Multistructure 3D-QSAR studies on a series of conformationally constrained butyrophenones docked into a new homology model of the 5-HT2A receptor

Article abstract of DOI:10.1021/jm070277a

The present study is part of a long-term research project aiming to gain insight into the mechanism of action of atypical antipsychotics. Here we describe a 3D-QSAR study carried out on a series of butyrophenones with affinity for the serotonin-2A recepto

Full text of DOI:10.1021/jm070277a

3242
J. Med. Chem. 2007, 50, 3242-3255
Multistructure 3D-QSAR Studies on a Series of Conformationally Constrained Butyrophenones
Docked into a New Homology Model of the 5-HT2A Receptor
Cristina Dezi,† Jose´ Brea,†,‡ Mario Alvarado,§ Enrique Ravin˜a,§ Christian F. Masaguer,§ Mar´ıa Isabel Loza,‡ Ferran Sanz,† and
Manuel Pastor*,†
Research Unit on Biomedical Informatics (GRIB), IMIM, UniVersitat Pompeu Fabra, Dr. Aiguader 88, E-08003 Barcelona, Spain,
and Department of Pharmacology, Department of Organic Chemistry, Faculty of Pharmacy, UniVersidade de Santiago de Compostela,
E-15782 Santiago de Compostela, Spain
ReceiVed March 11, 2007
The present study is part of a long-term research project aiming to gain insight into the mechanism of
action of atypical antipsychotics. Here we describe a 3D-QSAR study carried out on a series of butyrophenones
with affinity for the serotonin-2A receptor, aligned by docking into the binding site of a receptor model.
The series studied has two peculiarities: (i) all the compounds have a chiral center and can be represented
by two enantiomeric structures, and (ii) many of the structures can bind the receptor in two alternative
orientations, posing the problem of how to select a single representative structure for every compound. We
have used an original solution consisting of the simultaneous use of multiple structures, representing different
configurations, binding conformations, and positions. The final model showed good statistical quality (n )
426, r2 ) 0.84, q2
) 0.81) and its interpretation provided useful information, not obtainable from the
LOO
simple inspection of the ligand-receptor complexes.
Introduction
reflects the atypical profile (pKi 5-HT2A/pKi D2 >1.12).4 This
ratio led to the definition of the Meltzer’s index (MI) as a
screening criteria for candidate AAPD. Although the exceptional
therapeutic performance of clozapine, not completely achieved
by other AAPD that fulfill the MI criteria, has been recently
proposed to be extended to its interaction with a large number
of receptors in addition to 5-HT2A and D2 receptors.6
For a long time, our group has been involved in research
aiming to improve the understanding of the pharmacological
mechanisms of AAPD action and to discover novel compounds
with interest in psychosis therapy. Many of these studies have
been carried out on a series of conformationally restricted
butyrophenone analogues, still in active development, containing
compounds with high affinity for the 5-HT2A receptor,7-9 and
with particularly interesting pharmacological profiles.10 Current
computational studies carried out by our group on this series
aim to unveil the structural determinants of their receptor binding
affinity and selectivity and to identify the receptor binding
profile, which correlates with their therapeutic usefulness. These
studies, of which this work is a first report, involve building
homology models for receptors potentially involved in psychosis
and using them in a computational protocol that helps to unveil
the ligand-receptor interaction mechanisms.
Computational methods can be used in this context for
obtaining structural models of the ligand-receptor complexes,
but for a large series of compounds (as in this case), the visual
inspection of the complex structures is not helpful. Instead, 3D-
QSAR methodologies can use these receptor-docked ligand
structures to obtain models that extract from them the most
relevant information. This use of ligand-receptor complexes
asastartingpointfor3D-QSARhasbeenpreviouslyreported.11-18
It is well-known that one of the main drawbacks of 3D-QSAR
methods is the difficulty of obtaining suitable ligand alignments.
In this study in which a receptor structure has been built ad
hoc, the ligands can be docked into the receptor binding site,
and the resulting conformations can be used as aligned structures
for the 3D-QSAR analysis. The use of the receptor-docked
structures has the advantage of providing a more realistic
Drugs active at 5-HT2 receptors are used in the treatment of
depression, mania, anxiety, and schizophrenia. The blockade
of these serotonin receptors has been postulated to play a critical
role in the action of a new generation of antipsychotic drugs,
characterized by producing less extrapyramidal side effects
(EPSa), usually named atypical antipsychotic drugs (AAPD).
The involvement of 5-HT2 receptors in the pharmacological
profile of atypical antipsychotics is supported by a large number
of biological, pharmacological, and clinical studies.1-3 Unlike
classical antipsychotics like haloperidol, which mainly block
type 2 dopamine receptors (D2), clozapine and other AAPD are
relatively more potent at blocking serotonin-2A (5-HT2A) than
dopamine D2 receptors. This finding gave rise to the serotonin-
dopamine hypothesis, suggesting that for AAPD the blockade
of presynaptic 5-HT2A receptors is the predominant mechanism
in the nigrostriatal, mesocortical, and tuberoinfundibular dopam-
inergic pathways, where they increase dopamine release. This
effect will counteract the drugs action at D2 receptors, thus
decreasing the incidence of adverse EPS, cognitive deficits,
hyperprolactemia, and/or negative symptoms. In contrast, D2
blockade would prevail over 5-HT2A antagonism in the dopam-
inergic mesolimbic pathway, resulting in the mitigation of the
positive symptoms of psychosis.
Meltzer and co-workers4,5 suggested that the ratio between
the pKi of antipsychotic agents at the 5-HT2A and D2 receptors
* To whom correspondence should be addressed. Tel.: (+34) 933 160
512. Fax: (+34) 933 160 550. E-mail: manuel.pastor@upf.edu.
† Universitat Pompeu Fabra.
‡ Department of Pharmacology, Universidade de Santiago de Compostela.
§ Department of Organic Chemistry, Universidade de Santiago de
Compostela.
a Abbreviations: 5-HT, serotonin; AAPD, atypical antipsychotic drugs;
EPD, extrapyramidal side effects; D2, dopamine receptors of type 2; MI,
Meltzer’s index; 3D-QSAR, three-dimensional quantitative structure-
activity relationships; GPCR, G-protein coupling receptors; Rh, bovine
rhodopsin; MD, molecular dynamics; SRD/FFD, smart region definition/
fractional factorial design; PLS, partial least squares; LV, latent variables;
LOO, leave-one-out; CoMFA, comparative field analysis; BR, bacterior-
hodopsin; EM, energy minimization; MIF, molecular interaction fields.
10.1021/jm070277a CCC: $37.00 © 2007 American Chemical Society
Published on Web 06/19/2007

Studies on Conformationally Constrained Butyrophenones
Journal of Medicinal Chemistry, 2007, Vol. 50, No. 14 3243
representation of the ligands bioactive conformations and
normally does not introduce any difficulty in the methodology.
Unfortunately, in this particular series, these efforts led to
obtaining not one, but a set of candidate structures for every
compound, from which it was not obvious how to pick a single
representative structure. Most of the criteria that are used in
these circumstances to make a choice seem to us arbitrary or
dangerous, thus, we decided to build the 3D-QSAR models
using a balanced set of structures representing the different
species that bind the receptor. The procedure is unusual but
not entirely new. A similar approach has been reported in the
past in a different context19 with good results. Moreover, the
peculiar characteristics of the series studied offers an excellent
opportunity to validate the usefulness of the proposed approach
by interpreting the results using the structure of the receptor
and by comparing the results with those obtained using more
conservative approaches.7
the tetrahydrobenzothiophenylacetic acid 21 in 50% overall
yield. This was condensed with 4-(6-fluorobenzisoxazol-3-yl)-
piperidine in the presence of DCC and HOBt to give the
corresponding amide 22. This compound was protected by
ketalization with ethylene glycol and p-toluenesulfonic acid (p-
TsOH), the amide group was reduced to amine with LiAlH4,
and the ethylene ketal was cleaved in acidic medium to afford
the amine 24 in 54% yield from the acid 21. Further details
about the experimental procedures and data about the novel
compounds are available as Supporting Information.
Building of 5-HT2A Model. At present, the best starting
material for the homology building of G-protein coupling
receptor (GPCR) models is the crystal structure of bovine
rhodopsin (Rh),26 even if the low-sequence homology (∼20%)
between the 5-HT2 receptors and the Rh makes difficult the
application of standard homology modeling methods. In addi-
tion, the availability of only one experimental template might
introduce an undesirable bias in the modeling, making the
modeled structure “too similar” to the template. In this work,
we attempted to reach a certain compromise in-between making
use of the Rh template and the introduction of other sources of
structural information. The model was built introducing in the
MODELLER suite of programs27 a set of structural templates,
including the structure of Rh as well as other relevant structures.
The program yielded a consensus solution consisting of a set
of 25 candidate structures from which a single structure was
selected (Figure 1). A detailed description of the procedure used
for building and validating this model can be found in the
Experimental Section.
Results and Discussion
Synthesis. We will start by reporting the synthesis of seven
novel compounds added to the series subject of this study (Table
1; Chart 1). Pyridazinone derivatives 3, 5, and 6 were
synthesized as shown in Scheme 1: compounds 1 and 2 were
prepared by the reaction of 3-benzoylpropionic acid with
formaldehyde and heterocyclic amines under Mannich condi-
tions. The synthesis of the aminomethylpyridazinones 3 and 4
was readily accomplished in high yields by refluxing the
required 3-benzoyl-4-aminobutyric acid with hydrazine hydrate
in ethanol. Compounds 5 and 6 were obtained by alkylation of
the piperazine derivative 4 with 5-bromo-1-(p-fluorophenyl)-
1-pentanone,20 for compound 6, or 4-chloro-1,1-ethylenedioxy-
1-(p-fluorophenyl)butane21 followed by acid hydrolysis of the
ketal, for compound 5.
Compounds 9 and 10 were prepared (Scheme 2) from 3-(p-
toluenesulfonyloxymethyl)-1,2,3,4-tetrahydronaphthalen-1-
one 722 or 6,7-dimethoxy-3-(p-toluenesulfonyloxymethyl)-
1,2,3,4-tetrahydronaphthalen-1-one 8,23 respectively, via nucleo-
philic displacement of the tosylate with 1,3,8-triazaspiro[4.5]-
decan-4-one24 (for 9) or 1,3,8-triazaspiro[4.5]decane-2,4-dione25
(for 10) in N,N-dimethylformamide (DMF) and provided, after
bulb to bulb distillation of DMF and chromatography, the target
compounds 9 and 10 in moderate yields.
For the preparation of compound 16 (Scheme 3) 1-(2-
hydroxyethyl)piperazine was used as starting material. BOC
protection of the secondary amine followed by Swern oxidation
of the hydroxy group gave the unstable aldehyde 12 in 45%
yield. Aldol condensation with 4,5,6,7-tetrahydrobenzofuran-
4-one in the presence of LDA-HMPA and subsequent Pd-C
catalyzed hydrogenation afforded the piperazine derivative 14
in 46% yield. BOC-deprotection with TFA and finally direct
coupling with 4-fluorobenzoic acid by means of carboxylate
activation by 1,3-dicyclohexylcarbodiimide (DCC) in the pres-
ence of 1-hydroxybenzotriazole (HOBt) led to the target amine
16 in 70% yield (two steps).
The preparation of compound 24 (Scheme 4) was carried out
following a similar synthetic strategy reported for compound
69;7 Friedel-Crafts acylation of 2-pentylthiophene with succinic
anhydride in the presence of AlCl3, subsequent Clemmensen
reduction of the ketonic group of the ketoacid 17, and cyclization
with a 20:1 mixture of trifluoroacetic acid (TFA) and trifluo-
roacetic anhydride (TFAA) afforded the pentylbenzothiophene
19 in 40% yield (three steps). Alkylation of 19 with lithium
diisopropylamide at -78 °C, followed by quenching with ethyl
bromoacetate, gave ethyl ester 20 that was then hydrolyzed to
In the receptor model, the location of the binding pocket is
hinted by the position of the Asp3.32, which is known to
establish a charge-reinforced hydrogen bond with the protonated
nitrogen present in aminergic ligands. Before using this structure
for docking simulations, the binding pocket must be widened
and the side chains of the residues located there must be
reoriented. This operation was done by carrying out a prelimi-
nary docking of the 5-HT2A antagonist/inverse agonist ket-
anserin28,29 (see Chart 3) and extensive molecular dynamic (MD)
simulations using a stepwise strategy (as described in the
Experimental Section) until the observed ligand-receptor
interactions were consistent with experimental data. The reasons
for choosing ketanserin were twofold: on the one hand, it is a
well-known ligand of 5-HT2A for which many results of
mutagenesis experiments are available, thus allowing a valida-
tion of the proposed binding site using experimental data; on
the other hand, ketanserin exhibits a rather large chemical
similarity with the compounds of our series in terms of shape
and in terms of presence of similar functional groups at
equivalent positions.
The best 5-HT2A-ketanserin complex obtained is shown in
Figures 1 and 2. The binding site is located approximately
parallel to the helices, slightly tilted with respect to the axis of
the helices bundle and nearly parallel to the helix 3. The ligand
makes polar contacts mainly with residues in helices 3 and 7,
as well as with some residues of the extracellular loop 2.
Hydrophobic contacts are observed with residues of helices 5
and 6. The amino acids making the most relevant interactions
with ketanserin are Asp3.32, Ser3.36, Trp3.28, Phe6.51, Tyr7.43,
and Ser161. Table 2 lists the percentages of hydrogen bonding
between these residues and ketanserin, obtained by monitoring
a 1ns MD simulation (see the Experimental Section for details).
The comparison of the ketanserin complex with the results
of site-directed mutagenesis experiments shows a good agree-

3244 Journal of Medicinal Chemistry, 2007, Vol. 50, No. 14
Dezi et al.
Table 1. Chemical Structure and Binding Affinity Measured for the 5-HT2A Receptor
alkanone
fragmenta
amine
pKi 5-HT2A
pKi 5-HT2A
human
cmpd
code
fragmentb
rat
ref
3
5
6
9
10
16
24
25
QF0225B
QF0226B
QF0227B
QF0109B
QF0140B
QF0712B
QF0620B
QF0102B
QF0107B
(R)-QF0108B
(S)-QF0108B
QF0120V
(R)-QF0124B
(S)-QF0124B
QF0127B
(R)-QF0128B
(S)-QF0128B
QF0128V
QF0129V
QF0130V
QF0140V
(R)-QF0144B
(S)-QF0144B
QF0147V
QF0148V
QF0149V
QF0150V
QF0301B
QF0303B
QF0304B
QF0307B
QF0308B
QF0311B
QF0313B
QF0315B
QF0402B
QF0407B
QF0408B
QF0409B
QF0410B
QF0501B
QF0502B
QF0503B
QF0504B
QF0505B
QF0506B
QF0510B
QF0601B
QF0602B
QF0603B
QF0604B
QF0605B
QF0606B
QF0607B
QF0608B
QF0609B
QF0610B
QF0701B
QF0702B
QF0703B
QF0704B
QF0901B
QF0902B
QF1003B
QF1004B
QF1005B
QF1006B
QF1007B
QF1008B
QF1010B
QF1011B
VI
VI
VI
I
b
a
h
c
g
i
b
a
f
b
b
d
e
e
f
b
b
b
c
g
d
e
e
f
b
c
g
d
e
a
e
d
e
e
a
d
d
e
b
a
e
a
d
f
e
a
b
e
a
d
f
e
a
d
f
7.24 ( 0.02
6.56 ( 0.02
6.00 ( 0.01
<5
IV
<5
XVII
XVI
I
I
I
I
II
III
III
III
III
III
II
II
II
V
IV
5.70 ( 0.20
7.68 ( 0.13
7.75 ( 0.60
6.01 ( 0.20
9.33 ( 0.12
6.94 ( 0.31
38
23
23
23
39
23
23
23
23
23
39
39
39
39
23
23
39
39
39
39
40
41
42
41
8
26
(R)-27
(S)-27
28
(R)-29
(S)-29
30
(R)-31
(S)-31
32
33
34
35
(R)-36
(S)-36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
8.20 ( 0.37
7.53 ( 0.32
6.59 ( 0.29
7.79 ( 0.46
7.95 ( 0.41
6.16 ( 0.21
7.58 ( 0.19
8.83 ( 0.49
8.29 ( 0.25
<5
6.15 ( 0.13
6.65 ( 0.07
8.25 ( 0.23
7.36 ( 0.30
IV
V
V
V
6.33 ( 0.11
8.23 ( 0.14
<5
V
5.98 ( 0.15
7.39 ( 0.70
8.11 ( 0.80
7.14 ( 0.70
8.60 ( 0.80
6.71 ( 0.14
7.88 ( 0.70
8.42 ( 0.30
7.27 ( 0.10
7.04 ( 0.70
6.55 ( 0.80
8.06 ( 0.10
8.37 ( 0.80
6.04 ( 0.60
7.58 ( 0.14
7.34 ( 0.10
6.05 ( 0.21
<5
8.15 ( 0.13
7.37 ( 0.15
8.76 ( 0.20
8.84 ( 0.17
6.54 ( 0.13
6.84 ( 0.12
6.76 ( 0.11
7.24 ( 0.12
6.97 ( 0.13
5.82 ( 0.06
5.75 ( 0.10
7.93 ( 0.05
8.56 ( 0.20
6.72 ( 0.17
7.71 ( 0.90
8.97 ( 0.09
6.45 ( 0.10
7.95 ( 0.09
8.17 ( 0.18
7.29 ( 0.20
7.97 ( 0.03
6.02 ( 0.03
<5
VIII
VIII
VIII
VII
VII
IX
X
X
XXIII
XI
XI
41
41
42
43
43
43
43
43
7
7
7
8
7
7
7
7
7
7
8
7
7
XI
XI
XII
XII
XII
XII
XIII
XIII
XIII
XIV
XIV
XIV
XIV
XV
XV
XV
XV
XV
XIV
XVII
XVII
XVII
XVII
XVIII
XVIII
XIX
XIX
XIX
XIX
XIX
XIX
XIX
XIX
7
8
7
7
8
8
b
b
d
e
b
f
e
b
e
b
a
d
j
9.14 ( 0.11
8.26 ( 0.12
8
8
7
7
46
46
46
46
46
46
46
46
7.66 ( 0.05
5.08 ( 0.10
<5
f
k
h
83
5.90 ( 0.04

Studies on Conformationally Constrained Butyrophenones
Journal of Medicinal Chemistry, 2007, Vol. 50, No. 14 3245
Table 1 (Continued)
alkanone
amine
pKi 5-HT2A
pKi 5-HT2A
human
cmpd
code
fragmenta
fragmentb
rat
ref
84
85
86
87
88
89
90
91
92
QF2001B
QF2002B
QF2003B
QF2004B
QF2006B
QF2014B
QF3000B
QF3004B
QF3008B
XX
XX
XX
XX
l
m
e
b
d
m
d
e
<5
44
44
44
44
44
44
45
45
45
<5
8.04 ( 0.88
8.8 ( 0.88
6.20 ( 0.01
<5
<5
7.58 ( 0.16
8.86 ( 0.07
8.93 ( 0.15
XX
XXI
XXII
XXII
XXII
b
a See Chart 1. b See Chart 2.
Chart 1. Alkanone Fragments Present in Compounds of the Series
Scheme 1a
a Reagents and conditions: (i) 37% CH2O, HNRR, rt; (ii) N2H4‚H2O, EtOH, reflux; (iii) for 5, (a) 4-chloro-1,1-ethylenedioxy-1-(p-fluorophenyl)butane,
Na2CO3, KI, DMF, reflux; (b) 2 N HCl, reflux, and then rt; for 6, 5-bromo-1-(p-fluorophenyl)-1-pentanone, Na2CO3, KI, DMF, reflux.
Scheme 2a
ment, as the most important results30-33 (Table 3) can be easily
justified by the proximity of the ligand to the residues.
The observation of the complex (Figure 2) shows that the
side chain of Asp3.32 and the charged N of ketanserin interact
making a charge-reinforced hydrogen bond. The phenyl ring
of the Phe6.51 side chain adopts a conformation parallel to the
piperidine ring of ketanserin, where the aliphatic carbons directly
a Reagents and conditions: (i) 1,3,8-triazaspiro[4.5]decane-4-one (for 9)
or 1,3,8-triazaspiro[4.5]decane-2,4-dione (for 10), Et3N, IK, DMF, 85 °C.
linked to the charged ketanserin nitrogen (holding a positive
partial charge) can make favorable electrostatic interactions with

3246 Journal of Medicinal Chemistry, 2007, Vol. 50, No. 14
Dezi et al.
Scheme 3a
a Reagents and conditions: (i) [(CH3)3COCO]2O, CHCl3, rt; (ii) DMSO, oxalyl chloride, Et3N, CH2Cl2, -60 °C; (iii) n-BuLi, diisopropylamine, HMPA,
THF, -78 °C; (iv) H2, Pd-C, THF, rt; (v) TFA, CH2Cl2, rt; (vi) DCC, HOBt, CH2Cl2, rt.
Scheme 4a
a Reagents and conditions: (i) succinic anhydride, AlCl3, CH2Cl2, rt; (ii) Zn, HgCl2, HCl, H2O, toluene, AcOH, reflux; (iii) TFA, (CF3CO)2O, rt; (iv)
n-BuLi, diisopropylamine, BrCH2CO2Et, THF, -70 °C; (v) LiOH, THF, H2O, rt; (vi) 4-(6-fluorobenzisoxazol-3-yl)piperidine, DCC, HOBt, CH2Cl2, rt; (vii)
ethylene glycol, p-TsOH, toluene, reflux; (viii) (a) LiAlH4, Et2O, rt, (b) HCl, Et2O, reflux.
seem to participate in two additional polar interactions; one
between the remaining carbonyl oxygen and the hydroxyl of
Ser161, the other between one nitrogen and the Tyr7.43 phenolic
hydroxyl group. This last interaction seems to stabilize the
position of Tyr7.43 in an orientation in which it can establish
an additional hydrogen bond with Ser2.61.
With respect to the p-fluorobenzoyl moiety, the carbonyl
oxygen makes a hydrogen bond with the side chain of Ser3.36.
Deeper inside the binding pocket, the p-fluorobenzyl is im-
mersed in a hydrophobic pocket involving residues Val5.39 and
Leu163. The size of this hydrophobic pocket is delimited by
Phe6.52, a residue that is part of the so-called hydrophobic
toggle switch, a set of three residues (Phe6.51, Phe6.52, and
Trp6.48), which were known to participate in a coordinated
conformational change, potentially important for the activation
process.34-37 Apart from this role, the conformational changes
Figure 1. Best candidate structure of the 5-HT2A receptor model
of these three residues are very important for determining the
produced by MODELLER and used for subsequent analysis. To
size of the aforementioned hydrophobic pocket, which in turn
conditions the orientation of the p-fluorobenzoyl moiety. To
conclude, the fluorine substituent of the aromatic ring is located
close to the Ser5.43 and Ser5.46, which are residues that have
been postulated to be relevant for the serotonin binding.38,39
illustrate the location of the binding site, the graphic shows a molecule
of ketanserin docked according to the method explained in the text.
the phenyl ring of the Phe6.51 side chain. The result is a kind
of electrostatic “sandwich”, in which the ketanserin makes
favorable interactions with Asp3.32 on one side and with
Phe6.51 on the opposite side. The other oxygen atom of the
Asp3.32 side chain is at hydrogen-bonding distance of one of
the carbonyl oxygen atoms of the ketanserin benzouracil moiety,
which seems to be also involved in an additional hydrogen bond
with the indolic nitrogen of Trp3.28. The benzouracil moiety
Docking. The series subject of this study (Table 1) contains
previously published conformationally constrained butyro-
phenones,7,8,23,40-48 as well as the seven new, unpublished
compounds (3, 5, 6, 9, 10, 16, and 24), the preparation of which
was described above. The affinities of these 76 compounds for
the 5-HT2A receptor have been determined experimentally and

Studies on Conformationally Constrained Butyrophenones
Journal of Medicinal Chemistry, 2007, Vol. 50, No. 14 3247
Chart 2. NRR of Structures in Chart 1
nitrogen atoms were linked to aliphatic carbons, the one linked
to the alkanone fragment was chosen for consistency with other
members of the series, even if a tautomeric equilibrium between
species protonated in both positions can be expected. At the
end, 152 different structures were obtained and submitted to
the docking protocol (see Experimental Section for details).
The method was able to produce docked structures with a
fitting score (GOLDscore)51 above a reasonable prefixed cutoff
(35) for most of the compounds in the series (142 out of 152,
93%). Figure 3 shows, superimposed, the docked structures for
the R and for the S series. In the figure, the observation of both
series clearly shows both enantiomers bind in approximately
the same position, but some orientations are more frequent in
one series than in the other. For example, in the R series, many
compounds place the carbonyl group on the lower right side,
while in the S series, the fluorine substituent of the p-
fluorobenzoyl moiety is much more frequent on the upper right
side.
The analysis of the docking results revealed that most
structures can bind in two alternative orientations, one with the
alkanone group oriented toward the inner part of the binding
pocket and another orienting this group in the opposite direction.
In both orientations, the charged nitrogen can bind the Asp3.32
and the polar groups can make approximately equivalent
contacts. The presence of these two alternative binding modes
was also identified in a previous work7 in which a much earlier
version of the present series, containing only 25 compounds,
was docked into the binding site of a 5-HT2A receptor model,
even if the receptor modeling and the ligand docking followed
a completely different methodology. The authors of this work
justified this duplicity due to the presence of the same
pharmacophoric groups at both sides of the protonated amino
group and suggested that this finding “hints the possibility of
multiple binding modes”.
Chart 3. Structure of Ketanserin
The preference of the structures for binding in a certain
orientation follows no obvious rules. In some structures (like
R-38, R-61, or S-62), both orientations were present among the
three best solutions provided by GOLD, thus showing no strong
preference for any orientation. Other compounds seem to bind
only in one orientation, but the introduction of a small structural
change induces a preference for the other alternative. This is
the case for R-72 and R-75, which were inserted in opposite
directions even if their structures differ only in the substitution
of the furan ring by a thiophene (Figure 3). For some
compounds, like 75, the R- and S-enantiomers show a preference
for binding in opposite orientations (Figure 3). A particularly
interesting case are compounds 69 and 24. In the docking
simulation, both 69 enantiomers are found to bind with the
alkanone oriented toward the inner part of the binding pocket.
Compound 24, a derivative of 69 prepared ad hoc to prevent
the original binding mode (by adding a bulky pentane substituent
to the alkanone), was observed to dock in the opposite
orientation (Figure 3). The experimentally measured binding
affinity of 24 was only slightly lower than the binding affinity
of its parent (8.56 for 69 and 7.68 for 24), thus suggesting that
for compound 69 both binding orientations would also be
accessible and contribute to the experimental binding affinity,
something that probably applies to most of the compounds in
this series.
were listed in Table 1, expressed as pKi values. For QSAR
analysis purposes, inactive compounds were assigned an
arbitrary pKi value of 5. In this table, the values range from 5
to 9.14, covering more than four logarithmic units, and the
experimental errors (expressed as standard error of the mean)
fall typically under 0.5 logarithmic units.
Most of the compounds of this series share relevant structural
features with ketanserin, like the presence of a piperidine ring,
the linear shape with aromatic rings at both ends, and hydrogen-
bond acceptor and donor groups at an intermediate position.
This similarity justifies the use of the above-described ket-
anserin-conditioned binding pocket to carry out docking simula-
tions in this series. The objective was to obtain complexes that
highlight the more relevant features that determine the binding
affinity, but without carrying out extensive MD simulations.
With this aim, every compound of the series was docked into
the modeled receptor binding site using program GOLD 2.2.49
Unlike ketanserin, all the compounds in this series contain a
chiral center and, therefore, either their R or the S enantiomers
can be used for the docking. Both enantiomers were built in
2D and converted to 3D structures using CORINA 2.4.50 The
structures were all modeled with a formal positive charge
centered on the piperidine ring nitrogen. For the compounds
with a piperazine ring, the nitrogen linked to an aliphatic
substituent was preferred for protonation, and when both
In summary, these results seem to confirm the coexistence
of ligand molecules bound in two alternative orientations, the
proportion of which might be different for different structures,
considering also as such both enantiomeric forms of the same
compound. With all probability, none of the experimentally

3248 Journal of Medicinal Chemistry, 2007, Vol. 50, No. 14
Dezi et al.
Figure 2. Binding site of the 5-HT2A receptor complexed with a molecule of ketanserin. Only residues establishing relevant interactions are
depicted. The dashed lines represent hydrogen bonds and polar interactions described in the text.
Table 2. Percentage of Structures with a Geometry Compatible with a
the information contained in many complexes into a few
interpretable graphic plots. Moreover, the procedure would be
Hydrogen Bond between the Ketanserin and the Listed Residuesa
% hydrogen
bond
less influenced by the aforementioned small “random” docking
errors than the individual complex inspection, because 3D-
QSAR models highlight only correlations that are present in
several compounds.
residue
Asp3.32
Ser3.36
Ser161
Trp3.28
Tyr7.43
100
21
27
76
18
3D-QSAR. The output structures obtained from the docking
are an excellent material for building a 3D-QSAR model,
because they provide a more realistic representation of the ligand
bioactive conformation than any alignment obtained by structural
superimposition. When proceeding this way it must be remem-
bered that the spatial framework for the interpretation of the
resulting models is not the cluster of superimposed compounds,
but the structure of the receptor binding site, because this
structure was used to place the ligands within a consistent
reference frame.
a As observed during the 1 ns MD simulation described in the text.
Table 3. Residues Involved in the Binding of Ketanserin, According to
Mutagenesis Experiments, and the Variation of Ki Associated to Each
Mutation
Ki
interactions observed
in complex
mutation
variation > (fold)
Trp6.48Ala
100
part of the aromatic cluster,
delimiting binding site
charge-reinforced hydrogen
bond with charged piperidine
nitrogen atom
near the p-flurobenzyl moiety
hydrogen bonded to the
benzouracil moiety
In the present study, the biological activity values were
obtained from binding affinity measurements carried out (in the
vast majority of cases) using the compounds racemic mixture,
thus making available only one affinity value for the R and S
enantiomers present in the mixture. From a biological point of
view, both enantiomers are not equivalent and their binding
affinity might also be different. Indeed, in the few examples in
which the activity has been measured separately for the R and
S enantiomers, the values obtained were slightly different (see
Table 1).
Asp3.32Asn
75
Phe6.52Tyr
Tyr7.43Ala
73
20
Phe6.51Ala
12
Van der Waals contact with
piperidine ring
Asp2.50Asn
Trp1.34Ala
10
10
determined binding affinity constants listed in Table 1 represent
the affinity of a single binding mode, and in all instances, they
are a weighted average of the affinity values of the different
species present in the solution (corresponding to the two
alternative binding modes in different proportions for the two
enantiomers).
Apart from these mainly descriptive observations, our main
interest in the study was to identify in the complex structures
interaction patterns that can be correlated with the activity (e.g.,
polar interactions present in the most active compounds and
absent in the less active), but the preliminary inspection of the
complexes did not reveal any evident relationship. The direct
analysis of all these complexes would be cumbersome and prone
to introduce subjectivity in the observations. Another problem
associated with the use of a “rigid-receptor” docking method
that does not take into account the flexibility of the binding
site (GOLD), is that the complex structures often contain small
errors and inaccuracies, the observation of which can produce
misleading interpretations.
Obviously, the lack of separate affinity values for both
enantiomers is a problem for the QSAR modeling. A potential
solution used in a previous work7 is to select one of the
enantiomers and carry out the whole 3D-QSAR procedure using
only this structure. The alternative solution proposed here is to
include both enantiomers in the model as separate structures,
but associate them to the single binding affinity value measured
for the racemic mixture. By proceeding so, the binding affinity
used corresponds to neither of the individual enantiomers and
can be considered an average of both enantiomers true values.
The observation of the differences in binding affinity between
R and S enantiomers in the few examples reported in Table 1
shows that these differences are not large (0.67, 0.17, 1.25, 0.89)
but certainly not negligible. From a statistical point of view, it
is important to notice that the use of both values will introduce
“noise” into the model but not bias, because the true binding
affinity values should deviate from the estimate in opposite
directions. Such a model could not be expected to have a perfect
fitting, because part of the model lack-of-fit will be produced
by this “noise” and, therefore, the model r2 value will be lower
than the r2 obtained for a regular model. From the point of view
of the interpretation, a model obtained in this way could not be
For all these reasons, we decided to use 3D-QSAR methods.
The 3D-QSAR models can highlight the structural differences
that correlate with binding affinity differences, thus summarizing

Studies on Conformationally Constrained Butyrophenones
Journal of Medicinal Chemistry, 2007, Vol. 50, No. 14 3249
Figure 3. From top to bottom: superimposition of all the docking solutions for the R enantiomers (left) and the S enantiomers (right); side by side
view of the docking solution for the R (left) and S (right) enantiomers obtained for compounds 72, 75, 87, 69, and 24. In the case of compound 72,
both enantiomers are inserted in the same orientation, while for 75, the R and S enantiomers are inserted in opposite directions. Notice that the only
difference between 72 and 75 is the substitution of the oxygen by a sulfur in the heteroaromatic ring of the alkanone moiety. In compound 87, both
enantiomers orient the acceptor and donor groups in opposite directions. Both enantiomers of compound 69 are inserted in the same direction, but
the incorporation of a pentane substituent to produce compound 24 induces a change to the alternative orientation, with no relevant change in the
binding affinity.
Table 4. Statistical Parameters of the 3D-QSAR Models Obtained
expected to explain “fine effects”, because no structural differ-
model LV objects var.
r2
q2
SDEPLOO q2
q2
2RG
ence between the R and S complexes could be correlated with
differences in their binding affinities (the same binding affinity
is used for representing both).
LOO
5RG
M1
M2
3
3
142
426
1441 0.90 0.78
1579 0.84 0.81
0.55
0.52
0.77 0.73
0.80 0.79
Accordingly, the first 3D-QSAR model (M1) was built
including both the R and S enantiomers for every compound.
The enantiomer structures were represented by the docking
solution with the highest fitting scores. Their binding affinities
were represented by the experimental measurements listed in
Table 1 corresponding, for the vast majority of the cases, to
the affinity of the racemic mixture, but for a few exceptions in
which separate values for either enantiomers were available.
The GRID/GOLPE methodology14 was used. Every structure
was first analyzed using GRID52 to obtain molecular interaction
fields (MIF) with three probes, DRY, O, and N1, representing,
respectively, a hydrophobic, a hydrogen bond acceptor, and a
hydrogen bond donor group. These MIF were imported into
the program GOLPE53 together with the affinity values. The
variables were pretreated and scaled before using the partial
least-squares (PLS) regression analysis and the SRD/FFD54
variable selection method. At the end, the PLS model obtained
with three latent variables (LV) showed good statistical quality
(r2 ) 0.90, q2LOO ) 0.78, see Table 4 and Figure 4). To further
validate the model and to estimate its true predictive ability,
we have split the series into a training and a test set. The objects
assigned to the test set (15 structures, approximately 10% of
the whole set) were selected applying the Most Descriptive
Compounds algorithm,55 as it was implemented in the GOLPE
software,53 on the space of the first three principal components
extracted from the Principal Component Analysis of the whole
series. Then the whole 3D-QSAR modeling procedure was
carried out as described above (PLS modeling and SRD/FFD
variables selection) using only the compounds of the training
set, while the compounds in the test set were used only for the
predictions. In this case, the Standard Deviation of Error of
Prediction (SDEP) obtained for the test set was of 0.53, and
the external r2 between experimental and predicted values was
of 0.80. These values are perfectly comparable with the values
obtained by standard cross-validation methods and further
confirm the good predictive ability of M1.
Even if the model described above seems suitable for our
purposes we decided to go one step further and apply to the
problem of the multiple binding modes the same solution applied
to address the presence of two enantiomers. In the same way
that two enantiomers were used to characterize each compound,
it is possible to use more than one docking solution to
characterize each enantiomer. For those compounds in which
the three docking solutions are rather similar the use of multiple
structures will have very little effect (the median of the RMSD

3250 Journal of Medicinal Chemistry, 2007, Vol. 50, No. 14
Dezi et al.
Figure 4. Scatterplots representing the experimental versus calculated pKi values for the models M1 (left) and M2 (right).
was 0.62 for the solutions 1 and 2, 0.87 for the solutions 1-3,
and 0.9 for the solutions 2 and 3), but in some cases, these
solutions represent alternative binding modes. In our opinion,
the choice of the first solution to represent the docking structure
was arbitrary in these cases, much in the same sense that the
choice of a certain enantiomer would be arbitrary, because the
limited accuracy of the scoring functions question the reliability
of the solution ranking provided by the docking programs.
Moreover, most compounds probably interact with the receptors
in both orientations and in proportions that are difficult to
estimate. Therefore, the contemporary use of all these solutions
will provide a much more realistic picture of their true binding
ability. As in the previous case, we must be ready to accept a
small decrease in the model fitting, lower r2 values, and a less
detailed description of fine effects, but hopefully, the models
will be more robust and have better predictive ability, as it has
been reported in other similar situations.19
the M1 test set. As in the previous case, the structures in the
test set were left out of the analysis, which was repeated from
the beginning using only the compounds in the training set. To
obtain comparable results with M1, the predictions obtained for
the three docking solutions used to characterize each compound
in the test set were averaged to provide a single estimate. The
results of such predictions were good (SDEP ) 0.52, r2 ) 0.81)
and slightly better than those obtained with M1, thus confirming
the improved predictive ability of M2 over M1, as suspected.
The comparison of the experimental versus calculated plots
for M1 and M2 represented in Figure 4 shows that in M2 the
objects exhibit a larger spread, and as it was expected, not all
structures fit well into the model. The coefficient plots obtained
for M1 and M2 look rather similar and contain essentially the
same information. The main difference observed is that the plots
for M2 are “cleaner”, devoid of the small regions produced by
the effect of single compounds that usually have no general
relevance. All these results suggest that the model M2, obtained
with multiple structures, can be a better alternative than M1,
but the similarity observed between their coefficient plots
indicate that the results of the M1 and M2 interpretation would
be qualitatively similar.
Before beginning the interpretation of the M2 coefficient
plots, we must remind the reader that the MIF assign negative
energy values to favorable probe-ligand interactions and
positive values to unfavorable (repulsive) probe-ligand interac-
tions. Positive field values represent mainly the molecular shape,
while negative values represent regions where the ligand can
make energetically favorable interactions with the binding site.
The PLS coefficient plots shown in Figure 5 contain both
positive and negative values for the three probes used. Positive
(yellow) coefficients might correspond to binding site regions
where the presence of negative fields have an inverse correlation
with activity or to regions where the presence of the molecular
bulk (positive field) correlates directly with the activity.
Conversely, negative (cyan) coefficients might represent regions
where the presence of negative fields correlate directly with the
activity or regions where the ligand positive fields correlate
inversely with the activity. These considerations add complexity
to the interpretation and require a careful analysis of the
coefficients, the field values produced for different compounds,
and the activity contribution plots. It should also be stressed
that the coefficient positions will make reference to the binding
site residues (even if the binding site has not been used directly
for obtaining the model), because the ligand alignment was
based on the docking solutions. Therefore, the reference
framework of the whole analysis is the structure of the binding
site instead of the superimposed ligand structures.
A second 3D-QSAR model (M2) was built using a similar
methodology, but in this case, including the first three solutions
provided by GOLD to characterize every enantiomer (six
structures per compound) amounted to a total of 426 structures.
At the end of the modeling procedure (see Experimental Section
for details), we obtained a PLS model with three LV of
remarkable quality (r2 ) 0.84 and a q2
Figure 4).
) 0.81, Table 4,
LOO
From a statistical point of view, the r2 of M2 is lower than
M1, as can be expected for the aforementioned reasons, but
only slightly (from 0.90 to 0.84), and the difference is small if
we consider that the number of objects included in the model
has been increased from 142 to 426. With respect to the values
of q2, they should be interpreted with care. In this particular
series, the application of leave-one-out (LOO) method can
produce overoptimistic results, because the removal of a single
structure will not completely remove the presence of the
compound from the reduced models. Therefore, it is not
surprising to obtain a higher q2LOO of 0.81 for M2 than the 0.78
value obtained for M1. A better estimation of the true predictive
ability of M1 and M2 can be obtained using more strict cross-
validation utilizing either two or five randomly assigned groups
(see Experimental Section). The results listed in Table 4 agree
with the previous results, indicating that both models have a
rather high predictive ability and that the incorporation of
multiple structures does not decrease the predictive ability of
the original model. To further confirm the predictive ability of
M2 and to compare it with M1, an external validation protocol
identical to the one described for M1 was carried out. In this
case, the test set included 45 compounds, corresponding to the
three best docking solutions obtained for the 15 structures of

Studies on Conformationally Constrained Butyrophenones
Journal of Medicinal Chemistry, 2007, Vol. 50, No. 14 3251
pocket. One is located at the bottom near the Ser5.43 and
presumably indicates that ligands able to make a favorable
interaction with this residue have higher binding constants. On
the other side of the receptor there is a much wider area
enclosing the side chain of Asn7.36 and close to other polar
side chains (Ser1.35, Ser2.61, Trp7.40). Even if the regions do
not overlap these residues, we must remember that the structure
represented here is only a static snapshot of the receptor, and
the above-mentioned residues could move their side chains to
establish interactions at these positions. This can also be the
case for an elongated region in the center of the pocket near to
the Ser3.36 and Trp6.48. In the conformation shown in Figure
5, the polar part of the Trp6.48 side chain could not overlap
this region, but this residue forms part of the toggle switch that
has been suggested to participate in a concerted conformational
change together with Phe6.51 and Phe6.52. During this con-
formational change the Trp6.48 rotates, orienting the polar
nitrogen toward the ligand and locating this atom precisely in
the region highlighted by the negative coefficients. With respect
to the positive coefficients, most of them describe binding site
regions that are overlapped by compounds with high binding
affinity. The two main positive regions are located on the
extremes of the binding site, indicating that the compounds with
the highest affinity are long and fill the whole binding pocket,
while shorter compounds usually show low affinity.
O Probe (Figure 5, Bottom). The negative regions observed
for this probe are smaller and less defined than those observed
for N1, probably because the series contains much fewer
hydrogen bond donor groups. The most important regions are
located near the oxygen atoms of Asp3.32 and have a straight-
forward interpretation. The negative regions located on the
central part of the binding pocket, near the hydrophobic side
chains of Ser161, Phe6.51, and Val7.39, seem to represent the
interactions of some compounds that contain a spiranic hydan-
toin ring (g in Chart 2), like 40 shown in Figure 5. In these
compounds, one of the hydrogen bond donor atoms of g
generates two regions for potential hydrogen bond interactions,
one near the side chain of Ser161 and the other oriented toward
the Val7.39, both of which are shown in the coefficient plots
even if only one of them actually represents a plausible
interaction. The positive regions are rather similar to the positive
regions observed for the N1 probe and can be interpreted in a
similar way.
Figure 5. Coefficient plots obtained in model M2 for different blocks
of variables representing (from top to bottom) the probes DRY, N1,
and O. Positive coefficient values are represented in yellow and negative
coefficient values are represented in cyan. Every coefficient plot
included the structure of a representative ligand (72 for DRY and N1
and 40 for O), as well as the structures of the more relevant residues,
as mentioned in the text. Even if the structure of the binding site was
not used for obtaining the 3D-QSAR model, the coefficients represented
should reflect ligand-receptor interactions, which are relevant for the
ligand binding affinity.
To summarize, the model stresses the importance of some
of the polar interaction already recognized in the description of
the ketanserin complex, like those with Asp3.32, Trp3.28,
Ser161, and Ser 3.36, but adding some original information:
(i) highlights some biologically relevant interactions with
hydrogen bond donor residues in both extremes of the pocket
(Ser5.43 on one side, Ser1.35 and/or Asn7.36 on the other side)
and a putative interaction with the indolic nitrogen of Trp6.48
when its side chain adopts an alternative conformation; (ii)
shows the importance of the hydrophobic interactions with
residues located at the bottom of the pocket (Leu163) as well
as in the center (Phe6.51, Val7.39); (iii) stresses the importance
of the ligand size, indicating that ligands with high affinity tend
to completely fill the cavity; and (iv) indicates that the few
compounds in the series with hydrogen bond donor groups
located in the alkanone moiety can make interactions that lead
to an increase in the binding affinity, probably with Ser161.
DRY Probe (Figure 5, Top). The highest negative coef-
ficients were located in two areas, one at the bottom of the
pocket near Leu163 and the other in the center near the side
chains of Val7.39 and Phe6.51. The values of the coefficients
indicate that the presence of hydrophobic groups in these
positions increases the binding affinity, which is consistent with
the hydrophobic nature of the neighbor residues. The main
positive region obtained for this probe is located near the main
chain of Trp6.48, indicating that the presence in this region of
hydrophobic groups is detrimental for the binding. A similar
interpretation can be given to a small positive region located
near to the indolic nitrogen of Trp3.28 or to the polar atoms of
the main chain near to the Gly160.
N1 Probe (Figure 5, Middle). Not surprisingly, most
negative regions overlap some of the residues able to act as
hydrogen bond donors. This is the case of a large region in the
center that extends between the oxygen atoms of Asp3.32 and
the indolic nitrogen of Trp3.28. Also in the center, there is a
smaller region near the side chain of Ser161. Apart from these,
there are two large negative regions in both extremes of the
These results are in agreement with others obtained in our
group in a much earlier work7 in which a CoMFA model56 was
obtained for a subset of 24 compounds of the present series. In
this work, the model was obtained using only one enantiomer,

3252 Journal of Medicinal Chemistry, 2007, Vol. 50, No. 14
Dezi et al.
the most conserved residue in each helix was given the number
50, so for every helix, the most conserved residue was indicated
with the number of the helix followed by 50. For residues belonging
to loop regions, the sequence number was used instead.
one binding orientation, and carrying out a classical ligand
superimposition based on a few pharmacophoric points. The
quality of the best model obtained was not better than that of
M1 or M2 (r2 ) 0.88 and q2 ) 0.64), and the coefficient plots
are far less detailed than the ones shown in Figure 5, confirming
that the methodology presented here leads to results more useful
than those obtained with classic methods. Remarkably, the
interpretation of the previous model shows some points of
coincidence with the interpretation of M2, as it also points out
the importance of the hydrophobic regions in the center and in
the extremes of the ligand, the importance of the hydrophilic
regions at both ends, and hints the importance of the right ligand
size.
GPCR Modeling. (i) Sequence Alignment and Prediction of
Secondary Structure and Transmembrane Helices. The sequence
of 58 serotonin receptors cloned so far from different species
(excluding the subtype 3, because they are not GPCR) were
retrieved from the Swiss-Prot database59 and aligned with the
ClustalX software,60,61 using the PAM250 matrix. “Gap open” and
“gap elongation” penalties of 10 and 0.05, respectively, were used
in the alignment. The resulting multiple sequence alignment was
realigned with the sequence of Rh (OPSD_BOVIN), introducing
secondary structure information derived from the crystal structure
(residues belonging to the transmembrane alpha helices) to avoid
gaps within the seven helical segments.
Conclusions
The alignment was then manually refined to ensure a perfect
alignment of the highly conserved residues of GPCR superfamily,
according to Baldwin et al.,62 and looking for a consensus between
the predictions of transmembrane segments provided by TMAP,63
Swiss-Prot (http://www.expasy.ch), and TMPRED (http://www.
ch.embnet.org), as well as by considering that Arg and Lys residues
are often present at the membrane boundaries.64 Alternative lengths
for the helices were considered by taking into account the
experimental length of the Rh helices, the secondary structure
prediction by JPRED,65 and the sequence conservation.
(ii) Structural Alignment, Disulfide Bond Assignment, and
Model Building. To enrich the structural information given by the
Rh template with additional information, other membrane protein
structures stored in the “Membrane Proteins of Known 3D-
Structure” database66 were also considered, but with the condition
of being proteins sharing the GPCR folding (seven transmembrane
helices) and displaying a minimum homology with subtype 2
serotonin receptors of 20%. Only bacteriorhodopsin (BR) satisfied
both requirements, even if it is not a GPCR itself and presents major
structural differences between GPCR and BR.34 For this reason,
the use of the BR structure was limited to individual helices and
helices pairs to avoid introducing biases due to the different folding
of BR. Therefore, BR helices were separated into pairs and
superimposed to Rh using program STAMP.67 In addition and for
the same reasons, ideal alpha helices built de novo using the
sequence of human 5-HT2A were also superimposed to the Rh
structure and added to the templates library.
Regarding the modeling of the loops, suitable templates were
searched in the ArchDB loops database68 to enrich the structural
information provided by Rho. Template candidates must link a
couple of alpha helices, should have a similar number of residues
forming the loop ((2 residues), and the distance (between alpha
helices extremes) that braces the loops should be similar to those
found in rhodopsin ((5 Å). The modeling of the EL2 loop was
largely simplified by the presence of two important structural
constraints. First, the Cys163 and the Cys3.25 form a disulfide bond,
a feature highly common among GPCR receptors, and the secondary
structure prediction analysis suggested that a large part of the loop
adopts a â-sheet conformation. For the rest of the loops, suitable
templates were found and their structures were superimposed with
STAMP to the crystal structure of Rh to reach a consensus between
all the structures used.
The work presented here is a good example of the practical
limitations of well-established computational methods. The
analysis of the docked structures was not successful in uncover-
ing relevant relationships between the ligand structures and their
binding affinities, in part due to the large number of complexes
and in part due to the duality of the binding orientations
obtained. Then the application of 3D-QSAR methodologies was
envisaged as a potential solution, but its application was also
complicated by the presence of two enantiomers for each ligand
as well as by their binding diversity. One of the goals of this
article is to show how we have confronted these practical
problems, with the hope that other researchers can take
advantage of the proposed solutions.
The 3D-QSAR is a powerful methodology, and techniques
like the one presented here can widen its range of application.
It is noteworthy that our efforts were directed to building a
representation of the phenomena under study as realistic as
possible, even if we were aware that such models will have a
poorer fitting and lower statistical indexes than those obtained
with the standard methodologies. In our opinion, these “coarse
grain” QSAR models are more valuable, because they should
be able to identify more consistent relationships and provide
better predictions than standard “fine grain” methods. Indeed,
one of the reasons why the QSAR methodologies are being so
heavily criticized lately57 is probably the tendency to obtain “too
good” models with very high fitting in situations in which
experimental errors and methodological limitations make such
expectations unrealistic.
In our series, the simultaneous use of several receptor-docked
structures representing the same compound produces an interest-
ing model that compares favorably, from a statistical point of
view, with models obtained using classical methods, while being
presumably more robust and having more general validity. The
model is interpretable in terms similar to those used in structure-
based drug design studies, allowing to extract information that
can be easily translated as guidelines for the design of novel
compounds. Moreover, estimation of the models predictive
ability based on strict cross-validation methods and in external
datasets suggest that the predictions for structurally related
compounds will fall within an interval of (1.0. Even if this
cannot be considered a very accurate estimation, it is more than
enough for identifying potentially active or inactive compounds.
Indeed, preliminary results obtained for new, unpublished
compounds confirm the robustness and the reliability of our
approach.
3D models were then built using the MODELLER suite of
programs,27 which yielded 25 candidate models for the 5-HT2A
receptor final structure. From these candidates, the best structures
according to the MODELLER objective function and to visual
inspection were selected.
(iii) Assessing the Quality of the Structures. Models with
interruptions or gaps in the transmembrane regions, as identified
by visual inspection, were discarded. PROCHECK software69 was
used to assess the quality of the backbone dihedrals, resulting in
good quality parameters, with an excellent distribution of æ and ψ
angles in the Ramachandran plot. Also, the resulting models must
reproduce the correct orientation of the side chains for the set of
highly conserved amino acids in the GPCR superfamily.34-37 The
Experimental Section
Numbering of Residues. For residues belonging to helix regions,
the generalized numbering scheme proposed by Ballesteros and
Weinstein58 was used through the text. According to this scheme,

Studies on Conformationally Constrained Butyrophenones
Journal of Medicinal Chemistry, 2007, Vol. 50, No. 14 3253
conformation of these residues was set to the “inactive state”, which
probably is more appropriate for modeling the docking of antago-
nists and more consistent with the inactive state of the main template
structure (Rh).
Docking Simulation of Ketanserin. The docking of ketanserin
was carried out using the program GOLD.49 The structure of
ketanserin was docked in the active site of the 5-HT2A receptor,
forcing the interaction between ketanserin and the Asp3.32 side
chain with a harmonic force restraint. The rest of the parameters
of the program were set to their default values. The best docking
solution, according to the scoring function of GOLD, was then
refined using energy minimization (EM) and MD simulations, as
explained in the following paragraphs.
the human values were used, but in general, both values do not
show large differences (under 5% in all instances).
GRID/GOLPE Analysis. The structures corresponding to the
best docking solution (M1) or to the three best solutions (M2) were
imported into GRID52 where MIF were computed for all of them
using three probes (DRY, O, N1). All the MIF were computed using
a grid spacing of 1 Å. The analysis was made in a box of 22 × 36
× 25 nodes, containing 19 800 energy measures per structure. The
resulting MIF were then imported into program GOLPE 4.6.0,53
and the X matrix of variables obtained was pretreated, first applying
a zeroing of very small values (under 0.01), removing variables
with small standard deviation (under 0.05), and ill-conditioned
variables that take only two or three different values, one of which
is assigned to a single compound. Then a Block Unscaled Weight
scaling was applied to the whole matrix to equalize the importance
of the different blocks. To apply regression methods, the pKi values
listed in Table 1 were imported as the Y block. A first PLS model
was obtained and then the SRD/FFD variable selection method54
was applied, using as parameters for the SRD method 1000 seeds,
filtering with the PLS weights obtained for a 2 LV model. Two
consecutive FFD variable selection iterations were consecutively
applied using the groups obtained in the SRD procedure. For the
selection, LOO cross-validation and 2 LV were used. The statistical
parameters for the M1 and M2 models are shown in Table 4. The
q2 in this table were obtained using the classical LOO method and
two more strict methods (called 5RG and 2RG) consisting of
splitting the series randomly into either two (2RG) or five groups
(5RG), one of which is removed and predicted in turn until every
group has been removed once. This whole procedure is then
repeated either 20 (for the 5RG method) or 200 times (for the 2RG
method). As can be seen in Table 4, the q2 values obtained with
the most strict methods are rather similar to the LOO values.
Hardware. All the computations were carried out in Linux
workstations with Intel Pentium4 and Xeon CPUs. The visualization
of structures and complexes were made with VMD75 and our own
in-house developed software.
Energy Minimization (EM). The best docking solution was
submitted to EM with the SANDER classic module of AMBER6
suite of programs70 by applying 500 cycles of steepest descent
minimization, followed by conjugate gradient minimization until
the RMS gradient was lower than 0.001 kcal/mol‚Å. At the
beginning, only the positions of the hydrogen atoms were optimized
and then the whole complex (ligand and protein) was allowed to
relax. In all the minimizations, a harmonic constraint of 5 kcal/
mol‚Å2 was imposed to the CR trace, in order to maintain the overall
seven helices architecture. All the calculations were carried out
using the all atom force field of Cornell et al.,71 as implemented in
AMBER6, with a distance-dependent dielectric constant of 4r and
a cutoff for nonbonded interactions of 10 Å. The charges for the
ligand atoms were calculated using Gaussian 98,72 with a 6-31G**
basis set, and then the RESP procedure was applied as described
in literature.73 A formal charge of +1 was attributed to the ligand,
because the piperidine nitrogen of ketanserin is assumed to be
protonated at physiological pH, and the positively charged nitrogen
of the ligand is assumed to bind the negatively charged side chain
of Asp3.32.
Molecular Dynamics (MD). The minimized structure obtained
from the previous step was submitted to MD simulations using the
SANDER classic module of AMBER6 suite of programs during 1
ns at 300 °K. The time step was set to 2 fs, a value that is justified
by the application of the SHAKE algorithm.74 The nonbonded pair
interaction list was updated every 25 fs. For the aforementioned
reasons, we kept the harmonic force constant of 5 kcal/mol‚Å2 to
the CR trace, but the EL2 loop was left without restrictions. The
rest of the parameters were set to their default values. The MD
trajectories were analyzed with the CARNAL module of AMBER,
monitoring the formation of H-bonds, the profile for the total energy,
the temperature, as well as the RMS deviations of side chains. The
frames with the lowest total energy were selected, minimized, and
analyzed to detect the interaction between the ligand and the protein.
The best complex structure (with lower energy) was minimized
and used as the starting point for the ligand docking studies.
Ligand Structures. The structure of all the compounds in Table
1 were modeled in 2D in their protonated state (charge +1) and
then converted to 3D using program CORINA v2.4.50 Two
enantiomers (R and S) were obtained for every compound.
Ligand Docking. The ligands were docked within the modeled
binding site using the program GOLD.49 For every compound, the
program made 10 docking trials, and only the three best solutions
were retained. The fitness of the solutions were assessed using the
GOLDscore function.51 Solutions with a score under a prefixed
value of 35 were discarded. To obtain consistent orientations, a
distance constraint between the charged nitrogen of the ligand and
the Asp3.32 was defined. To compensate the binding site rigidity,
up to three bump contacts were allowed. As a test, this protocol
was used to dock the structure of the ketanserin, obtaining a solution
that showed a very low RMSD (0.58 Å) with respect to the MD
solution.
Acknowledgment. The authors acknowledge Nuria B.
Centeno, Hugo Gutie´rrez-de-Tera´n, and Fabien Fontaine for their
valuable scientific comments. The present work has been
supported in part by CICYT (SAF2002-04195-C03 and SAF2005-
08025-C03) and “Fundacio´ La Marato´ de TV3” grants. A Ph.D.
fellowship (FI) has been granted to C.D. by the Generalitat de
Catalunya.
Supporting Information Available: Experimental procedures,
spectroscopic data, and elemental analyses for compounds 1-6 and
9-24. This material is available free of charge via the Internet at
http://pubs.acs.org.
References
(1) Horacek, J. Novel antipsychotics and extrapyramidal side effects.
Theory and reality. Pharmacopsychiatry 2000, 33 (Suppl 1), 34-
42.
(2) Aghajanian, G. K.; Marek, G. J. Serotonin model of schizophrenia:
Emerging role of glutamate mechanisms. Brain Res. ReV. 2000, 31,
302-312.
(3) Reynolds, G. P. Receptor mechanisms in the treatment of schizo-
phrenia. J. Psychopharmacol. 2004, 18, 340-345.
(4) Meltzer, H. Y.; Matsubara, S.; Lee, J. C. Classification of typical
and atypical antipsychotic drugs on the basis of dopamine D-1, D-2,
and serotonin2 pKi values. J. Pharmacol. Exp. Ther. 1989, 251, 238-
246.
(5) Meltzer, H. Y.; Li, Z.; Kaneda, Y.; Ichikawa, J. Serotonin receptors:
Their key role in drugs to treat schizophrenia. Prog. Neuropsychop-
harmacol. Biol. Psychiatry 2003, 27, 1159-1172.
(6) Roth, B. L.; Sheffler, D. J.; Kroeze, W. K. Magic shotguns versus
magic bullets: Selectively nonselective drugs for mood disorders and
schizophrenia. Nat. ReV. Drug DiscoVery 2004, 3, 353-359.
(7) Ravina, E.; Negreira, J.; Cid, J.; Masaguer, C. F.; Rosa, E.; Rivas,
M. E.; Fontenla, J. A.; Loza, M. I.; Tristan, H.; Cadavid, M. I.; Sanz,
F.; Lozoya, E.; Carotti, A.; Carrieri, A. Conformationally constrained
butyrophenones with mixed dopaminergic (D(2)) and serotoninergic
Pharmacology. For most of the compounds, the binding was
measured using rat 5-HT2A receptors, except in a few for which
the affinity was measured in human receptors (see Table 1).
Radioligand binding assays in both human and rat were performed
by labeling 5-HT2A receptors with [3H]ketanserin, as previously
described.8 In the compounds for which both values were available,

3254 Journal of Medicinal Chemistry, 2007, Vol. 50, No. 14
Dezi et al.
(5-HT(2A), 5-HT(2C)) affinities: Synthesis, pharmacology, 3D-
QSAR, and molecular modeling of (aminoalkyl)benzo- and -thieno-
cycloalkanones as putative atypical antipsychotics. J. Med. Chem.
1999, 42, 2774-2797.
(26) 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.
(27) Sali, A.; Blundell, T. L. Comparative protein modelling by satisfaction
of spatial restraints. J. Mol. Biol. 1993, 234, 779-815.
(28) Egan, C.; Herrick-Davis, K.; Teitler, M. Creation of a constitutively
activated state of the 5-HT2A receptor by site-directed mutagenesis:
Revelation of inverse agonist activity of antagonists. Ann. N. Y. Acad.
Sci. 1998, 861, 136-139.
(29) Berg, K. A.; Harvey, J. A.; Spampinato, U.; Clarke, W. P.
Physiological relevance of constitutive activity of 5-HT2A and
5-HT2C receptors. Trends Pharmacol. Sci. 2005, 26, 625-630.
(30) Roth, B. L.; McLean, S.; Zhu, X. Z.; Chuang, D. M. Characterization
of two [3H]ketanserin recognition sites in rat striatum. J. Neurochem.
1987, 49, 1833-1838.
(8) Brea, J.; Rodrigo, J.; Carrieri, A.; Sanz, F.; Cadavid, M. I.; Enguix,
M. J.; Villazon, M.; Mengod, G.; Caro, Y.; Masaguer, C. F.; Ravina,
E.; Centeno, N. B.; Carotti, A.; Loza, M. I. New serotonin 5-HT-
(2A), 5-HT(2B), and 5-HT(2C) receptor antagonists: Synthesis,
pharmacology, 3D-QSAR, and molecular modeling of (aminoalkyl)-
benzo and heterocycloalkanones. J. Med. Chem. 2002, 45, 54-71.
(9) Brea, J.; Masaguer, C. F.; Villazon, M.; Cadavid, M. I.; Ravina, E.;
Fontaine, F.; Dezi, C.; Pastor, M.; Sanz, F.; Loza, M. I. Conforma-
tionally constrained butyrophenones as new pharmacological tools
to study 5-HT 2A and 5-HT 2C receptor behaviours. Eur. J. Med.
Chem. 2003, 38, 433-440.
(10) Brea, J.; Castro, M.; Loza, M. I.; Masaguer, C. F.; Ravina, E.; Dezi,
C.; Pastor, M.; Sanz, F.; Cabrero-Castel, A.; Galan-Rodriguez, B.;
Fernandez-Espejo, E.; Maldonado, R.; Robledo, P. QF2004B, a
potential antipsychotic butyrophenone derivative with similar phar-
macological properties to clozapine. Neuropharmacology 2006, 51,
251-262.
(31) Choudhary, M. S.; Craigo, S.; Roth, B. L. A single point mutation
(Phe340->Leu340) of a conserved phenylalanine abolishes 4-[125I]-
iodo-(2,5-dimethoxy) phenylisopropylamine and [3H]mesulergine but
not [3H]ketanserin binding to 5-hydroxytryptamine2 receptors. Mol.
Pharmacol. 1993, 43, 755-761.
(11) Waller, C. L.; Oprea, T. I.; Giolitti, A.; Marshall, G. R. Three-
dimensional QSAR of human immunodeficiency virus (I) protease
inhibitors. 1. A CoMFA study employing experimentally determined
alignment rules. J. Med. Chem. 1993, 36, 4152-4160.
(12) Waller, C. L.; Marshall, G. R. Three-dimensional quantitative
structure-activity relationship of angiotesin-converting enzyme and
thermolysin inhibitors. II. A comparison of CoMFA models incor-
porating molecular orbital fields and desolvation free energies based
on active-analog and complementary-receptor-field alignment rules.
J. Med. Chem. 1993, 36, 2390-2403.
(13) Cho, S. J.; Garsia, M. L.; Bier, J.; Tropsha, A. Structure-based
alignment and comparative molecular field analysis of acetylcho-
linesterase inhibitors. J. Med. Chem. 1996, 39, 5064-5071.
(14) Pastor, M.; Cruciani, G.; Watson, K. A. A strategy for the incorpora-
tion of water molecules present in a ligand binding site into a three-
dimensional quantitative structure-activity relationship analysis. J.
Med. Chem. 1997, 40, 4089-4102.
(32) Wurch, T.; Colpaert, F. C.; Pauwels, P. J. Chimeric receptor analysis
of the ketanserin binding site in the human 5-hydroxytryptamine1D
receptor: Importance of the second extracellular loop and fifth
transmembrane domain in antagonist binding. Mol. Pharmacol. 1998,
54, 1088-1096.
(33) Wurch, T.; Palmier, C.; Pauwels, P. J. Induction of a high-affinity
ketanserin binding site at the 5-hydroxytryptamine(1B) receptor by
modification of its carboxy-terminal intracellular portion. Biochem.
Pharmacol. 2000, 59, 1117-1121.
(34) Ballesteros, J. A.; Jensen, A. D.; Liapakis, G.; Rasmussen, S. G.;
Shi, L.; Gether, U.; Javitch, J. A. Activation of the beta 2-adrenergic
receptor involves disruption of an ionic lock between the cytoplasmic
ends of transmembrane segments 3 and 6. J. Biol. Chem. 2001, 276,
29171-29177.
(35) Bissantz C Conformational changes of G protein-coupled receptors
during their activation by agonist binding. J. Recept. Signal Transduct.
Res. 2003, 23, 123-153.
(15) Lozano, J. J.; Pastor, M.; Cruciani, G.; Gaedt, K.; Centeno, N. B.;
Gago, F.; Sanz, F. 3D-QSAR methods on the basis of ligand-receptor
complexes. Application of COMBINE and GRID/GOLPE method-
ologies to a series of CYP1A2 ligands. J. Comput.-Aided Mol. Des.
2000, 14, 341-353.
(16) Sippl, W. Receptor-based 3D QSAR analysis of estrogen receptor
ligands-merging the accuracy of receptor-based alignments with the
computational efficiency of ligand-based methods. J. Comput.-Aided
Mol. Des. 2000, 14, 559-572.
(17) Sippl, W.; Contreras, J. M.; Parrot, I.; Rival, Y. M.; Wermuth, C. G.
Structure-based 3D QSAR and design of novel acetylcholinesterase
inhibitors. J. Comput.-Aided Mol. Des. 2001, 15, 395-410.
(18) Sippl, W. Binding affinity prediction of novel estrogen receptor
ligands using receptor-based 3-D QSAR methods. Bioorg. Med.
Chem. 2002, 10, 3741-3755.
(19) Pastor, M.; Perez, C.; Gago, F. Simulation of alternative binding
modes in a structure-based QSAR study of HIV-1 protease inhibitors.
J. Mol. Graphics Modell. 1997, 15, 364-71, 389.
(36) Shapiro, D. A.; Kristiansen, K.; Weiner, D. M.; Kroeze, W. K.; Roth,
B. L. Evidence for a model of agonist-induced activation of
5-hydroxytryptamine 2A serotonin receptors that involves the disrup-
tion of a strong ionic interaction between helices 3 and 6. J. Biol.
Chem. 2002, 277, 11441-11449.
(37) Sylte, I. B. A.; Dahl, S. G. Ligand induced conformational states of
the 5-HT(1A) receptor. Eur. J. Pharmacol. 2001, 416, 33-41.
(38) Almaula, N.; Ebersole, B. J.; Ballesteros, J. A.; Weinstein, H.;
Sealfon, S. C. Contribution of a helix 5 locus to selectivity of
hallucinogenic and nonhallucinogenic ligands for the human
5-hydroxytryptamine2A and 5-hydroxytryptamine2C receptors: Di-
rect and indirect effects on ligand affinity mediated by the same locus.
Mol. Pharmacol. 1996, 50, 34-42.
(39) Shapiro, D. A.; Kristiansen, K.; Kroeze, W. K.; Roth, B. L.
Differential modes of agonist binding to 5-hydroxytryptamine(2A)
serotonin receptors revealed by mutation and molecular modeling
of conserved residues in transmembrane region 5. Mol. Pharmacol.
2000, 58, 877-886.
(20) Mewshaw, R. E.; Silverman, L. S.; Mathew, R. M.; Kaiser, C.;
Sherrill, R. G.; Cheng, M.; Tiffany, C. W.; Karbon, E. W.; Bailey,
M. A.; Borosky, S. A. Bridged gamma-carbolines and derivatives
possessing selective and combined affinity for 5-HT2 and D2
receptors. J. Med. Chem. 1993, 36, 1488-1495.
(21) Repke, D. B.; Clark, R. D.; Kluge, A. F.; Muchowski, J. M.;
Strosberg, A. M.; Lee, C. H.; Whiting, R. L. Syntheses and
hypotensive properties of substituted 2-aminotetralins. J. Pharm. Sci.
1985, 74, 37-39.
(40) Cortizo, L.; Santana, L.; Ravina, E.; Orallo, F.; Fontenla, J. A.; Castro,
E.; Calleja, J. M.; De Ceballos, M. L. Synthesis and antidopaminergic
activity of some 3-(aminomethyl)tetralones as analogs of butyrophe-
none. J. Med. Chem. 1991, 34, 2242-2247.
(41) Alvarado, M.; Coelho, A.; Masaguer, C. F.; Ravin˜a, E.; Brea, J.;
Pad´ın, J. F.; Loza, M. I. Synthesis and binding affinity of novel
3-aminoethyl-1-tetralones, potential atypical antipsychotics. Bioorg.
Med. Chem. Lett. 2005, 15, 3063-3066.
(42) Loza, M. I.; Ferreiro, T. G.; Sanz, F.; Lozoya, E.; Rodriguez, J.;
Manaut, F.; Verde, I.; Castro, E.; Fontenla, J. A.; Cadavid, I.
Antiserotoninergic activity of 2-aminoethylbenzocyclanones in rat
aorta: Structure-activity relationships. J. Pharm. Sci. 1993, 82, 513-
517.
(43) Fontenla, J. A.; Osuna, J.; Rosa, E.; Castro, M. E.; G-Ferreiro, T.;
Loza-Garcia, I.; Calleja, J. M.; Sanz, F.; Rodriguez, J.; et al. Synthesis
and atypical antipsychotic profile of some 2-(2-piperidinoethyl)-
benzocycloalkanones as analogs of butyrophenone. J. Med. Chem.
1994, 37, 2564-2573.
(22) Julia, S.; Bonnet, Y. Etude des ce´tones avec noyau cyclopropane I.
Les benzo- et naphtho- bicyclo- (0,1,4) hepte´nones. Bull. Soc. Chim.
Fr. 1957, 1340-1347.
(23) Caro, Y.; Torrado, M.; Masaguer, C. F.; Ravina, E.; Padin, F.; Brea,
J.; Loza, M. I. Chemoenzymatic synthesis and binding affinity of
novel (R)- and (S)-3-aminomethyl-1-tetralones, potential atypical
antipsychotics. Bioorg. Med. Chem. Lett. 2004, 14, 585-589.
(24) Metwally, K. A.; Dukat, M.; Egan, C. T.; Smith, C.; DuPre, A.;
Gauthier, C. B.; Herrick-Davis, K.; Teitler, M.; Glennon, R. A.
Spiperone: Influence of spiro ring substituents on 5-HT2A serotonin
receptor binding. J. Med. Chem. 1998, 41, 5084-5093.
(25) Wysong, C. L.; Yokum, T. S.; Morales, G. A.; Gundry, R. L.;
McLaughlin, M. L.; Hammer, R. P. 4-Aminopiperidine-4-carboxylic
acid: A cyclic alpha,alpha-disubstituted amino acid for preparation
of water-soluble highly helical peptides. J. Org. Chem. 1996, 61,
7650-7651.
(44) Ravina, E.; Fueyo, J.; Masaguer, C. F.; Negreira, J.; Cid, J.; Loza,
I.; Honrubia, A.; Tristan, H.; G-Ferreiro, T.; Fontenla, J. A.; Rosa,
E.; Calleja, J. M.; De Ceballos, M. L. Synthesis and affinities for
dopamine (D2) and 5-hydroxytryptamine (5-HT2A) receptors of
1-(benzoylpropyl)-4-(1-oxocycloalkyl-2-ethyl)-piperazines as cyclic
butyrophenone derivatives. Chem. Pharm. Bull. (Tokyo) 1996, 44,
534-541.

Studies on Conformationally Constrained Butyrophenones
Journal of Medicinal Chemistry, 2007, Vol. 50, No. 14 3255
(45) Masaguer, C. F.; Casariego, I.; Ravina, E. Conformationally restricted
butyrophenones with mixed dopaminergic (D[2]) and serotoninergic
(5-HT[2A]) affinities. Synthesis of 5-aminoethyl-and 6-aminomethyl-
4-oxotetrahydroindoles as potential atypical antipsychotics. Chem.
Pharm. Bull. 1999, 47, 621-632.
(46) Masaguer, C. F.; Formoso, E.; Ravin˜a, E.; Trista´n, H.; Loza, M. I.;
Rivas, E.; Fontenla, J. Butyrophenone analogues in the carbazole
series: Synthesis and determination of affinities at D2 and 5-HT2A
receptors. Bioorg. Med. Chem. Lett. 1998, 8, 3571-3576.
(47) Pita, B.; Masaguer, C. F.; Ravin˜a, E. A simple, efficient method for
regioselective synthesis of 7-aminomethyl-7,8-dihydro-6H-quinolin-
5-ones, new potential CNS agents. Tetrahedron Lett. 2000, 41, 9829-
9833.
(61) 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.
(62) Baldwin, J. M.; Schertler, G. F.; Unger, V. M. An alpha-carbon
template for the transmembrane helices in the rhodopsin family of
G-protein-coupled receptors. J. Mol. Biol. 1997, 272, 144-164.
(63) Persson, B.; Argos, P. Prediction of membrane protein topology
utilizing multiple sequence alignments. J. Protein Chem. 1997, 16,
453-457.
(64) Zhang, D.; Weinstein, H. Polarity conserved positions in transmem-
brane domains of G-protein coupled receptors and bacteriorhodopsin.
FEBS Lett. 1994, 337, 207-212.
(65) Cuff, J. A.; Clamp, M. E.; Siddiqui, A. S.; Finlay, M.; Barton, G. J.
JPred: A consensus secondary structure prediction server. Bioinfor-
matics 1998, 14, 892-893.
(66) White, S. H.; Wimley, W. C. Membrane protein folding and
stability: Physical principles. Annu. ReV. Biophys. Biomol. Struct.
1999, 28, 319-365.
(67) Russell, R. B.; Barton, G. J. Multiple protein sequence alignment
from tertiary structure comparison: assignment of global and residue
confidence levels. Proteins 1992, 14, 309-323.
(68) Espadaler, J.; Fernandez-Fuentes, N.; Hermoso, A.; Querol, E.; Aviles,
F. X.; Sternberg, M. J.; Oliva, B. ArchDB: Automated protein loop
classification as a tool for structural genomics. Nucleic Acids Res.
2004, 32, D185-D188.
(69) Laskowski, R. A.; MacArthur, M. W.; Moss, D. S.; Thornton, J. M.
PROCHECK: A program to check the stereochemical quality of
protein structures. J. Appl. Crystallogr. 1993, 26, 283-291.
(70) Case, D. A.; Pearlman, D. A.; Caldwell, J. M.; Cheatman, 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.
AMBER 6.0.; University of California, San Francisco, San Francisco,
CA, 1999.
(71) Cornell, W. D.; Cieplak, P.; Bayly, I. R.; Gould, I. R.; Merz, K. M.,
Jr.; Ferguson, D. M.; Spellmeir, D. C.; Fox, T.; Caldwell, J. W.;
Kollman, P. A. A second generation force field for the simulation of
proteins, nucleic acids, and organic molecules. J. Am. Chem. Soc.
1995, 117, 5179-5197.
(72) Frisch, M. J.; et al. Gaussian 98, revision A.7; Gaussian, Inc.;
Pittsburgh, PA, 1998.
(73) Bayly, C. I.; Cieplak, P.; Cornell, W.; Kollman, P. A. A well-behaved
electrostatic potential based method using charge restraints for
determining atom-centered charges: The resp model. J. Phys. Chem.
1993, 97, 10269.
(74) Miyamoto, S.; Kollman, P. A. An analytical version of the SHAKE
and RATTLE algorithm for rigid water molecules. J. Comput. Chem.
1992, 13, 952-962.
(75) Humphrey, W.; Dalke, A.; Schulten, K. VMD: Visual molecular
dynamics. J. Mol. Graph. 1996, 14, 33-8, 27-8.
(48) Ravina, E.; Casariego, I.; Masaguer, C. F.; Fontenla, J. A.; Mon-
tenegro, G. Y.; Rivas, M. E.; Loza, M. I.; Enguix, M. J.; Villazon,
M.; Cadavid, M. I.; Demontis, G. C. Conformationally constrained
butyrophenones with affinity for dopamine (D1, D2, D4) and
serotonin (5-HT2A, 5-HT2B, 5-HT2C) receptors: Synthesis of
aminomethylbenzo[b]furanones and their evaluation as antipsychotics.
J. Med. Chem. 2000, 43, 4678-4693.
(49) Verdonk, M. L.; Cole, J. C.; Hartshorn, M. J.; Murray, C. W.; Taylor,
R. D. Improved protein-ligand docking using GOLD. Proteins 2003,
52, 609-623.
(50) Gasteiger, J.; Rudolph, C.; Sadowski, J. Automatic generation of 3D-
atomic coordinates for organic molecules. Tetrahedron Comput.
Methodol. 1990, 3, 537-547.
(51) Jones, G.; Willett, P.; Glen, R. C.; Leach, A. R.; Taylor, R.
Development and validation of a genetic algorithm for flexible
docking. J. Mol. Biol. 1997, 267, 727-748.
(52) GRID 22; Molecular Discovery, Ltd.: London, U.K., 2006.
(53) GOLPE 4.6.0.; Multivariate Infometric Analysis: Srl. Perugia, Italy,
2003.
(54) Pastor, M.; Cruciani, G.; Clementi, S. Smart region definition: A
new way to improve the predictive ability and interpretability of three-
dimensional quantitative structure-activity relationships. J. Med.
Chem. 1997, 40, 1455-1464.
(55) Hudson, B. D.; Hyde, R. M.; Rah, E. r.; Wood, J.; Osman, J.
Parameter based methods for compound selection from chemical
databases. Quant. Struct.-Act. Relat. 2006, 15, 285-289.
(56) Cramer, R. D., III; Patterson, D. E.; Bunce, J. D. Comparative
molecular field analysis (CoMFA). 1. Effect of shape on binding of
steroids to carrier proteins. J. Am. Chem. Soc. 1988, 110, 5959-
5967.
(57) Doweyko, A. M. 3D-QSAR illusions. J. Comput.-Aided Mol. Des.
2004, 18, 587-596.
(58) Ballesteros, J.; Weinstein, H. Integrated methods for the construction
of three dimensional models and computational probing of structure-
function relations in G-protein coupled receptors. Methods Neurosci.
1995, 25, 366-428.
(59) Boeckmann, B.; Bairoch, A.; Apweiler, R.; Blatter, M. C.; Estreicher,
A.; Gasteiger, E.; Martin, M. J.; Michoud, K.; O’Donovan, C.; Phan,
I.; Pilbout, S.; Schneider, M. The SWISS-PROT protein knowledge-
base and its supplement TrEMBL in 2003. Nucleic Acids Res. 2003,
31, 365-370.
(60) Thompson, J. D.; Gibson, T. J.; Plewniak, F.; Jeanmougin, F.;
Higgins, D. G. The CLUSTAL_X windows interface: Flexible
strategies for multiple sequence alignment aided by quality analysis
tools. Nucleic Acids Res. 1997, 25, 4876-4882.
JM070277A