Content uploaded by Heena Bhojwani
Author content
All content in this area was uploaded by Heena Bhojwani on Mar 14, 2019
Content may be subject to copyright.
Current Computer-Aided Drug Design
Current
Computer-Aided Drug Design
ISSN: 1573-4099
eISSN: 1875-6697
Impact
Factor:
0.732
Send Ord ers for R eprints to reprints@benthamscience.ae
186
Current Computer-Aided Drug Design, 2017, 13, 186-207
RESEARCH ARTICLE
Pharmacophore and Docking Guided Virtual Screening Study for
Discovery of Type I Inhibitors of VEGFR-2 Kinase
Heena R. Bhojwani and Urmila J. Joshi*
Department of Pharmaceutical Chemistry, Prin. K.M. Kundnani College of Pharmacy, University of Mumbai, Mumbai,
Maharashtra, India
A R T I C L E H I S T O R Y
Received: July 01, 2016
Revised: September 01, 2016
Accepted: November 25, 2016
DOI:
10.2174/1386207319666161214 112536
Abstract: Background: Kinase domain of VEGFR-2 displays conformational flexibility which leads
to existence of two kinds of inhibitors viz. type I and type II inhibitors. They exhibit different binding
modes and this necessitates the development of separate pharmacophore models for them.
Methods: The virtual screening study for discovery of type I inhibitors of VEGFR-2 kinase was done
by using combined pharmacophore (generated using PHASE and validated by 3D-QSAR) and dock-
ing (Glide) based approach. Validated pharmacophore was used as preliminary filter followed by
docking. ADME properties were predicted for retrieved hits using QikProp.
Results: ADHRR.94 with statistical parameters r2
test 0.94, r2
training 0.99, SD 0.0766, r2 0.9861, F 283.3,
RMSE 0.2605, q2 0.8115 and Pearson’s R 0.9723was identified as the best pharmacophore hypothesis
for type I inhibitors of VEGFR-2 kinase. Virtual screening study was done for Asinex Elite Libraries
comprising of 104400 molecules using ADHRR.94, HTVS docking and XP docking that resulted in
twelve hits. Asinex ligand 5686 with docking score of -10.48kcal/mol was top-ranking hit. It made
two hydrogen bonding interactions with Cys 919, one as an acceptor and other as a donor, which are
characteristic of type I inhibitors. Additional interactions observed were -cation with Lys 868 and -
stacking with Phe 1047.Twelve hits had acceptable values for ADME properties.
Conclusion: Twelve hits with best obtained docking scores ranging from -10.48 to -7.23 kcal/mol and
mimicking characteristic type I inhibitor interactions were identified which could be probable inhibi-
tors of VEGFR-2.
Keywords: DFG-motif, docking, virtual screening, pharmacophore, quantitative structure-activity relationship, type I inhibitor,
VEGFR-2.
1. INTRODUCTION
Protein phosphorylation is a widely used mechanism
catalyzed by protein kinases that reversibly regulate the pro-
tein conformation and function [1]. Protein kinases phos-
phorylate tyrosine or serine-threonine residues and control
numerous energy dependent cellular processes such as cell
growth, differentiation, migration, and communication [2, 3].
Vascular Endothelial Growth Factor Receptor-2 (VEGFR-2)
is a receptor tyrosine kinase which is a major mediator of
these cellular processes and considered as an important tar-
get in the treatment of cancer [4-6].
VEGFR-2 consists of a single polypeptide chain that
spans the cellular membrane as is the case of several other
kinases [7]. It consists of extracellular and intracellular do-
mains [7, 8]. The extracellular domain consists of seven im-
munoglobulin-like domains which serve as a binding site
*Address correspondence to this author at the Department of Pharmaceuti-
cal Chemistry, Prin. K.M. Kundnani College of Pharmacy, University
of Mumbai, Mumbai, Maharashtra, India; Tel +91-9869612731;
Fax +91-22-2216 5282; E-mail: urmila.joshi1365@gmail.com
for the Vascular Endothelial Growth Factor (VEGF) while
the cytoplasmic or intracellular domain is folded into two
lobes, an N-terminal lobe and a C-terminal lobe. These two
lobes are joined by a segment called as the hinge region [7-
9]. The N-terminal lobe is smaller and contains a C-helix
including a functionally important glycine-rich nucleotide
binding loop [6-9]. The C-lobe is large and includes the cata-
lytic domain and the activation loop. Between these two
lobes lies the hydrophobic cleft also called as the catalytic
cleft where the ATP binds [5, 7, 9].
VEGFR-2 like other receptor tyrosine kinases displays
high conformational flexibility. The kinase domain can adopt
at least two conformations, the closed (inactive) and the open
(active) state [6, 8, 10]. The transition from inactive to the
active state involves a change in the orientation of the activa-
tion loop as well as C-helix [8, 10, 11]. The conformation
taken by the activation loop affects the orientation of Phe
1047 viz. ‘F’ of ‘DFG-motif’. In the ‘DFG-in’ viz. active
conformation Asp 1046 orients itself towards the ATP-
binding cleft and the Phe 1047 is buried in a hydrophobic
pocket or an allosteric pocket which lies adjacent to the
1875-6697/17 $58.00+.00 © 2017 Bentham Science Publishers
Personal Use Only
Not For Distribution
Pharmacophore and Docking Guided Virtual Screening Study Current Computer-Aided Drug Design, 2017, Vol. 13, No. 3 187
ATP-binding site [7, 8, 10, 11]. The flipping of the DFG-
motif relative to the active conformation causes the Phe 1047
side-chain to occupy the ATP-binding cleft and exposes the
hydrophobic pocket adjacent to ATP-binding site resulting in
the ‘DFG-out’ or the inactive conformation [7, 8, 10-12].
DFG-in/DFG-out states are also referred with respect to the
position of the C-helix [13]. The movement of C-helix
disrupts a salt bridge between a conserved Glu 885 and a
catalytic Lys 868 amino acid residue, which is critical for the
appropriate orientation of ATP in the catalytic site [7, 13].
Depending upon the conformational state of VEGFR-2,
the inhibitors are classified into type I and II [11, 14]. A ma-
jority of approved inhibitors of VEGFR-2 kinase are type I
inhibitors targeting the ATP binding pocket and forming one
to three hydrogen bonds with the hinge region [11]. The type
I inhibitors inhibit the kinases in DFG-in form but they are
capable of recognizing the DFG-out conformation as well
[11, 15, 16]. Type II inhibitors are also competitive inhibi-
tors of VEGFR-2 but in an indirect manner as they occupy
the allosteric pocket [10, 14, 16]. They form hydrogen bonds
with the Glu 885 of C-helix and Asp 1046 of DFG-motif
[14, 16]. However, they may extend and form additional
hydrogen bonds like type I inhibitors. Type II inhibitors spe-
cifically inhibit the kinase in DFG-out conformation [14].
Beyond these two preliminary types, there exist two more
types of inhibitors that have not been explored much [17].
The type I1/2 inhibitors bind exclusively in DFG-in con-
formation and bind to the ATP binding site like type I
compounds but they also interact with residues that are
partially involved in the Type II pharmacophore [16, 17].
The other unexplored category is of type III inhibitors
which do not occupy the ATP binding site and differ from
type I as they result in the displacement of C helix [17].
The inhibitors of VEGFR-2 reported in literature belong
to chemical classes such as quinazoline [18], thieno-
pyrimidine [19], furopyrimidine [20], indazole [18], substi-
tuted pyrazolone [21], biphenyl derivatives [22], indenopyr-
rolocarbazolone [23], 2,4-dihydrobenzoxazine [24], benzox-
azole [25] and benzimidazole [25]. Some of the clinically
successful VEGFR-2 inhibitors include sunitinib, sorafenib,
vandetanib, pazopanib and regorafenib while many other are
in clinical trials.
Pharmacophore models have been reported in the litera-
ture for various chemical classes of VEGFR-2inhibitors [26–
28]. Since the binding modes of type I and II VEGFR-2 in-
hibitors are different, a single pharmacophore model cannot
depict the structural features required for binding of these
two different types of inhibitors [14, 16]. Taking this into
consideration, we decided to do pharmacophore modeling
specifically for VEGFR-2 type I inhibitors and use it pre-
liminarily for virtual screening. The hits obtained from the
virtual screening were then submitted for docking analysis in
VEGFR-2 crystal structure present in DFG-in or active state
for interaction-based selection.
2. MATERIALS AND METHODS
2.1. Dataset
In the present study, a set of 35 type I inhibitors of
VEGFR-2 kinase were taken from the literature with their in-
vitro enzyme inhibitory data [17, 29]. These compounds,
reported in the literature, belonging to anilinoaryloxazole
class were used for pharmacophore modeling. No pharma-
cophore model has been reported for this class. The inhibi-
tory activity values of the compounds of the dataset in terms
of IC50 for the compounds belonging to dataset ranged from
15 nM to 1410 nM. The inhibitory activity was converted to
pIC50 and used in the study. The dataset was rationally di-
vided into a training set (26 compounds) and a test set (9
compounds) as reported in Table 1.
2.2. Ligand Preparation
Ligands were drawn using 2D-sketcher of Maestro 9.3
module of Schrodinger’s suite. The ligand preparation using
LigPrep 2.5 was done first by converting the structure format
from 2D to 3D. The hydrogen atoms were added to ensure an
all-atom structure and probable ionization state for ligands at
a user-defined pH of 7.4 was determined. Specified chirali-
ties were retained and no tautomers were generated. Con-
formers were generated by using ConfGen and energy-
minimized by applying Optimized Potential for Liquid
Simulations viz. OPLS 2005 force field from the Macro
Model options. A maximum of 1,000 conformers were gen-
erated per structure and 10,000 minimization steps were
used. The search mode for ConfGen sampling mode was set
to ‘rapid’. Default parameter for amide bonds was selected.
An energy cut-off 10kcal mol-1 was set as for elimination of
high energy conformers in comparison to conformer with the
lowest energy. The conformational search employed that an
RMSD value of 1.0 Å was used to eliminate redundant con-
formers [30].
2.3. Pharmacophore Model Generation
PHASE, Schrodinger, LLC, New York, NY, 2013 was
used to generate pharmacophore and 3D-QSAR models for
VEGFR-2 type I kinase inhibitors. A common pharma-
cophore hypothesis is an abstract description of molecular or
chemical features common to two or more active ligands
which are necessary for a ligand in order to bind with the
receptor. PHASE includes a set of six built-in features that
can be used to generate a pharmacophore model. These fea-
tures include a hydrogen bond acceptor (A), a hydrogen
bond donor (D), a hydrophobic group (H), and a negatively
ionizable (N), positively ionizable (P), and aromatic ring (R)
that define chemical features of ligands [31-33]. The pre-
pared ligands were imported. Their pIC50 ranged from 7.824
to 5.851. The dataset of prepared ligands was divided into a
set of active, intermediates and inactive by providing a
threshold for pIC50 as 7.640 for active and pIC50 as 6.070 for
inactive molecules. This division of dataset resulted into 5
active, 5 inactive and 25 moderately active molecules. The
dataset ligands were subjected to the create sites step which
resulted in the site feature frequencies (A, D, H, N, P and R)
for each conformer of each ligand [33]. In the find common
pharmacophore step, a variant list was obtained by specify-
ing the number of sites as five for the pharmacophores to be
generated. The set of features common to the five actives
was utilized to generate variant list [33]. Common pharma-
cophores were identified from this list of variants using a
tree-based partitioning technique that grouped similar
pharmacophores together according to their intersite
Personal Use Only
Not For Distribution
188 Current Computer-Aided Drug Design, 2017, Vo l. 13 , No. 3 Bhojwani and Joshi
Table 1. Molecular structures and pIC50 values of training and test set compounds using phase pharmacophore model ADHRR.94.
Compound Number Molecular Structure Pharm Set Actual pIC50 Predicted pIC50
(1). (T)
- 7.301 7.09
(2).
- 7.538 7.43
(3).
- 7.041 7.26
(4). (T)
active 7.796 7.48
(Table 1) Contd…
Personal Use Only
Not For Distribution
Pharmacophore and Docking Guided Virtual Screening Study Current Computer-Aided Drug Design, 2017, Vol. 13, No. 3 189
Compound Number Molecular Structure Pharm Set Actual pIC50 Predicted pIC50
(5).
active 7.824 7.82
(6).
active 7.657 7.53
(7).
- 7.119 7.09
(8).
- 7.108 7.11
(Table 1) Contd…
Personal Use Only
Not For Distribution
190 Current Computer-Aided Drug Design, 2017, Vo l. 13 , No. 3 Bhojwani and Joshi
Compound Number Molecular Structure Pharm Set Actual pIC50 Predicted pIC50
(9). (T)
- 7.602 7.28
(10).
- 7.102 7.18
(11).
- 7.051 7.11
(Table 1) Contd…
Personal Use Only
Not For Distribution
Pharmacophore and Docking Guided Virtual Screening Study Current Computer-Aided Drug Design, 2017, Vol. 13, No. 3 191
Compound Number Molecular Structure Pharm Set Actual pIC50 Predicted pIC50
(12). (T)
- 7.409 7.22
(13).
inactive 5.921 5.92
(14).
inactive 6.06 6.04
(15).
- 6.42 6.38
(Table 1) Contd…
Personal Use Only
Not For Distribution
192 Current Computer-Aided Drug Design, 2017, Vo l. 13 , No. 3 Bhojwani and Joshi
Compound Number Molecular Structure Pharm Set Actual pIC50 Predicted pIC50
(16). (T)
inactive 6.032 6.25
(17).
inactive 5.851 5.78
(18).
- 6.301 6.36
(19). (T)
- 6.143 6.45
(Table 1) Contd…
Personal Use Only
Not For Distribution
Pharmacophore and Docking Guided Virtual Screening Study Current Computer-Aided Drug Design, 2017, Vol. 13, No. 3 193
Compound Number Molecular Structure Pharm Set Actual pIC50 Predicted pIC50
(20).
inactive 6.06 6.14
(21).
- 6.301 6.27
(22).
- 6.854 6.87
(23).
- 6.769 6.78
(Table 1) Contd…
Personal Use Only
Not For Distribution
194 Current Computer-Aided Drug Design, 2017, Vo l. 13 , No. 3 Bhojwani and Joshi
Compound Number Molecular Structure Pharm Set Actual pIC50 Predicted pIC50
(24).
- 7 6.97
(25).
- 7.638 7.59
(26).
- 6.444 6.39
(27).
active 7.657 7.23
(Table 1) Contd…
Personal Use Only
Not For Distribution
Pharmacophore and Docking Guided Virtual Screening Study Current Computer-Aided Drug Design, 2017, Vol. 13, No. 3 195
Compound Number Molecular Structure Pharm Set Actual pIC50 Predicted pIC50
(28). (T)
- 6.851 6.86
(29).
- 7.444 7.38
(30).
- 7.032 7.13
(31). (T)
- 7.06 7.08
(Table 1) Contd…
Personal Use Only
Not For Distribution
196 Current Computer-Aided Drug Design, 2017, Vo l. 13 , No. 3 Bhojwani and Joshi
Compound Number Molecular Structure Pharm Set Actual pIC50 Predicted pIC50
(32).
- 7.091 7.14
(33).
- 7.102 7.14
(34). (T)
- 7.292 7.29
(35).
active 7.745 7.76
Note: (T) indicates molecule was included in test set.
Personal Use Only
Not For Distribution
Pharmacophore and Docking Guided Virtual Screening Study Current Computer-Aided Drug Design, 2017, Vol. 13, No. 3 197
distances [33]. The minimum distance allowed between two
features viz. the intersite distance was kept 2.0 Å. If the fea-
tures in the ligand were closer than this distance, then the
hypothesis was rejected. A tree depth of five was used and
therefore, the initial and final box sizes were set to be 32Å
and 1Å [33, 34]. A total 1670 five point common pharma-
cophore hypotheses belonging to different variants were
identified from a set of five active ligands and submitted to
scoring [33].
2.3.1. Scoring Pharmacophore Hypothesis
In this step, common pharmacophore hypotheses were
examined, and a scoring procedure was applied to identify
the pharmacophore from each surviving n-dimensional box
that yields the best alignment of the chosen actives [32, 33].
A surviving box contains a set of very similar pharmacopho-
res gathered from conformations of a minimum number of
active-set ligands, and certain of these ligands contribute
more than one pharmacophore to a box [32, 33]. Thus each
pharmacophore and its associated ligand were treated tempo-
rarily as a reference in order to assign a score [31-34]. Each
hypothesis was scored with respect to actives and keeping
the threshold for RMS deviation of the intersite distances of
any contributing ligand from those of the reference ligand as
1.2 Å. The threshold for the variation in the alignment of
vectors between any contributing ligand and the reference
ligand was kept as 0.5. The cutoff on the fraction or number
of hypotheses to be retained was set as top 10 percent [33].
The survival scores for active and inactive molecules were
calculated. The adjusted survival score was also calculated
by subtracting the multiple of survival score of inactive from
survival score of actives. The survival, survival-inactive and
adjusted scores are given in Table 2 [32-35].
2.3.2. Building QSAR Models
The hypotheses after surviving the scoring step were sub-
jected to 3D-QSAR studies. The QSAR model was devel-
oped for ligands belonging to the training set with a range of
activities as well as structural diversity [31, 33, 35, 36]. In
our study, the atom types used for QSAR model were hydro-
gen bond donors (D); hydrophobic/non-polar (H); hydrogen
bond acceptors/electron withdrawing (W) and others (X)
[31]. The training set molecules were aligned and placed in a
rectangular grid [31-33, 35]. This grid divided the space into
uniformly sized cubes of dimension 1A° on each side. These
cubes were occupied by the atoms or pharmacophore sites
that define each molecule. Each occupied cube gave rise to
one or more volume bits, where a separate bit is allocated for
each different category of atom/site that occupied the cube.
The total number of volume bits assigned to a given cube
was based on occupations from all training set molecules. A
molecule could be represented by a string of zeros and ones,
according to the cubes it occupied. This bit string was a col-
lection of binary valued 3D descriptors and these bits were
treated as a pool of independent variables for the purpose of
QSAR model development employing the partial least
squares (PLS) method. The number of these descriptors was
much greater than the number of molecules in training set
[31, 33].
2.3.3. Validation of Pharmacophore Models
The main purpose of developing the QSAR model was to
validate the generated model which would be statistically
robust. The dataset was divided into a training set (26 com-
pounds) and a test set (9 compounds) [31, 33, 35]. As men-
tioned in the above section, atom-based 3D-QSAR models
were generated for hypotheses using compounds in the train-
ing set. The strength of the developed pharmacophore hy-
potheses was validated by statistical parameters which in-
cluded the standard deviation of regression, Pearson’s corre-
lation coefficient (Pearson’s R), squared correlation coeffi-
cient (r2), q2 (r2 for test set), stability, and the variance ratio
(F) [31, 33, 35]. The predicted pIC50 at the 5th partial least-
squares (PLS) factor is shown in Table 1. Increasing the
number of PLS factors did not improve the statistics or pre-
dictive ability of the model. QSAR validation parameters
were used for selection of best pharmacophore hypothesis
[33].
2.4. Molecular Docking
The dataset used for the pharmacophore generation was
subjected to molecular docking studies. These studies were
done using Glide version 5.9software and the crystal struc-
ture of VEGFR-2 kinase (PDB code: 2P2H), complexed with
an inhibitor with ligand identifier 994 [37]. The protein data-
bank reports three mutations C817A, E990V and V916T for
VEGFR2 protein with PDB ID 2P2H. These have not been
converted to wild-type form. The protein was refined using
the protein preparation wizard. All the water molecules that
Table 2. Survival scores for the top ranked pharmacophore models.
Hypotheses ID Survival Survival-Inactive Adjusted Score Energy
ADHRR.94 6.293 1.453 4.84 4.86
ADHRR.26 6.224 1.552 4.672 4.86
ADRRR.19 6.038 1.601 4.437 2.077
AADHR.20 6.019 1.285 4.734 5.187
AAADR.61 5.609 1.307 4.302 5.135
AAADR.246 5.541 1.795 3.746 2.077
Note: Survival represents score for actives after aligning hypotheses on a pharmacophore; Survival-inactive represents score for inactives after aligning hypotheses on a pharma-
cophore; adjusted score is the score obtained on subtraction of multiple of survival-inactive score from survival active score; Energy represents relative energy of the reference ligand
in kcal/mol.
Personal Use Only
Not For Distribution
198 Current Computer-Aided Drug Design, 2017, Vo l. 13 , No. 3 Bhojwani and Joshi
did not form any interactions or were not a part of the active
site were deleted. Hydrogen atoms were added to the protein
which included the protons necessary to define the correct
ionization and tautomeric states of the amino acid residues.
The added hydrogen bonds were optimized. Protein structure
was energy minimized using the impact refinement module.
The steric clashes existing in the structures were removed
using OPLS 2005 force field. Minimization was terminated
at a default RMSD cutoff of 0.30 Å [31, 37].
After preparation of the protein, a receptor grid was gen-
erated using the receptor grid generation panel. VEGFR-2
protein structure and 994 were included as workspace entries
and ligand 994 was picked to be excluded from receptor grid
generation using the ‘receptor’ tab. Default vdW scaling
factor of 1.0 and charge scale factor of 0.25 were used. Glide
uses two boxes for receptor grid calculations. The two boxes
are enclosing or the outer box and bounding or the inner box.
Enclosing box was generated by specifying the centroid of
co-crystallized ligand as the centre for this box. Size section
of the ‘site’ tab provided options for the size of the enclosing
box. The default option of docking ligands similar in size to
the co-crystallized ligand was used. The centre for the en-
closing box acted as the centre for bounding box too. The
default setting of 10 Å on each side was used for the bound-
ing box from the advanced setting option of the ‘site’ tab in
the receptor grid generation panel. According to these speci-
fications for the centre and size the final enclosing box was
generated. No constraints were used [31, 37, 38].
The docking protocol was validated by removing the in-
hibitor from the complex, re-docking (as shown in Fig. 1)
and calculating root mean square deviation (RMSD). RMSD
value of 0.7 Å between the docked conformation of the in-
hibitor and native conformation depicted the accuracy of the
docking program. In order to further assess the accuracy, the
interactions reproduced on re-docking were checked with
that of the interactions produced in native conformation by
ligand 994 [31, 38]. It was observed that the conformation in
which 994 got re-docked produced exactly same hydrogen
bonding interactions with Cys 919 of hinge region. The ni-
trogen atom of triazine ring of 994 acted as hydrogen bond
donor from the backbone amide group of Cys 919 while the
small NH-linker of 994 acted as a hydrogen bond donor to
the backbone carbonyl of Cys 919. Hydrophobic interactions
were observed for triazine ring with region formed by Val
867, Ala 866, and Val 899(Supplementary Information Fig.
S1). An acceptable RMSD value of 0.7 and reproducibility
of important interactions indicated that the docking program
could be used [31, 38].
The molecules belonging to the dataset used in pharma-
cophore were docked using flexible mode of ligand docking.
The receptor grid generated for VEGFR-2 protein crystal
structure 2P2H was specified in the Glide docking panel.
The precision was docking was set to XP. XP descriptors
were selected to be written. Flexible mode of ligand docking
was used which allowed the conformations to be internally
generated during the docking process. The default parame-
ters for energy minimization specifying a distance-dependent
dielectric constant of 2.0 and a maximum number of 100
conjugate gradient steps were used. No constraints of simi-
larity scoring were applied [31, 37-39].
2.5. Virtual Screening
We adopted a two steps method for virtual screening.
Asinex Elite Libraries contains about 1, 04,400 commer-
cially available molecules and this was the database to be
screened. In the first step, the conformers were generated for
the database molecules and the best performing pharma-
cophore hypothesis was used as a query for screening data-
base molecules using the ‘Find Matches’ option in PHASE
software for identifying molecules that aligned well with the
query hypothesis [31, 33]. This gave us 8570 molecules
which mapped on all the five features of the pharmacophore
hypothesis used as query. In the second step, we subjected
top 10% of the hits (857 molecules) obtained by pharma-
cophore based screening for docking. The HTVS module
implemented in Glide was used for docking of the top hits
into the crystal structure of the VEGFR-2 kinase (PDB ID:
2P2H), co-crystallized with an inhibitor of type I. The G-
score obtained in docking was used to rank the hits further.
Now, 10% of the ranked hits (85 molecules), arising out of
HTVS docking were taken further for detailed docking stud-
ies using extra precision docking. The extra precision
method was used in the later stage as it helps to eliminate out
false positives as well as provides a better correlation be-
tween good poses and good scores. These studies helped us
to filter the hits additionally on the basis of XP G-score and
observe the interactions with the receptor.
2.6. QikProp Descriptors
These hits obtained from the database subjected to fur-
ther filtering using physicochemical/ molecular properties
that were calculated by QikProp. Default parameters were
used for QikProp [40].
3. RESULTS AND DISCUSSION
Literature reports the pharmacophore models for piperi-
done derivatives [41], quinoline derivatives [28], N-Phenyl-
N’-{4-(4-quinolyloxy) phenyl} urea and 4-aminopyrimidine-
5-carbaldehyde oxime [27] and thiazole-substituted pyra-
Fig. (1). Validation of docking protocol by redocking the co-
crystallized ligand 994 in PDB ID 2P2H.
GLY 841
PRO 839
GLY 837
LEU 840
PRO 1068
GLN 847
ILE 849
VAL 848
LEU 836
GLU 850
LYS 868
LYS 835
VAL 867
LEU 1049
ASP 852
ASP 1046
ARG 1027
ALA 866
CYS 1045
PHE 918
THR 864
VAL 865
CYS 919
PHE 1091
LE
GLY
LEU
LYS 920
LEU 1035
SER 1090
ILE 1034
LEU 1029
ASP 1828 ASN 1033
ALA 1030
ALA 1031
ASN 923
GLY 922
ARG 1032
TRP 1071
LYS 1070
VAL 1
LEU 834
ANG Y42 LTS 858
Personal Use Only
Not For Distribution
Pharmacophore and Docking Guided Virtual Screening Study Current Computer-Aided Drug Design, 2017, Vol. 13, No. 3 199
zolone [26] derivatives as VEGFR-2 inhibitors. Most of the
reported pharmacophore models do not take into considera-
tion the different binding modes of type I and II inhibitors.
The models reported in literature do not depict structural
features required for any one specific type of inhibitor.
Compounds belonging to anilinoaryloxazole class have been
reported to be type I kinase inhibitors of VEGFR-2 kinase
[17]. Keeping this in view, our study involved generation of
the pharmacophore model for type I inhibitors. The gener-
ated model was used as a primary filter in virtual screening
followed by docking and interaction studies.
3.1. Pharmacophore Modeling and QSAR Results
A number of common pharmacophore hypotheses were
generated for type I inhibitors of VEGFR-2 kinase using
PHASE module of the Schrodinger’s molecular modeling
software [35]. These hypotheses were generated using a set
of five active ligands which contained important structural
features majorly being hydrogen bond acceptor (A) and do-
nor (D) that happen to be crucial for binding at the active site
of the VEGFR-2 kinase domain [16, 18]. A total of 1670
five-point hypotheses belonging to 15 variants (AAAAD,
AAAAH, AAAAR, AAADH, AAADR, AAAHR, AAARR,
AADHR, AADRR, AAHRR, AARRR, ADHRR, ADRRR,
AHRRR and DHRRR) were generated. All these generated
hypotheses were evaluated on the basis of survival, survival -
inactive and adjusted survival score [35]. In general, a good
hypothesis should provide superior alignment with active
compounds and display the lowest possible relative confor-
mational energy values for reference ligand [33]. Besides, a
good hypothesis should be able to discriminate between ac-
tive and inactive compounds. Only 146 pharmacophore hy-
potheses were able to survive the stringent scoring procedure
and were ranked on the basis of these scores. The survival
scores of pharmacophore hypotheses have been given in Ta-
ble 2. Since hydrogen bond donor and acceptor happen to be
requisite pharmacophoric features for type I inhibitors of
VEGFR-2, the hypotheses possessing good survival scores
along with at least one donor or acceptor feature were sub-
ject to validation by QSAR analysis.
QSAR models were generated to establish the statistical
robustness of the pharmacophore models [31, 35]. Atom-
based 3D-QSAR models were developed using the partial
least-squares procedure [26]. This was initiated by partition-
ing of the dataset of compounds into a training set and test
set based on their chemical structure and activity-span. Three
active including the most active, three inactive including the
least active and twenty moderately active compounds with
structural variations were used in the training set. The test set
consisted of two active, two inactive and five moderately
active molecules. Additionally, we ensured that two
chemically most similar ligands were assigned to two dif-
ferent sets. This enabled us to introduce maximum diver-
sity in both training and test sets. The training set was used
for generation of 3D-QSAR models by regression analysis
[31, 33, 35]. Partial Least Square method was used for the
same. The optimum number of factors of the partial least-
squares based regression was found to be 5 as this yielded
minimum residual variation [33].
Parameters such as SD, r2, and the F-value were used to
evaluate the training set predictions while q2, RMSE and
Pearson-R were used to evaluate the test set predictions [33].
The acceptance criteria for selection was a higher F value
than the critical value, value of SD <0.35, RMSE < 0.4,
Pearson-R > 0.7, r2 > 0.8, and q2> 0.6. Since six of our hy-
potheses performed better than the normally accepted crite-
ria, with r2 > 0.9, and q2> 0.8, we selected these hypotheses
for further evaluation [42]. Top six hypotheses on the
basis of their fitting in the acceptance criteria, with their sta-
tistical parameters as mentioned in Table 3 were selected.
These top-ranked five-point hypotheses were ADHRR.94,
ADHRR.26, ADRRR.19, AADHR.20AAADR.61, and
AAADR.246.
It was not only important for a chosen hypothesis to be
statistically robust but it should describe the essential fea-
tures for binding as well. After scoring and validation, it was
much easier to select the most descriptive hypotheses as
one will have a practical number to evaluate. A close look
at these hypotheses indicated that AAADR.61 and
AAADR.261 possessed three acceptors and a donor. How-
ever, any type I inhibitor can form between one to three hy-
drogen bonds with the hinge region [11,16,18]. Therefore,
although statistically significant these hypotheses do not
reflect all the essential features of type I inhibitors and
were eliminated. Further, a two steps evaluation was done in
order to identify a hypothesis that can be used for virtual
screening. The fitness scores for inactives were checked.
ADRRR.19 was found to be having a fitness score of greater
Table 3. 3D-QSAR parameters for pharmacophore models generated.
Hypotheses ID SD r2 F RMSE q2 Pearson-R
ADHRR.94 0.0766 0.9861 283.3 0.2605 0.8115 0.9723
ADHRR.26 0.0931 0.9794 190.6 0.2534 0.8216 0.9692
ADRRR.19 0.0768 0.986 282.4 0.2375 0.8434 0.9451
AADHR.20 0.0871 0.982 218.3 0.2524 0.8231 0.967
AAADR.61 0.0912 0.9803 199 0.2475 0.8299 0.9588
AAADR.246 0.0932 0.9794 190.3 0.2427 0.8363 0.9544
Note: SD, the standard deviation of regression; r2, a coefficient of determination; F, the ratio of the model variance to the observed activity variance; q2, directly analogous to r2 but
based on the test set
p
redictions; RMSE, the RMS error in the test set
p
redictions; Pearson-R, value for the correlation between the
p
redicted and observed activit
y
for the test set.
Personal Use Only
Not For Distribution
200 Current Computer-Aided Drug Design, 2017, Vo l. 13 , No. 3 Bhojwani and Joshi
than 2.0 indicating a good alignment for the inactives (Table
S1, Supplementary Material). It appeared from this analysis
that ADRRR.19 in spite of being stable cannot distinguish
appropriately between actives and inactives, leading to its
elimination. The next step in evaluation was visual inspec-
tion of alignment of active molecules on a practical number
of remaining three hypotheses. When the actives were
aligned on ADHRR.94, AADHR.20 and ADHRR.26, it was
observed that the pharmacophore features of AADHR.20 and
ADHRR.26 did not map on the entire molecule whereas
ADHRR.94 could map the features over greater parts of ac-
tive molecules (Table S2, Supplementary Material).
This prompted us to select ADHRR.94 with r2
test 0.94,
r2
training 0.99, SD 0.0766, r2 0.9861, F 283.3, RMSE 0.2605,
q2 0.8115, and Pearson’s R 0.9723 as the best suitable for
virtual screening of type I inhibitors. The predicted pIC50
obtained from ADHRR.94 at the fifth partial least-squares
(PLS) factor is shown in Table 1. The plot of experimental
activity v/s phase predicted activity for ADHRR.94 with r2
test
0.94 and r2
training 0.99 are shown in Fig. (2a and b), respec-
tively. The distance between the pharmacophoric points has
been shown in Fig. (3) for the best pharmacophore hypothe-
sis ADHRR.94. The superposition of the most active com-
pound, least active compound, most fit compound and all
active compounds on the best pharmacophore hypothesis
ADHRR.94 has been given in Fig. (4).
Fig. (2). Scatter plot for the predicted and actual pIC50 values ob-
tained for dataset divided into test set (a) and training set (b) using
pharmacophore hypothesis ADHRR.94.
Fig. (3). Intersite distance (in Å) in the geometry of the pharma-
cophore hypothesis (ADHRR.94).
Note: Red spheres with vectors represent acceptor feature, blue
spheres with vectors represent donor feature, green spheres represent
hydrophobic group and orange rings represent aromatic feature.
(The color version of the figure is available in the electronic copy
of the article).
Fig. (4). Representation of superposition of (A) most fit, (B) most
active, (C) least active and (D) all active compounds on the phar-
macophore hypothesis ADHRR.94.
Note: Red spheres with vectors represent acceptor feature, blue
spheres with vectors represent donor feature, green spheres repre-
sent hydrophobic group and orange rings represent aromatic feature.
(The color version of the figure is available in the electronic copy
of the article).
A4
6.453
6.894
3.373
11.973
4.434
8.742
6.023
7.903
3.216
H8
R10
R11
D7
7.6
7.4
7.2
7
6.8
6.6
6.4
6.2
6
6 6.2 6.4 6.6 6.8 7 7.2 7.4 7.6 7.8 8
Actual Activity
8
7.5
7
6.5
6
5.5
5.5 6 6.5 7 7.5 8
2 (b)
Phase Predicted Activity
Actual Activity
2 (a)
Phase Predicted Activity
(A)
H8 R11
D7
A4
R10 R10
D7
R11
H8
(B)
A4
(C) (D)
R10
D7
R11
H8
A4
R10
D7
R11
H8
A4
Personal Use Only
Not For Distribution
Pharmacophore and Docking Guided Virtual Screening Study Current Computer-Aided Drug Design, 2017, Vol. 13, No. 3 201
3.2. Molecular Docking
Analysis of the active site VEGFR-2 indicates that bind-
ing of the inhibitor to the hinge region of the kinase via one
or multiple hydrogen bonds is the most important interaction
of the type I inhibitor [16, 31]. Docking of compounds be-
longing to dataset revealed that most of the active molecules
made one acceptor and one donor hydrogen bonding interac-
tion with Cys 919 of the hinge region of VEGFR-2. An addi-
tional hydrogen bonding interaction with Asn 923 was ob-
served for active molecules of the dataset. Compound 35
displayed -cation interaction with Lys 868. The study of
interactions revealed that four active molecules of dataset
had higher XP glide score than inactive molecules. The ac-
tive compounds of dataset used in pharmacophore generation
displayed a better binding as indicated by their glide energy
which is a combination of coulombic and Van der Waals
interactions. Fig. (5) shows the images of the active com-
pounds of the dataset docked into VEGFR-2 and Fig. (S2) of
supplementary material gives interactions for the same.
Fig. (5). Glide XP docking of active compounds of the dataset into the active site of crystal structure of VEGFR-2 (PDB ID: 2P2H) along
with (a) compound 4, (b) compound 5, (c) compound 27, (d) compound 6 and (e) compound 35 of the dataset. Active compounds represented
by green stick model. H-bond interactions are indicated with pink dotted lines. (The color version of the figure is available in the electronic
copy of the article).
Personal Use Only
Not For Distribution
202 Current Computer-Aided Drug Design, 2017, Vo l. 13 , No. 3 Bhojwani and Joshi
3.3. Virtual Screening
The statistically robust pharmacophore hypothesis,
ADHRR.94, was used for preliminary database screening for
retrieving the hits from Asinex Elite Libraries. The first step
consisted of using the best pharmacophore hypothesis viz.
ADHRR.94 as a query for screening the library containing 1,
04, 440 molecules. Pharmacophore-based screening resulted
in the retrieval of 8570 hits. Pharmacophore-based screening
is usually accompanied by docking in order to narrow down
the list of possible hits. Top 10% percent viz. 857 com-
pounds were subjected to virtual screening by HTVS dock-
ing. The molecules of the library were ranked on the basis of
glide score and top 10% viz. 85 compounds were subjected
to Glide-XP docking. At this point, we decided to take into
consideration both interaction patterns of hits with VEGFR-2
as well as their XP G-scores. Analysis of the interactions
made by actives used for pharmacophore generation indi-
cated atleast two H-bonding interaction, either both with Cys
919 or one with Cys 919 and another with Asn923 residues
of VEGFR-2 kinase. This was used as a criterion for reduc-
ing the number of hits. The glide score cut-off for the re-
trieval of hits was set to be more than -7.15 kcal/ mol after
taking into consideration the glide score for active com-
pounds of dataset used in pharmacophore generation. The
probable hits with their interactions are enlisted in Table 4
and ligand interactions are given in Fig. (S3) (Supplementary
Material). The docked images depicting hydrogen bonding
interactions for the 12 hits are given in Fig. (6). Asinex
ligand 5686 had an XP glide score of -10.48 kcal/ mol. It
made one hydrogen bond acceptor and one hydrogen bond
donor interaction with Cys 919. Other interactions made
were -cation with Lys 868 and - stacking with Phe 1047
[31].
3.4. Predicted ADME Properties
A molecular descriptor is a structural or physicochemical
property of a molecule. Several molecular properties are
often used to help predict the pharmacokinetic behavior of
potential drug-like compounds [43]. The 12 hits obtained
from Asinex Elite Libraries (Fig. 7) were analyzed for their
drug-likeness by assessing their physicochemical properties
and by applying Lipinski’s rule of five. The Lipinski’s rule
Table 4. Summary of docking, fitness score and PHASE predicted activity results for 12 best hits.
Asinex Elite
Library ID
Fitness
Score
PHASE Pre-
dicted Activity
XP Glide Score
(kcal/mol)
Glide
Evdw Glide Ecoul Glide Energy XP Hbond Interacting
Residues
5686 1.63 6.782 -10.48 -50.65 -4.60 -55.25 -1.29 Lys 868, Cys
919, Phe 1047
96939 1.58 6.633 -9.86 -41.16 -5.89 -47.05 -1.33
Cys 919, Phe
1047
5083 1.96 6.598 -9.82 -32.88 -14.58 -47.46 -1.55 Cys 919, Asp
1046
99248 1.78 6.364 -8.67 -38.76 -1.46 -40.23 -0.57
Lys 868, Cys
919, Phe 1047
85967 1.64 6.604 -8.42 -40.66 -2.82 -43.48 -0.78
Cys 919, Asp
1046
10679 1.69 6.398 -7.77 -37.08 -8.89 -45.97 -1.89
Cys 919, Arg
842
19811 1.65 6.300 -8.38 -44.93 -4.25 -49.18 -0.65
Glu 917, Cys
919, Phe 1047
19776 1.72 6.226 -8.57 -44.35 -4.26 -48.62 -0.67
Glu 917, Cys
919, Phe 1047
2090 1.94 6.588 -8.91 -38.20 -12.65 -50.85 -1.58
Cys 919, Asp
1046
35709 1.78 6.469 -7.43 -46.06 -2.19 -48.25 -0.70
Cys 919, Phe
1047
1420 1.63 6.604 -7.28 -31.60 -8.23 -39.83 -1.34
Thr 916, Cys
919, Asp 1046
97955 1.70 6.559 -7.63 -45.92 -3.03 -48.95 -1.20
Thr 916, Cys
919, Phe 1047
Note: Fitness score, a score that measures how well the matching pharmacophore site points align to those of the hypothesis; Phase predicted activity, Predicted activity from the 3D
QSAR model with n PLS factors; XP Glide score, Total GlideScore/ sum of XP terms; glide evdw, Van der Waals energy; glide ecoul, Coulomb energy; glide energy, Modified
Coulomb-van der Waals interaction energy; XP Hbond, Hydrogen-bonding term in the XP GlideScore.
Personal Use Only
Not For Distribution
Pharmacophore and Docking Guided Virtual Screening Study Current Computer-Aided Drug Design, 2017, Vol. 13, No. 3 203
Fig. (6). Molecular docking results: Binding orientations of 12 Asinex hits in the active site of VEGFR-2 (PDB ID: 2P2H). Screened hits
represented by green stick model. H-bond interactions are indicated with pink dotted lines. (The color version of the figure is available in the
electronic copy o f the article).
Personal Use Only
Not For Distribution
204 Current Computer-Aided Drug Design, 2017, Vo l. 13 , No. 3 Bhojwani and Joshi
Fig. (6). (continued): Molecular docking results: Binding orientations of 12 Asinex hits in the active site of VEGFR-2 (PDB ID: 2P2H).
Screened hits represented by green stick model. H-bond interactions are indicated with pink dotted lines. (The color version of the figure is
available in the electronic copy of the article).
Fig. (7). The 2D structures of the top 12 hits along with their Asinex Elite Libraries ID.
Personal Use Only
Not For Distribution
Pharmacophore and Docking Guided Virtual Screening Study Current Computer-Aided Drug Design, 2017, Vol. 13, No. 3 205
Table 5. Physicochemical descriptors and ADME properties of 12 hits.
Asinex Elite Li-
brary ID MW Do-
norHB AccptHB QPlogPo/w PSA QPlogS QPPCaco QPPMDCK % Human Oral
Absorption NRB
5686 365.41 2 7 2.73 103.29 -5.52 390.46 286.78 89.32 2
96939 359.42 2 6 3.25 86.67 -5.06 551.03 760.86 95.06 4
5083 392.46 3 7.7 2.57 115.52 -5.26 221.02 96.77 83.97 4
99248 378.45 1 6.5 3.53 88.49 -5.88 753.14 762.94 100 2
85967 365.43 2 5.5 3.99 78.82 -5.68 1663.13 857.32 100 5
10679 401.47 2 7.75 2.95 88.77 -4.99 152.72 71.8 83.30 4
19811 333.39 2 6.5 2.48 89.67 -4.44 485.50 350.32 89.52 4
19776 333.39 2 6.5 2.48 89.81 -4.45 487.01 351.53 89.57 4
2090 424.52 3 8.2 2.79 116.02 -5.52 214.53 159.64 85.00 5
35709 327.34 2 7.5 1.33 108.58 -2.31 381.54 335.72 80.93 5
1420 310.39 1 4.75 3.36 57.14 -4.42 1449.94 1237.65 100 5
97955 372.43 1 6.5 3.95 76.86 -5.03 1248.10 628.62 100 7
Note: MW - Molecular Weight (acceptable range 130.0 – 725.0).
Donor HB -Number of hydrogen bond donors (acceptable range 0.0 – 6.0).
Accpt HB - Number of hydrogen bond acceptors (acceptable range 2.0-20.0).
QPlogPo/w - Predicted octanol/water partition coefficient log p (acceptable range 2.0 to 6.5).
PSA – Van der Waals surface area of polar nitrogen and oxygen atoms and carbonyl carbon atoms. (acceptable range 7.0 -200.0).
QPlogS - Predicted aqueous solubility; S in mol/L (acceptable range -6.5 to 0.5).
QPPCaco - Predicted Caco-2 cell permeability in nm/s (acceptable range < 25 is poor and > 500 is great).
QPPMDCK - Predicted apparent MDCK cell permeability in nm/s (acceptable range < 25 is poor and > 500 is great).
% Human Oral Absorption - Percentage of human oral absorption (acceptable range < 25 % is poor and > 80 % is high).
NRB - Number of rotatable bonds.
for drug-like molecules states that the molecule should have
a molecular weight less than 500 Daltons, H-bond acceptors
less than 10 and H-bond donors and a logP less than 5 [44].
All the compounds identified as hits followed the Lipinski’s
rule of five. The permeability of a molecule depends on its
molecular size, ability to form hydrogen bonds, lipophilicity,
molecular shape and flexibility [43, 45]. Log P or the lipo-
philicity is an important factor which controls passive mem-
brane partitioning. An increase in log P enhances permeabil-
ity while reducing the solubility [45]. The selected hits had
the partition coefficient (QPlogPo/w) and water solubility
[QPlogS] that ranged between 1.33 to 3.99 and -6.22 to -
2.31, respectively [40]. Membrane permeation is a pre-
requisite for oral bioavailability. The hits were evaluated for
their Caco-2 and MDCK cell permeability in-silico as they
represent good models for permeability [40, 45]. Caco-2 cell
permeability (QPPCaco) for these hits ranged from 152.72 to
1663.22 while MDCK cell permeability (QPPMDCK)
ranged from 71.8 to 1237.64 [40]. Additionally, the pre-
dicted percentage human oral absorption for all compounds
ranged from 80.93 to 100 % [40]. Molecular flexibility de-
pends on the number of rotatable bonds and influences the
bioavailability. A compound having a number of rotatable
bonds less than 10 and a polar surface area less than or equal
to 140 Å2 is considered to possess good oral bioavailability
[43]. Identified hits were found to comply with the parame-
ters for the number of rotatable bonds as well as polar sur-
face area [40] as indicated in Table 5.
CONCLUSION
The six five-point hypotheses were generated from a
dataset of a total of thirty-five compounds belonging to type
I inhibitors and showing varying degree of inhibitory activity
against VEGFR-2 kinase after taking into consideration the
specific pharmacophoric features required for type I inhibi-
tors. These hypotheses were validated by using the 3D-
QSAR statistical parameters. ADHRR.94, a five-point hy-
pothesis consisting of one acceptor, one donor, one hydro-
phobic group and two ring features was identified as an op-
timum model with statistical parameters r2
test 0.94, r2
training
0.99, SD 0.0766, r2 0.9861, F 283.3, RMSE 0.2605, q2
0.8115 and Pearson’s R 0.9723 for type I inhibitors of
VEGFR-2 kinase. A virtual screening study was performed
for Asinex Elite Libraries comprising of 104400 molecules
using ADHRR.94, HTVS docking and XP docking that re-
sulted in twelve hits. Asinex ligand 5686 with docking score
of -10.48 kcal/ mol was top-ranking hit. It made two hydro-
gen bonding interactions with Cys 919, one as an acceptor
and other as a donor, which are characteristic of type I in-
hibitors. Additional interactions observed were -cation with
Lys 868 and - stacking with Phe 1047. The retrieved hits
were evaluated for their physicochemical properties and
drug-likeliness in-silico. Twelve hits had acceptable values
for the same. The findings of this study would be advanta-
geous in lead optimization for new type I inhibitors of the
VEGFR-2 kinase.
Personal Use Only
Not For Distribution
206 Current Computer-Aided Drug Design, 2017, Vo l. 13 , No. 3 Bhojwani and Joshi
CONFLICT OF INTEREST
The authors declare no conflict of interest, financial or
otherwise.
ACKNOWLEDGEMENTS
Department of Biotechnology (DBT), Government of In-
dia, for sanctioning the Research Grant (F. No.
BT/PR14373/Med/30/ 530/2010) and Schrodinger for ex-
tending the license of PHASE.
SUPPLEMENTARY MATERIAL
Supplementary material is available on the publisher’s
web site along with the published article.
REFERENCES
[1] Cohen, P. Protein kinases--the major drug targets of the twenty-
first century? Nat. Rev. Drug Discov., 2002, 1(4), 309-315.
[2] Duong-Ly, K.C.; Peterson, J.R. The human kinome and kinasei
inhibition as a therapeutic strategy. Curr. Protoc. Pharmacol.,
2013, 2, Unit 2.9.1-Unit 2.9.14.
[3] Cruzalegui, F. Protein kinases: From targets to anti-cancer drugs.
Ann. Pharm. Francaises., 2010, 68(4), 254-259.
[4] Zwick, E.; Bange, J.; Ullrich, A. Receptor tyrosine kinases as tar-
gets for anticancer drugs. Trends Mol. Med., 2002, 8(1), 17-23.
[5] Arora, A; Scholar, E.M. Role of tyrosine kinase inhibitors in cancer
therapy. J. Pharmacol. Exp. Ther., 2005, 315(3), 971-979.
[6] Roskoski, R. Vascular endothelial growth factor (VEGF) signaling
in tumor progression. Crit. Rev. Oncol. Hematol., 2007, 62(3), 179-
213.
[7] Roskoski, R. VEGF receptor protein-tyrosine kinases: Structure
and regulation. Biochem. Biophys. Res. Commun., 2008, 375(3),
287-291.
[8] Chioccioli, M.; Marsili, S.; Bonaccini, C.; Procacci, P.; Gratteri, P.
Insights into the conformational switching mechanism of the hu-
man vascular endothelial growth factor receptor type 2 kinase do-
main. J. Chem. Inf. Model., 2012, 52(2), 483-491.
[9] McTigue, M.A.; Wickersham, J.A.; Pinko, C.; Showalter, R.E.;
Parast V.C.; Tempczyk-Russell, A.; Gehring, M.R.; Mroczkowski,
B.; Chen-Chen, K.; Villafranca, J.E.; Appelt, K. Crystal structure
of the kinase domain of human vascular endothelial growth factor
receptor 2: A key enzyme in angiogenesis. Structure, 1999, 7(3),
319-330.
[10] Zhao, Z.; Wu, H.; Wang, L.; Liu, Y.; Knapp, S.; Liu, Q.; Gray,
N.S. Exploration of type II binding mode: A privileged approach
for kinase inhibitor focused drug discovery? ACS Chem. Biol.,
2014, 9(6), 1230-1241.
[11] Angiolini, M. Targeting the DFG-in kinase conformation: a new
trend emerging from a patent analysis. Future Med. Chem., 2011,
3(3), 309-337.
[12] Zuccotto, F.; Ardini, E.; Casale, E.; Angiolini, M. Through the
“gatekeeper door”: Exploiting the active kinase conformation. J.
Med. Chem., 2010, 53(7), 2681-2694.
[13] Palmieri, L.; Rastelli, G. C Helix displacement as a general ap-
proach for allosteric modulation of protein kinases. Drug. Discov.
Today, 2013, 18(7-8), 407-414.
[14] Liu, Y.; Gray, N.S. Rational design of inhibitors that bind to inac-
tive kinase conformations. Nat. Chem. Biol., 2006, 2(7), 358-364.
[15] Ghose, A.; Herbertz, T. Perspecti V. Knowledge based prediction
of ligand binding modes and rational inhibitor design for kinase
drug discovery. 2008, 51(17), 19-21.
[16] Huang, L.; Huang, Z.; Bai, Z.; Xie, R.; Sun, L.; Lin, K. Develop-
ment and strategies of VEGFR-2/KDR inhibitors. Future Med.
Chem., 2012, 4(14), 1839-1852.
[17] Lintnerova, L.; Garcia-Caballero, M.; Gregan, F.; Melichercik, M.;
Quesada, A.R.; Dobias, J.; Lac, J.; Salisova, M.; Bohac, A. A de-
velopment of chimeric VEGFR2 TK inhibitor based on two ligand
conformers from PDB: 1Y6A complex - Medicinal chemistry con-
sequences of a TKs analysis. Eur. J. Med. Chem., 2014, 72, 146-
159.
[18] Boyer, S.J. Small molecule inhibitors of KDR (VEGFR-2) kinase:
an overview of structure activity relationships. Curr. Top. Med.
Chem., 2002, 2(9), 973-1000.
[19] Musumeci, F.; Radi, M.; Brullo, C.; Schenone, S. Vascular endo-
thelial growth factor (VEGF) receptors: Drugs and new inhibitors.
J. Med. Chem., 2012, 55(24), 10797-10822.
[20] Miyazaki, Y.; Matsunaga, S.; Tang, J.; Maeda, Y.; Nakano, M.;
Philippe, R.J.; Shibahara, M.; Liu, W.; Sato, H.; Wang, L.; Nolte,
R.T. Novel 4-amino-furo[2,3-d]pyrimidines as Tie-2 and VEGFR2
dual inhibitors. Bioorg. Med. Chem. Lett., 2005, 15(9), 2203-2207.
[21] Tripathy, R.; Ghose, A.; Singh, J.; Bacon, E.R.; Angeles, T.S.;
Yang, S.X.; Albom, M.S.; Aimone, L.D.; Herman, J.L.; Mallamo,
J.P. 1,2,3-Thiadiazole substituted pyrazolones as potent
KDR/VEGFR-2 kinase inhibitors. Bioorg. Med. Chem. Lett., 2007,
17(6), 1793-1798.
[22] Wang, C.; Gao, H.; Dong, J.; Zhang, Y.; Su, P.; Shi, Y.; Zhang, J.
Biphenyl derivatives incorporating urea unit as novel VEGFR-2 in-
hibitors: Design, synthesis and biological evaluation. Bioorg. Med.
Chem., 2014, 22(1), 277-284.
[23] Gingrich, D.E.; Reddy, D.R.; Iqbal, M.A.; Singh, J.; Aimone, L.D.;
Angeles, T.S.; Albom, M.; Yang, S.; Ator, M.A.; Meyer, S.L.;
Robinson, C.; Ruggeri, B.A.; Dionne, C.A.; Vaught, J.L.; Mallamo,
J.P.; Hudkins, R.L. A new class of potent vascular endothelial
growth factor receptor tyrosine kinase inhibitors: structure-activity
relationships for a series of 9-alkoxymethyl-12-(3-hydroxy-
propyl)indeno[2,1-a]pyrrolo[3,4-c]carbazole-5-ones and the identi-
fication of CEP-521. J. Med. Chem., 2003, 46(25), 5375-5388.
[24] La, D.S.; Belzile, J.; Bready, J.V; Coxon, A.; DeMelfi, T.; Doerr,
N.; Estrada, J.; Flynn, J.C.; Flynn, S.R.; Graceffa, R.F.; Harriman,
S.P.; Larrow, J.F.; Long, A.M.; Martin, M.W.; Morrison, M.J.;
Patel, V.F.; Roveto, P.M.; Wang, L.; Weiss, M.M.; Whittington,
D.A.; Teffera, Y.; Zhao, Z.; Polverino, A.J.; Harmange, J.C. Novel
2,3-dihydro-1,4-benzoxazines as potent and orally bioavailable in-
hibitors of tumor-driven angiogenesis. J. Med. Chem., 2008, 51(6),
1695-1705.
[25] Potashman, M.H.; Bready, J.; Coxon, A.; DeMelfi Thomas M.;
DiPietro, L.; Doerr, N.; Elbaum, D.; Estrada, J.; Gallant, P.; Ger-
main, J.; Yan Gu , Harmange, J.C.; Kaufman, Kendall, R.;
Kim, J.L.; Kumar, G.N.; Long, A.M.; Neervannan, S.; Patel , V.F.;
Polverino, A.; Rose, P.; Plas, S.; Whittington, D.; Zanon, R.; Zhao,
H. Design, synthesis, and evaluation of orally active benzimida-
zoles and benzoxazoles as vascular endothelial growth factor-2 re-
ceptor tyrosine kinase inhibitors. J. Med. Chem., 2007, 50(18),
4351-4373.
[26] Ugale, V.G.; Patel, H.M.; Surana, S.J. Molecular modeling studies
of quinoline derivatives as VEGFR-2 tyrosine kinase inhibitors us-
ing pharmacophore based 3D QSAR and docking approach. Arab.
J. Chem., 2013, DOI: http://dx.doi.org/10.1016/j.arabjc.2013.
07.026.
[27] Rajagopalan, M.; Balasubramanian, S.; Ramaswamy, A.; Prakash
Mathur, P. Pharmacophore based 3D-QSAR modeling and free en-
ergy analysis of VEGFR-2 inhibitors. J. Enzyme Inhib. Med.
Chem., 2012, 28, 1-11.
[28] Pasha, F.A.; Muddassar, M.; Neaz, M.M.; Cho, S.J. Pharma-
cophore and docking-based combined in-silico study of KDR in-
hibitors. J. Mol. Graph. Model., 2009, 28(1), 54-61.
[29] Harris, P.A.; Cheung, M.; Hunter, R.N.; Brown, M.L.; Veal, J.M.;
Nolte, R.T.; Wang, L.; Liu, W.; Crosby, R.M.; Johnson, J.H.; Ep-
perly, A.H.; Kumar, R.; Luttrell, D.K.; Stafford, J.A. Discovery and
evaluation of 2-anilino-5-aryloxazoles as a novel class of VEGFR2
kinase inhibitors. J. Med. Chem., 2005, 48(5), 1610-1619.
[30] LigPrep, version 2.5, Schrodinger, LLC, New York, NY, 2013.
[31] Joshi, A.J.; Gadhwal, M.K.; Joshi, U.J. A combined approach
based on 3D pharmacophore and docking for identification of new
aurora A kinase inhibitors. Med. Chem. Res., 2014, 23(3), 1414-
1436.
[32] Dixon, S.L.; Smondyrev, A.M.; Knoll, E.H.; Rao, S.N.; Shaw,
D.E.; Friesner, R.A. PHASE: A new engine for pharmacophore
perception, 3D QSAR model development, and 3D database
screening: 1. Methodology and preliminary results. J. Comput.
Aided. Mol. Des., 2006, 20(10-11), 647-671.
[33] Phase, version 3.3, Schrodinger, LLC, New York, NY, 2013.
[34] Zhang, W.; Xia, S.; Ye, J.; Tang, Y.; Li, Z.; Zhu, W.; Cheng, J.
Structural features of GABAA receptor antagonists: Pharma-
cophore modeling and 3D-QSAR studies. Med. Chem. Res., 2013,
22(12), 5961-5972.
Personal Use Only
Not For Distribution
Pharmacophore and Docking Guided Virtual Screening Study Current Computer-Aided Drug Design, 2017, Vol. 13, No. 3 207
[35] Bhansali S.G.; Kulkarni, V.M. Pharmacophore generation, atom-
based 3D-QSAR, docking, and virtual screening studies of p38-
mitogen activated protein kinase inhibitors: pyridopyridazin-6-ones
( part 2 ). Res. Rep. Med. Chem., 2014, 2014(4), 1-22.
[36] Brogi, S.; Papazafiri, P.; Roussis, V.; Tafi, A. 3D-QSAR using
pharmacophore-based alignment and virtual screening for discov-
ery of novel MCF-7 cell line inhibitors. Eur. J. Med. Chem., 2013,
67, 344-351.
[37] Friesner, R.; Banks, J.L.; Murphy, R.B.; Halgren, T.; Klicic, J.J.;
Mainz, D.T.; Repasky, M.P.; Knoll, E.H.; Shelley, M.; Perry, J.K.;
Shaw, D.E.; Francis, P.; Shenkin, P.S. Glide: A new approach for
rapid, accurate docking and scoring. 1. Method and assessment of
docking accuracy. J. Med. Chem., 2004, 47(7), 1739-1749.
[38] Glide, version 5.9, Schrodinger, LLC, New York, NY, 2013.
[39] Friesner, R.A.; Murphy, R.B.; Repasky, M.P.; Frye, L.L.;
Greenwood, J.R.; Halgren, T.A.; Sanschagrin, P.C.; Mainz, D.T.
Extra precision glide: Docking and scoring incorporating a model
of hydrophobic enclosure for protein-ligand complexes. J. Med.
Chem., 2006, 49(21), 6177-6196.
[40] QikProp, version 3.8, Schrodinger, LLC, New York, NY, 2013.
[41] Rajappan, R.; Ramachandran, V.P.; Sreedhara, K.; Pai, R. Molecu-
lar docking , pharmacophore modeling and 3D-QSAR approach on
a series of piperidone derivatives as potential anticancer agents by
targeting the enzyme VEGFR-2 tyrosine kinase. Der Pharma
Chemica, 2015, 7(4), 155-163.
[42] Veerasamy, R.; Rajak, H.; Jain, A.; Sivadasan, S.; Varghese, C.P.;
Agrawal, R.K. Validation of QSAR models - strategies and impor-
tance. Int. J. Drug Des. Discovery, 2011, 2(3), 511-519.
[43] Veber, D.F.; Johnson, S.R.; Cheng, H.-Y.; Smith, B.R.; Ward,
K.W.; Kopple, K.D. Molecular properties that influence the oral
bioavailability of drug candidates. J. Med. Chem., 2002, 45(12),
2615-2623.
[44] Lipinski, C.A.; Lombardo, F.; Dominy, B.W.; Feeney, P.J. Ex-
perimental and computational approaches to estimate solubility and
permeability in drug discovery and development settings. Adv.
Drug Deliv. Rev., 2012, 64, 4-17.
[45] Refsgaard, H.H.F.; Jensen, B.F.; Brockhoff, P.B.; Padkjær, S.B.;
Guldbrandt, M.; Christensen, M.S. In silico prediction of mem-
brane permeability from calculated molecular parameters. J. Med.
Chem., 2005, 48(3), 805-811.
Personal Use Only
Not For Distribution
