N-[9-(ortho-Fluorobenzyl)-2-Phenyl-8-Azapurin-6-yl]-Amides as Potent and Selective Ligands for A1 Adenosine Receptors

Article abstract of DOI:10.1111/cbdd.12131

A series of N-[9-(ortho-fluorobenzyl)-2-phenyl-8-azapurin-6-yl]-amides were synthesized and tested for their affinity toward A₁, A_2A, and A₃ adenosine receptor subtypes. Biological results demonstrated that the introduction of a fluorine atom at the ortho position of the 9-benzyl group generally enhanced affinity toward A₁ subtype and did not significantly affect A_2A and A₃ affinity. Very interesting is the compound bearing a meta-fluorophenyl substituent on the carbonyl carbon of the amide group, which shows significantly high A₁/A_2A-A₃ selectivity. Compounds of this new series, together with the previously published analogs without the fluorine atom on the 9-benzyl group, constituted the starting dataset for the development of QSAR models. The models obtained were able to rationally describe the affinity trends resulting from biological testing and to enable investigation of the role of different substituents on the 8-azapurine scaffold, as well as the influence of the newly introduced fluorine atom on the benzyl moiety. The said QSAR models can also assist in the design of new compounds selectively active on A₁ adenosine receptors. Furthermore, a molecular docking study was carried out to assess hypothetical binding mode of N-[9-(ortho-fluorobenzyl)-2-phenyl-8-azapurin-6-yl]-amides to A₁ adenosine receptors.

Full text of DOI:10.1111/cbdd.12131

Chem Biol Drug Des 2013; 82: 22–38
Research Article
N-[9-(ortho-Fluorobenzyl)-2-Phenyl-8-Azapurin-6-yl]-
Amides as Potent and Selective Ligands for A1
Adenosine Receptors
ical and pathophysiological processes (1) mainly
addressed to tissue protection and repair. Four adenosine
receptors were identified and cloned: A1, A2A, A2B, and
A3. Up to now, many ligands targeting the adenosine
receptor subtypes have been synthesized and developed.
However, in the specific case of A1-selective antagonists,
none has reached a clinical application, despite the two
xanthines rolofylline and tonapofylline had been investi-
gated in clinical trials as novel therapeutic agents for the
treatment of acute and chronic heart failure (2,3). Never-
theless, the interest in the design of ligands, especially
non-xanthine ligands, with high activity and selectivity for
the A1 receptor subtype is always strong due to the inter-
esting therapeutic perspectives foreseen (4–6).
Alice Borghini*, Daniele Pietra, Michele Leonardi,
Irene Giorgi and Anna M. Bianucci
ꢀ
Dipartimento di Farmacia, Universita di Pisa, Via Bonanno
6, 56126, Pisa, Italy
*Corresponding author: Alice Borghini,
aliceborghini1974@yahoo.it
A series of N-[9-(ortho-fluorobenzyl)-2-phenyl-8-azapu-
rin-6-yl]-amides were synthesized and tested for their
affinity toward A1, A2A, and A3 adenosine receptor sub-
types. Biological results demonstrated that the intro-
duction of a fluorine atom at the ortho position of the
9-benzyl group generally enhanced affinity toward A1
subtype and did not significantly affect A2A and A3
affinity. Very interesting is the compound bearing a
meta-fluorophenyl substituent on the carbonyl carbon
of the amide group, which shows significantly high A1/
In the past, we obtained a number of ligands for A1 and
A3 adenosine receptors modifying the substituent at the 2,
6, and 9 positions of adenine and 8-azaadenine nuclei. A
phenyl ring at the 2 position and a benzyl group at the 9
position proved to be good substituents to ensure binding
with A1 and A3 receptors but not with A2A ones. For A1
receptors, we showed cycloalkyl, alkyl, and alkyl-phenyla-
mino groups, substituted or not with a hydroxyl group as
the best substituents on C6 (Figure 1A) (7,8). A phenyl-ure-
ido function at the same position shifts affinity toward A3
receptors (Figure 1B) (9). Furthermore, we introduced an
electron-withdrawing group at the ortho position of the 9-
benzyl substituent and a fluorine atom was found to be
the best one (Figure 1C) (10). More recently, we prepared
a series of 2-phenyl-9-benzyl-8-azaadenines having an
amido substituted moiety on C6 (Figure 1D); these com-
pounds demonstrated a very interesting affinity and selec-
tivity toward A1 receptors, either the ones having an alkyl
or cycloalkyl group bound to the carbonyl carbon or the
ones having a phenyl or a substituted phenyl at the same
position (11). A modification on the structure of this late
series, judged interesting by us, was the introduction of a
fluorine atom at the ortho position of the 9-benzyl group
(Figure 1E). In fact, this substituent had enhanced the
activity toward A1 receptors of 2-phenyl-9-benzyl-8-aza-
adenines and improved selectivity (10). Biological assays
on the new compounds, carried out to verify affinity
toward receptor subtypes, gave the expected results, in
agreement with those obtained by the previous series. So,
these molecules, together with the previously published
analogs without the fluorine atom on the 9-benzyl group,
A
2A-A3 selectivity. Compounds of this new series,
together with the previously published analogs without
the fluorine atom on the 9-benzyl group, constituted
the starting dataset for the development of QSAR
models. The models obtained were able to rationally
describe the affinity trends resulting from biological
testing and to enable investigation of the role of differ-
ent substituents on the 8-azapurine scaffold, as well as
the influence of the newly introduced fluorine atom on
the benzyl moiety. The said QSAR models can also
assist in the design of new compounds selectively
active on A1 adenosine receptors. Furthermore,
a
molecular docking study was carried out to assess
hypothetical binding mode of N-[9-(ortho-fluorobenzyl)-
2-phenyl-8-azapurin-6-yl]-amides to A1 adenosine
receptors.
Key words: 2-phenyl-8-azaadenines, A1 ligands, adenosine
receptors, molecular docking, N-[9-(ortho-fluorobenzyl)-2-phe-
nyl-8-azapurin-6-yl]-amides, QSAR, receptor selectivity
Received 17 December 2012, revised 12 February 2013 and
accepted for publication 25 February 2013
Adenosine is an important endogenous regulator present
in the extracellular space in high concentration during met-
abolically stressful conditions, such as in injury, ischemia,
and inflammation. It is a powerful signaling molecule that
participates in the regulation of a wide variety of physiolog-
22
ª 2013 John Wiley & Sons A/S. doi: 10.1111/cbdd.12131

Selective Ligands for A1 Adenosine Receptors
O
R
R
NH
N
NH
N
R
NH
N
NH
N
N
N
N
N
N
N
N
N
N
N
N
F
C
B
A
O
O
R
NH
NH
R
N
N
N
N
N
N
N
N
N
N
F
D
E
Figure 1: Design of the new compounds E starting from compounds A. The modifications introduced by us on the 2-phenyl-9-benzyl-8-
azaadenine nucleus, over the last few years, are highlighted [for references, see text].
constituted the starting dataset (DS) for the development
of QSAR models able to rationally describe the affinity
trends resulting from biological testing and investigate the
role of the different substituents on the 8-azapurine scaf-
fold, as well as the influence of the newly introduced fluo-
rine atom on the benzyl moiety. The relationship between
structural features of the compounds analyzed and affinity
trend toward A1 adenosine receptors was also assessed
by a molecular docking study. To do this, several 8-azapu-
rine derivatives were submitted to molecular docking to
establish their hypothetical interactions with amino acids
belonging to the binding pockets of a model of rat A1
adenosine receptor built with this purpose.
delocalized in aromatic rings. However, the yields were not
a problem because the starting compounds are not
expensive and we could obtain a sufficient amount of final
products to perform chemical analyses and biological
assays. Compounds 11 and 12 were reduced with hydro-
gen and 10% Pd/C at room pressure and temperature to
give the compounds 14 and 15.
Biological assays
Radioligand binding assays were carried out to assess the
affinity of all the new compounds toward A1, A2A, and A3
adenosine receptors. The results are summarized in
Table 1; when the percentage of inhibition (%i) resulted
<60% at 10 lM, compounds were considered inactive.
Biological results for compounds 10 and 14 were not
obtained, because precipitation occurred during the assay,
due to their low solubility both in DMSO and in the assay
buffer Tris–HCl 50 mM, pH 7.4. For tests involving A1
receptors, the radiolabeled antagonist [3H]DPCPX was
used, while the antagonist radioligand [3H]ZM241385 was
used in the experiments involving A2A receptors. In the
case of A3 receptors, the agonist radioligand [3H]NECA
was used as a probe. The results are given as Ki ꢀ SEM
(nM) and/or percentage inhibition of the radioligand. Control
binding is 0%, and non-specific binding is 100%.
The said QSAR models, along with the suggestions arising
from the molecular docking analysis preformed, can give a
valuable contribution to the rational design of other new
compounds selectively active on A1 adenosine receptors.
Methods and Materials
Chemistry
In Scheme 1 is depicted the general method to obtain
compounds 1bis-13 and 14–15.
2-phenyl-9-(2-fluorobenzyl)-8-azaadenine 1bis was obtain-
ed by a ‘one pot’ reaction (12) from malononitrile, 2-fluo-
robenzylazide, and benzonitrile in EtONa. Compound 1bis
reacted with the suitable acyl chlorides, prepared from the
corresponding acids and thionyl chloride, to give the
desired products 2–13. In some cases, the yield of the
reaction was not high, probably due to the very low basic
character of the N6 of 8-azaadenine, being the lone pair
Development of QSAR models
End-point selection
Compounds synthesized in this study were tested for
their A1, A2A, and A3 adenosine receptor affinity. More in
detail, receptor affinity was indicated with two different
Chem Biol Drug Des 2013; 82: 22–38
23

Borghini et al.
Table 1: Molecular structures and biological assay results of compounds toward A1, A2A, and A3 adenosine receptors
a
Percentage inhibition of specific binding of radioligand %i and Ki ꢀ SEM (nM)
of selected compounds
Compond ID
Structure
Adenosine A1 receptor
Adenosine A2A receptor
Adenosine A3 receptor
70%i
1bis
3.8 ꢀ 1.2
94 ꢀ 30
NH2
N
N
N
N
N
F
O
HN
R
N
N
N
N
N
F
R
2
9.2 ꢀ 6.2
292 ꢀ 85
70%i
3
4
0.71 ꢀ 0.3
1.4 ꢀ 0.8
72 ꢀ 18
68%i
74%i
190 ꢀ 62
5
6
2.1 ꢀ 1.5
5.5 ꢀ 1.5
413 ꢀ 17
224 ꢀ 53
62%i
65%i
O
O
7
8
2.3 ꢀ 1.1
43 ꢀ 9
91 ꢀ 35
77%i
46%i
Cl
25%i
F
9
5.5 ꢀ 1.5
17%i
n.d.
48%i
n.d.
10
n.d.
F
24
Chem Biol Drug Des 2013; 82: 22–38

Selective Ligands for A1 Adenosine Receptors
Table 1: continued
O
HN
R
N
N
N
N
N
F
R
NO2
11
28 ꢀ 1.6
32%i
59%
12
13
14
14 ꢀ 6
92 ꢀ 10
n.d.
30%i
18%i
n.d.
38%i
48%i
n.d.
NO2
I
NH
2
15
18 ꢀ 4
0%i
54%i
NH2
n.d., not determined.
aData are expressed as means from 2 to 3 independent experiments performed in duplicate; individual values varied <15%.
parameters: percentage inhibition of specific binding of
the radioligand (%i) and inhibition constant (Ki), as
described in the ‘Biological results’ section. All of the
assayed compounds resulted to be good ligands for the
A1 adenosine receptor, which enabled the measuring of
the Ki values. Consequently, in the case of the A1 recep-
tor subtype, the affinity of such compounds expressed in
terms of Ki values represented the proper end-point. In
contrast, most of the analyzed compounds were found to
be poorer ligands for the A2A and A3 receptors. In these
cases, Ki values were measured only for compounds
showing %i ꢁ 60% at 10 lM, while only %i values were
taken for all of them. Consequently, affinities of com-
pounds toward the A2A and A3 subtypes, expressed as
%i values, supplied the proper end-point values for such
receptor subtypes.
the same experimental conditions and with identical proto-
cols with respect to the compounds newly obtained, for
their affinity toward the different adenosine receptor sub-
types (11). This allowed us to assess the role played by
the said F atom on affinity toward the adenosine recep-
tors. Accordingly, the DS that represented the starting
point for the QSAR model construction was made up of
44 compounds in all.
Next, the end-point selected for the A1 receptor (Ki) was
converted into pKi (ꢂlogKi), whereas A2A and A3 end-
points (%i) were discretised into two classes (‘Inactive’ and
‘Active’) as follows:
ꢃ
the ‘Inactive’ class containing compounds with %i at
10 lM < 60%;
ꢃ
the ‘Active’ one containing compounds with %i at
Dataset building
10 lM ꢁ 60%.
Due to their structural similarity, 13 compounds synthe-
sized in the present work were analyzed together with 31
analogous, which do not have the fluorine atom on the
benzyl moiety at the 9 position. These latest compounds
had been previously synthesized and tested by us, under
Table 2 illustrates all the 44 compounds together with their
pKi values for the A1 receptor as well as the ‘Inactive’ or
‘Active’ class for A2A and A3 receptors.
Chem Biol Drug Des 2013; 82: 22–38
25

Borghini et al.
26
Chem Biol Drug Des 2013; 82: 22–38

Selective Ligands for A1 Adenosine Receptors
Molecular descriptor calculation
For each compound belonging to DS, molecular descrip-
tors were calculated starting from molecular structures
expressed as ‘molecular graphs’ (drawn in Symyx Draw
3.2; Symyx Technologies Inc., Santa Clara, CA, USA).
Such descriptors account for molecular features of the bi-
dimensional (2D) type. These descriptors were calculated
with the use of computer programs suitable for molecular
descriptor calculation, such as PADEL (13) and POWERMV (14).
These descriptors are commonly used in QSAR model
building and consist of constitutional descriptors, informa-
tion indices, topological indices, and charge descriptors. In
this work, they were added to more specific descriptors
developed and calculated ‘in house’. The latter descrip-
tors, successfully used in previous QSAR studies by us
(15), were designed to properly reflect steric hindrance
and electronegativity features of molecular moieties char-
acterizing the 6-amide substituent of the general azapurine
skeleton, which is supposed to directly interact with the
receptor pocket. Their calculation is briefly explained
below: the carbon atom of the amide carboxyl group at
the 6 position was considered as the starting center,
because it represents the ‘hinge’ point joining the fixed
portion of the molecule (that is, the general skeleton) to
the variable one. Then, the atoms constituting the nodes
1, 2, 3, 4, 5, 6, 7, and 8 were considered. The latter rep-
resent the nodes that can be found going from the ‘hinge’
nitrogen atom and moving away from it, as illustrated in
Figure 2 for the exemplar compounds 3 (Figure 2A) and
2k1 (Figure 2B).
Splitting of dataset into training set and test set
In this work, the splitting of DS into training set (TR) and
test set (TS) was carried out as follows: DS molecules
were first classified using a hierarchical clustering algo-
rithm.a In hierarchical clustering, data are iteratively parti-
tioned, until convergence is reached. Proceeding molecule
by molecule, each step of the clustering is characterized
by the creation of a tree that contains molecules at its
leaves and the whole DS at the root node. The method
used herein is a commonly known agglomerative method,
where molecules are added one by one and the forming
tree is opportunely updated at each step, assigning each
examined compound to the most appropriate leaf, that is,
cluster. At each stage, the method merges the two clus-
ters that are most similar to each other. The ‘choice’ of
the cluster to be assigned to a particular molecule was
made in this work using Ward’s method, which aggre-
gates instances such that the minimum increase in vari-
ance is obtained at each iteration within each cluster (16).
The threshold for the diameter of the clusters was selected
such that it provided in the hierarchy the minimum number
of clusters, all of them containing at least five compounds.
Doing this, it was possible to extract from each cluster at
least one compound while maintaining the ratio of about
20% between TS and TR. Such hierarchical clustering of
the DS provided two clusters. Individual TS compounds
Chem Biol Drug Des 2013; 82: 22–38
27

Borghini et al.
Table 3: Model validation on A1 adenosine receptors. Statistical parameters coming from training set, leave-one-out cross-validation and
test set, respectively
TR
LOO
TS
Residual sum of squares
SSRES
Multiple correlation coefficient
R2
4.787 Predictive residual sum of squares
PRESS
0.795 Multiple correlation coefficient Q2
5.891 Predictive residual sum of squares
PRESS
0.748 Multiple correlation coefficient R2
2.134
0.844
Number of predictor variables
Number of observations
Adjusted R2
Standard error of estimate
F-value
3
35
Number of predictor variables
Number of observations
3
35
Number of predictor variables
Number of observations
3
9
0.776 Adjusted Q2
0.724 Adjusted R2
0.436 Standard error of prediction
0.750
0.653
9.851
0.393 Standard error of prediction
60.269 F-value
49.732 F-value
Standard deviation error of
estimate
0.370 Standard deviation error of prediction
SDEP
0.410 Standard deviation error of prediction 0.487
SDEP
2
A
B
2
4
O
H
O
4
H
H4
1
1
3
5
H
H
3
2
2
4
5
N
H
N
4
H
3
5
H
3
3
4
H
6
4
O
5
4
H
4
H
5
5
H
6
6
H
6
H
Figure 2: Explanation of node assignment in the ‘node-wise’ molecular descriptor calculation, exemplified for compounds 3 (A) and 2k1
(B). They represent the nodes that can be found when going from the N6 and moving away from it.
were then extracted from each cluster by randomly select-
ing 20% of the compounds belonging to it. All the remain-
ing compounds made up the TR. As a result, the present
DS with 44 entities in all was split into a TR with 35 com-
pounds and a TS with nine compounds, as summarized in
Table 2.
prediction performance, when changes are introduced in
the TR and the model is retrained taking into account
these changes. Model robustness may be assessed by
leave-one-out cross-validation (LOO-CV). Finally, predictive
ability means how well the model predicts compounds
belonging to the external TS.
The most commonly used statistical parameters that give a
measure of performance of a MLR model, as is the case of
the A1 receptor, are those reported in Table 3. Concerning
the models developed for A2A and A3 receptors, the results of
the classification may be organized in confusion (or contin-
gency) matrices, wherein actual classes are reported in rows
and predicted classes are reported in columns. Thus, the
main diagonal represents the numbers of correctly classified
molecules, while numbers out of the main diagonal represent
misclassifications. From values included within confusion
matrices, many significant statistical parameters may be cal-
culated. Among them, sensitivity, specificity, concordance,
positive predictivity, negative predictivity, false-positive rate,
and false-negative rate constitute the so-called Cooper sta-
tistics (18). The remaining ones are also commonly used in
QSAR model validation. Such parameters refer to TR, LOO-
CV, and TS classifications.
QSAR model building
A method based on multiple linear regression (MLR) was
used for developing the model with regard to the A1 aden-
osine receptor subtype. A data mining approach that
makes use of the C4.5 decision tree (17) was used instead
for developing the QSAR model with regard to the A2A
and A3 adenosine receptor subtypes.
Statistical validation of the models
In this work, goodness of fit, robustness, and predictive
ability of the developed models were used as statistical
measures of performance of the models themselves.
Briefly, goodness of fit means how well the model predicts
compounds belonging to TR. Robustness indicates the
stability of the model and consequently is the one of its
28
Chem Biol Drug Des 2013; 82: 22–38

Selective Ligands for A1 Adenosine Receptors
Table 4: Confusion matrices concerning QSAR models for A2A (left) and A3 (right) adenosine receptors, calculated on the basis of training
set, LOO, and test set, respectively
2 9 2 contingency table
Predicted class
2 9 2 contingency table
Predicted class
A2A training set (n = 35)
Active
Inactive
A3 training set (n = 35)
Active
Inactive
Actual class
Active
Inactive
10
1
1
23
Actual class
Active
Inactive
21
2
1
11
2 9 2 contingency table
Predicted class
2 9 2 contingency table
Predicted class
A2A LOO (n = 35)
Active
Inactive
A3 LOO (n = 35)
Active
Inactive
Actual class
Active
Inactive
10
1
1
23
Actual class
Active
Inactive
21
2
1
11
2 9 2 contingency table
2 9 2 contingency table
Predicted class
Predicted class
A2A test set (n = 9)
Active
Inactive
A3 test set (n = 9)
Active
Inactive
Actual class
Active
Inactive
2
0
1
6
Actual class
Active
Inactive
4
1
1
3
ꢃ
Compounds 3 and 2o that are the compounds showing
Applicability domain of the models
the highest and the lowest affinity toward the receptor
modeled, respectively. This study was aimed at identifying
possible differences in the binding modes, which drive the
interaction of the same receptor with compounds
endowed with highly different affinities.
In the present work, the applicability domain of the models
obtained was estimated as follows. First of all, all TS mole-
cules were verified to possess a N-9-(ortho-fluorobenzyl)-
2-phenyl-8-azapurine substructure. This is because the
models were developed with the aim of screening new
libraries of compounds bearing such moieties. Secondly,
the algorithm of probability density function was applied to
the descriptors involved in the model.
Experimental
Chemistry
Molecular modeling study
Melting points were determined using a Reichert Kofler
hot-stage apparatus and are uncorrected. Infrared spectra
were recorded with a FT-IR spectrometer Nicolet/Avatar in
Nujol mulls. Routine nuclear magnetic resonance spectra
were recorded in DMSO-d6 solution on a Varian Gemini
200 spectrometer operating at 200 MHz. Mass spectra
were obtained on a Hewlett-Packard 5988 A spectrometer
(Hewlett-Packard, Palo Alto, CA, USA) using a direct injec-
tion probe and an electron beam energy of 70 eV. Evapo-
ration was performed in vacuo (rotary evaporator).
Analytical thin layer chromatography was carried out on
Merck 0.2-mm precoated silica gel aluminum sheets
(60 F-254). Flash-column chromatography was performed
using Merck Kieselgel 60 (230–400 mesh). Elemental anal-
yses were performed by our Analytical Laboratory.
A molecular docking study was performed to investigate,
with details at the molecular level, the interactions of N-[9-
(ortho-fluorobenzyl)-2-phenyl-8-azapurin-6-yl]-amides with
rat A1 adenosine receptors. The 3D structure of the rat A1
adenosine receptor was built by homology modeling using
the X-ray structure of the human A2A adenosine receptor
as a template, as described in the Experimental part. The
following compounds were selected to be submitted to
molecular docking:
ꢃ
DPCPX (dipropylcyclopentylxanthine) chosen as a refer-
ence compound; its interactions with the rat A1 adenosine
receptor are well known (such data can be widely retrieved
from the literature). This study was aimed at providing a
comparison of our model with published data.
ꢃ
All the nine compounds belonging to the same set used
9-(ortho-fluorobenzyl)-2-phenyl-8-azaadenine 1bis
To a stirred solution of sodium ethoxide obtained from
sodium (4.6 g, 0.02 g atom) and absolute ethanol
(50 mL), malononitrile (2.20 g, 0.033 mol) was added and
as TS for developing the above-mentioned QSAR models.
This part of the study served as a validation method for
the docking procedure used.
Chem Biol Drug Des 2013; 82: 22–38
29

Borghini et al.
the mixture was heated at reflux for 15 min. Then, a solu-
tion of 2-fluorobenzylazide (5.9 g, 0.058 mol) and benzo-
nitrile (20.6 g, 0.20 mol) in absolute ethanol (2 mL) was
slowly added and the mixture was refluxed for 7 h. The
solvent was partially evaporated at reduced pressure, and
the residue diluted with water and made acid to pH = 5
with 4 N acetic acid and the acid solution repeatedly
washed with hot ethyl ether. The desired product (3.57 g,
60% yield) was obtained by slow precipitation from the
acid layer. Mp: 215 °C (crystallized from EtOH 95°). IR
(per cm): 3468 (NH); 3317 (NH); 3175 (NH). 1H-NMR: d
8.70 (broad s, 1H, exch.), 8.40 (m, 3H, 2H arom + 1H
exch.), 7.53–7.23 (m, 7H, arom.), 5.88 (s, 2H, CH2-benzyl)
ppm. Elemental analysis: calcd for C17H13FN6 C 63.74, H
4.09, N 26.24; found C 63.61, H 4.05, N 26.15.
exch.), 8.52 (m, 2H, arom.), 7.59–7.20 (m, 7H, arom.),
6.00 (s, 2H CH2-benzyl) 2.85 (m, 1H, CHCO), 1.97–1.25
(m, 10H, CH2) ppm. MS (m/z): 430 (M+, 5%), 231
(100%); Elemental analysis: calcd for
C24H23FN6O C
66.69, H 5.39, N 19.52; found C 66.60, H 5.25, N
19.38.
6. Crystallized, yield: 32%. Mp 197 °C. IR (per cm):
3259 (NH), 1682 (C=O). 1H-NMR:
d 11.75 (s, 1H,
exch.), 8.53 (m, 2H, arom.), 8.08 (m, 1H, arom.), 7.70
(d, J = 2.8 Hz, 1H, arom.), 7.71–7.20 (m, 7H, arom.),
6.80 (m, 1H, arom.), 6.02 (s, 2H, CH2 -benzyl) ppm.
MS (m/z): 414 (M+, 14%), 104 (100%). Elemental analy-
sis: calcd for C22H15FN6O2 C 63.77, H 3.65, N 20.28;
found C 63.55, H 3.41, N 20.11.
7. Crystallized, yield: 10%. Mp 209–212 °C. IR (per cm):
3260 (NH), 1682 (C=O). 1H-NMR: d 8.68 (s, 1H, exch.),
8.53 (m, 2H, arom.), 7.89 (m, 1H, arom.), 7.60–7.10 (m,
9H, arom.), 6.01 (s, 2H, CH2-benzyl) ppm. MS (m/z): 414
(M+, 25%), 274 (100%). Elemental analysis: calcd for
C22H15FN6O2 C 63.77, H 3.65, N 20.28; found C 63.43,
H 3.37, N 20.19.
General method to obtain N-[9-(ortho-
fluorobenzyl)-2-phenyl-9H-8-azapurin-6-yl]-amides
2–13
To a solution of 8-azaadenine 1bis (160 mg, 0.5 mmol) in
100 mL of anhydrous toluene and some drops of triethyl-
amine, a solution of the suitable acyl chloride (2.5 mmol) in
10 mL of toluene was added dropwise, and then, the mix-
ture was refluxed for 24 h. After cooling, the mixture was
evaporated under reduced pressure, the residue diluted
with chloroform and the solution washed with 10% NaOH,
10% HCl, and water. After evaporation, the solid obtained
was crystallized from EtOH 95° or purified by flash chro-
matography.
8. Crystallized, yield: 11%. Mp 210 °C. IR (per cm): 3237
(NH), 1674 (C=O). 1H-NMR: d 11.98 (broad s, 1H, exch.),
8.50 (m, 2H, arom.), 8.16 (s, 1H, arom.), 8.04 (d, 1H,
J = 7.8 Hz, arom.), 7.80–7.21 (m, 9H, arom.), 6.04 (s, 2H,
CH2-benzyl) ppm. MS (m/z): 458 (M+, 10%), 460 (M+ +2,
4%) 149 (100%). Elemental analysis: calcd for
C24H16ClFN6O C 62.82, H 3.51, N 18.31; found C 63.12,
H 3.57, N 18.29.
2. Crystallized, yield 18%. Mp: 175 °C. IR (per cm): 3237
(NH), 1674 (C=O). 1H-NMR: d 11.80 (s, 1H, exch.), 8.50
(m, 2H, arom.), 7.59–7.20 (m, 7H, arom.), 5.99 (s, 2H,
CH2-benzyl), 2.36 (m, 1H, CHCO), 0.98 (m, 4H, CH2)
ppm. MS (m/z): 388 (M+, 6%), 304 (100%). Elemental
analysis: calcd for C21H17FN6O C 64.90, H 4.41, N 21.64;
found C 64.82, H 4.53, N 21.62.
9. Flash-chromatographed (eluent CH2Cl2-MeOH 99:1),
yield: 23%. Mp 185 °C; IR (per cm): 3249 (NH), 1689
(C=O). 1H-NMR: d 11.98 (s, 1H, exch.), 8.48 (m, 2H,
arom.), 7.94 (m, 2H, arom.), 7.60–7.25 (m, 9H, arom.),
6.04 (s, 2H, CH2-benzyl) ppm. MS (m/z): 442 (M+, 5%),
322 (100%). Elemental analysis: calcd for C24H16F2N6O
C 65.16, H 3.65, N 19.00; found C 65.28, H 3.67, N
19.18.
3. Flash-chromatographed (eluent CHCl3), yield 10%. Mp
187–190 °C. IR 3261 (NH) 1683 (C=O). 1H-NMR: d 11.34
(s, 1H, Exch.), 8.48 (m, 2H, arom.), 7.59–7.18 (m, 7H,
arom.), 5.98 (s, 2H, CH2-benzyl), 3.73 (m, 1H, CHCO),
2.35–1.87 (m, 4H, CH2.), 1.20 (m, 2H, CH2) ppm. MS (m/
z): 404 (M+ +1, 6%), 91 (100%). Elemental analysis: calcd
for C22H19FN6O C 65.66, H 4.76, N 20.88; found C
65.81, H 4.55, N 20.73.
10. Flash-chromatographed (eluent CHCl3), yield: 58%.
Mp 243 °C. IR (per cm): 3259 (NH), 1678 (C=O). 1H-
NMR: d 11.90 (s, 1H, exch.), 8.48 (m, 2H, arom.), 8.18
(q, J = 5.5 Hz, 2H, arom.), 7.59–7.24 (m, 9H, arom.),
6.02 (s, 2H, CH2-benzyl) ppm. MS (m/z): 442 (M+, 3%),
104 (100%). Elemental analysis: calcd for C24H16F2N6O C
65.16, H 3.65, N 19.00; found C 65.18, H 3.71, N
19.14.
4. Crystallized, yield 39%. Mp 185–187 °C. IR (per cm):
3236 (NH), 1685 (C=O). 1H-NMR: d 11.46 (s, 1H, exch.),
8.54 (m, 2H, arom.), 7.60–7.23 (m, 7H, arom.), 6.00 (s,
2H, CH2- benzyl), 3.20 (m, 1H, CHCO) 1.95–1.64 (m, 8H,
CH2) ppm. MS (m/z): 416 (M+, 90%), 273 (100%). Elemen-
tal analysis: calcd for C23H21FN6O C 66.33, H 5.08, N
20.18; found C 66.60, H 5.15, N 20.32.
11.
Flash-chromatographed
(eluent
CH2Cl2-MeOH
99.5:0.5), yield: 52%. Mp 188–200 °C. IR (per cm):
3227 (NH), 1682 (C=O). 1H-NMR: d 12.31 (broad s, 1H,
exch.), 8.93 (s, 1H, arom.), 8.51 (m, 4H, arom.), 7.88 (t,
J = 8.0 Hz, 1H, arom.), 7.60–7.22 (m, 7H, arom.), 6.03
(s, 2H, CH2-benzyl) ppm. MS (m/z): 469 (M+, 4%), 149
5. Crystallized, yield: 10%. Mp 180–184 °C. IR (per cm):
3251 (NH), 1689 (C=O). 1H-NMR:
d
11.41 (s, 1H,
(100%). Elemental analysis: calcd for C24H16FN7O3
C
30
Chem Biol Drug Des 2013; 82: 22–38

Selective Ligands for A1 Adenosine Receptors
Radioligand binding assays
61.41, H 3.44, N 20.89; found C 61.33, H 3.40, N
20.66.
Membranes of rat cerebral cortex that express A1 aden-
osine receptors were prepared using the method
described by Lohse et al. (19) with slight modifications.
Male Wistar rat brain cortex was homogenized in 10 vol-
umes of ice-cold 0.32 M sucrose, 20 mM Tris–HCl buffer,
pH 7.4 with 30 strokes in Dounce homogenizer. The
homogenate was centrifuged at 1000 9 g for 10 min to
remove the nuclear fraction, and the resulting superna-
tant was centrifuged at 30 000 9 g for 30 min. The pel-
let was resuspended using 10 strokes in Dounce
homogenizer in 10 volumes of ice-cold 5 mM Tris–HCl
buffer, pH 7.4 for 30 min. After 60 strokes in Dounce
homogenizer, the resulting synaptosomal membranes
were preincubated for 30 min at 37 °C with 2 U/mL of
adenosine deaminase to remove endogenous adenosine.
The membrane suspension was then centrifuged at
48 000 9 g for 30 min, and the resulting pellet was
resuspended in 10 volumes of 50 mM Tris–HCl buffer,
pH 7.4, and stored at 80 °C until binding assays were
made.
12. Flash-chromatographed (eluent CHCl3), yield 61%. Mp
258–260 °C. IR (per cm): 3303 (NH), 1705 (C=O). 1H-
NMR: d 12.26 (s, 1H, exch.), 8.43–8.27 (m, 6H, arom.),
7.57–7.25 (m, 7H, arom.), 6.03 (s, 2H, CH2-benzyl) ppm.
MS (m/z): 470 (M+, 100%). Elemental analysis: calcd for
C
24H16FN7O3 C 61.41, H 3.44, N 20.89; found C 61.39,
H 3.31, N 20.76.
13. Crystallized, yield: 20%. Mp 220 °C. IR (per cm): 3225
(NH), 1669 (C=O). 1H-NMR: d 11.94 (s, 1H, exch.), 8.46
(m, 2H, arom.), 7.85 (dd, J = 8.4 Hz, 2H, arom.),
7.58–7.24 (m, 9H, arom.), 6.02 (s, 2H, CH2-benzyl). MS
(m/z): 550 (M+, 5%), 104 (100%). Elemental analysis: calcd
for C24H16FIN6O C 52.38, H 2.93, N 15.27; found C
52.18, H 2.79, N 15.44.
N-[9-(ortho-fluorobenzyl)-2-phenyl-9H-8-azapurin-
6-yl]-amides 14 and 15
Compounds 11 and 12 (0.066 mmol) were hydrogenated
over 10% Pd/C (20 mg) in THF and isopropanol 3:1
(100 mL) at room pressure and temperature. After filtra-
tion, the solvent was evaporated to give the desired
compounds that were flash-chromatographed on silica
gel using CH2Cl2/MeOH 99:1 as eluent for compound 14
and cyclohexane-ethyl acetate 3:2 as eluent for com-
pound 15.
For displacement experiments involving A1 adenosine
receptors, rat cortex membranes (40 mg of protein) were
incubated at 25 °C for 60 min with [3H]DPCPX 0.5 nM
(Kd = 0.4 nM) and increasing concentrations of the com-
pounds, in a final volume of 0.4 mL of Tris–HCl buffer.
Non-specific binding was measured in the presence of
100 lM CPA. Binding reactions were terminated by dilu-
tion with ice-cold 50 mM Tris–HCl buffer, pH 7.4.
Samples were then filtered through Whatman GF/C
glass-fiber filters using a Brandel cell harvester. Filters
were washed three times with 4 mL of the same
buffer. Bound radioactivity was measured in a liquid
scintillation counter (1600 TR Packard) after the
addition of 4 mL of scintillation liquid (Emulsifier-Safe;
Hewlett-Packard, Palo Alto, CA, USA).
14. Yield: 25%. Mp 149 °C. IR (per cm): 3349 (NH), 3211
(NH), 1680 (C=O). 1H-NMR: d 11.62 (s, 1H, exch.), 8.48
(m, 2H, arom.), 7.59–7.23 (m, 11H, arom. + 1H exch.),
6.86 (s, 1H, arom.) 6.01 (s, 2H, CH2-benzyl), 5.90 (s, 1H,
exch.) ppm. MS (m/z): 439 (M+, 7%), 109 (100%). Elemen-
tal analysis: calcd for C24H18FN7O C 65.60, H 4.13, N
22.31; found C 65.38, H 4.00, N 22.19.
Slightly different conditions were set in the case of bind-
ing displacement experiments regarding A2A and A3
adenosine receptors. Membranes of Chinese hamster
ovary cells expressing recombinant human A2A or A3
receptors were prepared as previously described (8).
Membranes (40 mg of protein) were incubated with [3H]
ZM241385 6 nM (Kd = 4 nM) in the experiments involving
the A2A subtype and with [3H]NECA 15 nM (Kd = 150 nM)
in the ones involving the A3 subtype and the compounds
to be assayed, at fixed concentration (10 lM) or at
increasing concentrations of the compounds in duplicate,
in a final volume of 0.4 mL of Tris–HCl buffer for 120 min
at 25 °C. Non-specific binding was measured in the
presence of 100 lM NECA in the case of A2A and
100 lM R-PIA in the case of A3 binding assay. Samples
were handled as mentioned before.
15. Yield: 31%. Mp 185–190 °C. IR (per cm): 3385 (NH),
3228 (NH) 1668 (C=O). 1H-NMR: d 11.34 (s, 1H, exch.),
8.52 (m, 2H, arom.), 7.89 (d J = 8.6 Hz, 2H, arom.),
7.59–7.20 (m, 7H, arom.), 6.63 (m, 4H, 2H arom. + 2H
exch.), 6.00 (s, 2H, CH2-benzyl). MS (m/z): 439 (M+, 5%),
104 (100%). Elemental analysis: calcd for C24H18FN7O C
65.60, H 4.13, N 22.31; found C 65.48, H 4.03, N 22.25.
Biological assays
Materials
[3H]DPCPX (120 Ci/mmol) was purchased from Amersham
Pharmacia, Amersham, UK, and [3H]NECA (20.6 Ci/mmol)
was from PerkinElmer Life Sciences. [3H]ZM241385
(27.4 Ci/mmol) was from Tocris Cookson. DPCPX, NECA,
and CPA were from Sigma-Aldrich (St. Louis, MO, USA).
All other chemicals used, at analytical grade, were from
standard commercial sources.
Binding parameters were calculated by GRAPHPAD PRISM
(GRAPHPAD, San Diego, CA, USA).
Chem Biol Drug Des 2013; 82: 22–38
31

Borghini et al.
Molecular docking
The 3D structure of the rat A1 adenosine receptor was
retrieved from MODBASE database (20) where it had
been built by homology modeling using the X-ray struc-
ture of the human A2A adenosine receptor (PDB entry:
3EML) as
a template. The overall sequence identity
between the two proteins was 49%. Model refinement
was then preformed using the MODREFINER program (21)
Figure 3: A graphical view of the tree model developed for A2A
adenosine receptors.
to generate
a refined full-atom model from C-alpha
traces, with improved global and local qualities. The
refinement performed by ModRefiner is a two-step pro-
cedure. First, the algorithm constructs
a main-chain
A3 receptors. Among them, only compounds having a cy-
cloalkyl or a furan group bound to the carbonyl carbon
showed appreciable affinity values toward A2A receptors,
while none of the new compounds showed affinity toward
A3 receptors. Accordingly, such new compounds possess
good selectivity properties toward A1 receptor with respect
to A2A and A3 receptors. These results are summarized in
Table 1.
model from the C-alpha trace with an acceptable back-
bone topology and main-chain hydrogen (H)-bonding
network. Next, side-chain atoms are added onto the
backbone conformation and optimized with the use of a
composite physics-based and knowledge-based force
field. Hydrogen atoms were then added, and disulfide
bonds between Cys80 and Cys169 and between
Cys260 and Cys263 were built using YASARA View ver-
sion 12.11.25. Receptor model was then energy mini-
mized using YASARA Energy Minimization server (22).
The MolProbity server (23) was then used to validate the
physical quality of the A1 receptor model. MolProbity
provided an MPscore for the receptor model, which is a
log-weighted combination of the number of structural
outliers, if any, that have values outside the region of
standard protein structures, including rotamer outliers,
torsion-angle outliers, and steric clashes.
QSAR study of A1 adenosine receptor ligands
The model developed for the A1 adenosine receptor is
defined by the following MLR equation:
pKi ¼ &141:2432 ꢂ 0:0739ðꢀ0:0664Þ
ꢄ N atom node 6 ꢂ 0:1038ðꢀ0:01893Þ
ꢄ SaaCH þ 83:81ðꢀ11:1057Þ ꢄ WBN EN L 0:50
The three descriptors involved in the model (each one of
them reported with its regression coefficient within the
equation above) are the following:
The software AUTODOCKTOOLS version 1.5.4 revision 30 was
used to add Gasteiger charges, select the flexible side-
chain residues, and save the structures in pdbqt file for-
mat. For each of the ligands to be subjected to molecular
docking, the search for the minimum energy conformation
was first performed using VEGAZZ RELEASE 2.0.8.60 (http://
www.vegazz.net/). Then, VINA version 1.1.2 (24) was used
to perform molecular docking. The docking grid was set to
ꢃ
N_atom_node_6 is the number of atoms in the sixth
node counted starting from the N6 according to Figure 2.
ꢃ
SaaCH, abbreviation for ‘Sum of atom-type E-State::
CH:2D’ belongs to the electrotopological state indices pro-
ꢁ
ꢁ
ꢁ
a size of 14 A 9 12 A 9 24 A and placed in such a way
that it could embrace the extracellular portion of the recep-
tor. The GraphPad Prism software was used to calculate
the Pearson correlations by comparing the experimental
pKi values with the predicted Vina affinity scores. All fig-
ures illustrating the molecular structures were produced
using YASARA View.
posed by Hall and Kier (25–27).
ꢃ
WBN_EN_L_0.50, which stands for WeightedBurden-
Number_Electronegativity_LowestEigenv_050, is a modifi-
cation of the Burden number (28).
In brief, it represents the lowest eigenvalue obtained from
the Burden connectivity matrix, with electronegativity on
the diagonal, properly weighted.
Results and Discussion
Biological results
From the equation above, it is possible to observe that
both descriptors N_atom_node_6 and SaaCH negatively
contribute to affinity toward A1 receptors. The first descrip-
tor, N_atom_node_6, reflects features related to the steric
hindrance of the substituent on the carbonyl carbon, at a
distance of six nodes from the N6 and its values indicate
that higher steric hindrances correspond to lower affinity
values. See for example, compound 2k1, with
Biological results demonstrated that the synthesized com-
pounds generally possess a very interesting affinity and
selectivity toward A1 receptors. All compounds having a
cycloalkyl group and compounds having a furan or a
substituted phenyl group bound to the carbonyl carbon
showed affinity values in the low nanomolar range toward
the said receptor subtype. On the contrary, the new com-
pounds are in general not significantly active on A2A and
32
Chem Biol Drug Des 2013; 82: 22–38

Selective Ligands for A1 Adenosine Receptors
Table 5: The most significant statistical parameters obtained
from values included within confusion matrices, for both A2A (left)
and A3 (right) models, each calculated on the basis of TR, LOO,
and TS, respectively
A2A
TR
A3
LOO
0.91
TS
0.67
TR
LOO
0.95
TS
0.80
Sensitivity (true
positive rate)
Specificity (true
negative rate)
Concordance or
accuracy
Positive
predictivity
Negative
predictivity
0.91
0.95
0.96
0.94
0.91
0.96
0.04
0.96
0.94
0.91
0.96
0.04
1.00
0.89
1.00
0.86
0.00
0.85
0.91
0.91
0.92
0.15
0.85
0.91
0.91
0.92
0.15
0.75
0.78
0.80
0.75
0.25
Figure 4: A graphical view of the tree model developed for A3
adenosine receptors.
False-positive
(over-
classification)
rate
False-negative
(over-
classification)
rate
ues in compounds with high affinity toward A1 receptors,
such as 2a–2d that have cycloalkyl groups bound to the
carbonyl carbon.
0.09
0.09
0.33
0.05
0.05
0.20
Error rate
0.06
0.06
0.11
0.09
0.09
0.22
The model was developed on compounds belonging to
TR and validated by means of LOO-CV and using the
external TS. Statistical parameters resulting from TR, LOO-
CV, and TS validations are reported in Table 3.
No-model error
rate, NOMER%
Prior probability
of active class
Prior probability
of inactive class
Prior proportional
probability of
active class
Prior proportional
probability of
inactive class
31.43 31.43 33.33 62.86 62.86 55.56
0.09
0.04
0.31
0.09
0.04
0.31
0.33
0.17
0.33
0.05
0.08
0.63
0.05
0.08
0.63
0.20
0.25
0.56
QSAR study of A2A adenosine receptor ligands
The QSAR model developed on A2A adenosine receptor is
a J48 pruned tree and may be drawn as follows:
0.69
0.69
0.67
0.37
0.37
0.44
N_atom_node_6 = 0: Active (11/1)
N_atom_node_6 > 0: Inactive (24/1)
TR, training set; TS, test set.
N_atom_node_6 = 4 and pKi = 1523 ꢀ 500, while com-
pounds having no substituents or cycloalkyl and furan
groups bound to the carbonyl carbon have
N_atom_node_6 = 0 and high affinity toward A1 receptors.
The electrotopological SaaCH generally decreases at
increasing values of pKi on A1 receptors. Again, the same
compounds having no substituents or cycloalkyl and furan
groups on the carbonyl carbon show the lowest values of
the said descriptor while possessing a good affinity toward
A1 receptor. Unlike the first descriptor, SaaCH is strongly
affected by the presence or absence of the fluorine atom
on the 9-benzyl group, because all of the F-containing
compounds belonging to TR have values of it distributed
in the lowest half of a scale ordered according to increas-
ing values, and this is again in agreement with their affinity
observed toward A1 receptors. The third descriptor,
WBN_EN_L_0.50, gives instead a positive contribution to
pKi. Indeed, it goes from relatively low values for inactive
compounds such as 2o and 2p, to reach the highest val-
Number of leaves: 2; size of the tree: 3. A graphical repre-
sentation of the tree above is illustrated in Figure 3. The
numbers in parentheses represent the number of com-
pounds predicted to belong to the corresponding class
(left) and the number of misclassified molecules (right). The
descriptor N_atom_node_6 is able to correctly classify
most of the examined compounds as ‘active’ or ‘inactive’
toward A2A adenosine receptors, with only two misclassifi-
cations over 35 compounds in all. This indicates that com-
pounds bearing any substituent six or more bonds further
from the carbonyl carbon than are inactive, suggesting
that the relatively small size of the A2A binding pocket does
not allow for large substituents at the considered position.
Figure 2 itself exemplifies compound 3, with N_atom_
node_6 = 0, and compound 2k1, with N_atom_node_6 = 4.
The first compound is active on both A1 and A2A receptors,
while the second one is less active on A1 receptors and even
inactive on A2A
.
Chem Biol Drug Des 2013; 82: 22–38
33

Borghini et al.
Compound 2o
Compound 3
A
D
Interaction energy = –8.8 kcal/mol
Interaction energy = –7.7 kcal/mol
B
E
Interaction energy = –8.3 kcal/mol
Interaction energy = –7.0 kcal/mol
C
F
Interaction energy = –7.9 kcal/mol
Interaction energy = –6.6 kcal/mol
Figure 5: Depiction of the three best docking poses of compounds 3 (A–C) and 2o (D–F), which are representative of different possible
conformation families. In each instance, the amino acid residues of the A1 adenosine receptor interacting with the ligand are labeled, and
the energy scores provided by Vina are reported in kcal/mol. Hydrophobic and p–p interactions between ligand and amino acid residues
on the receptor are illustrated as green and red lines, respectively.
34
Chem Biol Drug Des 2013; 82: 22–38

Selective Ligands for A1 Adenosine Receptors
Scheme 1: General method for the synthesis of compounds 1–13 and 14–15.
Again, the model was developed on compounds belong-
ing to a TR and validated by means of LOO-CV and exter-
nal TS. Confusion matrices for this model are illustrated in
Table 4, left column, while the most significant statistical
parameters are reported in Table 5.
negativity of the whole molecule has to be properly bal-
anced with respect to its size. As expected, it appears to
be highly affected by the presence of the strongly elec-
tronegative F atom. Inactive compounds such as 2l and
2g (without F) have low values of this descriptor, while its
highest value is calculated for 6 and 7 (with F), which are
active compounds. The second descriptor is SP-0,
meaning simple path order 0, and belonging to the class
QSAR study of A3 adenosine receptor ligands
The QSAR model developed on A3 adenosine receptors is
a J48 pruned tree as well. Only two descriptors are
enough to classify the examined compounds as ‘active’ or
‘inactive’ toward this receptor subtype. The tree is
reported below:
of the ChiPathDescriptors (31). Being
a topological
descriptor, it highlights the importance of having a proper
value of steric hindrance. It is worth noting that in this
specific case, all compounds, bearing a nitro substituent
at any position of the phenyl ring and any multiple substi-
tution on the same ring, fall into the range
22.622745–24.363232 wherein compounds are correctly
predicted as inactive. Also, in this case, the model was
developed on compounds belonging to a TR and vali-
dated by means of LOO-CV and external TS. Confusion
matrices for this model are illustrated in Table 4, right
column, while the most significant statistical parameters
are reported in Table 5.
ETA_BetaP_s ꢅ 0.64706: Inactive (5/0)
ETA_BetaP_s > 0.64706
| SP-0 ꢅ 22.622745: Active (20/2)
| SP-0 > 22.622745
Molecular docking
| | SP-0 ꢅ 24.363232: Inactive (7/1)
| | SP-0 > 24.363232: Active (3/0)
The model for rat A1 adenosine receptor obtained was
checked out by assessing its agreement with the well-rec-
ognized experimental binding mode of DPCPX, as a preli-
minary first validation step. Such a model showed to be in
good agreement with the literature data.
Number of leaves: 4; size of tree: 7. A graphical repre-
sentation of the tree above is illustrated in Figure 4.
Again, the numbers in parentheses represent the number
of compounds predicted to belong to the corresponding
class (left) and the number of misclassified molecules
(right). The first descriptor is ETA_BetaP_s, an extended
topochemical atom descriptor, which reflects the measure
of electronegative atom count of the molecule relative to
molecular size 2D (29,30). The presence of a threshold
value of the said descriptor, below which compounds are
inactive toward A3 receptors, indicates that the electro-
As a further validation step, the docking procedure applied
to the A1 receptor model was tested by assessing the cor-
relation between the experimental pKi values of the nine
ligands belonging to the TS and their predicted affinity
scores provided by the Vina software (see the Experimen-
tal section), as the mean of the best three docking poses
for each ligand. This correlation coefficient has a value of
R2 = 0.52.
Chem Biol Drug Des 2013; 82: 22–38
35

Borghini et al.
Each of the TS compounds resulted to interact with A1
receptor in different conformation families. By way of
example, compounds 3 and 2o, showing the highest
and the lowest affinity values toward A1, respectively,
were selected for a more detailed discussion. Their three
best docking poses, having the spatial arrangements
corresponding to the lowest interaction energies, are
reported as representative of three optimal conformation/
orientation families found; they are illustrated in Fig-
ure 5A–C (compound 3) and Figure 5D–F (compound
2o). In particular, Figure 5A shows that 3, in its first
is generally related to the presence of the fluorine atom,
such an increase being especially marked in compounds
5, 7, and 9. On the contrary, the affinity toward A2A
and A3 receptors is not significantly affected by the fluo-
rine atom. The higher A1 affinity of F-containing com-
pounds combined with their substantially unchanged
affinity toward A2A and A3 increases the selectivity in all
of them.
Based on the structural and biological data obtained,
QSAR models were developed to rationalize the observed
affinity trends and to find out other molecular features,
beyond the presence/absence of fluorine atom, which
can drive the affinity toward the different adenosine
receptor subtypes. Descriptors reflecting both electroneg-
ativity and steric properties were involved in model build-
ing and were calculated both for the whole molecule and
for the substituent bound to the carbonyl carbon at the 6
position only. The latter descriptors were calculated on
the basis of an original ‘node-wise’ representation of the
said substituent (see Figure 2) where the graph nature of
the molecules is emphasized. In the QSAR models devel-
oped for all the three receptor subtypes, the descriptor
related to node 6 (N_atom_node_6) appears to be as
important as descriptors reflecting steric and electronega-
tivity features related to the whole molecule (SaaCH,
WBN_EN_L_0.50, ETA_BetaP_s, and SP-0). However,
descriptors relating to the whole molecule may also be
considered to reflect the influence exerted by the substi-
tuent bound to the carbonyl carbon and the presence or
absence of the F atom at the ortho position of 9-benzyl
ring on the properties of the whole molecule. Indeed, the
remaining portion is identical for all the compounds under
investigation.
docking pose, establishes
a hydrophobic interaction
between its 2-phenyl ring and ILE270 and a p–p interac-
tion between its 9-F-benzyl group and HIS278. Fig-
ure 5B illustrates 3 in its second best docking pose and
the interactions established by its 9-F-benzyl group with
VAL62 (hydrophobic) and HIS278 (p–p). Figure 5C
depicts the third docking pose of 3, which establishes a
hydrophobic interaction between its 2-phenyl ring and
ILE270 and a p–p interaction between N7 of the 8-azap-
urine nucleus and PHE171. Similarly, Figure 5D shows
compound 2o in its first docking pose and the interac-
tions established by its 2-phenyl ring with LEU250
(hydrophobic) and PHE171 (p–p). Figure 5E illustrates the
second best pose of 2o, which establishes a hydropho-
bic interaction between the OMe moiety of the R substi-
tuent on the carbonyl carbon bound to N6 and ILE69
and a p–p interaction between the phenyl ring of the
same
R substituent and HIS278. Finally, Figure 6F
depicts the third docking pose of 2o, the 2-phenyl ring
of which only interacts with ARG157 by a hydrophobic
interaction.
Conclusion
Other than rationally describing the observed behavior of
the newly synthesized compounds toward the adenosine
receptor subtypes, the QSAR models obtained can repre-
sent a helpful tool in the design of new compounds
endowed with high affinity and selectivity toward the A1
subtype. According to the findings above, for an adeno-
sine receptor ligand, having a N-(9-benzyl-2-phenyl-8-
azapurin-6-yl)-amide structure, to show a high A1/A2A-A3
selectivity, it should preferentially have a fluorine atom at
the ortho position of the 9-benzyl group; furthermore, opti-
mal substituents bound to the carbonyl carbon at the 6
position are cycloalkyl groups, furan, and phenyl rings,
especially monosubstituted by halogens, amino, or nitro
groups. On the contrary, different phenyl substituents such
as alkoxy, as well as the presence of multiple substitution
on the said phenyl ring, decrease affinity toward A1 recep-
tors.
This article presents a series of N-[9-(ortho-fluorobenzyl)-
2-phenyl-8-azapurin-6-yl]-amides that were synthesized
and tested for their affinity toward A1, A2A, and A3
adenosine receptor subtypes. Many of them show a
good selectivity for A1 with respect to A2A and A3
receptors. In this regard, special attention should be
focused on compound 9, which bears a meta-fluorophe-
nyl substituent bound to the carbonyl carbon and shows
a highly significant A1/A2A-A3 selectivity (three orders of
magnitude) due to lack of affinity toward A2A (17%i) and
A3 (48%i) subtypes. This property is extraordinarily inter-
esting; indeed, 8-(noradamantan-3-yl)-1,3-dipropylxan-
thine (also known as rolofylline or KW-3902), an A1-
selective reference compound, has a value of 890 for
A1/A2A selectivity (32). By comparing the series of N-(9-
benzyl-2-phenyl-8-azapurin-6-yl)-amides previously syn-
thesized by us (11), it is possible to directly estimate the
effects of the introduction of the F atom on the benzyl
group at the 9 position. Indeed, several of the com-
pounds in reference (11) are analogous to compounds
reported in this article, except for the presence of the
said F atom. An increase in affinity toward A1 receptors
Finally, molecular docking showed that the aromatic moie-
ties of the ligands analyzed may interact with the model of
rat A1 receptor at different binding pockets and that these
interaction typologies are mainly of hydrophobic and p–p
nature.
36
Chem Biol Drug Des 2013; 82: 22–38

Selective Ligands for A1 Adenosine Receptors
of new trisubstituted 8-azaadenines with high affinity
for A1 adenosine receptors. Eur J Med Chem;42:1–9.
11. Giorgi I., Leonardi M., Pietra D., Borghini A., Massarelli
I., Ciampi O., Bianucci A.M. (2009) Synthesis, biologi-
cal assays and QSAR studies of N-(9-benzyl-2-phenyl-
8-azapurin-6-yl)-amides as ligands for A1 adenosine
receptors. Bioorg Med Chem;17:1817–1831.
Acknowledgment
This research was supported by the Italian Ministero
dell’Istruzione, dell’Universita e della Ricerca (MIUR). The
authors are grateful to International Society for Drug Devel-
opment SpA (ISDD), Milano (Italy), for the financial support
granted to research team.
ꢀ
12. Biagi G., Franchi M., Giorgi I., Livi O., Scartoni V.
(1989) ‘One pot’ synthesis of 2-substituted 9-(2′-
hydroxy-3′-aminopropyl)-8-azahypoxanthines and 8-
azaadenines (5-substituted 3-(2′-hydroxy-3′-aminopro-
pyl) 7-amino and 7-hydroxy-3H-1,2,3-triazolo[4,5-d]
pyrimidines). J Heterocycl Chem;26:39–43.
Conflict of Interest
The authors declare no conflict of interest.
References
13. Yap C.W. (2011) PaDEL-descriptor: an open source
software to calculate molecular descriptors and finger-
prints. J Comput Chem;32:1466–1474.
14. Liu K., Feng J., Young S.S. (2005) PowerMV: a soft-
ware environment for molecular viewing, descriptor
generation, data analysis and hit evaluation. J Chem
Inf Model;45:515–522.
15. Borghini A., Crotti P., Pietra D., Favero L., Bianucci
A.M. (2010) Chemical reactivity predictions: use of data
mining techniques for analyzing regioselective azidoly-
sis of epoxides. J Comput Chem;31:2612–2619.
16. Ward J.H. (1963) Hierarchical grouping to optimize an
objective function. J Am Stat Assoc;58:236–244.
17. Witten I.H., Frank E. (2005) Chapter 6.1 decision trees.
In: Gray J., editor. Data Mining. Practical Machine
Learning Tools and Techniques, 2nd edn. San Fran-
cisco, CA, USA: Morgan Kaufmann Publisher; p.
189–199.
1. Fredholm B.B., Ijzerman A.P., Jacobson K.A., Klotz
K.N., Linden J. (2001) International Union of Pharma-
cology. XXV. Nomenclature and classification of adeno-
sine receptors. Pharmacol Rev;53:527–552.
2. Voors A.A., Dittrich H.C., Massie B.M., DeLucca P.,
Mansoor G.A., Metra M., Cotter G., Weatherley B.D.,
Ponikowski P., Teerlink J.R., Cleland J.G.F., O’Connor
C.M., Givertz M.M. (2011) Effects of the adenosine A1
receptor antagonist rolofylline on renal function in
patients with acute heart failure and renal dysfunction.
J Am Coll Cardiol;57:1899–1907.
3. Abraham W.T. (2010) Paper presented at: the Euro-
pean Society of Cardiology Heart Failure Congress,
Late Breaking Clinical Trials; May 30. Berlin, Germany.
4. Jacobson K.A., Gao Z.-G. (2006) Adenosine receptors
as therapeutic targets. Nat Rev Drug Discov;5:247–264.
5. Hocher B. (2010) Adenosine A1 receptor antagonists in
18. Cooper J.A., Saracci R., Cole P. (1979) Describing the
validity of carcinogen screening tests. Br J Can-
cer;39:87–89.
clinical
research
and
development.
Kidney
Int;78:438–445.
6. Schenone S., Brullo C., Musumeci F., Bruno O., Botta
M. (2010) A1 receptor ligands: past, present and future
trends. Curr Top Med Chem;10:878–901.
19. Lohse M.J., Lenschow V., Schwabe U. (1984) Interac-
tion of barbiturates with adenosine receptors in rat
brain.
Naunyn-Schmiedeberg’s
Arch
Pharma-
7. Biagi G., Giorgi I., Livi O., Scartoni V. (1996) N(9)-
substituted 2-phenyl-N(6)-benzyl-8-azaadenines: A1
adenosine receptor affinity. A comparison with the cor-
responding N(6)-substituted 2-phenyl-N(9)-benzyl-8-
azaadenines. Farmaco;51:395–399.
8. Biagi G., Giorgi I., Leonardi M., Livi O., Pacchini F.,
Scartoni V., Costa B., Lucacchini A. (2003) New N6- or
N(9)-hydroxyalkyl substituted 8-azaadenines or ade-
nines as effective A1 adenosine receptor ligands. Eur J
Med Chem;38:801–810.
col;326:69–74.
20. Pieper U., Webb B.M., Barkan D.T., Schneidman-Du-
hovny D., Schlessinger A., Braberg H., Yang Z. et al.
(2011) MODBASE, a database of annotated compara-
tive protein structure models and associated
resources. Nucleic Acids Res;39:465–474.
21. Xu D., Zhang Y. (2011) Improving the physical realism
and structural accuracy of protein models by a two-
step atomic-level energy minimization. Biophys
J;101:2525–2534.
9. Biagi G., Bianucci A.M., Coi A., Costa B., Fabbrini L.,
Giorgi I., Livi O., Micco I., Pacchini F., Santini E., Leo-
nardi M., Ahmad Nofal F., LeRoy Salerni O., Scartoni
V. (2005) 2,9-Disubstituted-N6-(arylcarbamoyl)-8-aza-
adenines as new selective A3 adenosine receptor
antagonists: synthesis, biochemical and molecular
modelling studies. Bioorg Med Chem;13:4679–4693.
10. Giorgi I., Bianucci A.M., Biagi G., Livi O., Scartoni V.,
Leonardi M., Pietra D., Coi A., Massarelli I., Ahmad
Nofal F., Fiamingo F.L., Anastasi P., Giannini G. (2007)
Synthesis, biological activity and molecular modelling
22. Krieger E., Joo K., Lee J., Lee J., Raman S., Thomp-
son J., Tyka M., Baker D., Karplus K. (2009) Improving
physical realism, stereochemistry, and side-chain
accuracy in homology modeling: four approaches that
performed
well
in
CASP8.
Proteins;77(Suppl
9):114–122.
23. Chen V.B., Arendall W.B. III, Headd J.J., Keedy D.A.,
Immormino R.M., Kapral G.J., Murray L.W., Richard-
son J.S., Richardson D.C. (2010) MolProbity: all-atom
structure validation for macromolecular crystallography.
Acta Crystallogr D Biol Crystallogr;66:12–21.
Chem Biol Drug Des 2013; 82: 22–38
37

Borghini et al.
24. Trott O., Olson A.J. (2010) AutoDock Vina: improving
the speed and accuracy of docking with a new scoring
function, efficient optimization and multithreading. J
Comput Chem;31:455–461.
25. Kier L.B., Hall L.H. (1990) An electrotopological-state
index for atoms in molecules. Pharm Res;7:801–807.
26. Hall L.H., Kier L. (1995) Electrotopological state indices
for atom types: a novel combination of electronic,
topological, and valence state information. J Chem Inf
Comput Sci;35:1039–1045.
30. Roy K., Das R.N. (2011) On some novel extended to-
pochemical atom (ETA) parameters for effective encod-
ing of chemical information and modelling of
fundamental physicochemical properties. SAR QSAR
Environ Res;22:451–472.
31. Kier L.B., Hall L.H. (1976) Molecular Connectivity in
Chemistry and Drug Research. New York: Academic
Press.
32. Nonaka H., Ichimura M., Takeda M., Kanda T., Shi-
mada J., Suzuki F., Kase H. (1996) KW-3902, a selec-
tive high affinity antagonist for adenosine A1 receptors.
Br J Pharmacol;117:1645–1652.
27. Kier L.B., Hall L.H. (1999) Molecular Structure Descrip-
tion: The Electrotopological State. London: Academic
Press.
28. Burden F.R. (1989) Molecular identification number for
Note
substructure searches.
J
Chem Inf Comput
Sci;29:225–227.
aJ. Demsar, B. Zupan, Orange: from experimental machine
learning to interactive data mining. Website http://www.
ailab.si/orange/
29. Roy K., Ghosh G. (2004) QSTR with extended topo-
chemical atom indices. 2. Fish toxicity of substituted
benzenes. J Chem Inf Comput Sci;44:559–567.
38
Chem Biol Drug Des 2013; 82: 22–38