
NHCO-3Cl-Ph
NHCO-3OCH3-Ph
NHCONH-Ph
NHCONH-Ph
NHCONH-Ph
NHCONH-Ph
NHCONH-Ph
NHCONH-Ph
NHCONH-Ph
NHCO-(4-Cl-Ph)
NHCO-(2,4-CH3-Ph)
NHCO-(3-CH3-Ph)
methyl
3m
3n
3oa
3pa
3qa
4a
4b
4c
N
CH
CH
CH
H
H
H
H
H
H
H
Cl
H
H
H
H
methyl
methyl
ethyl
ethyl
ethyl
ethyl
n-propyl
ethyl
methyl
ethyl
propyl
ethyl
COO-ethyl
COO-ethyl
COO-ethyl
COO-ethyl
COO-ethyl
COO-propyl
COO-propyl
COO-propyl
COO-ethyl
COO-ethyl
COO-ethyl
COOCH2CH2OH
COO-ethyl
COO-ethyl
COO-ethyl
methyl
methyl
O-propyl
O-ethyl
4d
4e
methyl
ethyl
O-ethyl
O-ethyl
4f
ethyl
O-ethyl
4g
4h
4ia
4ja
4ka
4la
4ma
5a
5ba
ethyl
ethyl
methyl
propyl
methyl
ethyl
ethyl
methyl
O-ethyl
O-propyl
S-ethyl
S-ethyl
S-ethyl
S-ethyl
S-ethyl
O-ethyl
S-ethyl
ethyl
ethyl
ethyl
H
methyl
methyl
ethyl
a Compounds included in the test set.
At the same time, a raw model of hA3AR was generated by
homology modeling, using bovine rhodopsin as a template,30
and several molecular interaction fields (MIFs) were calculated
for each transmembrane helix with the aim of localizing the
minimum-energy interaction points.
The three screened pharmacophores were then subjected, one
by one, to a manual docking procedure into the raw binding
site, which was adjusted as well, to achieve the highest
superposition between each centroid of the pharmacophoric
elements and a corresponding local minimum of the MIFs. Such
a procedure yielded a unique alignment for each pharmacophore
so that 12 ligand-protein complexes could be modeled on the
basis of four reference compounds with the purpose of refining
the receptor model. Each complex was in fact relaxed (according
to a protocol fully described in this report) and evaluated in
terms of energies and “feature correspondence” to pick both

Pharmacophore Based Receptor Modeling
Journal of Medicinal Chemistry, 2006, Vol. 49, No. 14 4087
Table 2. Cost, Statistical Parameters, and Composition Features
Associated with the First 10 Pharmacophoric Hypotheses Generated by
Catalyst
statistical parameter
Hypo costa rmsd r2 r2
composition featureb
1
2
3
4
5
tr-set
test-set
1
2
3
4
5
6
7
8
9
10
163.3 1.05 0.933 0.744 Hyd Hyd Hyd RA
167.8 1.17 0.917 0.806 Hyd Hyd Hyd HBA PI
168.7 1.18 0.915 0.839 Hyd Hyd Hyd HBA HBA
170.4 1.22 0.909 0.854 Hyd Hyd Hyd RA
172.1 1.26 0.903 0.827 Hyd Hyd Hyd RA
174.0 1.28 0.900 0.852 Hyd Hyd Hyd HBA HBA
174.9 1.31 0.895 0.738 Hyd Hyd Hyd RA HBA
175.0 1.30 0.897 0.840 Hyd Hyd Hyd HBA HBA
175.2 1.31 0.895 0.809 Hyd Hyd Hyd HBA HBA
175.6 1.32 0.893 0.760 Hyd Hyd Hyd RA
PI
HBA
HBA
HBA
a All costs are reported in bits. Fixed cost ) 155.3 is the cost of a
theoretical ideal hypothesis that is able to perfectly predict activities. Null
cost ) 290.8 is the cost of a hypothesis that gives no correlation between
experimental and predicted activities. b Hyd: hydrophobic. RA: ring
aromatic. PI: positive ionizable. HBA: hydrogen bond acceptor.
Figure 1. Compound 1k. The red sphere shows the positive ionizable
feature (PI) located in front of the guanidine moiety.
investigation. In the first place, the prediction of the test set
(r2test-set) was used as the selection standard. To improve this
rule, hypotheses were discarded, which estimated the activity
of the most potent compound (1k) out of the correct 10-2 nM
range. We believe indeed it was crucial that the activity of this
compound was well estimated for a hypothesis to be considered
reliable. Finally, the first two hypotheses of Table 2 were
discarded, since they contained a doubtful positive ionizable
feature (PI). A guanidine scaffold was recognized by Catalyst
in several of our hA3AR antagonists so that in a few hypotheses
a centroid was located in an area between N5, C5 and N6 of
the pyrazolotriazolopyrimidine nucleus (see the case of com-
pound 1k, shown in Figure 1), indicating the presence of a
potentially positively ionizable group. In short, the software was
not able to discriminate between a guanidine system substituted
with electron-withdrawing groups, which, as in our case, is not
protonated at physiological pH, and a basic unsubstituted
guanidine group (fairly protonated).
the optimum hA3AR model and the best performing pharma-
cophore out of the three options investigated.
The selected pharmacophore was used afterward to export
the relative alignment of four further compounds. The predictive
power of the ligand-receptor model chosen was finally
established by finding a quantitative correlation between
experimental and theoretical values of free energies of binding,
calculated by the application of a scoring function.
Finally, some novel putative A3 antagonists were designed,
synthesized, and biologically evaluated to test the reliability of
our combined modeling strategy.
Results and Discussion
Pharmacophore Generation. hA3AR antagonists are char-
acterized by great structural diversity, making it difficult to find
a common chemical pattern.9-21,31 Nevertheless, certain common
electronic and steric features have already been reported in the
literature based on a combination of ab initio calculations,
electrostatic potential map comparison, and steric and electro-
static alignment (SEAL) analyses.31 In this study we exploited
the ability of the Catalyst 4.6 software package32 to find a
common alignment for a comprehensive set of 55 A3 antagonists
(see Table 1) with the aim of explaining most of their structure-
activity relationships.
The entire set of compounds, with activity data spanning 5
orders of magnitude (from 10-2 to 103 nM), was divided into a
training set of 38 compounds and a complementary test set of
17 compounds (see Table 1), following Catalyst’s guidelines.33
In particular, (i) derivative 1k (the most potent and selective
A3 antagonist ever reported)12 was included in the training set
because the software pays particular attention to the most active
compound while generating the chemical feature space and (ii)
the types and relative positions of the chemical features
(substitution patterns) shown by the molecules were maximized
because the program recognizes the molecules as collections
of chemical features, not as assemblies of atoms or bonds.
Ten hypotheses were collected during a first pharmacophore
generation, whose details are reported in Table 2. The 12.3 bit
cost range gained over the models suggested the existence of a
strong signal generated by the training set; moreover, a
difference greater than 100 bits between the cost of each
hypothesis and the null cost was indicative of more than 90%
of true correlation.33
On the basis of such considerations, hypotheses 4 (HYPO1
in the following) and 6 (HYPO2 in the following) of Table 2
were chosen for further evaluation. HYPO1 and HYPO2
exhibited, respectively, correlation coefficients r2 ) 0.909 and
r2 ) 0.900 for the training set and r2 ) 0.854 and r2 ) 0.852
for the test set. The activity of compound 1k was well estimated
by both HYPO1 (Kcalc ) 5.6 × 10-2 nM) and HYPO2 (Kcalc
)
8.4 × 10-2 nM). In Figure 2 the entire set of features of both
pharmacophores is displayed superposed to this derivative. Both
hypotheses were characterized by five features and shared a
common scheme consisting of three hydrophobics at the vertexes
of a triangle (HYD1, HYD2, and HYD3). They differed for a
hydrogen bond acceptor (HBA1) pointing toward opposite
directions and for a hydrogen bond acceptor (HBA2), which
replaced, in HYPO2, the ring aromatic (RA) found in HYPO1.
Experimental and calculated (estimated or predicted by
Catalyst) A3 affinity values of all the compounds used in the
computational studies are shown in Table 3, while the regression,
based on HYPO1, of experimental versus estimated/predicted
affinities is shown in Figure 3.
As shown in Table 3 with regard to HYPO1 and HYPO2,
the ratio between calculated and experimentally measured Ki
values of the compounds (error columns) was generally better
than 10-fold and in most cases better than 3-fold. Taking into
account this result and considering that the strong affinity values
shown by 1k might signify that it possesses all the crucial groups
involved in ligand-receptor interactions, we could assume that
both HYPO1 and HYPO2 accounted for relevant interactions
between antagonists and hA3AR. Furthermore, the soundness
A careful analysis was performed on the generated hypotheses
with the aim of selecting the most significant ones for further

4088 Journal of Medicinal Chemistry, 2006, Vol. 49, No. 14
Tafi et al.
Figure 2. HYPO1 (a) and HYPO2 (b) superposed to compound 1k. Pharmacophoric features are color-coded: sky blue for hydrophobic (HYD),
green for hydrogen bond acceptor (HBA), and orange for aromatic ring (RA).
Figure 3. Regression of experimental versus estimated/predicted A3
affinities (nM) based on HYPO1 (logarithmic scale), for each member
of the training and test sets.
Figure 4. HYPO3 superposed to compound 1k. Pharmacophoric
features are color-coded: sky blue for hydrophobic (HYD), green for
hydrogen bond acceptor (HBA), and magenta for hydrogen bond donor
(HBD).
of both HYPO1 and HYPO2 hypotheses was supported by
literature reports that stressed that a wide π-π interaction
involving the pyrazolotriazolopyrimidine system and the C2-
furyl substituent might be crucial for A3 antagonism,31 while
hydrophilic interactions are probably responsible for A3 selectiv-
ity.
the vertexes of a triangle. No more RA features were present
in the model, while two hydrophilic interactions (one HBA and
one HBD) appeared at one side of the triangle. This hypothesis,
like HYPO1 and HYPO2, was able to estimate/predict the
activities of both training and test sets in a satisfactory manner
(see Table 3), showing indeed a correlation coefficient r2 )
0.91 for the training set and r2 ) 0.87 for the test set. Moreover,
while two inactive compounds (4a and 1b) were completely
“mis-predicted” by HYPO3, most of the other compounds were
fairly well-predicted with an error lower than (3.
In any case, on the basis of what has been already reported
by some of us about the importance of a hydrogen bonding
donor (HBD) interaction between the NHCONH group present
on most of the compounds considered in this study and hA3-
AR,9,12 we then pushed Catalyst to generate new hypotheses
with an HBD feature. Accordingly, we set up a possible
adjustment to correct our computational protocol (see the
Experimental Section for details). Catalyst’s default spacing
value (that is, the minimum distance between the location of
actual features) was hypothesized to be the critical control
parameter that had not worked properly for our set of com-
pounds. As a result of this effort, 10 new HBD-endowed
hypotheses were collected whose details are reported in Table
4, which presented similar statistical data with respect to the
ones previously generated. Among those hypotheses possessing
the HBD feature, the one ranked 9th (HYPO3 in the following)
was chosen for further investigation according to the aforemen-
tioned selection rules.
HYPO1, HYPO2, and HYPO3 were then exported and
manually docked into the receptor model as described later in
this article.
Receptor Modeling and Structure Optimization. All the
information regarding the primary structure of the human A3
receptor and the subdivision into transmembrane, cytoplasmatic,
and extracellular domains was obtained from the GPCR Data
Bank.34 A raw structure of hA3AR was obtained through
molecular modeling, using bovine rhodopsin as a template.30
The receptor-template superposition was carried out maintain-
ing the maximum analogy between them and choosing the
regions with a conserved or semiconserved sequence. The
alignment was studied on several adenosine receptors by means
of the ClustalW program35 and was guided by the highly
In Figure 4 the entire set of features of HYPO3 is displayed
superposed to compound 1k. Once again, HYPO3 presented
the recurring scheme of three hydrophobic interactions lying at

Pharmacophore Based Receptor Modeling
Journal of Medicinal Chemistry, 2006, Vol. 49, No. 14 4089
Table 3. Experimental (Ki,exp) and Calculated (Ki,calcd) Affinity Values
(nM) of All the Compounds Used in the 3D-QSAR Studies
Table 4. Cost, Statistical Parameters, and Composition Features
Associated with the Second 10 Pharmacophoric Hypotheses Generated
by Catalyst
HYPO1
compd Ki,exp Ki,calcd error
-2.6
-28.0 160
HYPO2
HYPO3
statistical parameter
Hypo costa rmsd r2 r2
composition featureb
Ki,calcd
8.4
error
Ki,calcd
6.6
error
1
2
3
4
5
tr-set
test-set
1a
1b
1c
40
15
-4.8
-6.0
-22.0
3600 130
0.4
-22.0 160
-3.5
3.3
1.5
2.8
2.2
-3.0
-1.2
1.2
1
2
3
4
5
6
7
8
9
10
176.9 1.05 0.934 0.821 Hyd Hyd Hyd Hyd HBD
178.2 1.08 0.930 0.811 Hyd Hyd Hyd Hyd HBA
183.0 1.20 0.913 0.822 Hyd Hyd Hyd HBD RA
183.7 1.18 0.916 0.821 Hyd Hyd Hyd HBA HBA
184.7 1.23 0.908 0.800 Hyd Hyd Hyd HBA HBA
185.1 1.23 0.908 0.857 Hyd Hyd Hyd HBA HBD
185.2 1.24 0.906 0.743 Hyd Hyd Hyd HBA RA
185.2 1.21 0.912 0.795 Hyd Hyd Hyd HBA HBA
185.3 1.22 0.909 0.870 Hyd Hyd Hyd HBA HBD
186.1 1.17 0.918 0.738 Hyd Hyd Hyd HBA HBD
0.12
1.0
-3.3
7.2
1.6
1.4
3.1
1.6
1.2
1.3
0.11
0.46
1.3
0.092 -4.4
1d
1e
1f
1g
1h
1i
0.14
0.86
0.56
0.3
2.1
0.16
0.6
0.41
0.86
0.49
0.58
0.67
2.9
1.0
-1.1
1.9
1.4
0.81
0.93
0.9
0.19
0.79
1.6
0.65
0.69
0.14
0.74
-3.1
0.093 -1.7
0.33
0.062 1.6
1j
-1.8
1k
1la
1ma
1na
1oa
1pa
1qa
1ra
2o
2b
2c
2d
2ea
3a
3b
3c
3d
3e
3f
3g
3h
3i
3j
3k
3l
3m
3n
3oa
3pa
3qa
4a
4b
4c
4d
4e
0.04
613
0.29
348
0.2
0.65
0.15
30
85
3
0.65
7.7
13.9
17
200
96
69
310
770
150
76
1200 1000
490
51
260
180
82
200
360
240
0.056 1.4
0.084 2.1
96
3.3
45
0.18
4500
-13
-6.4
11
7.9
-1.2
11.0
3.8
-7.1
2.9
-2.8
2.6
-13.0 8.4
13.0
11.0
2
3.3
4.4
-2.7
-3.1
1.1
99
-6.2
a All costs are reported in bits. Fixed cost ) 155.34 is the cost of a
theoretical ideal hypothesis that is able to perfectly predict activities. Null
cost ) 290.81 is the cost of a hypothesis that gives no correlation between
experimental and predicted activities. b Hyd: hydrophobic. RA: ring
aromatic. PI: positive ionizable. HBA: hydrogen bond acceptor.
-1.6
0.11
3400
-2.7
13.0
2800
0.17
7.1
0.57
4.2
240
1.1
1.7
0.61
180
180
390
310
300
120
250
160
130
680
130
120
94
9.8
0.057 -3.5
0.069 -2.9
12
19.0
2.5
-9.6
1.4
-2.5
2.9
-8.6
1.8
6.3
0.17
2.6
200
7.9
9.7
1.1
-11.0
2.3
0.38
3.1
120
1.2
1.9
0.9
24
Table 5. Mutational Analysis for the Human A3 Receptor Antagonists
Interaction
2.6
5.7
8.7
1.1
region
A3AR
mutational results
TM3
L 4-5
TM6
H95
A: reduction of affinitya
A: reduction of affinitya
A: reduction of affinitya
A: modest variationa
140
170
350
210
69
9.9
9.9
1.7
2.2
K152
W243
L244
S247
N250
H272
Y282
110
300
500
120
98
130
160
210
6.2
1.5
5.2
1.7
A: modest reduction of affinitya
A: loss of bindinga
2.9
-3.2
130
190
150
170
960
180
150
76
-2.4
TM7
E: reduction of affinitya
F: reduction of affinityb
-6
-4
1.1
1
2.2
a See ref 40. b See ref 39.
2.7
1.7
-1.2
-2.2
2.9
-2.7
-2.8
1.9
2.4
-1.5
-2
-1.7
-3.8
2.3
-2.8
-3.4
1.6
2.1
-1.3
1.5
-40.0 100
-2
1.1
3.9
1.0
12.0
8.2
2.4
-1.2
-2.8
2.9
-3.5
-4.3
1.9
230
150
97
the TM3 and TM7 had to be rotated respectively by 60°
clockwise and 90° counterclockwise (extracellular point of view)
to let them turn toward the intrahelical channel, thereby allowing
the interaction of Hys95 and Hys272 with the ligands. On the
other hand, in agreement with the findings of Gouldson et al.,37
rotations and translations of the transmembrane (TM) domains
are important steps in a ligand-receptor interaction process in
different GPCRs. Because of the antagonist profile of the ligands
considered, possible rearrangements of the receptor in an
activated form able to interact with agonists were not taken into
account. The hA3AR model so obtained was optimized through
the procedure fully described in the Experimental Section, and
the seven helices were then disassembled.
64
52
42
160
480
230
120
130
420
280
370
160
290
190
180
1.4
-1.9
-1.3
-44.0
-1.9
1.4
3.4
1.1
12.0
4.0
4500 140
-32.0 110
210
180
43
120
8.3
19
86
-2.5
-1.1
3.6
1.1
9.8
110
200
170
120
100
160
19
130
100
130
98
170
180
140
NCc
NCc
NCc
110
260
150
130
100
75
170
150
130
81
4f
4g
94
5.0
4h
4ia
4ja
4ka
4la
4ma
5a
7.9
20
40
5.0
16
2.0
130
91
6.5
2.7
6.5
3.1
100
110
97
5.2
3.3
Individual molecular interaction fields (MIFs) were calculated
for every transmembrane helix by means of the GRID program38
with the aim of determining energetically favorable interaction
sites on each one of them. Several probes were used at this
stage (see Table 6 for details) to investigate the ligand-receptor
binding mode.
33.3
194
188
13.4
380
219
5.1
2.0
34
200
80
86
1.1
-2.4
6.4
-1.4
-1.9
13
-2.1
-1.6
NCc
NCc
NCc
-2.0
-2.2
9.2
-2.7
-1.6
NCc
NCc
NCc
86
120
140
140
NCc
NCc
NCc
160
130
5.1
5.4
8.7
-2.4
-1.7
1.0
2.7
-3.9
5ba
16b
17b
18b
After the seven helices were reassembled, the receptor was
displayed together with the points of local minimum energy of
the MIFs located at their proper position in 3D space. At this
point, Catalyst hypotheses were manually docked one by one,
coupling this operation with a manual rearrangement of the
relative positions and orientations of the helices in order to (a)
superpose each pharmacophoric feature found by Catalyst on a
minimum of the corresponding MIF (see Table 6) and (b) retain
the crucial interactions between the pharmacophore features and
the most important residues, as highlighted by site directed
mutagenesis studies.39-42
a Compounds included in the test set. b New A3 antagonist designed,
synthesized and biologically evaluated, to test the reliability of the
computational approach. c Not calculated.
conserved amino acid residues (see the Supporting Information),
including the D/ERY motif (D/E3.49, R3.50, and Y3.51), the
two Pro residues P4.50 and P6.50, and the NPXXY motif in
the TM7 (N7.49, P7.50, and Y7.53).36
The homology model directly obtained using bovine rhodop-
sin as a template was not able to take into account all the
mutagenesis data reported concerning antagonists of hA3AR (see
Table 5). In particular, in contrast with these data, Hys95 and
Hys272 did not point toward the intrahelical channel. Therefore,
In this manner, three models of the receptor-antagonist
complex were obtained (MODEL1, MODEL2, MODEL3) that
differed from each other for the binding mode of the ligands as

4090 Journal of Medicinal Chemistry, 2006, Vol. 49, No. 14
Tafi et al.
Figure 5. From left to right, the three receptor complexes of hA3AR with 1k (MODEL1, MODEL2, and MODEL3) are shown, each one superposed
on the alignment between the complementary pharmacophoric hypothesis and the MIFs (displayed as spheres). In sky-blue the minima obtained
with the DRY probes are represented. In green are those calculated with O and N:d probes. In orange are those derived from the calculation with
the C1d probe, and in magenta is the one obtained with the NHd probe.
Table 6. Details of the Probes Used for MIF Calculationsa
In MODEL2, the overlap between HBA1 and the O probe
minimum as well as the correspondence between HYD3 and
the lipophilic cleft H3 were maintained. However, different from
MODEL1, HYD1 corresponded to the lipophilic cleft H2 while
HYD2 corresponded to a lipophilic interaction with Leu246,
close to the cleft H1. Regarding HBA2, this feature (not present
in HYPO 1) was found to overlap an N:d probe minimum,
suggesting a possible hydrogen bonding interaction between 1k
(N6 of the aromatic core) and the backbone NH of His95.
Finally, the features of HYPO3 determined a completely
different disposition of the pharmacophore inside MODEL3 (see
Figure 5): HYD1 was accommodated into a lipophilic cleft
delimited by Met146, Phe182, and Trp185 (roughly correspond-
ing to the H3 cleft); HYD3 matched a DRY minimum next to
Leu246 and His272; HYD2 corresponded to an energy-favorable
lipophilic interaction with Trp185. Regarding electrostatic
interactions, HBA1 corresponded to an H-bond with Tyr254
and HBD1 matched an H-bond between the CdO of Asn250
and the NH of 1k.
The 12 generated complexes (obtained from the docking of
1j, 1k, 2c, and 3a into the three MODELs) were then subjected
to a relaxation protocol (see Experimental Section).
In MODEL1 all four ligands established the expected
interactions with His95, Trp243, His272 and formed one H-bond
with Asn250.
In MODEL2 the lipophilic interactions with His95, Trp243,
and His272 were detected for all four ligands, while only three
ligands out of the four (1j, 1k, and 2c) possessed the crucial
hydrogen bond with Asn250 (see Table 6). Furthermore, none
of the complexes showed the hydrogen bond predicted by
HYPO2 between the N6 nitrogen of the aromatic core and the
backbone NH of His95 (HBA2 feature).
corresponding
catalyst feature
probe
brief description
hydrogen
H
HYD
HYD, RA
RA
DRY
C1d
N:d
N1d
NHd
N1+
N1:
hydrophobic probe
sp2 CH aromatic or vinyl
sp2 N with lone pair
HBA
sp2 amine NH cation
sp2 NH with lone pair
sp3 amine NH cation
sp3 NH with lone pair
neutral flat NH2, e.g., amide
sp2 amine NH2 cation
sp3 amine NH2 cation
sp3 NH2 with lone pair
trimethylammonium cation
alkylhydroxyl OH group
ether or furan oxygen
sp2 carbonyl oxygen
PI, HBD
HBA, HBD
PI, HBD
HBA, HBD
HBD
PI, HBD
PI, HBD
HBA, HBD
PI
N2
N2d
N2+
N2:
NM3
O1
HBA, HBD
HBA
HBA
OC2
O
a The corresponding catalyst features are reported in the last column.
defined by the three pharmacophoric hypotheses (HYPO1,
HYPO2, and HYPO3).
To determine which of the three models was the most reliable
one, derivatives 1j, 1k, 2c, and 3a were chosen among those
used in the 3D-QSAR studies and manually docked into the
receptor, using the three alignment rules and conformations
defined by the corresponding pharmacophoric hypotheses.
Figure 5 shows the three models (compound 1k is displayed
as reference) together with the alignments between selected local
minima of the MIFs (calculated as described above) and the
three pharmacophoric hypotheses HYPO1, HYPO2, and HY-
PO3.
In MODEL1, the aromatic ring feature RA (matched by the
core of 1k) resulted in being superimposed on one of the C1d
probe minima, highlighting a π-π interaction with Trp243. In
fact, this residue has been widely reported as playing an
important role in antagonist recognition.40-42 HBA1 mapped
profitably an O probe minimum, digging up a possible hydrogen
bond between the CdO of the ligand and the side chain NH of
Asn250. Notably, the mutation of this residue causes loss of
affinity for both agonists and antagonists.40-42 Finally, HYD1,
HYD2, and HYD3 resulted in being superimposed on three
DRY minima and were accommodated into three receptor
hydrophobic clefts (H1, H2, and H3) delimited, respectively,
by (i) Phe239, Trp243, Hys272; (ii) Leu102, Phe239, Trp243;
and (iii) Hys95, Phe182, Trp185, Tyr254.
In MODEL3 all the ligands interacted with His95, Trp243,
His272, and Asn250; however, as a result of the optimization
procedure, both compounds 2c and 3a acted as a hydrogen bond
acceptor with respect to Asn250, in contrast to the pharma-
cophoric hypothesis (HYPO3) which predicted an HBD feature
(HBD1).
Therefore, although mutagenesis data were generally verified
for all three models, only MODEL1 was able to fulfill both
mutagenesis data and the Catalyst predicted pharmacophoric
framework for all the compounds tested.
The interaction energies of all 12 complexes were then
calculated by subtracting the energy of the separate ligand and
receptor from the energy of the receptor-ligand complex, and

Pharmacophore Based Receptor Modeling
Journal of Medicinal Chemistry, 2006, Vol. 49, No. 14 4091
Table 7. Interaction Energies of Compounds 1i, 1k, 2c, and 3a in the
Three Considered Alignment Models: MODEL1, MODEL2, and
MODEL3
interaction energies (kJ/mol)
compd
MODEL 1
MODEL 2
MODEL 3
1j
-55.9
-61.4
-58.5
-55.2
-52.1
-53.4
-53.5
-47.0
-53.5
-56.8
-55.7
-47.6
1k
2c
3a
Table 8. A3 Affinity (Ki,), Experimental Binding Free Energy (Gexptl),
and AutoDock Binding Free Energy (Gcalcd) into the Three Models of
Compounds 1k, 1j, 2c, 2d, 3a, 3i, 4a, and 4h
Gcalcd (kcal/mol)
compd
Ki (nM)
Gexptl (kcal/mol)
MOD 1
MOD 2
MOD 3
1k
1j
2c
2d
3a
3i
0.04
0.60
0.65
7.66
17
1180
4470
7.94
-14.18
-12.57
-12.53
-11.07
-10.59
-8.08
-12.05
-12.84
-12.80
-10.81
-11.78
-9.22
-11.10
-12.79
-12.09
-9.30
-9.58
-7.72
-9.00
-8.57
-11.29
-11.00
-11.64
-9.46
-11.24
-10.03
-7.71
4a
4h
-7.29
-10.00
-10.54
-11.04
-11.42
the values calculated are reported in Table 7. Generally it was
not possible to find a quantitative correlation between calculated
values and affinities of the compounds mainly because of the
lack of the solvation and entropic terms. However, in this case,
only the interaction of the same ligand, though in different
orientations, in the same receptor was to be evaluated, and
therefore, the solvation and entropic contributions could be
considered approximately constant. As shown in Table 7,
MODEL1 yielded the highest interaction energy values for all
four ligands. Although the difference in energies among the three
models was small and probable within the error of the method,
the fact that the best interaction energy for all ligands was found
with MODEL1 strengthens the hypothesis that this model could
be considered the most reliable one.
In a third check, an assessment of the predictive power of
the three models was carried out. The free energies of binding
of the complexes with several compounds were calculated by
means of the AutoDock 3.0 scoring function.43 This docking
application proved in fact to be reliable in many studies present
in the literature,44-46 since its free energy function, based on
the principles of QSAR, has been parametrized using a large
number of protein-inhibitor complexes for which both structure
and inhibition constants were known. Accordingly, compounds
2d, 3i, 4a, and 4h were docked in turn, applying the protocol
already discussed for compounds 1j, 1k, 2c, and 3a.
Figure 6. Experimental versus calculated (AutoDock) binding energy
of compounds 1k, 1j, 2c, 2d, 3a, 3i, 4a, and 4h.
Figure 7. Superposition of the complexes of hA3AR with the most
active ligands 1j, 1k, and 2c (left) and with the less active ones 2d,
3a, 3i, 4a, and 4h (right). For sake of clarity, only residues under
consideration are reported.
As shown in Table 8 and Figure 6, a reasonable quadratic
correlation between experimental and calculated values of free
energy of binding was found for MODEL1 (r2 ) 0.69). In
contrast, a weaker free energy correlation was obtained for
MODEL2 and MODEL3 (values of 0.56 and 0.57, respectively).
All these results suggested MODEL1 as the most reliable one,
and for this reason it was selected for further analysis.
With the aim of achieving a qualitative knowledge of the
features of hA3AR, the eight relaxed complexes of compounds
1k, 1j, 2c, 2d, 3a, 3i, 4a, and 4h were compared among each
other through the superposition of the seven helices. The
outcome of such a procedure is shown in Figure 7. The most
evident result was that the receptor model was able to vary the
width and shape of its three hydrophobic clefts in order to
accommodate different ligands. In particular, Trp185 could act
as a gate, making the H2 and H3 clefts wider or narrower. In
the case of compounds showing Ki < 1 nM (left part of Figure
7), H2 was very small, thus allowing a perfect matching with
the ligand, while the wider H3 was able to accommodate long
chains such as phenylcarbamoyl moieties. On the other hand,
for compounds showing Ki > 1 nM (right part of Figure 7),
Trp185 was in an “open” conformation so that weaker contacts
were possible with the ligands. Therefore, the model seemed
to be able to discriminate between very active compounds
(Ki < 1 nM) and less active ones (Ki > 1 nM).
Regarding the interactions between 1k and the hA3 receptor,
beyond the above-mentioned interactions and already suggested
by site directed mutagenesis studies, residue Tyr254 also played
an important role in our A3 model because it formed an H-bond
with the pyridyl ring of the ligand (see Figure 8). This interaction

4092 Journal of Medicinal Chemistry, 2006, Vol. 49, No. 14
Tafi et al.
Figure 8. Side view of hA3AR complexing 1k into its intrahelical region (MODEL1, left) and details of relevant binding interactions (right).
Scheme 1a
could justify the improved activity of compounds such as 1k,
presenting a proper hydrogen-bonding acceptor group in that
region (4-pyridyl) and the irreversible inhibition by p-fluoro-
sulfonylpyrazolo[4,3-e]-1,2,4-triazolo[1,5-c]pyrimidine deriva-
tives.47 Furthermore, because of the fact that Tyr254 is a
nonconserved residue in other adenosine receptors, this interac-
tion could also explain the high selectivity shown by 1k.
Design and Synthesis of hA3AR Antagonists. The analysis
of the recognition geometry of compound 1k in MODEL1
suggested some structural changes to be made on the pyrazo-
lotriazolopyrimidines, with the aim of improving their water
solubility and ADME properties as well as the activity/selectivity
profile. In particular, a small hydrophilic pocket bordered by
two serine residues (Ser242 and Ser275) lay empty near the
H1 cleft of the receptor. Hence, the introduction of a short alkyl
chain at N8 of 1k, terminating with a hydrophilic group, was
hypothesized to verify the possibility of binding interactions
within this pocket.
In contrast, it appears that the interactions with His98, Trp243,
Asn250, Tyr254, and His272 (see Figure 8) are to be conserved;
therefore, the tricyclic system of 1k together with the pyridine-
urea moiety should be maintained.
Compounds 16-18 (Scheme 2) to be proposed for the
synthesis were then designed and virtually evaluated in silico
(applying the whole modeling procedure described above).
The synthesis of compounds 16-18 was performed using the
synthetic strategy depicted in Schemes 1 and 2. Alkylation of
the 2-(furan-2-yl)-7H-pyrazolo[4,3-e]-1,2,4-triazolo[1,5-c]pyri-
midin-5-ylamine 610,48 with 2-iodoethanol or benzyl-3-bro-
mopropyl ether in DMF and 60% NaH afforded compounds
748 and 8, which were separated from the corresponding N7
isomers by flash chromatography.
The free hydroxylic group of the N8 isomer of compound 7
was then protected by reaction with benzyl bromide to furnish
derivative 948 (Scheme 1). As reported in Scheme 2, the free
amino group at the 5 position of compounds 8 and 9 was
converted into the corresponding ureas 13-15 by treatment with
freshly prepared 3- or 4-pyridyl isocyanates 12a or 12b,
respectively. The preparation of the intermediates 12a,b is
described in the literature48-50 and illustrated in Scheme 2. The
commercially available nicotinoyl and isonicotinoyl hydrazides
(10a and 10b) were converted into the corresponding acyl azides
(11a and 11b) by reaction with sodium nitrite and aqueous HCl.
The azides were then converted into the isocyanates 12a and
12b by Curtius rearrangement induced by heating the azides in
dry toluene for 2 h. The crude isocyanates obtained were added
to the tricyclic compounds 8 and 9 dissolved in dry THF, and
a Reagents: (i) 2-iodoethanol, 60% NaH, room temp; (ii) benzyl
3-bromopropyl ether, 60% NaH, 100 °C, 12 h; (iii) benzyl bromide, 60%
NaH, room temp, 5 h.
the mixture was then refluxed for 5-8 h to afford compounds
13-15. The final products 16-18 were achieved by deprotecting
the ether function by treatment with HCO2NH4 and 10% Pd/C
in dry acetone at reflux.
Compounds 13-18 were tested for their affinity toward
human A1, A2A, A2B, and A3 adenosine receptors (see Experi-
mental Section). The values measured for compounds 16-18
are shown in Table 9, to be compared with the theoretical values
predicted for the A3 receptor by both the AutoDock scoring
function (notably, the scoring function hypothesized an activity
between 0.5 and 2 nM for these ligands) and Catalyst (Table
3).
Compound 17 resulted in being the most promising selective
A3 antagonist. The docking of this derivative into MODEL 1 is
shown in Figure 9. The expected hydrogen-bonding interaction
between the hydroxyl substituent at N8 and Ser275 is high-
lighted. Even if the affinity of 17 for the A3 receptor dropped
about 2 orders of magnitude with respect to compound 1k, its

Pharmacophore Based Receptor Modeling
Journal of Medicinal Chemistry, 2006, Vol. 49, No. 14 4093
Scheme 2a
a Reagents: (i) NaNO2, aqueous HCl, 0 °C, 1 h; (ii) toluene, 80 °C, 2 h; (iii) THF, reflux 5-8 h; (iv) acetone, HCO2NH4, 10% Pd/C, reflux 12 h. 10a,
11a, 12a: X ) CH, Y ) N. 10b, 11b, 12b: X ) N, Y ) CH. 13: X ) N, Y ) CH, R ) 2-benzyloxyethyl. 14: X ) N, Y ) CH, R ) 3-benzyloxypropyl.
15: X ) CH, Y ) N, R ) 2-benzyloxyethyl. 16: X ) N, Y ) CH, R′ ) 2-hydroxyethyl. 17: X ) N, Y ) CH, R′ ) 3-hydroxypropyl. 18: X ) CH, Y
) N, R′ ) 2-hydroxyethyl.
Table 9. Affinity Values of Compounds 16-18
Ki (nM)
as well as to an entropy penalty associated with the high number
of degrees of freedom of the alkyl chain. In any case, there is
quite a good affinity prediction by our hA3AR model.
a
b
c
d
e
compd
hA1
hA2A
>1000 >1000 5.1 (4.1-6.5)
(96%) (71%)
hA2B
hA3
predicted hA3
1.78
16
>1000
(74%)
Conclusions
In the research reported here we performed a pharmacophoric
study using the software Catalyst, which yielded three different
common feature hypotheses for antagonists of the human A3
receptor. The statistical parameters of the three models, i.e., high
cost differences with respect to null costs, suggested a possible
reliability of all of them; moreover, they all showed the ability
to predict affinity data for a test set of compounds, in good
agreement with experimental data. The three pharmacophores
referred to a recurring scheme consisting of three hydrophobic
interactions lying at the vertexes of a triangle. They seemed
particularly good in handling pyrazolotriazolopyrimidine deriva-
tives, the most potent class of A3 antagonists ever reported. On
the other hand, our ligand based approach alone did not converge
on a unique model and only the use of molecular modeling
techniques allowed us to choose the most reliable pharmaco-
phore.
17
18
350 ( 30 >1000 >1000 2.0 (1.7-2.4)
(95%) (73%)
>1000 >1000 34 (28-40)
(93%) (99%)
0.78
1.58
>1000
(92%)
a Displacement of specific [3H]DPCPX binding at human A1 receptors
expressed in CHO cells. b Displacement of specific [3H]ZM 241385 binding
at human A2A receptors expressed in CHO cells. c Displacement of specific
[3H]MRE 2029F20 binding at human A2B receptors expressed in CHO cells.
d Displacement of specific [3H]MRE 3008F20 binding at human A3 receptors
expressed in CHO cells. e Predicted A3 Ki affinity (using our A3 model
and AutoDock scoring function).
The construction of a human G-protein-coupled receptor
model through a homology procedure solely based on the bovine
rhodopsin structure is quite an unreliable task because of usually
low homology percentages and the high degree of mobility of
the helices.37 The use of mutagenesis data is nowadays an
important improvement in the procedure because it allows us
to take into account residues experimentally found to be
necessary for interaction. The relative positions of these residues,
however, remain unknown and can only be hypothesized
through the docking into the receptor of a ligand, which retains
all these interactions. Generally the alignment for this docking
is to be manually performed. Differently, in the procedure
described herein the use of a pharmacophoric model representing
the activity data of a lot of ligands allowed the building of a
receptor model in which the relative positions and distances
between important residues were determined by the distances
between pharmacophoric features.
Figure 9. Compound 17 docked into the putative binding site.
receptor subtype selectivity was completely conserved, and this
could confirm the importance of the pyridine ring for the A3
selectivity. As regards the affinity drop, this could be due to
partial damage of the π-π stacking interaction with Trp243
due to the formation of the new hydrogen bond (see Figure 9)
As a whole, our combined modeling strategy slightly differed
from canonical procedures. The building of 3D-QSAR models,

4094 Journal of Medicinal Chemistry, 2006, Vol. 49, No. 14
Tafi et al.
All MM minimizations were performed with either Polak-Ribier
conjugate gradient or steepest descent as minimizers and with a
threshold value of 0.05 kJ/(Å‚mol) as the convergence criterion.
The temperature was set at 300 K, and the time step was 1.0 fs in
MD simulations.
All graphical manipulations and visualizations were performed
by means of the InsightII,55 UCSF-CHIMERA,56 and WebLab
Viewer57 programs.
in fact, helped us to exploit a set of active molecules to highlight
statistically relevant features in ligand-receptor interactions,
and such information was crucial for generating a reliable model
of hA3AR when interacting with antagonists (MODEL1). The
information produced through docking studies, in turn, allowed
us to select the best performing pharmacophore model (HYPO1)
among a set of plausible hypotheses.
MODEL1 seemed to be able to explain different modes of
binding of very active compounds with respect to less active
ones and also to reproduce free energies of binding with good
approximation. The model was also able to explain the selectiv-
ity of 1k toward hA3AR due to the presence of the nonconserved
residue Tyr254 and therefore suggested that a mutagenesis study
on this residue could be of great importance to find out the
molecular features determining the selectivity at the AR
subtypes.
The reliability of our synergistic approach was tested by the
rational design and synthesis of a series of novel compounds.
Biological assays indicated that reasonably good activity values
were reached and, at the same time, water solubility was
enhanced compared to results of previously synthesized com-
pounds.
The alignment of several adenosine receptors was studied with
the ClustalW program using the Blosum algorithm, with a gap open
penalty of 10 and a gap extension penalty of 0.05. From the
ClustalW alignment, the structure of the seven-TM helices of hA3-
AR and the first intracellular loop were constructed directly from
the coordinates of the corresponding amino acids in rhodopsin by
means of the Modeller program.58 Through Maestro interface, the
TM3 and TM7 were rotated respectively 60° clockwise and 90°
counterclockwise (extracellular point of view) to let Hys95 and
Hys272 turn toward the intrahelical channel. Because the amino
acid length differs from the template and for the rotation of TM3
and TM7, the other loops were constructed by means of the “loop
optimization method” of Modeller, applying the “very_slow” loop
refinement method. The model was subjected to a preliminary
minimization and to 400 ps of MD (after 50 ps of equilibration).
The final structure was then minimized. When MD simulations were
carried out in the gas phase, all the R carbons of the TM of the
protein were blocked by means of decreasing force constants to
simulate the stabilizing presence of the membrane around the
receptor. For the first 200 ps, restraints with a force constant of 10
kcal/(mol‚Å2) were applied to CR, and for the remaining 200 ps
these restraints were gradually reduced to 1 kcal/(mol‚Å2).
The refinement of the ligand-protein complexes was initially
performed by means of a total of 400 ps of MD. All the R carbons
of the TMs and the main ligand-receptor interactions were
constrained during the trajectory by decreasing the force constants.
In detail, an initial restraint with a force constant of 10 kcal/(mol‚
Å2) was applied on the R carbons. This force constant decreased
during the entire MD, and in the last 200 ps a value of 0.1 kcal/
(mol‚Å2) was applied. As regards the H bond ligand-receptor
interactions, suggested by the HBA1, HBA2, and HBD1 features
found by Catalyst, a restraint of 50 kcal/(mol‚Å2) was applied in
order to stabilize ligand-receptor complex structures maintaining
all these interactions. At the end of the MD simulation, three steps
of minimization were applied on the average structure obtained
during the last 100 ps of the MD run. During these three steps a
restraint of 0.1 kcal/(mol‚Å2) was applied on the R carbons, while
with regard to the main ligand-receptor interactions, in the first
two steps a restraint of 25 and 10 kcal/(mol‚Å2) was applied and
in the last one the restraints were removed.
Experimental Section
Pharmacophore Generation. The literature was searched for
compounds showing the widest array of chemical features and the
most homogeneous biological data.9-21,31 Fifty-five compounds (see
Table 1) with activity data spanning over 5 orders of magnitude
(from 10-2 to 103 nM) were selected and divided into a training
set (38 compounds) and a complementary test set (17 compounds).
Biological data for all the inhibitors were reported as Ki.
Each compound was built using the 2D-3D sketcher implemented
in Catalyst and then submitted to a conformational search. The “best
conformer generation” method, which makes use of the Poling
algorithm to reduce conformational redundancies, ensuring good
space coverage,51 was preferred and we found up to 250 conformers
within 20.0 kcal/mol above the global minimum.
Because of the presence of great chemical functionalization in
all compounds, the generator was constrained to generate sets of
hypotheses bearing at least the following five features: hydrophobic
(HYD), hydrogen bond acceptor (HBA), hydrogen bond donor
(HBD), ring aromatic (RA), and positively ionizable atom (PI).
All compounds (each one with its conformational model) were
put into a spreadsheet and associated with their affinity constants
with the default uncertainty of 3. Because of the fact that the most
active compound (1k) was the only one showing a Ki in the 10-2
nM range, the set of “active molecules” used by the software in
the very first part of the calculation to generate hypotheses was
reduced by diminishing from 3 to 2 the uncertainty parameter
associated with 1k. Otherwise, the software might have attached
more statistical importance to ligands presenting lower affinity
values just because many compounds carry a greater amount of
information than a single one does.33
The quantitative evaluation of the free energy of binding of the
12 complexes was performed by means of the AutoDock scoring
function,43 using the Lamarckian genetic algorithm. The region of
interest used by AutoDock was defined considering compound 1k
docked in hA3AR as a center group. In particular, a grid of 40, 54,
and 50 points in the x, y, and z directions was built centered on the
center of mass of 1k. A grid spacing of 0.375 Å and a distance-
dependent function of the dielectric constant were used for the
calculation of the energetic maps.
During the second generation of hypotheses, two more control
parameters were changed from their default value; the MinFeatDist
value was decreased from 300 to 100 (1 Å), and the Variable
Tolerance value was set at 1. The first adjustment allowed the
generator to retrieve hypotheses with a minimum interfeature
distance down to 1 Å, while the second change allowed the
tolerance of each feature to vary during generation to avoid overlap
between features.
Molecular Modeling. Molecular mechanics (MM) and molecular
dynamics (MD) calculations were performed using the AMBER
force field52 as implemented in the MacroModel software package,53
using a “distance-dependent” dielectric constant of 4.0. Electrostatic
charges for the set of ligands were calculated with the RHF/AM1
semiempirical calculation and RESP program.54
General Chemistry. Reaction courses and product mixtures were
routinely monitored by thin-layer chromatography (TLC) on silica
gel (precoated F254 Merck plates) and visualized with aqueous
potassium permanganate or ethanolic ninhydrin solutions. Infrared
spectra (IR) were measured on a Perkin-Elmer 257 instrument. 1H
NMR were determined in CDCl3 or DMSO-d6 solutions with a
Bruker AC 200 spectrometer. Peak positions are given in parts per
million (δ) downfield from tetramethylsilane as internal standard,
and J values are given in Hz. Light petroleum refers to the fractions
boiling at 40-60 °C. Melting points were determined on a Buchi-
Tottoli instrument and are uncorrected. Chromatography was
performed with Merck 60-200 mesh silica gel. All products
1
reported showed H NMR spectra in agreement with the assigned
structures. Organic solutions were dried over anhydrous sodium

Pharmacophore Based Receptor Modeling
Journal of Medicinal Chemistry, 2006, Vol. 49, No. 14 4095
Table 10. Affinity Values of Compounds 13-15
Ki (nM)
sulfate. Elemental analyses were performed by the microanalytical
laboratory of Dipartimento di Chimica, University of Ferrara and
were within (0.4% of the theoretical values for C, H, and N.
Syntheses. 8-(3-Benzyloxypropyl)-2-(furan-2-yl)-8H-pyrazolo-
[4,3-e]-1,2,4-triazolo[1,5-c]pyrimidin-5-ylamine (8). To a solution
of 6 (0.6 g, 2.4 mmol) in dry DMF (10 mL) was added 60% NaH
(0.12 g, 1 mol equiv), and the suspension was stirred at 0 °C for
10 min. Benzyl 3-bromopropyl ether (0.42 mL, 1 mol equiv) was
added in small portions, and the mixture was heated at 100 °C for
12 h. The solvent was removed at reduced pressure, and the residue
was dissolved in water (100 mL) and extracted with EtOAc (3 ×
30 mL). The organic layer was dried (Na2SO4) and evaporated under
vacuum. The residue obtained was purified by chromatography
(EtOAc, 100%) to afford the N8 isomer 8 as pale-yellow solid (0.36
a
b
c
d
compd
hA1
hA2A
hA2B
hA3
13
>1000
(71%)
>1000
(74%)
>1000
(85%)
>1000
(97%)
>1000
(89%)
>1000
(82%)
>1000
(65%)
>1000
(66%)
>1000
(58%)
27 (23-32)
14 (12-16)
9.0 (8.1-9.9)
14
15
a Displacement of specific [3H]DPCPX binding at human A1 receptors
expressed in CHO cells. b Displacement of specific [3H]ZM 241385 binding
at human A2A receptors expressed in CHO cells. c Displacement of specific
[3H]MRE 2029F20 binding at human A2B receptors expressed in CHO cells.
d Displacement of specific [3H]MRE 3008F20 binding at human A3 receptors
expressed in CHO cells.
1
g, 77%): mp 173-5 °C; H NMR (CDCl3) δ 2.24 (m, 2H), 3.49
(t, 2H, J ) 6.1), 4.45 (s, 2H), 4.52 (t, 2H, J ) 6.2), 5.97 (bs, 2H),
6.58 (m, 1H), 7.30 (m, 6H), 7.61 (m, 1H), 8.17 (s, 1H). Anal.
(C20H19N7O2) C, H, N.
for 30 min at 48000g. The membrane pellet was resuspended in
50 mM Tris-HCl buffer at pH 7.4 for A1 adenosine receptors, in
50 mM Tris-HCl, 10 mM MgCl2 at pH 7.4 for A2A adenosine
receptors, and in 50 mM Tris HCl, 10 mM MgCl2, 1 mM EDTA
at pH 7.4 for A3 adenosine receptors.
Binding of [3H]DPCPX to CHO cells transfected with the human
recombinant A1 adenosine receptor was performed according to
the method previously described by Varani et al.60 Displacement
experiments were performed for 120 min at 25 °C in 200 µL of
buffer containing 1 nM [3H]DPCPX, 20 µL of diluted membranes
(50 µg of protein/assay), and at least six to eight different
concentrations of examined compounds. Nonspecific binding was
determined in the presence of 10 µM of CHA, and this is always
e10% of the total binding.
Binding of [3H]ZM 241385 to CHO cells transfected with the
human recombinant A2A adenosine receptors (50 µg of protein/
assay) was performed according to Varani et al.60 In competition
studies, at least six to eight different concentrations of compounds
were used and nonspecific binding was determined in the presence
of 1 µM ZM 241385 for an incubation time of 60 min at 25 °C.
Binding of [3H]MRE 2029F20 cells transfected with the human
recombinant A2B adenosine receptors was performed essentially with
the method described by Varani et al.60 In particular, assays were
carried out for 60 min at 25 °C in 100 µL of 50 mM Tris-HCl
buffer, 10 mM MgCl2, 1 mM EDTA, 0.1 mM benzamidine, pH
7.4, 2 IU/mL adenosine deaminase containing 40 nM [3H]MRE
2029F20, diluted membranes (20 µg of protein/assay), and at least
six to eight different concentrations of tested compounds. Nonspe-
cific binding was determined in the presence of 100 µM NECA
and was always e30% of the total binding.
Binding of [3H]MRE 3008F20 to CHO cells transfected with
the human recombinant A3 adenosine receptors was performed
according to Varani et al.60 Competition experiments were carried
out in duplicate in a finale volume of 250 µL in test tubes containing
1 nM [3H]MRE 3008F20, 50 mM Tris-HCl buffer, 10 mM MgCl2,
pH 7.4, and 100 µL of diluted membranes (50 µg protein/assay),
and at least six to eight different concentrations of examined ligands
for 120 min at 4 °C. Nonspecific binding was defined as binding
in the presence of 1 µM of MRE3008 F20 and was about 25% of
total binding.
General Procedure for 5-[[(3(4)-Pyridyl)amino]carbonyl]-
amino-8-(2-benzyloxyethyl)-2-(furan-2-yl)-8H-pyrazolo[4,3-e]-
1,2,4-triazolo[1,5-c]pyrimidines (13 and 15). Amino compound
948 (0.2 g, 5.3 mmol) was dissolved in dry THF (10 mL), and the
freshly prepared 3(4)-pyridyl isocyanates49,50,59 12a,b (5 mol equiv)
were added. The mixture was refluxed under argon for 5-8 h. Then
the solvent was removed under reduced pressure and the residue
was purified by flash chromatography (EtOAc/MeOH, 8:2) to afford
compounds 13 and 15 as solids.
5-[[(4-Pyridyl)amino]carbonyl]amino-8-(3-benzyloxypropyl)-
2-(furan-2-yl)-8H-pyrazolo[4,3-e]-1,2,4-triazolo[1,5-c]pyrimi-
dine (14). Amino compound 8 (0.18 g, 4.6 mmol) was dissolved
in dry THF (10 mL), and the freshly prepared 4-pyridyl isocyan-
ate49,50,59 12b (0.35 g, 5 mol equiv) was added. The mixture was
refluxed under argon for 8 h. Then the solvent was removed under
reduced pressure and the residue was purified by flash chroma-
tography (EtOAc/MeOH, 8:2) to afford compound 14 as a pale-
1
yellow solid (0.13 g, 64%): mp 178-180 °C; H NMR (DMSO-
d6) δ 3.31 (m, 2H), 3.65 (s, 2H), 4.37-4.43 (m, 4H), 6.50 (m,
1H), 7.19 (m, 6H), 7.33 (d, 2H, J ) 5.3), 7.54 (d, 2H, J ) 6), 8.08
(s, 1H), 8.25 (d, 1H), 9.2 (bs, 1H), 11.39 (s, 1H). Anal.(C26H23N9O3)
C, H, N.
General Procedure for Compounds 16-18: O-Debenzylation.
To a solution of compounds 13-15 (0.4 mmol) in dry acetone (20
mL) was added HCO2NH4 (8 mol equiv) and 10% Pd/C (0.5 mmol),
and the resulting mixture was heated at reflux for 12 h. The solution
was cooled, and the catalyst was removed by filtration. The solvent
was evaporated at reduced pressure, and the residue was washed
with water (25 mL). The aqueous layer was extracted with EtOAc
(3 × 20 mL), and the recombined organic phases were dried (Na2-
SO4) and evaporated under reduced pressure. The residue was
purified by flash chromatography (EtOAc/MeOH, 8:2) to afford
compounds 16-18.
Biology Experiments. All synthesized compounds have been
tested for their affinity to human A1, A2A, A2B, and A3 adenosine
receptors. Beyond the three compounds suggested by the compu-
tational studies we also tested compounds 13-15, and they showed
quite good activity (see Table 10). The affinity values were
determined by receptor binding assays at human A1, A2A, A2B, and
A3 adenosine receptor subtypes cloned in Chinese hamster ovary
(CHO) cells using [3H]DPCPX, [3H]ZM 241385, [3H]MRE 2029F20,
and [3H]MRE 3008F20, respectively. Bound and free radioactivity
were separated by rapid filtration through Whatman GF/B glass-
fiber filters that were washed three times with ice-cold buffer. The
filter bound radioactivity was counted in a Beckman LS-1800
spectrometer (efficiency of 55%).
Data Analysis. The protein concentration was determined
according to a Bio-Rad method61 with bovine albumin as a standard
reference. Inhibitory binding constants, Ki, were calculated from
IC50 according to the Cheng-Prusoff equation.62 A weighted
nonlinear least-squares curve-fitting program LIGAND63 was used
for computer analysis of saturation and inhibition experiments. Data
are expressed as the geometric mean with 95% or 99% confidence
limits in parentheses.
Human Cloned Adenosine Receptor Binding Assay. The cells
were grown and maintained in Dulbecco’s modified Eagle’s
medium with nutrient mixture F12 without nucleosides at 37 °C in
5% CO2/95% air. The cells were washed with phosphate-buffered
saline and scraped from flasks in ice-cold hypotonic buffer (5 mM
Tris-HCl, 2 mM EDTA, pH 7.4). The cell suspension was
homogenized with a Polytron, and the homogenate was centrifuged
Acknowledgment. This work was supported by the Minis-
tero dell’Istruzione dell’Universita` e della Ricerca (MIUR):
Program FIRB 2003, Protocol RBNE034XSW_005; Program
PRIN 2003, Protocol 2003054595_002; Program PRIN 2004,
Protocol 2004037521_002. M.B. thanks the Merck Research

4096 Journal of Medicinal Chemistry, 2006, Vol. 49, No. 14
Tafi et al.
(18) Li, A. H.; Moro, S.; Forsyth, N.; Melman, N.; Ji, X. D.; Jacobson,
K. A. Synthesis, CoMFA analysis, and receptor docking of 3,5-diacyl-
2,4-dialkylpyridine derivatives as selective A3 adenosine receptor
antagonists. J. Med. Chem. 1999, 42, 706-721.
(19) Xie, R.; Li, A. H.; Ji, X. D.; Melman, N.; Olah, M. E.; Stiles, G. L.;
Jacobson, K. A. Selective A3 adenosine receptor antagonists: water-
soluble 3,5-diacyl-1,2,4-trialkylpiridinium salts and their oxidative
generation from dihydropyridine precursors. J. Med. Chem. 1999,
42, 4232-4238.
(20) Van Rhee, A. M.; Jiang, J. L.; Melman, N.; Olah, M. E.; Stiles, G.
L.; Jacobson, K. A. Interaction of 1,4-dihydropyridine and pyridine
derivatives with adenosine receptors: selectivity for A3 receptors. J.
Med. Chem. 1996, 39, 2980-2989.
(21) Jiang, J. L.; Van Rhee, A. M.; Chang, L.; Patchornik, A.; Ji, X. D.;
Evans, P.; Melman, N.; Jacobson, K. A. Structure-activity relation-
ships of 4-(phenylethynyl)-6-phenyl-1,4-dihydropyridines as highly
selective A3 adenosine receptor antagonists. J. Med. Chem. 1997,
40, 2596-2608.
(22) Jalaie, M.; Erickson, J. A. Homology model directed alignment
selection for comparative molecular field analysis: Application to
photosystem II inhibitors. J. Comput.-Aided Mol. Des. 2000, 14, 181-
197.
(23) Bro¨er, B. M.; Gurrath, M.; Ho¨ltje, H. D. Molecular modelling studies
on the ORL1-receptor and ORL1-agonists. J. Comput.-Aided Mol.
Des. 2003, 17, 739-754.
(24) Poulsen, A.; Bjørnholm, B.; Gundertofte, K.; Pogozheva, I. D.;
Liljefors, T. Pharmacophore and receptor models for neurokinin
receptors. J. Comput.-Aided Mol. Des. 2003, 17, 765-783.
(25) Pogozheva, I. D.; Przydzial, M. J.; Mosberg, H. J. Homology
modeling of opioid receptor-ligand complexes using experimental
constraints. Am. Assoc. Pharm. Sci. J. 2005, 7, E434-E448 and
references therein.
(26) Evers, A.; Gohlke, H.; Klebe, G. Ligand supported homology
modelling of protein binding-sites using knowledge-based potentials.
J. Mol. Biol. 2003, 334, 327-345.
(27) Evers, A.; Klebe, G. Ligand-supported homology modelling of
G-protein coupled receptor sites: Models sufficient for successful
virtual screening. Angew. Chem., Int. Ed. 2004, 43, 248-251.
(28) Di Santo, R.; Tafi, A.; Costi, R.; Botta, M.; Artico, M.; Corelli, F.;
Forte, M.; Caporuscio, F.; Angiolella, L.; Palamara, A. T. Antifungal
agents. 11. N-Substituted Derivatives of 1-[(aryl)(4-aryl-1H-pyrrol-
3-yl)methyl]-1H-imidazole: Synthesis, anti-Candida activity, and
QSAR studies. J. Med. Chem. 2005, 48, 3019-3031.
(29) Ferrarini, P. L.; Betti, L.; Cavallini, T.; Giannaccini, G.; Lucacchini,
A.; Manera, C.; Martinelli, A.; Ortore, G.; Saccomanni, G.; Tucci-
nardi, T. Study on affinity profile toward native human and bovine
adenosine receptors of a series of 1,8-naphthyridine derivatives. J.
Med. Chem. 2004, 47, 3019-3031.
(30) Palczewski, K.; Kumasaka, T.; Hori, T.; Behnke, C. A.; Motoshima,
H.; Fox, B. A.; Le Trong, I.; Teller, D. C.; Okada, T.; Stenkamp, R.
E.; Yamamoto, M.; Miyano, M. Crystal structure of rhodopsin: A
G protein-coupled receptor. Science 2000, 289, 739-745.
(31) Moro, S.; Li, A. H.; Jacobson, K. A. Molecular modeling studies of
A3 adenosine antagonists: structural homology and receptor docking.
J. Chem. Inf. Comput. Sci. 1998, 38, 1239-1248.
Laboratories (2004 Academic Development Program Chemistry
Award). A.T., C.B., M.B., and F.C. thank the “Centro Univer-
sitario per l’Informatica e la Telematica” of the University of
Siena.
Supporting Information Available: Details of the alignment
procedure of the adenosine receptors and bovine rhodopsin amino
acid sequences and elemental analysis data of the newly synthesized
derivatives 13-18. This material is available free of charge via
the Internet at http://pubs.acs.org.
References
(1) Zhou, Q. Y.; Olah, M. E.; Johnson, R. A.; Stiles, G. L.; Clivelli, O.
Molecular cloning and characterization of an adenosine receptor: the
A3 adenosine receptor. Proc. Natl. Acad. Sci. U.S.A. 1992, 89, 7432-
7436.
(2) Jacobson, K. A. Adenosine A3 receptors: novel ligands and paradoxi-
cal effects. Trends Pharmacol. Sci. 1998, 19, 184-191.
(3) Van Schaick, E. A.; Jacobson, K. A.; Kim, H. O.; IJzerman, A. P.;
Danhof, M. Hemodynamic effects and histamine release elicited by
the selective adenosine A3 receptor agonist Cl-IB-MECA in conscious
rats. Eur. J. Pharmacol. 1996, 308, 311-314.
(4) Hannon, J. P.; Pfannkuche, H. J.; Fozard, J. R. A role for mast cells
in adenosine A3 receptor-mediated hypotension in the rat. Br. J.
Pharmacol. 1995, 115, 945-952.
(5) Ramkumar, V.; Stiles, G. L.; Beaven, M. A.; Ali, H. The A3AR is
the unique adenosine receptor which facilitates release of allergic
mediators in mast-cells. J. Biol. Chem. 1993, 268, 16887-16890.
(6) Von Lubitz, D. K. J. E.; Carter, M. F.; Deutsch, S. I.; Lin, R. C. S.;
Mastropaolo, J.; Meshulam, Y.; Jacobson, K. A. The effects of
adenosine A3 receptor stimulation on seizures in mice. Eur. J.
Pharmacol. 1995, 275, 23-29.
(7) Von Lubitz, D. K. J. E.; Lin, R. C. S.; Popik, P.; Carter, M. F.;
Jacobson, K. A. Adenosine receptors: pharmacology, structure-
activity relationships, and therapeutic potential. J. Med. Chem. 1992,
35, 407-422.
(8) Beaven, M. A.; Ramkumar, V.; Ali, H. Adenosine A3 receptors in
mast cells. Trends. Pharmacol. Sci.1994, 15, 13-14.
(9) Baraldi, P. G.; Cacciari, B.; Romagnoli, R.; Spalluto, G.; Klotz, K.
N.; Leung, E.; Varani, K.; Gessi, S.; Merighi, S.; Borea, P. A.
Pyrazolo[4,3-e]1,2,4-triazolo[1,5-c]pyrimidine derivatives as highly
potent and selective human A3 adenosine receptor antagonists. J. Med.
Chem. 1999, 42, 4473-4478.
(10) Baraldi, P. G.; Cacciari, B.; Romagnoli, R.; Spalluto, G.; Moro, S.;
Klotz, K. N.; Leung, E.; Varani, K.; Gessi, S.; Merighi, S.; Borea,
P. A. Pyrazolo[4,3-e]1,2,4-triazolo[1,5-c]pyrimidine derivatives as
highly potent and selective human A3 adenosine recptor antago-
nists: influence of the chain at the N8 pyrazole nitrogen. J. Med.
Chem. 2000, 43, 4768-4780.
(11) Baraldi, P. G.; Cacciari, B.; Moro, S.; Spalluto, G.; Pastorin, G.; Da
Ros, T.; Klotz, K. N.; Varani, K.; Gessi, S.; Borea, P. A. Synthesis,
biological activity, and molecular modeling investigation of new
pyrazolo[4,3-e]1,2,4-triazolo[1,5-c]pyrimidine derivatives as human
A3 adenosine receptor antagonists. J. Med. Chem. 2002, 45, 770-
780.
(12) Maconi, A.; Pastorin, G.; Da Ros, T.; Spalluto, G.; Gao, Z.g.;
Jacobson, K. A.; Baraldi, P. G.; Cacciari, B.; Varani, K.; Moro, S.;
Borea, P. A. Synthesis, biological properties, and molecular modeling
investigation of the first potent, selective, and water-soluble human
A3 Adenosine receptor antagonist. J. Med. Chem. 2002, 45, 3579-
3582.
(13) Kim, Y. C.; Ji, X. D.; Jacobson, K. A. Derivatives of the triazolo-
quinazoline adenosine antagonist (CGS15943) are selective for the
human A3 receptor subtype. J. Med. Chem. 1996, 39, 4142-4148.
(14) Van Muijlwijk-Koezen, J. E.; Timmerman, H.; Link, R.; van der Goot,
H.; IJzerman, A. P. A novel class of adenosine A3 receptor ligands.
1. 3-(2-Pyridinyl)isoquinoline derivatives. J. Med. Chem. 1998, 41,
3987-3993.
(15) Van Muijlwijk-Koezen, J. E.; Timmerman, H.; Link, R.; van der Goot,
H.; IJzerman, A. P. A novel class of adenosine A3 receptor ligands.
2. Structure affinity profile of a series of isoquinoline and quinazoline
compounds. J. Med. Chem. 1998, 41, 3994-4000.
(16) Van Muijlwijk-Koezen, J. E.; Timmerman, H.; van der Goot, H.;
Menge, W. M. P. B.; von Drabbe Ku¨nzel, J. F.; de Groote, M.;
Ijzerman, A. P. Isoquinoline and quinazoline urea analogues as
antagonists for the human adenosine A3 receptor. J. Med Chem. 2000,
43, 2227-2238.
(17) Li, A. H.; Moro, S.; Melman, N.; Ji, X. D.; Jacobson, K. A.
Structure-activity relationships and molecular modeling of 3,5-
diacyl-2,4-dialkylpyridine derivatives as selective A3 adenosine
receptor antagonists. J. Med. Chem. 1998, 41, 3186-3201.
(32) Catalyst 4.6; Accelrys, Inc. (9685 Scranton Road, San Diego, CA
92121).
(33) Catalyst 4.6 manual at http://www.accelrys.com.
(34) GPCRDB: Information System for G Protein-coupled receptors
(GPCRs). www.gpcr.org.
(35) Thompson, J. D.; Higgins, D. G.; Gibson, T. J. CLUSTAL W:
improving the sensitivity of progressive multiple sequence alignment
through sequence weighting, position-specific gap penalties and
weight matrix choice. Nucleic Acids Res. 1994, 22, 4673-4680.
(36) Fredriksson, R.; Lagerstro¨m, M. C.; Lundin, L. G.; Schio¨th, H. B.
The G-protein-coupled receptors in the human genome form five main
families. Phylogenetic analysis, paralogon groups, and fingerprints.
Mol. Pharmacol. 2003, 63, 1256-1272.
(37) Gouldson, P. R.; Snell, C. R.; Reynolds, C. A. A new approach to
docking in the â2-adrenergic receptor that exploits the domain
structure of G-protein-coupled receptors. J. Med. Chem. 1997, 40,
3871-3886.
(38) Cruciani, G.; Goodford, P. GREATER Graphical Interface for GRID,
version, 1.1.7 GRIB, UPF/IMIM; Molecular Discovery Ltd.: Bar-
celona, Spain, 2001-2003; http://www.moldiscovery.com.
(39) Chen, A.; Gao, Z. G.; Barak, D.; Liang, B. T.; Jacobson, K. A.
Constitutive activation of A3 adenosine receptors by site-directed
mutagenesis. Biochem. Biophys. Res. Commun. 2001, 284, 596-601.
(40) Gao, Z. G.; Chen, A.; Barak, D.; Kim, S. K.; Muller, C. E.; Jacobson,
K. A. Identification by site-directed mutagenesis of residues involved
in ligand recognition and activation of the human A3 adenosine
receptor. J. Biol. Chem. 2002, 277, 19056-19063.

Pharmacophore Based Receptor Modeling
Journal of Medicinal Chemistry, 2006, Vol. 49, No. 14 4097
(41) Gao, Z. G.; Kim, S. K.; Biadatti, T.; Chen, W.; Lee, K.; Barak, D.;
Kim, S. G.; Johnson, C. R.; Jacobson, K. A. Structural determinants
of A3 adenosine receptor activation: nucleoside ligands at the agonist/
antagonist boundary. J. Med. Chem. 2002, 45, 4471-4484.
(42) Jacobson, K. A.; Gao, Z. G.; Chen, A.; Barak, D.; Kim, S. A.; Lee,
K.; Link, A.; Van Rompaey, P.; Van Calenberg, S.; Liang, B. T.
Neoceptor concept based on molecular complementarity in GPCRs:
a mutant adenosine A3 receptor with selectively enhanced affinity
for amine-modified nucleosides. J. Med. Chem. 2001, 44, 4125-
4136.
(43) Morris, G. M.; Goodsell, D. S.; Halliday, R. S.; Huey, R.; Hart, W.
E.; Belew, R. K.; Olson, A. J. Automated docking using a Lamarckian
genetic algorithm and empirical binding free energy function. J.
Comput. Chem. 1998, 19, 1639-1662.
(44) Costantino, G.; Macchiarulo, A.; Campioni, E.; Pellicciari, R.
Modeling of poly(ADP-ribose)polymerase (PARP) inhibitors. Dock-
ing of ligands and quantitative structure-activity relationship analysis.
J. Med. Chem. 2001, 44, 3786-3794.
(45) Tanczos, A. C.; Palmer, R. A.; Potter, B. S.; Saldanha, J. W.; Howlin,
B. J. Antagonist binding in the rat muscarinic receptor. A study by
docking and X-ray crystallography. Comp. Biol. Chem. 2004, 28,
375-385.
(46) Tuccinardi, T.; Ferrarini, P. L.; Manera, C.; Ortore, G.; Saccomanni,
G.; Martinelli, A. Cannabinoid CB2/CB1 selectivity. Receptor
modeling and automated docking analysis. J. Med. Chem. 2006, 49,
984-994.
(47) Baraldi, P. G.; Cacciari, B.; Moro, S.; Romagnoli, R.; Ji, X.; Jacobson,
K. A.; Gessi, S.; Borea, P. A.; Spalluto, G. Fluorosulfonil- and bis-
(â-chloroehyl)amino-phenylamino functionalized pyrazolo[4,3-e]1,2,4-
triazolo[1,5-c]pyrimidine derivatives: irreversible antagonists at the
human A3 adenosine receptor and molecular modelling studies. J.
Med. Chem. 2001, 44, 2735-2742.
(51) Smellie, A.; Teig, S.; Towbin, P.; Poling, P. Promoting conformational
variation. J. Comput. Chem. 1995, 16, 171-187.
(52) Case, D. A.; Pearlman, D. A.; Caldwell, J. W.; Cheatham, T. E., III;
Ross, W. S.; Simmerling, C. L.; Darden, T. A.; Merz, K. M.; Stanton,
R. V.; Cheng, A. L.; Vincent, J. J.; Crowley, M.; Tsui, V.; Radmer,
R. J.; Duan, Y.; Pitera, J.; Massova, I.; Seibel, G. L.; Singh, U. C.;
Weiner, P. K.; Kollman, P. A. AMBER 6; University of California:
San Francisco, CA, 1999.
(53) MacroModel, version 7.2; Schrodinger Inc., 1999; http://
www.schroedinger.com.
(54) Pearlman, D. A.; Case, D. A.; Caldwell, J. W.; Ross, W. S.;
Cheatham, T. E., III; Ferguson, D. M.; Seibel, G. L.; Chandra Singh,
U.; Weiner, P. K.; Kollman, P. A. AMBER 4.1; University of
California: San Francisco, CA, 1995.
(55) InsightII (98) Molecular Modeling System, MSI; Accelrys Software,
Inc.; http://www.accelrys.com.
(56) Huang, C. C.; Couch, G. S.; Pettersen, E. F.; Ferrin, T. E. Chimera:
an extensible molecular modelling application constructed using
standard components. Pac. Symp. Biocomput. 1996, 1, 724; http://
www.cgl.ucsf.edu/chimera.
(57) WebLab Viewer Pro 3.7; Accelrys Inc.: San Diego, CA.
(58) Fiser, A.; Do, R. K.; Sali, A. Modeling of loops in protein structures.
Protein Sci. 2000, 9 1753-1773.
(59) Hyden, S.; Wilbert, G. Pyridine isocyanates. Chem. Ind. (London)
1967, 33, 1406-1407.
(60) Varani, K.; Merighi, S.; Gessi, S.; Klotz, K. N.; Leung, E.; Baraldi,
P. G.; Cacciari, B.; Romagnoli, R.; Spalluto, G.; Borea, P. A. [3H]-
MRE 3008F20: A novel radioligand for the pharmacological and
biochemical characterization of human A3 adenosine receptors. Mol.
Pharmacol. 2000, 57, 968-975.
(61) Bradfor, M. M. A rapid and sensitive method for the quantification
of microgram quantities of protein utilizing the principle of protein
dye-binding. Anal. Biochem. 1976, 72, 248-254.
(48) Baraldi, P. G.; Bovero, A.; Fruttarolo, F.; Romagnoli, R.; Tabrizi,
M. A.; Preti, D.; Varani, K.; Borea, P. A.; Moorman, A. R. New
strategies for the synthesis of A3 adenosine receptor antagonists.
Bioorg. Med. Chem. 2003, 11, 4161-4169.
(62) Cheng, Y. C.; Prusoff, W. H. Relationship between the inhibition
constant (Ki) and the concentration of inhibitor which causes 50%
inhibition (IC50) of an enzymatic reaction. Biochem. Pharmacol. 1973,
22, 3099-3108.
(63) Munson, P. J.; Rodbard, D. Ligand: a versatile computerized
approach for the characterization of ligand binding system. Anal.
Biochem. 1980, 107, 220-239.
(49) Singha, C. N.; Dixit, N.; Sathyanarayana, D. N. 1H and 13C NMR
spectra of some asymmetric N,N-dipyridyl ureas: spectral assign-
ments and molecular conformations. J. Chem. Soc., Perkin Trans. 2
1997, 157-162.
(50) Curtius, T.; Mohr, E. Transformation of nicotinic acid to beta-
aminopyridine. Ber. 1898, 31, 2493-2495.
JM051112+