Content uploaded by Anuradha Sharma
Author content
All content in this area was uploaded by Anuradha Sharma
Content may be subject to copyright.
Mol Divers (2012) 16:803–823
DOI 10.1007/s11030-012-9394-x
FULL-LENGTH PAPER
Molecular docking and receptor-specific 3D-QSAR studies
of acetylcholinesterase inhibitors
Pran Kishore Deb ·Anuradha Sharma ·
Poonam Piplani ·Raghuram Rao Akkinepally
Received: 13 May 2012 / Accepted: 27 August 2012 / Published online: 21 September 2012
© Springer Science+Business Media B.V. 2012
Abstract The reversible inhibition of acetylcholinesterase
(AChE) has become a promising target for the treatment
of Alzheimer’s disease (AD) which is mainly associated
with low in vivo levels of acetylcholine (ACh). The avail-
ability of AChE crystal structures with and without a lig-
and triggered the effort to find a structure-based design of
acetylcholinesterase inhibitors (AChEIs) for AD. The major
problem observed with the structure-based design was the
feeble robustness of the scoring functions toward the corre-
lation of docking scores with inhibitory potencies of known
ligands. This prompted us to develop new prediction mod-
els using the stepwise regression analysis based on consen-
sus of different docking and their scoring methods (GOLD,
LigandFit, and GLIDE). In the present investigation, a data-
set of 91 molecules belonging to 9 different structural classes
of heterocyclic compounds with an activity range of 0.008 to
281,000 nM was considered for docking studies and develop-
ment of AChE-specific 3D-QSAR models. The model (M1)
developed using consensus of docking scores of scoring
functions viz. Glide score, Gold score, Chem score, ASP
score, PMF score, and DOCK score was found to be the best
(R2=0.938,Q2=0.925,R2
pred =0.919,R2m(overall)=
0.936) compared to other consensus models. Docking studies
Electronic supplementary material The online version of this
article (doi:10.1007/s11030-012-9394-x) contains supplementary
material, which is available to authorized users.
P. K. Deb ·A. Sharma ·P. Piplani ( B
)
Pharmaceutical Chemistry Division, University Institute
of Pharmaceutical Sciences (UIPS) and Centre of Advanced
Study in Pharmaceutical Sciences (UGC-CAS), Panjab University,
Chandigarh 160 014, India
e-mail: ppvohra28in@yahoo.co.in
R. R. Akkinepally
Medicinal Chemistry Division, University College of Pharmaceutical
Sciences, Kakatiya University, Warangal 506009, India
revealed that the molecules with proper alignment in the
active site gorge and the ability to interact with all the crucial
amino acid residues, in particular by forming π–πstacking
interactions with Trp84 at the catalytic anionic site (CAS) and
Trp279 at peripheral anionic site (PAS), showed augmented
potencies with consequent improvement in patient cognition
and reduced the formation of senile plaques associated with
AD. Further, the descriptors that signify the association of
the ligands with the receptor as well as ADME properties of
the ligands were also analyzed by means of the set of ligands
that have been pre-positioned with respect to a receptor after
docking analysis and considered as independent variables to
generate a linear model (M3 and M4) using a stepwise multi-
ple linear regression method to get additional insight into the
physicochemical requirements for effective binding of lig-
ands with AChE as well as for prediction of AChE inhibition.
The developed AChE-specific prediction models (M1–M4)
satisfactorily reflect the structure–activity relationship of the
existing AChEIs and have all the potential to facilitate the
process of design and development of new potent AChEIs.
Keywords Acetylcholinesterase inhibitors ·Molecular
docking ·Consensus scoring ·3D-QSAR ·
ADME prediction
Introduction
Alzheimer’s disease (AD) is a multifactorial disease that
affects millions of elderly people and is clinically character-
ized by the progressive loss of memory and other cognitive
impairments [1,2]. It has been observed that the formation
of extracellular senile plaques consisting of aggregated amy-
loid-β-peptide (Aβ) deposits, τ-protein aggregation, oxida-
tive stress, dyshomeostasis of biometals, and low levels of
123
804 Mol Divers (2012) 16:803–823
acetylcholine (ACh) plays significant roles in the pathophys-
iology of the disease [3].
The proven and currently accepted therapeutic approach
for the treatment of AD involves the inhibition of the enzyme
acetylcholinesterase (AChE) which is responsible for the
hydrolysis of ACh, thereby raising the levels of ACh in the
synaptic cleft [4]. As a consequence, four acetylcholines-
terase inhibitors (AChEIs) belonging to different chemical
groups have been approved by the Food and Drug Adminis-
tration (FDA) for the symptomatic treatment of mild to mod-
erate stages of AD, such as tacrine, donapezil, rivastigmine,
and galantamine [5]. The most common adverse effects of
these drugs include nausea and vomiting, which are linked
to the presence of excess cholinergic neurons. Less com-
mon secondary adverse effects include bradycardia, mus-
cle cramps, decreased appetite, and weight loss as well as
increased gastric acid production [6].
A lot of 3D structures of AChE bound with different inhib-
itors have been determined by X-ray crystallography from
native Torpedo californica (TcAChE) and humans (hAChE)
as well (shown in the Table S1 in Supplementary material).
All these structures significantly enhance our understanding
of the structural elements of AChE [7–18].
The crystal structure of TcAChE [19] revealed that its
active site is buried at the bottom of a narrow gorge, about
20 Å deep, lined with conserved aromatic residues and the
active site gorge is only ∼5Å wide at a bottleneck formed by
the van der Waals surfaces of Tyr121 and Phe330. Several
kinetic models for AChE proposed that the binding pocket
of AChE consists of two substrate-binding sites, the catalytic
anionic site (CAS), near the bottom of the active site gorge,
and the peripheral anionic site (PAS), near its entrance. The
binding of ligands at the PAS affects catalytic activity [20].
In the catalytic anionic subsite (CAS), it has been proposed
that the choline moiety of ACh is stabilized principally via
a cation–πinteraction with Trp84, and it also interacts with
Glu199 and Phe330. The catalytic site is the binding site
of classical AChE inhibitors, such as tacrine and huperzine,
which have been studied thoroughly [19]. The esteratic sub-
site in TcAChE contains a typical serine-hydrolase catalytic
triad, Ser200-His440-Glu327. A substantial contribution to
ACh binding within the active site also arises from stabil-
ization of the carbonyl oxygen within the oxyanion hole,
Gly118, Gly119, and Ala201, and of the acetyl group in the
“acyl-pocket,” Trp233, Phe288, Phe290, and Phe331 [17].
The PAS contains three principal amino acids, Trp279,
Tyr70, and Asp72. Biochemical studies have indicated that
AChE promotes amyloid fibril formation by interaction thro-
ugh the peripheral anionic site of the enzyme (PAS), giving
stable AChE-Aβcomplexes, which are more toxic than sin-
gle Aβpeptides [21]. The fact that AChE accelerates Aβ
aggregation and that this effect is sensitive to PAS block-
ers has led to the development of dual inhibitors of both
catalytic active site (CAS) and PAS. These compounds are
promising disease-modifying AD drug candidates because
they can simultaneously improve cognition and slow the rate
of Aβ-elicited neurodegeneration [20,22]. To date, donepe-
zil is the sole dual binding site AChE inhibitor approved for
the treatment of AD. The X-ray crystallographic structure of
the complex between TcAChE and donepezil (PDB: 1EVE)
[9] shows that the elongated structure of donepezil spans the
entire length of the enzyme active site gorge forming a variety
of interactions with specific residues such as aromatic stack-
ing interactions between the benzyl and indanone moieties
and the indole rings of Trp84 and Trp279 (Trp86 and Trp286
in hAChE) at the catalytic and peripheral sites, respectively,
and the cation–πinteraction between the protonated piperi-
dine nitrogen and the phenyl ring of Phe330. As a result of
its dual binding site character, donepezil at 100 µM is able to
inhibit by 22% the AChE-induced aggregation of Aβ[23].
Several attempts have been made recently in generat-
ing the predictive models on various targets using ligand
and structure-based methods for identifying the right mol-
ecules from virtual screening [24–28]. Interestingly, none of
the reported prediction models could satisfactorily explain
the physicochemical requirements for effective binding of
the diverse class of heterocyclic compounds with AChE. In
spite of several efforts, including cholinergic and non-cho-
linergic approaches, an efficient strategy for designing new
drugs for the treatment of AD is still lacking. In pursuance
of our ongoing efforts to design and develop novel AChEIs,
we attempted to compare the biological activity of reported
as well as marketed AChEIs with their docking scores, and
a perusal of the results indicated poor correlation. This may
be due to the artifacts present in each scoring function as all
these are based on various assumptions and simplifications
[29]. Hence, an attempt was made to develop new predic-
tion models based on the ligand-receptor interaction field and
trained satisfactorily to corroborate the experimental results
with the pharmacophoric and structural features of AChEIs.
The results correctly reflect the structure–activity relation-
ship (SAR) of the existing AChEIs and indicated a reason-
ably good potential for them as virtual screening tools, and
the details of our methodology and results are described in
this communication.
Material and methods
Dataset
In the present work, a dataset of 91 compounds belonging
to different classes of human AChEIs, such as tacrine-8-
hydroxyquinoline hybrids [30], tacripyrines [31], donepezil–
tacrine hybrids derivatives [32], tetrahydroacridine [33],
benzofuran-based hybrids [34], rivastigmine analogues [35],
tacrine-Melatonin hybrids [36], carbamates of tetrahydro-
123
Mol Divers (2012) 16:803–823 805
furobenzofuran and methanobenzodioxepine [37], and
diamine diamides [38], has been used for docking and QSAR
studies. The activity among these compounds ranges from
0.008 to 281,000nM (Table 1). The biological activity data
[IC50 (nM)] were converted to logarithmic scale [pIC50
(nM)] and then used for subsequent QSAR analyses as
the response variable (dependent variable). The biological
activity values and structural features of the compounds are
presented in Table 1. It is worth mentioning that donepezil–
tacrine hybrid derivatives (compounds 19–27), which bear
the stereo center containing an indanone unit of donepezil,
were prepared in racemic form. It has been observed that both
the enantiomorphs display similar pharmacologic and phar-
macokinetic profiles and span the entire AChE gorge with a
common pattern of interaction. Hence, the activity was taken
without enantiomeric separation [32]. In the present work,
R-configurations of compounds 19–27 were used for dock-
ing studies and further calculations. Similarly, for rivastig-
mine analogues (compounds 42–48), cis-isomers were used
for docking studies.
Descriptors
To develop the consensus models (M1 and M2) and per-
form the QSAR studies, eleven scoring functions and twenty
four docking descriptors were investigated. The eleven scor-
ing functions include one from GLIDE (Glide score) [39,
40], three from GOLD [41] (Gold score [42], Chem score
[43], ASP score [44]), and seven from LigandFit [45](PMF
(potential of mean force) [46], -PLP1 [47], -PLP2 [47], Jain
[48], LigScore-1 [49], LigScore-2 [49], Dock score [50]).
Docking descriptors were obtained from different scoring
functions, viz. four from Gold score, seven from Chem score,
Tabl e 1 Dataset ligands with biological activities used for docking studies and generation of prediction models (M1–M4)
N N
H
N
OH3C
O
H3CX
R
(19–26)
n
N
O
H3C
O
H3CO
(27)
N
NHR
(1–3)
R
N
HN
H3C
C2H5
NH2
O
(5–18)
N
HN
Cl
NHR
(28–32)
O
O N
CH3
R
n
(33–40)
O N CH3
H3C
N
CH3
H3C
CH3
O
(41)
X
N
ON
H
R
H3C
O
H
H
(42–48)
N
HN
NH
H
N
O
X
(49–55)
R
n
O
N
CH3
H
N
N
H
H3C
O
(58)
O
N
H
R
X
Y
H3C
OH
(66–72)
O
N
R
H
OO
O
H3C
(73–74)
O
N
R
H
OO
O
H3C
(75–77)
OCH3
N
C2H5R
N
O
CH3
C2H5
O
O
n
n
(78–82)
OCH3
N
C2H5
RN
OCH3
C2H5
nn
(83–89)
O
N
O
H3C
HO
CH3
(90) (91)
OCH3
N
C2H5
N
O
O
n
Donepezil Rivastigmine
Galantamine
N
NH2
(4)
O
N
R
H
X
Y
H
H3C
O
(56–57)
ON
R
H
X
Y
H
H3C
O
(59–65)
Physostigmi ne
CH3
CH3
Tacrine
123
806 Mol Divers (2012) 16:803–823
Tabl e 1 continued
C
I
nYXR
.oNgiL 50(nM) pIC50 (nM) Reference
1
C
9
H
18
N
H
N
OH
-- -- -- 5.5 8.26
[29]
2
C
8
H
16
N
H
N
OH
CH
3
-- -- -- 0.5 9.30
3
C
9
H
18
N
H
N
OH
Cl
-- -- -- 1.0 9.00
464
.6
053
---
-
-
-
--
5-- -- -- 122 6.91 [30]
6
F
-- -- -- 193 6.71
7
F
3
C
-- -- -- 226 6.65
8
O
2
N
-- -- -- 338 6.47
9
NO2
-- -- -- 191 6.71
10
NO
2
-- -- -- 309 6.51
11
CH
3
-- -- -- 169 6.77
12 -- -- -- 71 7.15
13
H
3
CO
-- -- -- 234 6.63
14
OCH
3
-- -- -- 58 7.24
15
OCH
3
-- -- -- 105 6.98
123
Mol Divers (2012) 16:803–823 807
Tabl e 1 continued
16
OCH3
OCH3
-- -- -- 45 7.35
17
N
-- -- -- 158 6.80
18
N
-- -- -- 223 6.65
19
H
O
-- 2 4.04 8.39 [31]
20
H
O
-- 3 0.88 9.05
21
Cl
O
-- 2 0.67 9.17
22
Cl
O
-- 3 0.27 9.57
23
H
H
,
H
-- 2 5.13 8.29
24
H
H
,
H
-- 3 2.16 8.67
25
Cl
H
,
H
-- 2 2.60 8.59
26
Cl
H
,
H
-- 3 1.06 8.97
27 49.76.11----
-
---
C
I
nYXR
.oNgiL 50(nM) pIC50 (nM) Reference
28
NH
O
C
3
H
6
-- -- --
2.15
8.67
[32]
29
NH
O
C
4
H
8
-- -- -- 1.65 8.78
30
NH
O
C
5
H
10
-- -- -- 1.54 8.81
31
NH
O
C
6
H
12
-- -- -- 2.57 8.59
32
H
-- -- 21.5 7.67
33
H
-- -- 7 32600 4.49 [33]
34
CH
3
O
-- -- 7 17400 4.76
123
808 Mol Divers (2012) 16:803–823
Tabl e 1 continued
35
O
-- -- 7 40700 4.39
36
O
CH
3
-- -- 7 127000 3.90
37
O
CH3
-- -- 7 10500 4.98
38
O
OCH
3
-- -- 7 127000 3.90
39
O
OCH
3
-- -- 7 281000 3.55
40
O
OCH
3
OCH
3
-- -- 7 177000 3.75
41 25.5
0303-
-------
42
CH
3
O
-- -- 30.0 7.52 [34]
43
CH
3
CH
2
-- -- 17.3 7.76
44
CH
3
S
-- -- 8.11 8.09
*45
CH
CH
3
O
-- -- 1870 5.73
46
(CH
2
)
6
CH
3
O
-- -- 20.3 7.69
47
C
2
H
5
O
-- -- 420 6.38
48
C
2
H
5
CH
2
-- -- 393 6.41
49
HH
-- 6 0.5 9.30 [35]
50
Cl6
H
-- 6 0.1 10.00
51
Cl
8
H
-- 5 0.87 9.06
52
diCl6,8
H
-- 6 0.008 11.1
53
H
-OCH
3
-- 6 0.65 9.19
54
diCl6,8
-OCH
3
-- 6 0.04 10.40
55
H
-OH
-- 5 0.45 9.35
56
C
2
H
5
O
O
100 7.00 [36]
C
I
nYXR
.oNgiL 50(nM) pIC50 (nM) Reference
123
Mol Divers (2012) 16:803–823 809
Tabl e 1 continued
57
H
3
C
O
O
-- 20 7.70
58
CH
3
NCH
3
NCH
3
-- 28 7.55
59
C
2
H
5
NCH
3
NCH
3
-- 94 7.03
60
H
3
C
NCH
3
NCH
3
-- 10 8.00
61
CH
3
CH
3
NCH
3
NCH
3
-- 760 6.12
62
CH
3
NCH
3
O
-- 27 7.57
63
C
2
H
5
NCH
3
O
-- 82 7.09
64
H
3
C
NCH
3
O
-- 13 7.89
65
C
3
H
7
NCH
3
O
-- 3860 5.41
66
C
2
H
5
O O
-- 486 6.31
67
H
3
C
O O
-- 36 7.44
68
C
3
H
7
O O
-- 6700 5.17
69
CH
3
NCH
3
NCH
3
-- 9890 5.00
70
H
3
C
NCH
3
NCH
3
-- 5510 5.26
71
CH
3
NCH
3
O
-- 56 7.25
72
H
3
C
NCH
3
O
-- 142 6.85
73
C
2
H
5
-- -- -- 610 6.21
74
H
3
C
-- -- --
38
7.42
75
C
2
H
5
-- -- -- 2420 5.62
76
H
3
C
-- -- -- 97 7.01
77
C
3
H
7
-- -- -- 3430 5.46
C
I
nYXR
.oNgiL 50(nM) pIC50 (nM) Reference
123
810 Mol Divers (2012) 16:803–823
Tabl e 1 continued
79
NH HN
-- -- 5 4160 5.38
80
NH HN
-- -- 5 5470 5.26
81
N
H
N
H
-- -- 5 8550 5.07
82
NN
-- -- 5 4.83 8.32
83
NN
OO
-- -- 5 68.9 7.16
84
N
N
O
O
-- -- 5 15.9 7.80
85
NN
OO
-- -- 5 1.41 8.85
78
NN
-- -- 5 25.7 7.59 [37]
86
NN
O
-- -- 5 72.4 7.14
87
N
O
N
O
O
O
-- -- 5 14.0 7.85
88
N
O
N
O
O
O
-- -- 5 7.70 8.11
89
N
N
OO
O
O
-- -- 5 0.37 9.43
90
0
7.
50102---
-
--
91 7
2.
7
5
.35
6
---
-
--
C
I
nYXR
.oNgiL 50(nM) pIC50 (nM) Reference
aCis-Rconfiguration was used for docking studies and further calculations
123
Mol Divers (2012) 16:803–823 811
Tabl e 2 Docking methods
(software), their scoring
functions with type of scoring,
and the various descriptors used
for generating Model 1 (M1)
and Model 2 (M2)
PLP piecewise linear potential,
PMF potential of mean force,
ASP astex statistical potential
Docking
methods
Scoring functions Type of scoring Scoring function descriptors
LigandFit PMF Knowledge-based –
-PLP1, -PLP2, Jain, LigScore1
dreiding, LigScore2 dreiding
Empirical –
Dock score Force field –
GLIDE Glide score Empirical Hbond, vdW, Coul, Emodel,
CvdW, Intern, Electro, PhobEn,
PhobEnHB
Gold score Force field S(hb_ext), S(vdw_ext), S(hb_int),
S(int)
GOLD Chem score Empirical DG, S(hbond), S(metal), S(lipo),
H(rot), DE(clash), DE(int)
ASP Knowledge-based ASP, S(Map), ASP_DE(clash),
ASP_DE(int)
Tabl e 3 Various ligand and structure-based descriptors used for generating QSAR Models (M3 and M4)
Types of descriptors Descriptors
Embrace descriptors Total energy without constraints, Valence energy, vdW energy, Electrostatic energy, Solvation energy, Constraint energy
QikProp descriptors #Stars, #amine, #amidine, #acid, #amide, #rotor, #rtvFG, CNS, mol_MW, dipole, SASA, FOSA, FISA, PISA,
WPSA, volume, donorHB, accptHB, dipˆ2/V, ACxDNˆ.5/SA, glob, QPpolrz, QPlogPC16, QPlogPoct, QPlogPw,
QPlogPo/w, QPlogS, CIQPlogS, QPlogHERG, QPPCaco, QPlogBB, QPPMDCK, QPlogKp, IP(ev), EA(eV),
metab, QPlogKhsa, Human oral absorption, Percent human oral absorption, SAfluorine, SAamideO, PSA,
#NandO, Lipinski’s Rule of five, Jorgensen’s Rule of three, #ringatoms, #in34, #In56, #noncon, #nonHatm, Jm
four from ASP score, and six from Glide score (G-score).
Further, six embrace energetics descriptors and 50 ADME
descriptors were obtained from the Embrace module [51] and
QikProp module [52] of Schrodinger software, respectively,
for the development of models M3 and M4. The categorical
list of various descriptors used in the development of QSAR
models is presented in Tables 2and 3.
Validation methods
The aim of any QSAR modeling is that the developed model
should be strong enough to be capable of making accurate
and reliable predictions of biological activities of new com-
pounds. The robustness of the QSAR models is generally
verified by means of different types of validation criteria
such as (i) internal validation or cross-validation, (ii) valida-
tion by dividing the dataset into training and test compounds,
(iii) data randomization or Y-scrambling, and (iv) true exter-
nal validation by application of model on new external data
[53,54]. In the present work, due to the lack of a true exter-
nal evaluation set, the total dataset (n=91) was divided
into training set (n=69) and test (external evaluation) set
(n=22) (75 and 25%, respectively, of the total number
of compounds) based on clusters obtained from K-means
clustering [55] applied on docking scores of eleven scoring
functions (Table S2 in Supplementary material). The whole
dataset was clustered into five subgroups from each of which
25 % of compounds were selected as members of the testset.
The serial numbers of compounds under different clusters
are shown in Table S3 in Supplementary material.
The 3D-QSAR models developed from the training set
were internally cross validated using the leave-one-out
(LOO) method with LOO-Q2metric [53,54]. Prediction error
sum of squares (PRESS) is a standard index to measure the
accuracy of a modeling method based on the cross validation
technique [53,56]. The cross-validated squared correlation
coefficient R2(LOO-Q2)was calculated based on the PRESS
and SSY (sum of squares of deviations of the experimental
values from their mean) using Eq. (1).
Q2=1−PRESS
SSY =1−(yobs(training)−ycal(training))2
(yobs(training)−¯ytraining)2
(1)
In Eq. (1), ¯ytraining represents the average activity value of
the training set, while Yobs(training)and Ycal(training)represent
observed and calculated activity values, respectively, of train-
ing set compounds. Often, a high Q2value (Q2>0.5) is
considered as a proof of the high predictive ability of the
model [53,56].
However, internal validation does not ascertain that the
model will perform well on a new set of data. Thus, the mod-
els developed from the training set were externally validated
based on external predictions for the test set molecules using
the predictive R2(R2
pred)metric[53,56]. The predictive R2
(R2
pred)values were calculated according to the following
equation:
123
812 Mol Divers (2012) 16:803–823
R2
pred =1−(yobs(test)−ypred(test))2
(yobs(test)−ytraining)2(2)
In Eq. (2), Ypred(test)and Yobs(test)indicate predicted and
observed activity values, respectively, of the test set com-
pounds and ¯ytraining indicates the mean activity value of the
training set compounds. The value of R2
pred for an acceptable
model should be more than 0.5.
Further, the models were also validated using an additional
parameter R2
m(defined as R2∗(1−R2−R2
0)), R2, and
R2
0being squared correlation coefficients between observed
and predicted values of the compounds), which penalizes a
model for large differences between observed and predicted
values [53,56]. Two variants of R2
mparameter, R2
m(LOO)and
R2
m(test), which penalize a model more strictly than Q2and
R2
pred, respectively, were calculated [57]. In case of good pre-
diction, where the predicted activity values lie in close prox-
imity to the observed data, the R2value will be very near
the R2
0value. Consequently, an ideal prediction of activity
is characterized by a value of R2
mequal to that of R2. When
the value of R2
mis above 0.5, the model may be considered
satisfactory. The R2
m(LOO)and R2
m(test)metrics are used for
detecting proximity of the predicted activity data to those of
the observed ones for the training and test sets (external set),
respectively. Besides these, the R2
m(overall)metric [57]was
also calculated which ascertains the overall model predictiv-
ity based on the predicted property values of the whole data-
set (both training and test sets). The R2
m(overall)metric helps
to identify the best model from among comparable models,
especially when different models show different patterns in
internal and external predictivity.
Computational details
Three molecular docking programs have been used in this
study, GLIDE v9.1 [39,40], GOLD v4.0 [41], and Ligand-
Fit [45]. A further six embrace energetics descriptors and
fifty ADME descriptors were obtained from Embrace and
QikProp modules, respectively, of Schrodinger software [51,
52]. The prediction models were developed using the step-
wise multiple linear regression method by employing MINI-
TAB software [58,59]. All calculations were made with a
DELL precision desktop workstation T3400 running an Intel
Core2 Duo Processor, a 4GB RAM, a 250 GB hard disk, and
NVidia Quodro FX 4500 graphics card.
Preparation of protein, ligand structures, and docking
protocols
The activities had been focused on a hAChE structure,
but docking studies were carried out using the crystal
structure of TcAChE which includes a co-crystallized
inhibitor donepezil (E2020) [9] and was taken from the
Protein Data Bank (PDB ID: 1EVE) for protein prepara-
tion.
GLIDE The multi-step Schrodinger’s protein preparation
wizard tool (PPrep) has been used for protein preparation
which was minimized using OPLS-2005 force field with a
polack-ribiere conjugate gradient (PRCG) algorithm. Lig-
Prep module was used for ligand preparation. The ligands
were minimized by means of molecular mechanics force
fields (OPLS-2005) with a default setting. Using the best
10 docking poses, the corresponding scores have been eval-
uated in extra precision mode (Glide XP) for each ligand. To
soften the potential for non-polar parts of the receptor, we
scaled van der Waals radii of the receptor atoms by 1.00 with
partial atomic charge 0.25. The G-score (Glide score) and
six docking descriptors were calculated for each of the best
docked pose [39,40].
GOLD The protein and ligand preparations were car-
ried out following the same procedure as mentioned under
GLIDE. Taking the prepared protein and ligand, GOLD
docking calculations were performed using all the default
standard set parameters. For each of the 10 independent
genetic algorithm (GA) runs, with a selection pressure 1.1,
100,000 GA operations were performed on a set of five
islands with a population size of 100 individuals. Default
operator weights were used for crossover, mutation, and
migration of 95, 95, and 10, respectively. Default cut off
values were employed with 2.5Å for hydrogen bonds and
4.0Å for vdW. To speed up the calculations, the GA dock-
ing was terminated when the top three solutions were within
1.5Å RMSD. All other values were set to the default.
GOLD score, Chem score, and ASP scores were calculated
[41].
LigandFit All hydrogen atoms were included to the pro-
tein which was subjected to minimization using steepest
descent (gradient <0.1) and conjugate gradient algorithms
(gradient <0.01) using the CHARMm force field. The active
site was defined within a 10Å radius from the center of
the bound ligand. The “Prepare Ligands” module was used
for final preparation of ligands from libraries for dock-
ing. The ligands were minimized by the CHARMm force
field with a default setting. Docking was performed with
monte carlo simulations using the CFF95 force field. The
grid resolution was set to 0.5Å (default), and the ligand-
accessible grid was defined such that the minimum distance
between a grid point and the protein is 2.0Å for hydro-
gen and 2.5Å for heavy atoms. The grid extends from the
defined active site to a distance of 5Å in all directions.
The top 10 conformations were saved after rigid body min-
imizations of 1,000 steps. Dockscore, Ligscore1, and Lig-
score2, PLP1, PLP2, JAIN, PMF, and Dock scores were
calculated for each of the 10 saved ligand conformations
[45].
123
Mol Divers (2012) 16:803–823 813
Consensus models
Three molecular docking programs have been used in this
study, GOLD v4.0, LigandFit, and GLIDE v5.6. In all, eleven
scoring functions and 24 docking descriptors were inves-
tigated (Table 2). The stepwise multiple linear regression
method was used to obtain consensus models (M1 and M2)
with Minitab software.
Ligand and structure-based descriptors
The ligand and structure-based descriptors panel in Maestro
provide a convenient interface to several Schrödinger pro-
grams, which are used to generate ligand and structure-based
descriptors for a structure-based QSAR model of ligand bind-
ing to a receptor [51,52]. The focus is on generating descrip-
tors for a set of ligands that are docked to a receptor. The
program Embrace module of MacroModel operates on the
ligand and the receptor. The other program QikProp oper-
ates on the ligand as well as both on the ligand and recep-
tor. The descriptors extracted from Embrace are energetic
properties related to ligand binding and QikProp generates
ADME properties. These descriptors were used as an input
to the QSAR model generation (M3 and M4) with the help
of statistical software Minitab.
To set up the calculation, a pose viewer file (generated
after docking with Glide) was used to consider the recep-
tor and source of ligands. After choosing the receptor and
ligands, using a pose viewer file (pv.maegz), both programs
(Embrace and QikProp) that generate the descriptors were
run with default options that were chosen to produce reason-
able descriptors.
Embrace descriptors
To study the association of the ligands with the receptor fur-
ther, the automated mechanism of Multi-Ligand Bimolecu-
lar Association with Energetics (Embrace) was used. With
Embrace, complexes can be studied using simple minimi-
zations or conformational searches. An Embrace minimiza-
tion is a type of multiple minimization in which each of
the specified pre-positioned ligands is minimized, in turn,
with the receptor. Embrace minimization calculations can
be performed in two modes: Interaction Mode, in which the
interaction between each ligand and the substrate is stud-
ied, and Energy Difference Mode, in which energy changes
upon association are estimated. Embrace calculates ligand–
receptor binding energies by molecular mechanics energy
minimization of the complex and the separated receptor and
ligand, with or without continuum solvation. In the present
study, the Embrace calculation was run in an energy dif-
ference mode. In this mode, the calculation was performed
first on the receptor, then on the ligand, and finally on the
complex. The energy difference was then calculated using
the equation [51]:
E=Ecomplex −Eligand −Eprotein
A total of six descriptors were generated from the above cal-
culation (Table 3) that signifies the association of the ligands
with the receptor. These descriptors were used as input to
the QSAR model generation (M3) with the help of statistical
software Minitab.
QikProp descriptors
QikProp is a quick, accurate, easy-to-use absorption, distri-
bution, metabolism, and excretion (ADME) prediction pro-
gram design to produce 50 descriptors (Table 3) related to
ADME. QikProp predicts physically significant descriptors
and pharmaceutically relevant properties of organic mole-
cules, either individually or in batches. These descriptors
were used for the generation of QSAR model M4, includ-
ing properties like skin permeability and octanol/water parti-
tion coefficients, and counts of important functional groups.
QikProp has two modes: normal mode and fast mode. In
fast mode, certain time-consuming calculations are omitted,
some properties are not predicted, and some have different
values. In the present study, QikProp was run in normal pro-
cessing mode with default options [52].
Minitab software
The prediction models were developed using the step-
wise multiple linear regression method (MINITAB software)
based on forward selection and backward elimination tech-
niques for inclusion and rejection of descriptors. The selec-
tion of the significant descriptors for developing the model
was done using “stepping criteria” (F) with F=4 for inclu-
sion and F=3.9 for exclusion [58,59].
Results and discussion
The main objective of this work is to generate structure-based
quantitative models using a consensus of different dockings
and their scoring methods which can be useful in identify-
ing potent AChEIs. For generating the consensus models,
three docking protocols have been used such as LigandFit,
GLIDE, and GOLD. Before generating the models, these
protocols have been validated by reproducing the bound nat-
ural substrate conformation [co-crystalized ligand donepe-
zil (E2020)] in the crystal structure of TcAChE [9]. The
co-crystal natural substrate was taken out of the active site
and docked again. The top 3 docking configurations were
taken into consideration to validate the results and the RMSD
was calculated (Table S4 in Supplementary material) for
123
814 Mol Divers (2012) 16:803–823
each configuration in comparison with the co-crystallized lig-
and, donepezil. The acceptable results (RMSD within 0.32–
0.62 Å) indicated that the docked configurations have similar
binding positions and orientations within the binding site and
are similar to the co-crystal structure (Fig. 3a; Table S4 in
Supplementary material), which illustrates the fact that the
docking protocols used could successfully generate the crys-
tal donepezil–AChE complex precisely.
Further mechanism of multi-ligand bimolecular associa-
tion with energetic (embrace) was used to study the asso-
ciation of the ligands with the receptor. Moreover, ADME
descriptors were also used to get better insight into the phys-
icochemical requirements for effective binding of ligands
with AChE.
The corresponding amino acid residues which have been
hypothesized to be crucial at the binding site of hAChE (PDB
ID: 1B41) [8] and TcAChE (PDB ID: 1EVE) [9] were found
to be almost similar [24]. In particular, the crucial interacting
amino acid residues at PAS and CAS are the same in both
hAChE and TcAChE. Further, as the use of better resolved
structure (up to 2.5Å) enables proper insight into the ligand
receptor interactions, it was felt worthwhile to use the crystal
structure of TcAChE (2.5Å) rather than hAChE (2.95Å).
Moreover, the crystal structure of TcAChE (PDB ID:
1EVE) with bound co-crystal ligand donepezil (E2020) was
preferred to carry out the docking studies as it (ligand) is
the only FDA-approved dual binding site AChE inhibitor
for the treatment of AD. Due to the lack of a proper under-
standing of the persistence and roles of individual solvent
molecules in and around the AChE binding sites, all water
molecules were removed from the structures. In the pres-
ent study, 91 compounds belonging to 9 different classes of
reported AChEIs [30–38] along with their biological activ-
ities from a range 0.008 to 281,000nM (Table 1) obtained
by screening on human AChE inhibition were considered
for the docking experiments. These compounds were geo-
metrically optimized and docked into the binding site of the
prepared AChE crystal structure using all the three dock-
ing methods. The pose (best conformation of docked mol-
ecules) selection of docked ligands was done based on the
closest alignment to the original donepezil ligand as well as
the interaction of ligands with key amino acid residues such
as Trp84, Glu199, Phe330 at the catalytic anionic subsite
(CAS) and Trp279, Tyr70, Asp72 at PAS. All the three types
of scoring functions were calculated by means of three dock-
ing protocols including knowledge-based, force field-based,
and empirical scores (Table 2)[39–50]. While correlating
docking scores with inhibitory potencies of known ligands
(AChEIs), feeble robustness of scoring functions toward pre-
diction was observed and their R2value with respect to activ-
ity was <0.371 (Tables S5, S6, S7, and S8 in Supplementary
material). Hence, these individual scoring functions were not
useful to identify true hits which prompted us to develop
consensus prediction models for calculating the AChE inhib-
itory activities of these compounds. Two consensus models
were developed using the stepwise multiple linear regres-
sion analysis: One is based on consensus of different scoring
functions of three docking methods and the other is based on
the individual descriptors of each docking protocol obtained
by docking of 91 AChEIs.
The Model 1 (M1) was generated using different scor-
ing functions obtained from all the three docking protocols.
As with individual scoring functions, it was difficult to find
a good correlation coefficient (R2)between docking scores
and pIC50 values (Table S5 Supplementary material); all the
eleven scoring functions were investigated as consensus to
generate the predictive model using the stepwise multiple
linear regression method.
The Model 1 (M1) and the corresponding statistics are
shown below:
pIC50 =1.4036 −(0.25 ×Glide score)
+(0.1058 ×Gold score)
−(0.321 ×Chem score)+(0.148 ×ASP score)
+(0.0405 ×Dock score)
−(0.0093 ×PMF)(M1)
ntraining =69 R2=0.938,Q2=0.925 R2
adj =0.935,
s=0.427,PRESS=14.308 R2
m(LOO)=0.938,ntest =22,
R2
pred =0.919,R2
m(test)=0.919,R2
m(overall)=0.936
As shown in the Model 1 (M1), the consensus of six types
of scoring functions was found to be very significant where
Glide score, Gold score, Chem score, ASP score, Dock score,
and PMF refer to computed affinities from Glide score, Gold
score, Chem score, ASP score, Dock score, and PMF meth-
ods, respectively. Glide score incorporates the coulomb and
vdW interaction energies between the ligand and the receptor
and also introduces a solvation model [53–55]. It has nega-
tive regression coefficient toward the binding affinity. GOLD
score is made up of five components: protein–ligand hydro-
gen bond energy (external H-bond), van der Waals (vdW)
energy (external vdW), ligand internal vdW energy (inter-
nal vdW), ligand torsional strain energy (internal torsion),
and ligand intramolecular hydrogen bond energy (internal
H-bond) [57]. It has a favorable contribution toward the bind-
ing affinity. Chem score recognizes favorable hydrophobic,
hydrogen-bonding, and metal-ligation interactions and also
penalizes steric clashes. It uses a simple rotatable-bond term
to treat conformation entropy effects arising from restricted
motion of the ligand [43]. It has an unfavorable contribution
toward binding affinity as evidenced by negative regression
coefficient.
ASP score is the atom–atom potential derived from a data-
base of protein–ligand complexes [44]. Traditional scoring
123
Mol Divers (2012) 16:803–823 815
Fig. 1 Correlation between observed and calculated/predicted activity using model 1 based on scoring functions (M1) for atraining set and btest
set
Tabl e 4 Comparative table of statistical qualities of different models (M1–M4)
Eq. no. Model type Model quality Internal validation
parameters
External validation
parameters
Overall validation
parameters
R2R2
adj sQ2PRESS R2
m(LOO)R2
pred R2
m(test)R2
m(overall)
(M1) Consensus
scoring
function
0.938 0.935 0.427 0.925 14.308 0.938 0.919 0.919 0.936
(M2) Descriptors of
scoring
function
0.856 0.850 0.651 0.834 32.003 0.856 0.824 0.823 0.845
(M3) Embrace
energetic
descriptors
0.864 0.856 0.637 0.844 29.955 0.864 0.865 0.865 0.859
(M4) ADME
(QikProp)
descriptors
0.903 0.896 0.542 0.866 25.745 0.903 0.804 0.803 0.884
functions are based on force field or on regression, where
parameters are derived from a set of experimental binding
affinities and structures. ASP score uses a different approach;
information about the frequency of interaction between lig-
and and receptor atoms is gathered by analyzing existing
ligand–protein structures in the PDB and this information is
used to generate statistical potentials. The empirical parame-
ters used in the scoring function are hydrogen bond energies,
atom radii, polarizabilities, torsion potentials, and hydrogen
bond directionalities. It has a favorable contribution toward
the binding affinity as evidenced by positive regression coef-
ficient.
Dock score is the sum of internal energy of the ligand and
the interaction energy of the ligand with receptor [50]. The
interaction energy is taken as the sum of the van der Waals
energy and electrostatic energy. It has a positive contribution
toward the binding affinity. PMF is calculated by summing
pairwise interaction terms over all interatomic pairs of the
receptor–ligand complex along with metal ion interactions
and halogen potentials [46]. It has a detrimental contribu-
tion toward the affinity as evidenced by negative regression
coefficient.
This consensus score model shows an improvement in pre-
diction of activity in training as well as test set as compared to
the individual scoring functions (Tables S2 and S9 in Supple-
mentary material). The above model explains 93.5% of the
variance (adjusted coefficient of variation-R2
adj). The leave-
one-out predicted variance (Q2)was found to be 92.5 %.
The predictive potential of this model was determined by
predicted R2(R2
pred)of the test set compounds and it was
found to be 91.9% (Fig. 1a, b). The R2
mvalues for the train-
ing, test, and overall sets were found to be 0.938, 0.919, and
0.936, respectively (Table 4), which indicate the robustness
of the fit and suggested that the calculated pIC50 (Table S9
in Supplementary material) based on scoring functions is
reliable.
The 24 descriptors obtained from GOLD and GLIDE
docking methods, which signify the electronic and van der
Waal, hydrogen bonding, and internal energy between lig-
ands and receptor, were considered as independent variables
to generate a linear model (M2) using the stepwise multiple
linear regression method for prediction of AChE inhibition
(pIC50).
pIC50 =2.496 −(0.767 ×vdw)
−(1.144 ×DE(int))+(0.334 ×DE(clash))
+(1.25 ×H(rot))+(0.158 ×S(hb_ext)) (M2)
123
816 Mol Divers (2012) 16:803–823
Fig. 2 Correlation between observed and calculated/predicted activity using model 2 based on descriptors of scoring functions (M2) for atraining
set and btest set
ntraining =69,R2=0.856,Q2=0.834,R2
adj =0.850,
s=0.651,PRESS=32.003,R2
m(LOO)=0.856,ntest =22,
R2
pred =0.824,R2
m(test)=0.823,R2
m(overall)=0.845
In this model, five types of descriptors were found to be
very significant for predicting the biological activity. Vdw
includes van der Waals energy, calculated with reduced net
ionic charges on groups with formal charges. It has an unfa-
vorable contribution toward the binding affinity as evidenced
by negative regression coefficient; and, the lesser the vdW
energy, the better the activity as evident from the compounds
2,3,20–22,49–55, etc., whereas increase in the value of
it may decrease the activity as is the case for compounds
5–18,27,32,41–48,61–77,79–81, and 91. DE(int) includes
internal torsional strain and it has a detrimental effect toward
binding affinity such as compound number 42,43,44,45,
etc. DE(clash) includes protein–ligand clashes which depend
upon hydrogen bonding. It has a favorable effect as evi-
denced by positive regression coefficient. This implies that
increase in the strength of hydrogen bonding may increase
the binding affinity evident from the compounds such as 2–3,
20–26,49–55,89, etc. H(rot) indicates rotatable H-bond-
ing and has a favorable contribution as evidenced by posi-
tive regression coefficient. S(hb_ext) includes protein–ligand
hydrogen bond energy and has a favorable effect on the
binding affinity as evident from the compounds 20–26,
28–31,49–55, etc., whereas the decrease in the hydrogen
bond energy resulted in the decrease in affinity as indicated
from the compounds 33–40 where no significant hydrogen
bond interactions were observed between the ligands and
receptor. This has been already supported by the previous
descriptor DE(clash). The quality of the fit can be judged by
the value of the squared correlation coefficient (R2), which
was 0.856 for the dataset. Figure 2a, b graphically show
the quality of fit. The regression model developed in this
study with statistical parameters of Q2=0.834,R2
pred =
0.824,R2
m(test)=0.823,R2
m(overall)=0.845 is convincing
enough for its subsequent use for the prediction of AChE
inhibition (pIC50) of the AChEIs (Tables 4, S9 in Supple-
mentary material).
Before carrying out the binding mode analysis, redocked
conformation for donepezil in TcAChE was compared
relative to its original co-crystallized conformation. As
shown in Fig. 3a, in the active site, benzyl ring of donepe-
zil formed a π–πinteraction with the indole ring of Trp84
and in the peripheral binding site (PAS), the indanone ring of
donepezil formed a π–πstacking interaction with the indole
ring of Trp279. The charged nitrogen of the piperidine ring of
donepezil undergoes a cation–πinteraction with the phenyl
ring of Tyr334. The O-atom of the indanone moiety and
the protonated NH of piperidine ring formed a significant
H-bond interaction with NH of Phe288 and OH group of
Tyr334 at a distance of 2.319 and 3.341Å, respectively. The
same protocol was used for docking other molecules of the
dataset. The observed differences in activities of different
molecules of the dataset could be explained in terms of
their binding orientations and spatial arrangement toward the
indole ring of Trp84 and Trp279 at the catalytic pocket and
peripheral site of the AChE. Tacrine, the first FDA-approved
drug for the treatment of AD, was found to inhibit AChE
through its binding to the CAS where the NH2group formed
a significant H-bond interaction with the O-atom of Phe330 at
a distance of 2.180Å (Fig. 3b), while the tacrine-8-hydroxy-
quinoline hybrids (compounds 1–3of the dataset) were able
to bind with both CAS and PAS of AChE. In particular,
8-hydroxyquinoline moiety was found to interact with the
amino acid residues at PAS and showed a significant π–π
stacking interaction with the indole ring of Trp279 which
could probably be the reason for their ability to inhibit Aβ
fibril formation. The introduction of methyl and chloro sub-
stituents at the 2nd and 5th position of the quinolin-8-ol ring
of compound 2and 3of the dataset was found to enhance the
selectivity as well as affinity. The binding mode analysis of
tacripyrines or tacrine-dihydropyridine hybrids (compounds
5–18 of the dataset) revealed that they were successfully
docked at the PAS, but could not properly occupy the CAS.
As shown in Fig. 3c, the most active tacripyrine (compounds
16) stacked well against the indole ring of Trp279 at PAS
and showed H-bond interactions between the NH2group and
O-atom of propanone group at the 1st and 3rd position of the
molecule with Ser286 and Tyr70 at a distance of 2.402 and
123
Mol Divers (2012) 16:803–823 817
Fig. 3 a Binding orientation of redocked co-crystal E2020 (donepezil)
in the binding site (CAS and PAS) of the crystal structure of AChE (PDB
ID: 1EVE) showing H-bond interactions with the amino acid residues
Tyr334, Phe288 and π-stacking interaction with Tyr334 and Trp279.
bBinding orientation of compound 4(tacrine) of the dataset within the
binding site (CAS and PAS) showing H-bond interactions with Phe330
and π-stacking interaction with Tyr334. cBinding orientation of com-
pound 16 (tacripyrine: tacrine-dihydropyridine hybrid) of the dataset
within the PAS binding site of AChE showing H-bond interactions with
Tyr70 and Ser286 at a distance of 1.988 and 2.402 Å, respectively, and
π-stacking interaction with Trp279. dHypothetical binding motif of
compound 20 (donepezil–tacrine hybrid) of the dataset with tacrine
moiety firmly bound to CAS showing π-stacking interaction with Trp84
and the indanone moiety bound to PAS showing π-stacking interaction
with Trp279 along with the piperidine linker showing π-stacking inter-
action with Tyr334. eHypothetical binding orientation of compound 21
(donepezil–tacrine hybrid) of the dataset with chlorotacrine moiety
firmly bound to CAS showing π-stacking interaction with Trp84 with
chlorine atom embedded in a small hydrophobic pocket and the inda-
none moiety bound to PAS showing π-stacking interaction with Trp279
along with the piperidine linker showing π-stacking interaction with
Tyr334. fHypothetical binding orientation of compound 27 with tetra-
hydroacridine and carbazole moieties showing π-stacking interaction
with Trp84 of CAS and Trp279 of PAS, respectively
1.988Å, respectively. The donepezil–tacrine hybrids (com-
pounds 19–27 of the dataset) were able to dock well in CAS,
PAS, and mid-gorge binding sites and were found to be
highly potent human AChE inhibitors (Fig. 3d, e). The length
of the tether that connects the two constituting fragments of
the hybrids, i.e., indanone (or indane derivatives) moiety of
donepezil and the tacrine (or 6-chlorotacrine) unit, has a rele-
vant effect on the arrangement of the hybrid along the gorge,
leading to proper orientation of the tacrine moiety within
the AChE active site. As shown in Fig. 3d, e, the tacrine
moiety of compound 20 and 21 of the dataset firmly bound
to the CAS and stacked against the aromatic ring of Trp84,
whereas the indanone/indane ring showed a significant π–π
stacking interaction with the indole ring of Trp279 at PAS.
Moreover, the piperidine ring of the linker, which is aligned
along the gorge, protonated at physiological pH, formed π–π
stacking and H-bond interaction with the aromatic ring and
OH group of Tyr334, respectively. Finally, the chlorine atom
of compound 21 was found to occupy a small hydropho-
bic pocket of the receptor which could probably be the rea-
son to enhance the potency of these heterodimers having the
6-chlorotacrine unit relative to their unsubstituted counter-
parts. Similarly, the tacrine-carbazole hybrids (compounds
28–32 of the dataset) were also docked well in CAS, PAS, and
mid-gorge binding sites and exhibited significant inhibition
at human AChE (Fig. 3f). In particular, the tetrahydroacridine
and carbazole moieties of compound 29 could interact with
Trp84 of the catalytic pocket and Trp279 of the PAS with a
123
818 Mol Divers (2012) 16:803–823
Fig. 4 a Binding orientation of compound 34 (benzofuran deriva-
tive) in the binding site (CAS and PAS) of the crystal structure of
AChE (PDB ID: 1EVE) showing H-bond interactions with Phe288
(2.148 and 2.342Å) and benzofuran moiety showing π-stacking inter-
action with Trp279. bBinding orientation of compound 44 (rivastig-
mine analogue) of the dataset within the binding site (CAS and PAS)
showing H-bond interactions with Phe288 (2.314Å) and π-stack-
ing interaction with Tyr334. cBinding orientation of compound 50
(tacrine–melatonin hybrid) of the dataset within the binding site show-
ing H-bond interactions with Asp72, Tyr334, and Phe288 at a distance
of 2.939, 3.814, and 1.839Å, respectively, and indole moiety show-
ing π-stacking interaction with Trp279. dHypothetical binding motif
of compound 52 (tacrine–melatonin hybrid) of the dataset within the
binding site (CAS and PAS) showing H-bond interactions with Asp72,
Tyr334, Phe288, and Phe290 at a distance of 2.401, 2.871, 2.048, and
2.071Å, respectively, and indole moiety showing π-stacking interac-
tion with Trp279. eHypothetical binding orientation of compound 69
(tetrahydrobenzofuran derivative) of the dataset showing H-bond inter-
actions with Tyr334 and Phe288 (4.302 and 1.949Å) and π-stacking
interaction with Trp279. fHypothetical binding orientation of com-
pound 89 (diamine diamide) within the binding site (CAS and PAS)
showing H-bond interactions with Asp72, Phe288, Arg289, and Asp285
at a distance of 2.068, 2.131, 2.188, and 2.454Å, respectively, and
π-stacking interaction with Trp279 at PAS
significant π–πstacking interaction (Fig. 3f). One of the NH
group of the alkyl linker moiety formed a strong H-bonding
interaction with the carboxyl group of Asp72 (1.979Å) and
the chlorine atom of the tacrine moiety found to occupy a
small hydrophobic pocket of the receptor similar to donepe-
zil–tacrine hybrids. In the case of benzofuran-based hybrids
and rivastigmine analogues (compounds 33–41 and 42–48 of
the dataset), they were unable to show any significant π–π
stacking interaction with Trp84 of the catalytic pocket and
Trp279 of the PAS which probably led to the decrease in their
potency toward AChE (Fig. 4a, b). The tacrine-melatonin
hybrids (compounds 49–55 of the dataset), most potent com-
pounds of the dataset, were found to appropriately occupy
and interact with the Trp84 of the catalytic pocket and Trp279
of the PAS. The presence of chlorine atom in position 6 of
tacrine moiety (compound 50,Fig.4c) improved the potency,
whereas further addition of the chlorine atom in position
8 along with position 6 of tacrine moiety (compound 52;
Fig. 4d) caused a 10-fold increase in potency and selectivity.
The methylene linker aligned well in the mid-gorge, and the
indole moiety showed significant aromatic π-stacking inter-
action with Trp279 of the PAS (Fig. 4c, d). Additional signifi-
cant H-bond interactions were also observed between the NH
groups of tacrine and the carbonyl group (C=O) of methylene
linker with the carboxyl group of Asp72 and the NH group
of Phe288 at a distance of 2.939 and 1.839Å, respectively.
In the case of compound 52, some other significant H-bond
interactions were also observed between the NH group of
indole moiety and methylene linker with the carboxyl group
of Tyr334 and Phe290 at a distance of 2.871 and 2.071Å,
respectively, which might be the reason for its increased
potency. Also, the chlorine atom of these compounds was
found to occupy a hydrophobic pocket of the receptor which
probably led to synergestic enhancement of the potency rel-
ative to their unsubstituted counterparts. The carbamates of
tetrahydrobenzofuran (compounds 56–77 of the dataset) did
123
Mol Divers (2012) 16:803–823 819
not show significant potency compared to other analogues.
All these compounds occupied the mid-gorge region of the
receptor and showed H-bond interactions with Phe288 and
Tyr334 and moderate π-stacking interactions with Phe330
and Trp279 (Fig. 4e). Finally, the diamine diamides (com-
pounds 78–91 of the dataset) with a rigid linker aligned prop-
erly in CAS, PAS, and mid-gorge region of the receptor and
interacted with all the important amino acid residues. In par-
ticular, the compound 89 with the most rigid linker was able
to firmly occupy PAS by forming aromatic π-stacking inter-
actions with Trp279 (Fig. 4f). It has also shown significant
H-bond interactions with several amino acid residues of the
enzyme mid-gorge region such as Asp285, Phe288, Arg289,
and also with Asp72 at a distance of 2.454, 2.131, 2.188, and
2.068 Å which could account for its higher inhibitory potency
as compared to other analogues.
Docking studies as well as the prediction models reveal
that tacrine–melatonin hybrid derivatives and donepezil–
tacrine hybrids act as dual binding inhibitors (act at catalytic
and peripheral site) and aligned properly in the active site
gorge which results in the increase in their potency (higher
docking score) compared to simple tacrine derivatives, which
act only at catalytic site. It has been observed that the phenyl
ring of Tyr70 is almost perpendicular to the Trp279 ring and
forms a blocking wall to prevent the ligand ring from mov-
ing away from the position where it forms a π–πinteraction
with the indole ring of Trp279 toward the PAS.
Further, to study the association of the ligands with the
receptor, the automated mechanism of Multi-Ligand Bimo-
lecular Association with Energetics (Embrace) [51] has been
used. The six embrace energetics descriptors that signify the
association of the ligands with the receptor were obtained
from Embrace module (Schrodinger) by means of the set of
ligands (docked ligands) that have been pre-positioned with
respect to a receptor after docking analysis and considered as
independent variables to generate a linear model (M3) using
the stepwise multiple linear regression method for prediction
of AChE inhibition (pIC50).
pIC50 =5.434 −(0.00992 ×Electrostatic_energy)
−(0.0193 ×vdW_energy)
−(0.0315 ×Valence_energy)
+(0.0098 ×Total_energy)(M3)
ntraining =69,R2=0.864,Q2=0.844,R2
adj =0.856,
s=0.637,PRESS =29.955,R2
m(LOO)=0.864,ntest =22,
R2
pred =0.865,R2
m(test)=0.865,R2
m(overall)=0.859
In the above QSAR model (M3), the embrace elec-
trostatic_energy, vdW_energy, Valence_energy, and Total_
energy_without_constraints include coulomb energy
difference, van der Waals energy difference, valence energy
difference, and ligand binding energy, respectively.
Electrostatic energy has an unfavorable contribution as
evidenced by negative regression coefficient. It means that
an increase in the electrostatic energy may decrease the affin-
ity as in the case of compounds 1,2,3,4, etc. Valence
energy and van der Waals energy have an unfavorable con-
tribution toward binding affinity as evidenced by negative
regression coefficient, and an increase in the value of it may
decrease the activity as in the case of compounds 9,10,11,
etc. Total energy without constraints has a favorable contri-
bution toward binding affinity and it indicates that the ligand
which binds more tightly to its receptor has more activity. The
regression model developed was found to be reliable statisti-
cally ( R2=0.864,Q2=0.844,R2
m(LOO)=0.864,R2
pred =
0.865,R2
m(test)=0.865,R2
m(overall)=0.859) (Fig. 5a, b) and
consequently used for prediction of AChE inhibition (pIC50)
of the AChEIs (Tables 4, S9 in Supplementary material).
Moreover, 50 descriptors obtained using QikProp mod-
ule [52] that signifies the ADME properties of the ligands
were considered as independent variables to generate a lin-
ear model (M4) using the stepwise multiple linear regression
method to get additional insight into the physicochemical
requirements for effective binding of ligands with AChE as
well as for prediction of AChE inhibition (pIC50).
pIC50 =3.886 +(0.336 ×non con)+(0.0294 ×FISA)
+(0.0278 ×WPSA)−(0.0315 ×SAflourine)
−(0.114 ×metab)(M4)
ntraining =69,R2=0.903,Q2=0.866,R2
adj =0.896,
s=0.542,PRESS =25.745,R2
m(LOO)=0.903,ntest =22,
R2
pred =0.804,R2
m(test)=0.803,R2
m(overall)=0.884
The noncon is the number of ring atoms (e.g., sp3C) not
able to form conjugated aromatic systems. It has a positive
contribution toward the binding affinity. It has been observed
that the compounds should have the optimum number of non-
con value (noncon = 4–6) to show the affinity as observed
from the compounds 28–31,49–55,82,85,89, etc. Interest-
ingly, an increase in noncon value was not found to be respon-
sible for any further increase in activity as evident from the
compounds 19–27(noncon = 11–12), whereas the decrease in
noncon value (noncon = 0–3) caused a tremendous decrease
in activity as is the case with the compounds 33–41,65,
68–70,77,80–81, etc. FISA is a hydrophilic component of
the SASA (SASA on N, O atoms, and H on heteroatoms;
recommended value 7.0–330.0) [52] and WPSA is a weakly
polar component of the SASA (halogens, P, and S; recom-
mended value 0.0–175.0) [52]. SASA is the total solvent
accessible surface area (SASA) in square angstroms using a
probe with a 1.4 Å radius. FISA and WPSA have a favorable
123
820 Mol Divers (2012) 16:803–823
Fig. 5 Correlation between observed and calculated/predicted activity using model 3 based on embrace energetic descriptors (M3) for atraining
set and btest set
Fig. 6 Correlation between
observed and
calculated/predicted activity
using model 4 based on ADME
(QikProp) descriptors (M4) for
atraining set and btest set
contribution toward affinity as evidenced by the positive
regression coefficient. It indicates that the compounds with
increased hydrophilicity and a weakly polar halogen atom
show better activity and it can be evident from the data-
set where compounds such as 21–26,50–54 with Cl-atom
attached to the benzene ring of tacrine moiety show better
activity as compared to other analogs.
SAfluorine is the solvent accessible surface area of fluo-
rine atoms and has an unfavorable contribution toward bind-
ing affinity. This shows that the presence of fluorine atom
may cause a decrease in activity as evident from the least
active compounds 6and 7of the dataset. The metab is the
number of likely metabolic reactions and has a detrimental
contribution toward activity.
The QSAR model developed based on ADME descrip-
tors is well predictive and statistically significant as ensured
from the value of R2=0.903,Q2=0.866,R2
m(LOO)=
0.903,R2
pred =0.804,R2
m(test)=0.803,R2
m(overall)=0.884,
which is much higher and satisfactory than the stipulated
value of 0.5 (Fig. 6a, b; Tables 4, S9 in Supplementary mate-
rial).
Conclusions
In the present study, 4 new receptor-specific (AChE) pre-
diction models have been developed and used in predict-
ing binding affinities of AChEIs. Among them, the scoring
function (Model 1, M1) developed using consensus of dock-
ing scores of scoring functions viz. Glide score, Gold score,
Chem score, ASP score, PMF score, and DOCK score was
found to be the best according to the statistical parameters
of R2=0.938,Q2=0.925,R2
pred =0.919,R2m(overall)=
0.936. This consensus score model shows an improvement
in prediction of activity in training as well as test sets as
compared to the individual scoring functions. Moreover, the
consensus model developed based on the docking descriptors
(M2) with the statistical parameters of R2=0.856,Q2=
0.834,R2
pred =0.824,R2
m(overall)=0.845 was also found to
be satisfactory enough for its subsequent use for prediction of
AChE inhibition (pIC50). It has been observed from the
Model 2 that the hydrogen bonding between the active site
amino acid residues of protein and ligand plays an important
role for showing the better affinity as indicated by the descrip-
tors DE(clash), H(rot), and S(hb_ext) with positive regres-
sion coefficient. Docking studies indicate that the amino
acid residues Trp84 at the CAS and Trp279 at PAS play
a crucial role by forming π–πstacking interactions with
the dual binding inhibitors and enhancing their potencies to
increase patient cognition (by increasing levels of acetyl-
choline) and reduce the formation of senile plaques. Among
all the 91 docked compounds, the tacrine-melatonin hybrid
derivatives and donepezil–tacrine hybrids with an appropri-
ate linker were found to be aligned properly in the active
site gorge and act as dual binding inhibitors by interacting
with all the crucial amino acid residues at the catalytic and
peripheral site which results in the increase in their potencies
compared to other molecules of the dataset. Further, the six
embrace energetics descriptors that signify the association of
the ligands with the receptor were also analyzed by means of
the set of ligands that have been pre-positioned with respect
to a receptor after docking analysis and considered as inde-
pendent variables to generate a linear model (M3) using the
stepwise multiple linear regression method. This Model 3
123
Mol Divers (2012) 16:803–823 821
(M3) was also found to be convincing enough statistically
(R2=0.864,Q2=0.844,R2
m(LOO)=0.864,R2
pred =
0.865,R2
m(test)=0.865,R2
m(overall)=0.859) and conse-
quently used for prediction of AChE inhibition of the AC-
hEIs. Finally, descriptors that signify the ADME properties
of the ligands were considered as independent variables to
generate a linear model (M4) with significant statistical value
(R2=0.903,Q2=0.866,R2
m(LOO)=0.903,R2
pred =
0.804,R2
m(test)=0.803,R2
m(overall)=0.884) using the
stepwise multiple linear regression method to get additional
insight into the physicochemical requirements for effective
binding of ligands with AChE as well as for prediction of
AChE inhibition. It has been observed that compounds with
the optimum number of ring atoms (e.g., sp3C) that are
not able to form conjugated aromatic systems along with
increased hydrophilicity and a weakly polar halogen atom
such as Cl-atom attached at strategic position of the mol-
ecules show better activity as compared to other analogs.
These AChE-specific prediction models (M1–M4) satisfac-
torily reflect the SAR of the existing AChEIs and could be
useful to predict the range of activities for new molecules
and have all the potential to facilitate the process of design
and development of new potent AChEIs.
Acknowledgments One of the authors (PKD) thanks the CSIR,
New Delhi, for awarding the Senior Research Fellowship (SRF)
[F. No. 09/135/(0534)/2008/EMR-I]. The authors thankfully acknowl-
edge the Chairman, the University Institute of Pharmaceutical Sciences
(UIPS), Panjab University (PU), Chandigarh, for providing facilities.
The authors also thank Dr. Ravikumar M., application scientist, Schro-
dinger, India, for providing valuable suggestions.
References
1. Lahiri DK, Farlow MR, Greig NH, Sambamurti K (2002) Cur-
rent drug targets for Alzheimer’s disease treatment. Drug Dev Res
56:267–281. doi:10.1002/ddr.10081
2. Inestrosa NC, Dinamarca MC, Alvarez A (2008) Amyloid–cholin-
esterase interactions. Implications for Alzheimer’s disease. FEBS
J 275:625–632. doi:10.1111/j.1742-4658.2007.06238.x
3. Querfurth HW, LaFerla FM (2010) Alzheimer’s disease. N Engl J
Med 362:329–344. doi:10.1056/NEJMra0909142
4. Pepeu G, Giovannini MG (2009) Cholinesterase inhibitors
and beyond. Curr Alzheimer Res 6:86–96. doi:10.2174/
156720509787602861
5. Smith DA (2009) Treatment of Alzheimer’s disease in the long-
term care setting. Am J Health Syst Pharm 66:899–907. doi:10.
2146/ajhp070622
6. Selkoe DJ (1999) Translating cell biology into therapeutic
advances in Alzheimer’s disease. Nature 399:A23–A31. doi:10.
1038/399a023
7. Carletti E, Colletier JP, Dupeux F, Trovaslet M, Masson P,
Nachon F (2010) Structural evidence that human acetylcholines-
terase inhibited by tabun ages through O-dealkylation. J Med Chem
53:4002–4008. doi:10.1021/jm901853b
8. Kryger G, Harel M, Giles K, Toker L, Velan B, Lazar A, Kronman
C, Barak D, Ariel N, Shafferman A, Silman I, Sussman JL
(2000) Structures of recombinant native and E202Q mutant
human acetylcholinesterase complexed with the snake-venom
toxin fasciculin-II. Acta Crystallogr D 56:1385–1394. doi:10.
1107/S0907444900010659
9. Kryger G, Silman I, Sussman JL (1999) Structure of acetylcho-
linesterase complexed with E2020 (Aricept): implications for the
design of new anti-Alzheimer drugs. Structure 7:297–307. doi:10.
1016/S0969-2126(99)80040-9
10. Raves ML, Harel M, Pang YP, Silman I, Kozikowski AP, Suss-
man JL (1997) Structure of acetylcholinesterase complexed with
the nootropic alkaloid, (-)-huperzine A. Nat Struct Biol 4:57–63.
doi:10.1038/nsb0197-57
11. Greenblatt HM, Kryger G, Lewis T, Silman I, Sussman JL
(1999) Structure of acetylcholinesterase complexed with (-)-galan-
thamine at 2.3 Å resolution. FEBS Lett 463:321–326. doi:10.1016/
S0014-5793(99)01637-3
12. Harel M, Schalk I, Ehret-sabatier L, Bouet F, Goeldner M, Hirth
C, Axelsen PH, Silman I, Sussman JL (1993) Quaternary ligand
binding to aromatic residues in the active-site gorge of acetylcho-
linesterase. Proc Natl Acad Sci USA 90:9031–9035. doi:10.1073/
pnas.90.19.9031
13. Doucet-Personeni C, Bentley PD, Fletcher RJ, Kinkaid A, Kryger
G, Pirard B, Taylor A, Taylor R, Taylor J, Viner R, Silman I, Suss-
man JL, Greenblatt HM, Lewis T (2001) A structure-based design
approach to the development of novel, reversible AChE inhibitors.
J Med Chem 44:3203–3215. doi:10.1021/jm010826r
14. Millard CB, Koellner G, Ordentlich A, Shafferman A, Silman I,
Sussman JL (1999) Reaction products of acetylcholinesterase and
VX reveal a mobile histidine in the catalytic triad. J Am Chem Soc
121:9883–9884. doi:10.1021/ja992704i
15. Greenblatt HM, Guillou C, Guenard D, Argaman A, Botti S, Badet
B, Thal C, Silman I, Sussman JL (2004) The complex of a bivalent
derivative of galanthamine with torpedo acetylcholinesterase dis-
plays drastic deformation of the active-site gorge: implications for
structure-based drug design. J Am Chem Soc 126:15405–15411.
doi:10.1021/ja0466154
16. Rydberg EH, Brumshtein B, Greenblatt HM, Wong DM, Shaya
D, Williams LD, Carlier PR, Pang YP, Silman I, Sussman
JL (2006) Complexes of alkylene-linked tacrine dimers with Tor-
pedo californica acetylcholinesterase: binding of bis(5)-tacrine
produces a dramatic rearrangement in the active-site gorge. J Med
Chem 49:5491–5500. doi:10.1021/jm060164b
17. Harel M, Quinn DM, Nair HK, Silman I, Sussman JL (1996) The
X-ray structure of a transition state analog complex reveals the
molecular origins of the catalytic power and substrate specificity
of acetylcholinesterase. J Am Chem Soc 118:2340–2346. doi:10.
1021/ja952232h
18. Ravelli RB, Raves ML, Ren Z, Bourgeois D, Roth M, Kroon J,
Silman I, Sussman JL (1998) Static Laue diffraction studies on ace-
tylcholinesterase. Acta Crystallogr D 1:1359–1366. doi:10.1107/
S0907444998005277
19. Sussman JL, Harel M, Frolow F, Oefner C, Goldman A, Toker
L, Silman I (1991) Atomic structure of acetylcholinesterase from
Torpedo californica: a prototypic acetylcholine-binding protein.
Science 253:872–879. doi:10.1126/science.1678899
20. Silman I, Sussman JL (2008) Acetylcholinesterase: How is struc-
ture related to function. Chem Biol Interact 175:3–10. doi:10.1016/
j.cbi.2008.05.035
21. Reyes AE, Chacon MA, Dinamarca MC, Cerpa W, Morgan C,
Inestrosa NC (2004) Acetylcholinesterase-Aβcomplexes are
more toxic than Aβfibrils in rat hippocampus: effect on rat beta-
amyloid aggregation, laminin expression, reactive astrocytosis,
and neuronal cell loss. Am J Pathol 164:2163–2174. doi:10.1016/
S0002-9440(10)63774-1
22. Castro A, Martınez A (2006) Targeting beta-amyloid pathogenesis
through acetylcholinesterase inhibitors. Curr Pharm Des 12:4377–
4387. doi:10.2174/138161206778792985
123
822 Mol Divers (2012) 16:803–823
23. Bartolini M, Bertucci C, Cavrini V, Andrisano V (2003) β-Amy-
loid aggregation induced by human acetylcholinesterase: inhibi-
tion studies. Biochem Pharmacol 65:407–416. doi:10.1016/S0006-
2952(02)01514-9
24. Guo J, Hurley MM, Wright JB, Lushington GH (2004) A docking
score function for estimating ligand-protein interactions: applica-
tion to acetylcholinesterase inhibition. J Med Chem 47:5492–5500.
doi:10.1021/jm049695v
25. Nagaraju A, Laurent L, Janet G, Vassilios P (2006) 3D QSAR stud-
ies of AChE inhibitors based on molecular docking scores and
CoMFA. Bioorg Med Chem Lett 16:6277–6280. doi:10.1016/j.
bmcl.2006.09.030
26. Chaudhary SS, Roy KK, Shakya N, Saxena G, Sammi SR, Nazir A,
Nath C, Saxena AK (2010) Novel carbamates as orally active ace-
tylcholinesterase inhibitors found to improve scopolamine-induced
cognition impairment: pharmacophore-based virtual screening,
synthesis, and pharmacology. J Med Chem 53:6490–6505. doi:10.
1021/jm100573q
27. Santos JSO, Lesnard A, Agondanou JH, Dupont N, Godard AM,
Stiebing S, Rochais C, Fabis F, Dallemagne P, Bureau R, Rault
S (2010) Virtual screening discovery of new acetylcholinesterase
inhibitors issued from CERMN chemical library. J Chem Inf Model
50:422–428. doi:10.1021/ci900491t
28. Gupta S, Mohan CG (2011) 3D-pharmacophore model based vir-
tual screening to identify dual-binding site and selective acetylcho-
linesterase inhibitors. Med Chem Res 20:1422–1430. doi:10.1007/
s00044-010-9373-7
29. Lakshmi Narayana B, Pran Kishore D, Balakumar C, Venkata Rao
K, Rajwinder K, Raghuram Rao A, Narashima Murthy J, Raviku-
mar M (2012) Molecular modeling evaluation of non-steroidal aro-
matase inhibitors. Chem Biol Drug Des 79:674–682. doi:10.1111/
j.1747-0285.2011.01277.x
30. Fernandez-Bachiller MI, Perez C, Gonzalez-Munoz GC, Conde S,
Lopez MG, Villarroya M, Garcıa AG, Rodriguez-Franco
MI (2010) Novel tacrine-8-hydroxyquinoline hybrids as multi-
functional agents for the treatment of alzheimer’s disease, with
neuroprotective, cholinergic, antioxidant, and copper-complexing
properties. J Med Chem 53:4927–4937. doi:10.1021/jm100329q
31. Marco-Contelles J, Leon R, de los Rios C, Samadi A, Bartolini M,
Andrisano V, Huertas O, Barril X, Luque FJ, Rodrıguez-Franco
MI, Lopez B, Lopez MG, Garcıa AG, Carreiras MC, Villarroya
M (2009) Tacripyrines, the first tacrine-dihydropyridine hybrids,
as multitarget-directed ligands for the treatment of Alzheimer’s
disease. J Med Chem 52:2724–2732. doi:10.1021/jm801292b
32. Camps P, Formosa X, Galdeano C, Gomez T, Munoz-Torrero
D, Scarpellini M, Viayna E, Badia A, Victoria-Clos M, Camins
A, Pallas M, Bartolini M, Mancini F, Andrisano V, Estelrich J,
Lizondo M, Bidon-Chanal A, Luque FJ (2008) Novel donepezil-
based inhibitors of acetyl- and butyrylcholinesterase and acetyl-
cholinesterase-induced β-amyloid aggregation. J Med Chem 51:
3588–3598. doi:10.1021/jm8001313
33. Rosini M, Simoni E, Bartolini M, Cavalli A, Ceccarini L, Pascu N,
McClymont DW, Tarozzi A, Bolognesi ML, Minarini A,
Tumiatti V, Andrisano V, Mellor IR, Melchiorre C (2008) Inhibi-
tion of acetylcholinesterase, β-amyloid aggregation and NMDA
receptors in alzheimer’s disease: a promising direction for the
multi-target-directed ligands gold rush. J Med Chem 51:4381–
4384. doi:10.1021/jm800577j
34. Rizzo S, Riviere C, Piazzi L, Bisi A, Gobbi S, Bartolini M,
Andrisano V, Morroni F, Tarozzi A, Monti JP, Rampa
A (2008) Benzofuran-based hybrid compounds for the inhibi-
tion of cholinesterase activity, β-amyloid aggregation, and Aβ
neurotoxicity. J Med Chem 51:2883–2886. doi:10.1021/
jm8002747
35. Bolognesi ML, Bartolini M, Cavalli A, Andrisano V, Rosini M,
Minarini A, Melchiorre C (2004) Design, synthesis and biological
evaluation of conformationally restricted rivastigmine analogues.
J Med Chem 47:5945–5952. doi:10.1021/jm049782n
36. Rodrıguez-Franco MI, Fernandez-Bachiller MI, Perez C,
Hernandez-Ledesma B, Bartolome B (2006) Novel tacrine-
melatonin hybrids as dual-acting drugs for alzheimer disease,
with improved acetylcholinesterase inhibitory and antioxidant
properties. J Med Chem 49:459–462. doi:10.1021/jm050746d
37. Luo W, Yu QS, Kulkarni SS, Parrish DA, Holloway HW, Tweedie
D, Shafferman A, Lahiri DK, Brossi A, Greig NH (2006) Inhibition
of human acetyl- and butyrylcholinesterase by novel carbamates
of (-)- and (+)-tetrahydrofurobenzofuran and methanobenzodioxe-
pine. J Med Chem 49:2174–2185. doi:10.1021/jm050578p
38. Tumiatti V, Milelli A, Minarini A, Rosini M, Bolognesi ML, Micco
M, Andrisano V, Bartolini M, Mancini F, Recanatini M, Cavalli A,
Melchiorre C (2008) Structure-activity relationships of acetylcho-
linesterase noncovalent inhibitors based on a polyamine backbone.
4. Further investigation on the inner spacer. J Med Chem 51:7308–
7312. doi:10.1021/jm8009684
39. GLIDE Maestro, version 9.1 (2010) Schrodinger, LLC, New York
40. Friesner RA, Banks JL, Murphy RB, Halgren TA, Klicic JJ, Mainz
DT, Repasky MP, Knoll EH, Shelley M, Perry JK, Shaw DE, Fran-
cis P, Shenkin PS (2004) Glide: a new approach for rapid, accurate
docking and scoring. 1. Method and assessment of docking accu-
racy. J Med Chem 47:1739–1749. doi:10.1021/jm0306430
41. GOLD, version 4.0 (2000) Cambridge Crystallographic Data Cen-
tre, Cambridge
42. Jones G, Willett P, Glen RC, Leach AR, Talor R (1997) Develop-
ment and validation of a genetic algorithm for flexible docking. J
Mol Biol 267:727–748. doi:10.1006/jmbi.1996.0897
43. Eldridge MD, Murray CW, Auton TR, Paolini GV, Mee
RP (1997) Empirical scoring functions: I. the development of a
fast empirical scoring function to estimate the binding affinity of
ligands in receptor complexes. J Comput Aided Mol Des 11:425–
445. doi:10.1023/A:1007996124545
44. Mooij WTM, Verdonk ML (2005) General and targeted statisti-
cal potentials for protein–ligand interactions. Proteins 61:272–287.
doi:10.1002/prot.20588
45. LIGANDFIT, Discovery Studio Modeling Environment Release,
version 2.1 (2008) Accelrys Software Inc, San Diego
46. Muegge I (2006) PMF scoring revisited. J Med Chem 49:5895–
5902. doi:10.1021/jm050038s
47. Gehlhaar DK, Verkhivker GM, Rejto PA, Sherman CJ, Fogel DB,
Freer ST (1995) Molecular recognition of the inhibitor AG-1343
by HIV-1 Protease: conformationally flexible docking by evolu-
tionary programming. Chem Biol 2:317–324. doi:10.1016/1074-
5521(95)90050-0
48. Ajay NJ (2006) Scoring functions for protein-ligand docking. Curr
Protein Pept Sci 7:407–420. doi:10.2174/138920306778559395
49. Krammer A, Kirchhoff PD, Jiang X, Venkatachalam CM, Wald-
man M (2005) LigScore: a novel scoring function for predicting
binding affinities. J Mol Graph Model 23:395–407. doi:10.1016/j.
jmgm.2004.11.007
50. Meng EC, Shoichet BK, Kuntz ID (1992) Automated docking with
grid-based energy evaluation. J Comput Chem 13:505–524. doi:10.
1002/jcc.540130412
51. MacroModel, version 9.8 (2010) Schrödinger, LLC, New York
52. QikProp, version 3.3 (2010) Schrödinger, LLC, New York
53. Roy K (2007) On some aspects of validation of predictive quan-
titative structure-activity relationship models. Expert Opin Drug
Discov 2:1567–1577. doi:10.1186/1752-153X-2-S1-P9
54. Deb PK, Balakumar C, Rao AR, Roy PP, Roy K (2011) QSAR
of adenosine receptor antagonists: exploring physicochemical
requirements for binding of pyrazolo[4,3-e]-1,2,4-triazolo[1,5-
c]pyrimidine derivatives with human adenosine A3receptor sub-
type. Bioorg Med Chem Lett 21:818–823. doi:10.1016/j.bmcl.
2010.11.094
123
Mol Divers (2012) 16:803–823 823
55. Everitt BS, Landau S, Leese M (2001) Cluster analysis. Arnold,
London
56. Roy PP, Roy K (2008) On some aspects of variable selection for
partial least squares regression models. QSAR Comb Sci 27:302–
313. doi:10.1002/qsar.200710043
57. Roy K, Paul S (2009) Exploring 2D and 3D QSARs of 2,4-
Diphenyl-1,3-oxazolines for Ovicidal Activity Against Tetrany-
chus urticae. QSAR Comb Sci 28:406–425. doi:10.1002/qsar.
200810130
58. Darlington RB (1990) Regression and linear models. McGraw-
Hill, New York
59. Mitra I, Saa A, Roy K (2009) Quantitative structure–activity rela-
tionship modeling of antioxidant activities of hydroxybenzalac-
etones using quantum chemical, physicochemical and spatial
descriptors. Chem Biol Drug Des 73:526–536. doi:10.1111/j.1747-
0285.2009.00801.x
123
