LEADOPT: An automatic tool for structure-based lead optimization, and its application in structural optimizations of VEGFR2 and SYK inhibitors

Article abstract of DOI:10.1016/j.ejmech.2015.02.019

Lead optimization is one of the key steps in drug discovery, and currently it is carried out mostly based on experiences of medicinal chemists, which often suffers from low efficiency. In silico methods are thought to be useful in improving the efficiency of lead optimization. Here we describe a new in silico automatic tool for structure-based lead optimization, termed LEADOPT. The structural modifications in LEADOPT mainly include two operations: fragment growing and fragment replacing, which are restricted to carry out in the active pocket of target protein with the core scaffold structure of ligand kept unchanged. The bioactivity of the newly generated molecules is estimated by ligand efficiency rather than a commonly used scoring function. Twelve important pharmacokinetic and toxic properties are evaluated using SCADMET, a program for the prediction of pharmacokinetic and toxic properties. LEADOPT was first evaluated using two retrospective cases, in which it showed a very good performance. LEADOPT was then applied to the structural optimizations of the VEGFR2 inhibitor, sorafenib, and the SYK inhibitor, R406. Though just several compounds were synthesized, we have obtained some compounds that are more potent than sorafenib and R406 in enzymatic and functional assays. All of these have validated, at least to some extent, the effectiveness of LEADOPT.

Full text of DOI:10.1016/j.ejmech.2015.02.019

European Journal of Medicinal Chemistry 93 (2015) 523e538
Contents lists available at ScienceDirect
European Journal of Medicinal Chemistry
journal homepage: http://www.elsevier.com/locate/ejmech
Original article
LEADOPT: An automatic tool for structure-based lead optimization,
and its application in structural optimizations of VEGFR2 and SYK
inhibitors
,
,
, b,
Guo-Bo Li a 1, Sen Ji a 1, Ling-Ling Yang a 1, Rong-Jie Zhang a, Kai Chen a, Lei Zhong a,
Shuang Ma a, Sheng-Yong Yang a
,
*
a State Key Laboratory of Biotherapy/Collaborative Innovation Center of Biotherapy, West China Hospital, West China Medical School, Sichuan University,
Sichuan 610041, China
b College of Chemical Engineering, Sichuan University, Sichuan 610041, China
a r t i c l e i n f o
a b s t r a c t
Article history:
Lead optimization is one of the key steps in drug discovery, and currently it is carried out mostly based
on experiences of medicinal chemists, which often suffers from low efficiency. In silico methods are
thought to be useful in improving the efficiency of lead optimization. Here we describe a new in silico
automatic tool for structure-based lead optimization, termed LEADOPT. The structural modifications in
LEADOPT mainly include two operations: fragment growing and fragment replacing, which are restricted
to carry out in the active pocket of target protein with the core scaffold structure of ligand kept un-
changed. The bioactivity of the newly generated molecules is estimated by ligand efficiency rather than a
commonly used scoring function. Twelve important pharmacokinetic and toxic properties are evaluated
using SCADMET, a program for the prediction of pharmacokinetic and toxic properties. LEADOPT was first
evaluated using two retrospective cases, in which it showed a very good performance. LEADOPT was then
applied to the structural optimizations of the VEGFR2 inhibitor, sorafenib, and the SYK inhibitor, R406.
Though just several compounds were synthesized, we have obtained some compounds that are more
potent than sorafenib and R406 in enzymatic and functional assays. All of these have validated, at least to
some extent, the effectiveness of LEADOPT.
Received 7 September 2014
Received in revised form
3 January 2015
Accepted 12 February 2015
Available online 14 February 2015
Keywords:
Structure-based drug design (SBDD)
Lead optimization
Fragment growing
Fragment replacing
Ligand efficiency
© 2015 Published by Elsevier Masson SAS.
1. Introduction
combinatorial chemistry, as well as virtual screening [2e6]. How-
ever, at present the lead optimization is carried out mostly based on
Discovery of drug candidates is a bottleneck in drug research
and development (R&D). The whole procedure for the discovery of
drug candidates often includes two phases: lead discovery and lead
optimization [1]. Currently the lead discovery is no longer a big
problem or at least not a key challenge due to that many new
technologies for the lead discovery have been well established and
applied, such as high-throughput screening, biophysical screening,
experiences of medicinal chemists, which often suffers from low
efficiency. The lead optimization has now become a key challenge
in not only the discovery of drug candidates but also the whole
process of drug R&D.
Computational methods are often thought to be helpful in
improving the efficiency of lead optimization [7e10]. Thus, some in
silico methods have been established and applied in the lead
optimization. Among these methods, three-dimensional quantita-
tive structure activity relationship (3D-QSAR) methods, for
example, comparative molecular field analysis (CoMFA) [11] and
comparative molecular similarity indices analysis (CoMSIA) [12],
are the most widely used ones. Numerous studies have demon-
strated that 3D-QSAR methods are very useful in lead optimization
[13e16]. Nevertheless, the current 3D-QSAR methods as well as
other commonly used in silico methods for lead optimization are
often just based on small-molecule ligands, implying ignoring
Abbreviations: R&D, research and development; SBDD, structure-based drug
design; LE, ligand efficiency; 3D-QSAR, three-dimensional quantitative structure
activity relationship; ADMET, absorption, distribution, metabolism, excretion, and
toxicity; PDB, Protein Data Bank; JAK1, Janus kinase 1; MPS1, monopolar spindle 1;
VEGFR2, vascular endothelial growth factor receptor 2; SYK, spleen tyrosine kinase.
* Corresponding author.
E-mail address: yangsy@scu.edu.cn (S.-Y. Yang).
These authors contributed equally to this work.
1
http://dx.doi.org/10.1016/j.ejmech.2015.02.019
0223-5234/© 2015 Published by Elsevier Masson SAS.

524
G.-B. Li et al. / European Journal of Medicinal Chemistry 93 (2015) 523e538
Table 1
structural information of target protein. This situation may lead to
the theoretically optimized molecules having no activity due to
bumps between molecules and target protein atoms. Structure-
based lead optimization approaches are expected to be able to
overcome these shortcomings, in which the structure of ligand is
modified to enhance the binding affinity based on the known
interaction mode between target protein and ligand [17e21].
Although these approaches have been applied in lead optimization
for a long time, they are often used manually by medicinal chem-
ists, which still suffer inevitably from low efficiency. We also
noticed that very recently a commercial module for structure-
based ligand optimization has been issued in MOE by Chemical
Computing Group Inc. However, its implementation details are not
known, and it likely does not include the evaluation of pharma-
cokinetic and toxic properties for the generated molecules, which is
also important for lead optimization.
Here we present an automatic tool for structure-based lead
optimization, termed LEADOPT. LEADOPT has the following char-
acteristics that make it a practical tool in lead optimization. First, a
large number of derivatives of lead compound that have the same
binding mode with the active pocket of target protein can be
created. These compounds can be accommodated in the active
pocket without any atom bump with the target protein. Second,
LEADOPT builds up new molecules using an efficient fragment-
based strategy. Such a strategy can help avoid producing unrea-
sonable molecular structures and to some extent keep the synthetic
accessibility of the derived compounds. Third, ligand efficiency (LE)
rather than scoring function is used as a measure to sort the newly
generated molecules; LE has been considered as an effective
strategy to help narrow focus to lead compounds with optimal
combinations of physicochemical properties and pharmacological
properties [22,23]. Fourth, a number of pharmacokinetic and toxic
properties of the generated molecules are evaluated, which makes
the lead candidates pharmacologically acceptable. We shall in the
following describe the details for the algorithm of LEADOPT. Then,
LEADOPT will be evaluated with two retrospective cases. Finally, it
will be applied to the structural optimizations of sorafenib, an in-
hibitor of vascular endothelial growth factor receptor 2 (VEGFR2),
and R406, an inhibitor of spleen tyrosine kinase (SYK).
Bond cleavage types used for producing fragments from molecules.
Rule# Cleavage type Example
Cleavage marksa
1
Amide
2
3
Ester
Urea
4
5
Sulphonamide
Acyclic amine
6
7
Cyclic aminee
aliphatic carbon
Lactam nitrogene
aliphatic carbon
8
Acyclic ether
Acyclic sulfide
Aromatic ether
9
10
11
Cyclic aminee
aromatic carbon
12
13
Aromatic nitrogene
aliphatic carbon
Aromatic nitrogene
aromatic carbon
2. Methods and materials
14
15
Aromatic carbone
aromatic carbon
2.1. Fragment library construction
Olefin
Since LEADOPT adopts a fragment-based approach for structural
modifications to derive new molecular structures, a good fragment
library is quite important for the quality of derived molecules. To
establish such a fragment library, we first collected a total of 17,858
drug or drug-like molecules from CMC database, ChEMBL database
[24], and DrugBank database [25]. Then, an in-house program
written by us in C/Cþþ programming language was used to auto-
matically build the fragment library. The process is briefly
described as follows.
a
*-, nucleophilic mark; *þ, electrophilic mark; *, aromatic mark; **, olefin mark.
(3) Four types of connection marks, including electrophilic mark
(*þ), nucleophilic mark (*-), aromatic mark (*), and olefin
mark (**), were used to label the cleaved terminal(s) for each
fragment. Some examples are given in Supporting
Information Fig. S1.
(4) Any duplicate or fragments containing more than 14 heavy
atoms or more than 3 terminal marks were removed.
(1) Each input molecule was detected for whether it contains
any of the fifteen types of chemical bonds defined in Table 1;
these chemical bond types were derived from common
chemical reactions, which is for making the generated frag-
ments easy to be used in later chemical synthesis [26].
(2) Molecules that contain at least one chemical bond defined in
Table 1 were cleaved into fragments at the defined chemical
bonds. If a terminal fragment cleaved is only a small group,
such as methyl, ethyl, propyl, or butyl, the cleaving operation
will not be performed anymore.
We finally obtained a fragment library consisting of 6877 unique
fragments, including 5106 fragments with a single terminal mark,
1587 fragments with two terminal marks, and 184 fragments with
three terminal marks. The defined terminal marks will be used in
the structural modification process in LEADOPT, which will be
described in the next section.

G.-B. Li et al. / European Journal of Medicinal Chemistry 93 (2015) 523e538
525
Table 2
Five bond connection types.
2.2. Algorithms of LEADOPT
Fig. 1 schematically depicts the workflow of LEADOPT. LEADOPT
is launched through inputting a ligandereceptor complex struc-
ture, which may be prepared from either experimental X-ray
crystal structure or molecule docking. In LEADOPT, the ligand and
protein receptor, which are hydrogenated, correspond to two
separated input files. For the protein receptor, some important
molecules such as metal atoms and water molecules can be kept in
the binding pocket. Subsequently, LEADOPT executes the following
four steps: binding site analysis, structural modification, evaluation
of binding affinity and ligand efficiency, and evaluation of phar-
macokinetic and toxic properties.
Rule#
1
Connection type
Example
Connected bond
Nucleophilice
electrophilic
2
3
4
5
Aromatic carbone
aromatic carbon
Electrophilice
aromatic carbon
Nucleophilice
aromatic carbon
Olefin carbone
olefin carbon
2.2.1. Binding site analysis
In LEADOPT, new molecules are built up within the constraints
of the protein binding site. Thus the size and shape of the binding
site are critical for later structural modifications. For a given
protein-ligand complex, LEADOPT will define a rectangular box
covering the ligand and all residues around the ligand in 6 Å (see
Supporting Information Fig. S2). The rectangular box was further
divided into small equal-sized square grids according to its vertex
coordinates. The grid spacing was set as 0.5 Å. A hydrogen atom was
used as a probe on each grid to check whether the grid is accessible.
If the probe bumps with the protein, the grid will be labeled as ‘E’; if
not, the grid will be labeled as ‘V’. The grid information will be used
in the structural modification process.
while it is not allowed to connect to the fragment terminal with
other terminal marks such as nucleophilic mark. These defined
connection rules can help avoid the generation of unreasonable
molecules and to some extent improve the synthetic accessibility.
The procedure of ‘fragment growing’ is briefly described as
follows (see Fig. 2A): (i) detecting whether there is certain vacuity
around the reference compound according to the grid information;
if no, the ‘fragment growing’ will stop; (ii) identifying possible
connection points and determining their corresponding connection
marks as that used for labeling the fragment terminals; (iii) con-
necting each fragment in the fragment library onto one of the
connection points according to the five defined connection rules
(see Table 2); (iv) adjusting the orientation for each fragment and
checking whether the fragment bumps with the protein; (v)
repeating step iii and iv for other possible connection points.
The ‘fragment replacing’ procedure involves the following steps
(see Fig. 2B): (i) identifying the core scaffold for the reference
compound; (ii) determining the connection bonds between the
2.2.2. Structural modification
Two main types of structural modification strategies, including
‘fragment growing’ and ‘fragment replacing’, were adopted to
construct new molecules. In the structural modification process,
five bond connection rules were used according to the fragment
terminal marks (see Table 2). For example, the fragment terminal
with a nucleophilic mark can be linked to the fragment terminal
with an electrophilic mark or aromatic mark (Rule 1 and 3, Table 2),
Fig. 1. The workflow of LEADOPT.

526
G.-B. Li et al. / European Journal of Medicinal Chemistry 93 (2015) 523e538
Fig. 2. Schematic illustrations for the two main types of structural modifications: (A) ‘fragment growing’ and (B) ‘fragment replacing’.
core scaffold and other parts of the reference compound; (iii)
selecting one connection bond, cleaving it, and labeling the
connection point for the core scaffold; (iv) connecting each frag-
ment in the fragment library onto one of the connection points
according to the connection rules; (v) adjusting the orientation for
each fragment and checking whether the fragment bumps with the
protein; (vi) recording the top ranking fragments for the connec-
tion point; (vii) repeating step iii to vi for all the connection points;
(viii) assembling all the top ranking fragments onto the corre-
sponding points on the core scaffold. An example for ‘fragment
replacing’ is given in Supporting Information Fig. S3.
Besides, we introduced another structural modification strategy
termed ‘fragment cyclization’ to create more structurally diverse
molecules. This strategy relies on the ‘fragment growing’ and
‘fragment replacing’. During a ‘fragment growing’ or ‘fragment
replacing’ process, if a fragment can approach the defined core
scaffold within a certain distance, it will be cyclized with the core
scaffold to form a new scaffold. There have two situations for the
‘fragment cyclization’ operation in LEADOPT (see Supporting
Information Fig. S4). First, when the fragment collides with the
core scaffold by a pair of hydrogen atoms, these two hydrogen
atoms will be deleted and a methyl group will be used to bridge the
fragment and core scaffold. Second, when the fragment collides
with the core scaffold by a pair of heavy atoms, one of the heavy
atoms will be deleted and another heavy atom will be connected
directly. Such a structural modification strategy could create new
chemical scaffolds and generate structurally diverse derivatives. All
the newly generated molecules by the modification strategies
described above will be submitted to the following evaluation.
matching are not calculated, since they are difficult to be accurately
desolv
calculated. Besides, two other descriptors, including K
recꢀSAlnonSA
representing the ratio of the polar and nonpolar surface areas of the
desolv
recꢀSARA
protein binding site and K
describing the solvent accessible
surface area of the protein binding site, were also discarded,
because these two descriptors each are a constant value for one
target protein. A total of 42 descriptors were finally used for the
establishment of this version of ID-Score (see Supporting
Information Table S1). The validation results are given in
Supporting Information Fig. S5.
Ligand efficiency, which is defined as the binding energy per
atom of a ligand to its receptor [28,29], is then calculated based on
the calculated ID-Score value.
LE ¼ 1:4*ðID ꢀ Score valueÞ=N
where N is the number of non-hydrogen atoms. In LEADOPT, LE
rather than scoring function is used as a measure to sort the newly
generated molecules.
2.2.4. Evaluations of pharmacokinetic and toxic properties
The evaluations of pharmacokinetic and toxic properties were
carried out using SCADMET, which is a prediction program of
pharmacokinetic and toxic properties developed using a support
vector machine (SVM) method [30e33]. In LEADOPT, 12 types of
pharmacokinetic and toxic properties are predicted, including hu-
man oral bioavailability, caco-2 cell permeability, in vivo clearance,
human intestine absorption, human plasma protein binding rate,
pregnane X receptor ligand, half lethal concentration (LC50),
aqueous solubility (logS), mitochondria toxicity, genotoxicity, hu-
man ether-a-go-go-related gene (hERG) toxicity, and teratogenicity.
The establishment and evaluation of these prediction models are
well documented in Ref. [30e33]. For each of the generated mol-
ecules, LEADOPT calculates an ADMET score (ADMET-score); if a
compound passes the evaluation of anyone of the pharmacokinetic
and toxic properties, a value of 1 is assigned to the ADMET-score,
otherwise 0 is assigned. The overall ADMET-score value is a sum
of all the individual values for the 12 pharmacokinetic and toxic
properties, which will be used to indicate the overall ADMET fea-
tures of the generated derivatives.
2.2.3. Evaluation of binding affinity and ligand efficiency
The protein-ligand binding affinity is assessed by using the ID-
Score method. ID-Score was derived from the established
comprehensive set of descriptors related to protein-ligand in-
teractions. These descriptors cover nine categories, including
hydrogenebonding interaction, van der Waals interaction,
p-sys-
tem interaction, electrostatic interaction, metal-ligand bonding
interaction, entropic loss effect, desolvation effect, shape matching,
and surface property matching. A total of 2278 crystal protein-
ligand complexes were used as the training set, and a modified
support vector regression (SVR) method was used to establish the
relationship between descriptors and experimental binding affin-
ities. Details about the establishment of ID-Score please see
Ref. [27]. Here we used a modified version of ID-Score. In this
version of ID-Score, the descriptors regarding surface property
2.2.5. Implementation of LEADOPT
The source code for the implementation of LEADOPT was writ-
ten in C/Cþþ programming language under the LINUX operating
system. The LEADOPT program is available free of charge to not-for-

G.-B. Li et al. / European Journal of Medicinal Chemistry 93 (2015) 523e538
527
profit institution upon request from the corresponding author.
2.4.5. Western blot analysis
After treatment with different concentrations of compound B4
or R406 for 6 h at 37 ꢂC, MV4-11 cells were harvested, washed with
ice-cold physiological saline, and lysed with RIPA lysis buffer
(Beyotime, China) containing 1% cocktail (SigmaeAldrich). Then,
the cell lysates were separated by SDSꢀPAGE and electro-
transferred onto PVDF membranes (Millipore). The PVDF mem-
branes were incubated with each antibody and detected according
to the immunoblot analysis principle. All the antibodies used in this
immunoblot analysis were purchased from Cell Signaling Tech-
nology, with the exception of the anti-FLT3 antibody, which was
obtained from Abcam.
For the HUVEC immunoblot studies, the cells were serum
starved overnight in EBM-2 medium. After that, the cells were
incubated with vehicle, compound A4 or sorafenib for 2 h, followed
by treatment with 50 ng/mL recombinant human VEGF (Lonza Inc.)
for 10 min. The cells were harvested and lysed in RIPA buffer with
1% cocktail. Proteins were separated by gel electrophoresis on 5%e
10% SDS-PAGE gels and probed with antibodies. All of the anti-
bodies were used at a 1:1000 dilution, and the horseradish
peroxidase-coupled secondary antibodies (Zhong Shan Golden
Bridge Biotechnology, China) were used at 1:5000.
2.3. Chemistry
All reagents and solvents were purchased from commercial
sources (Adamas-beta Switzerland, Acros Organics USA, Sigma-
eAldrich Switzerland) and used without further purification. Re-
actions were monitored by thin-layer chromatography (TLC)
analysis (Merck, 0.2 mm silica gel 60 F254 on glass plates). 1H NMR
and 13C NMR spectra were recorded on a Bruker AV-400 spec-
trometer at 400 and 100 MHz, respectively. Chemical shifts are
reported in parts per million (ppm) relative to tetramethylsilane
(TMS), and multiplicities are designed as s (singlet), d (doublet), t
(triplet), m (multiplet), or br s (broad singlet). Mass spectroscopy
analyses were performed with an Agilent 1100 series LCꢀMS in-
strument with UV detection at 254 nm in low-resonance electro-
spray mode (ESI). All of the target compounds were purified to
>95% purity, as determined by high-performance liquid chroma-
tography (HPLC). HPLC analysis was performed on a Waters 2695
HPLC system with the use of
(4.6 mm ꢁ 250 mm, 5 um).
a
Kramosil C18 column
2.4. Enzymatic and functional assays
2.4.6. In vivo live fluorescent zebrafish assay
2.4.1. In vitro kinase inhibitory assays
Transgenic zebrafish (FLK-1:EGFP) embryos were grown and
maintained in accordance to the same protocols as given in
Ref. [34]. Compound B4 and R406 were prepared initially as
10 mmol/L stock solution in dimethyl sulfoxide (DMSO), and then
were diluted in different assay concentrations with fish water, and
0.3% DMSO served as control. The embryos were distributed to 24-
well plates (10 embryos in each well). Then the embryos were
exposed to compound solution and included at 28.5 ꢂC from 15 h
postfertilization to 31 h postfertilization. In this end, the zebra-
fishes were anesthetized with 0.01% tricaine and imaged using a
fluorescence microscope (Carl Zeiss Microimaging Inc.) equipped
with an AxioCam MRc5 digital CCD camera (Carl Zeiss Micro-
imaging Inc.).
All the kinase inhibitory assays in this study were tested via the
Kinase Profiler service provided by Millipore Company. The IC50
values for compounds were determined from doseeresponse
curves obtained from the assays at 10 different concentrations of
each compound. All the assays were test twice and the mean values
of the replicates were calculated.
2.4.2. Cell lines and cell culture
The Ramos B and MV4-11 cell lines were obtained from the
American Type Culture Collection (ATCC). Ramos B cells were
grown in RPMI-1640 medium supplemented with 10% fetal bovine
serum (FBS; Caoyuan lvye, Huhht, China) in 5% CO2 at 37 ꢂC. MV4-
11 was maintained in IMDM medium according to ATCC guidelines.
Human umbilical vein endothelial cells (HUVECs) were isolated
from human umbilical cord veins using a standard operating pro-
cedure and grown in EBM-2 basal medium (Cat. No.cc-3156, Lonza
Inc.).
3. Results and discussion
3.1. Development of LEADOPT
LEADOPT is an automatic tool developed for structure-based
lead optimization. The overall workflow of LEADOPT is schemati-
cally depicted in Fig. 1. A detailed description of the algorithms of
LEADOPT has been given in the Methods and Materials section.
Here we just make a short summary for the workflow of LEADOPT.
LEADOPT is started through input of the structure of
ligandereceptor complex, which can be from either experimental
X-ray crystal structure or molecule docking. Firstly, LEADOPT per-
forms an analysis to the binding site of receptor, which includes
identifying the active pocket, partitioning the active pocket into
small equal-sized square grids, and checking which grids are
accessible with a probe. Secondly, LEADOPT carries out the struc-
tural modification, which is the most crucial step in the whole lead
optimization process. A fragment library containing 6877 frag-
ments that were obtained through cleaving known drug molecules
was established in advance. Based on the established fragment li-
brary, two types of structural modification operations, fragment
growing and fragment replacing, were used to generate new de-
rivatives. Finally, LEADOPT performs evaluations of bioactivity and
LE, and evaluations of pharmacokinetic and toxic properties of the
new derivatives. The bioactivity evaluation was performed by
estimating the binding affinity of ligand with the target protein,
which was carried out using a modified version of ID-Score. LE was
2.4.3. Wound healing assay
HUVECs were cultured to confluence in 24 well plates and
wounded by using a sterilized yellow pipette tip to make a straight
scratch. Cells were rinsed gently with sterile PBS, and then PBS was
replaced with EGM-2 medium (EBM-2 basal medium plus recom-
binant human VEGF) containing vehicle, compound A4 or sorafenib
(Selleck Inc). After 18 h, pictures were taken using an OLYMPUS
digital camera attached to a light microscope.
2.4.4. Cell viability assays
Cell viability was measured using MTT assay. The Ramos B cells
and MV4-11 cells were seeded in a 96-wellplate at (2ꢀ5) ꢁ 104 cells
per well in RPMI-1640 and IMDM medium, respectively, with 10%
FBS and incubated overnight. Different concentrations of com-
pound B4 or R406 (Selleck Inc.) were added to the cells and incu-
bated at 37 ꢂC for 72 h. Then, 20
added to each well. After the incubation for 2e4 h at 37 ꢂC, 50
acidified SDS (20%, w/v) was added to lyse the oxidative product.
Finally, the light absorption of the dissolved cells was measured at
570 nm on Multiskan MK3 (Thermo Scientific, USA). Each assay was
performed in triplicate. The IC50 values were calculated using
GraphPad Prism software.
m
L of 5 mg/mL MTT solution was
mL of

528
G.-B. Li et al. / European Journal of Medicinal Chemistry 93 (2015) 523e538
Table 3
20 representative derivatives generated by LEADOPT in the structural optimization of JAK1 inhibitor (T1) together with their most similar compounds of the known JAK1
inhibitors that are more potent than T1.
ID
1
Compounds generated by LEADOPT
Chemical structure
The most similar JAK1 inhibitors that are more potent than T1
Calculated
LE
ADMET
Score
Chemical structure
MeasuredLEa
Tcb
1
0.437244
7
0.64
2
3
0.436997
0.462091
7
7
0.61
0.64
0.979592
0.953488
4
5
0.492473
0.446427
7
8
0.61
0.64
0.94
0.931818
6
0.437396
9
0.64
0.930233
7
0.434816
10
0.61
0.924528

G.-B. Li et al. / European Journal of Medicinal Chemistry 93 (2015) 523e538
529
Table 3 (continued )
ID
Compounds generated by LEADOPT
Chemical structure
The most similar JAK1 inhibitors that are more potent than T1
Calculated
LE
ADMET
Score
Chemical structure
MeasuredLEa
Tcb
8
0.447304
7
0.64
0.913043
9
0.437135
9
0.64
0.911111
10
0.439435
9
0.61
0.9
11
12
0.484464
0.434759
9
9
0.55
0.55
0.8909
0.8908
13
14
0.435165
0.52583
10
0.55
0.64
0.8905
7
0.883721
15
16
0.453018
0.449766
8
7
0.55
0.55
0.881356
0.872727
(continued on next page)

530
G.-B. Li et al. / European Journal of Medicinal Chemistry 93 (2015) 523e538
Table 3 (continued )
ID
Compounds generated by LEADOPT
Chemical structure
The most similar JAK1 inhibitors that are more potent than T1
Calculated
LE
ADMET
Score
Chemical structure
MeasuredLEa
Tcb
17
0.445717
0.460586
8
7
0.64
0.87234
18
0.64
0.866667
19
20
0.44639
8
7
0.61
0.64
0.867925
0.866667
0.445855
a
The experimentally measured LE values were selected from Ref. [35].
The Tanimoto coefficients were calculated with the aid of ‘MDLPublicKeys’ fingerprints implemented in Discovery Studio 3.1.
b
then calculated and used for sorting the generated molecules. The
evaluations of pharmacokinetic and toxic properties were per-
formed using SCADMET, which is a prediction program of phar-
macokinetic and toxic properties developed using the support
vector machine method. After the evaluations of bioactivity/LE and
pharmacokinetics/toxicity properties, the program will output a
number of derivatives of the lead compound together with their
calculated LEs and ADMET-scores.
Then, we performed a structural similarity analysis through
calculating the Tanimoto coefficients (Tc) between the generated
derivatives and known JAK1 inhibitors that are more potent than T1
using MDLPublicKeys fingerprints implemented in Discovery Stu-
dio 3.1. The results showed that 55 derivatives have a Tc value larger
than 0.85, and, particularly, one derivative has a Tc value of 1; two
compounds bearing a large value of Tc are generally thought to
have a very similar bioactivity. Table 3 lists the top 20 representa-
tive derivatives generated by LEADOPT according to the Tc value.
MPS1 is a central component of the spindle assembly check-
point signal, which has been linked to many human diseases
especially cancers. Naud et al. recently identified a series of 1H-
Pyrrolo[3,2-c]pyridine derivatives as inhibitors of MPS1 through
structure-based drug design, and obtained a crystal structure of
MPS1with one inhibitor (T2), N-(3,4-dimethoxyphenyl)-2-(1H-
pyrazol-4-yl)-1H-pyrrolo[3,2-c] pyridin-6-amine (PDB code: 4C4E)
[36]. Here the 1H-pyrrolo[3,2-c]pyridin-6-amine moiety of T2 was
defined as the core scaffold (Table 4). After running LEADOPT,
95,419 derivatives were generated, of which 8643 derivatives have
an LE value larger than 0.4150 (an LE value of T2) and an ADMET-
score value larger than 11 (an ADMET-score value of T2). Again,
we performed a similarity analysis between these derivatives and
known MPS1 inhibitors whose experimentally measured LE values
larger than T2. A total of 86 derivatives have a Tc value larger than
0.85, and two completely identical compounds with known MPS1
inhibitors were generated (Tc ¼ 1). Table 4 lists 20 representative
derivatives for MPS1 generated by LEADOPT.
3.2. Retrospective case studies
To test the performance of LEADOPT, we carried out retrospec-
tive case studies on two selected targets: JAK1 and MPS1; these
targets were chosen because crystal structural information of these
targets and structureeactivity relationships (SARs) of their ligands
were all publicly available.
JAK1 is a key intracellular mediator of helical cytokine signaling
pathways, which plays a central role in inflammation, immune
function and hematopoiesis [35]. Very recently, Zak et al. reported a
series of selective JAK1 inhibitors and the crystal structure of JAK1
in complex with an inhibitor (T1), 2-methyl-1-(piperidin-4-yl)-1,6-
dihydroimidazo[4,5-d]pyrrolo[2,3-b]pyridine (PDB code: 4EHZ)
[35]. In this case study, we tried to use LEADOPT to perform
structural modifications for T1. The 1,6-dihydroimidazo[4,5-d]pyr-
rolo[2,3-b]pyridine motif of T1 was choose as the core scaffold
(highlighted in red in Table 3). The running of LEADOPT led to the
generation of a total of 40,961 derivatives. Among them, 4898 de-
rivatives have an LE value larger than 0.4343 (an LE value of T1) and
an ADMET-score value larger than 7 (an ADMET-score value of T1).
The above two retrospective case studies demonstrated, to a
certain degree, the effectiveness of LEADOPT. In the following

G.-B. Li et al. / European Journal of Medicinal Chemistry 93 (2015) 523e538
531
Table 4
20 representative derivatives generated by LEADOPT in the structural optimization of MPS1 inhibitor (T2) together with their most similar compounds in the known JAK1
inhibitors that are more potent than T2.
ID
Compounds generated by LEADOPT
Chemical structure Calculated
The most similar MPS1 inhibitors that are more potent than T2
ADMET
Score
Chemical structure
Measured
LEa
Tcb
1
LE
1
2
3
0.471995
11
12
11
0.4627
0.4491
0.4806
0.446064
0.464795
1
0.975
4
5
0.447191
0.44145
11
11
0.4491
0.4806
0.958333
0.95122
6
7
0.453121
0.417374
11
11
0.4806
0.3751
0.95
0.938776
8
0.442617
0.451198
0.456824
0.492806
0.488157
0.417272
0.428762
11
11
11
12
11
11
11
0.4806
0.4806
0.4806
0.4806
0.4806
0.4627
0.4627
0.928571
0.926829
0.926829
0.926829
0.926829
0.916667
0.914894
9
10
11
12
13
14
15
16
17
0.418037
0.415137
0.434185
11
12
11
0.4491
0.4806
0.4806
0.914894
0.906977
0.904762
(continued on next page)

532
G.-B. Li et al. / European Journal of Medicinal Chemistry 93 (2015) 523e538
Table 4 (continued )
ID
Compounds generated by LEADOPT
The most similar MPS1 inhibitors that are more potent than T2
Chemical structure
Calculated
LE
ADMET
Score
Chemical structure
Measured
LEa
Tcb
18
0.439537
11
0.4806
0.904762
19
20
0.904762
0.449756
11
12
0.4806
0.4806
0.904762
0.902439
a
The experimentally measured LE was calculated by: LE ¼ (ꢀ1.4log (experimental IC50))/(n heavy atoms), where experimental IC50 values were selected from Ref. [36].
The Tanimoto coefficients were calculated with the aid of ‘MDLPublicKeys’ fingerprints implemented in Discovery Studio 3.1.
b
Fig. 3. The chemical structures and inhibitory activities of sorafenib and compounds A1eA6 against VEGFR2.
sections, we shall apply LEADOPT to the structural optimizations of
particularly solid tumors [37]. Several VEGFR2 inhibitors have been
approved by the US Food and Drug Administration (FDA) for
treatment of solid tumors [38]. Among them, sorafenib is the first
marketed small molecule VEGFR2 inhibitor [39]. However, its po-
tency against VEGFR2 is relatively poorer compared with drugs
VEGFR2 inhibitor sorafenib, and SYK inhibitor R406. The aim of
which is mainly to further evaluate the performances of LEADOPT
by these two studies. In addition, we also hope to obtain more
potent VEGFR2 and SYK inhibitors due to their potential thera-
peutic values in treating related diseases.
approved recently (a half maximal inhibitory concentration (IC50
)
of sorafenib against VEGFR2 is 90 nM) [39]. Here we utilized LEA-
DOPT to perform structural optimization to improve its potency.
The structure of sorafenib-VEGFR2 complex was taken from the
Protein Data Bank (PDB code: 4ASD) [38]. The 1-(4-hydroxyphenyl)
urea motif of sorafenib was defined as the core scaffold (see Fig. 3),
which was kept unchanged in the optimization process.
3.3. Prospective study 1: the optimization of VEGFR2 inhibitor
sorafenib
The VEGF receptor 2 (VEGFR2), a type III receptor tyrosine ki-
nase, is a key mediator of angiogenesis. It has been recognized as an
important target for diseases associated with angiogenesis,
LEADOPT generated a total of 94,071 molecules. Among them,

G.-B. Li et al. / European Journal of Medicinal Chemistry 93 (2015) 523e538
533
Scheme 1. The general synthetic route to compounds A1eA6. Reagents and conditions: (a) t-BuOK, DMF, 80e110 ꢂC, 38% to quantitative; (b) for F1, F3: acetonitrile, reflux; for F2:
toluene, reflux, quantitative.
64,412 compounds have an LE value larger than 0.3639 (an LE value
of sorafenib) and an ADMET-score value larger than 9 (an ADMET-
score value of sorafenib). From them, we selected four compounds
(Fig. 3, A1, A2, A3, and A4) to synthesize. Besides, we also selected
another two compounds (Fig. 3, A5 and A6) that have an LE value
less than 0.3639. The six compounds were selected because they
have never been reported before and can be very easily synthesized
without too much synthetic efforts in our current experimental
conditions.
3.3.1. Chemical synthesis for compounds A1-A6
The target compounds A1-A6 were synthesized following the
procedures illustrated in Scheme 1. Firstly, 4-aminophenol
(10 mmol, 1.0 equiv) forms 4-aminophenolate in a potassium tert-
butyl alcohol solution (12 mmol, 1.2 equiv, potassium tert-butyl
alcohol in 5 ml/10 mmol dimethyl formamide) during 0.5 h. Se-
lective nucleophilic substitution reaction of the oxygen anion of 4-
aminophenolate with the chloride of heterocyclic compounds
(H1eH4, 1.0 equiv) resulted in the corresponding intermediates
containing an ether bond at 80 ꢂCe110 ꢂC with the yields ranging
from 38% to quantitative, since the nucleophilicity of the oxygen
anion is relatively stronger than that of nitrogen in 4-
aminophenolate. Secondly, the intermediates were respectively
reacted with the isocyanates (F1eF3, 1.1 equiv) that were readily
synthesized using a similar method as that described previously
[40], producing the target compounds A1-A6 with excellent yields.
For details about compounds A1-A6 see Supporting Information.
3.3.2. Inhibitory activities of compounds A1-A6 against VEGFR2
The inhibitory activities of compounds A1-A6 were measured
with gold-standard 33P radio labeled technology. The results are
present in Fig. 3. Three of the four compounds that have an LE value
larger than 0.3639 (A1, A3, and A4) have indeed shown a higher
potency against VEGFR2 than sorafenib. The most active com-
pound, A4, showed a 5-fold improvement of VEGFR2 inhibition
compared with sorafenib (IC50: 17 nM of A4 vs 90 nM of sorafenib).
Fig. 4A shows the interaction mode of the most active compound,
A4, with the VEGFR2 kinase. The introduced fragment by LEADOPT,
5-(tert-butyl)isoxazole, forms very intensive hydrophobic
Fig. 4. The binding mode of compound A4 with VEGFR2.

534
G.-B. Li et al. / European Journal of Medicinal Chemistry 93 (2015) 523e538
Fig. 5. The anti-angiogenesis effects of compound A4 in HUVECs and transgenic zebrafish embryos assays and its anti-angiogenesis mechanism of action. (A) Compound A4
inhibited HUVEC cells migration in wound healing assay; (B) Compound A4 inhibited zebrafish embryonic angiogenesis; (C) Compound A4 inhibited the phosphorylation of VEGFR2
and ERK1/2 in VEGF-stimulated HUVECs.
interactions with Ile1044, Leu1019, Ile892, and Leu889, which is the
most important factor contributing to the high potency of com-
pound A4 against VEGFR2 (see Fig. 4B). Compounds A5 and A6,
which have a lower LE value than sorafenib, indeed showed a lower
inhibitory activity against VEGFR2. Collectively, these results vali-
dated, at least to some extent, the effectiveness of LEADOPT.
3.3.3. Functional evaluations of compound A4
It has been widely accepted that VEGFR2 is the most crucial
regulator of angiogenesis. We thus assessed the effects of com-
pound A4 on angiogenesis using in vitro functional assays and
in vivo transgenic zebrafish assays. Firstly, wound healing assays
were performed to evaluate the influence of compound A4 on the
migration ability of human umbilical vein endothelial cells
(HUVEC). The results showed that compound A4 could dose-
Fig. 6. The chemical structures and inhibitory activities of compounds B1eB4 against SYK.

G.-B. Li et al. / European Journal of Medicinal Chemistry 93 (2015) 523e538
535
Scheme 2. The general synthetic route to compounds B1eB4. Reagents and conditions: (c) for F4, F5, F6: DIPEA, ethyl alcohol, 90e100 ꢂC, 64%e91%; (d) for 3,4,5-trimethoxyaniline:
DIPEA, n-butyl alcohol, 90e120 ꢂC, 70%e95%; for 4-(4-methylpiperazin-1-yl)aniline: HCl, dimethyl formamide, 120e125 ꢂC, 32%e61%.
dependently prevent HUVECs migration (see Supporting
Information Fig. S6) and significantly reduce the number of
migrating cells at a concentration of 5 mM. By comparison, sorafenib
showed a slightly weak suppression effects in the wound healing
assay. Secondly, transgenic zebrafish assays were used to directly
evaluate the anti-angiogenesis ability of compound A4. The results
showed that compound A4 could almost entirely inhibit the growth
of the intersegmental blood vessel of zebrafish at a concentration of
0.25 mM (Fig. 5B), while 0.25 mM sorafenib just led to a suppression
of about 60%. Finally, western blot analysis was adopted to examine
the ability of compound A4 to inhibit the phosphorylation of
VEGFR2 and downstream ERK1/2 in HUVECs. As shown in Fig. 5C,
compound A4 could effectively inhibit the phosphorylation of
VEGFR2 and ERK1/2. Taken together, all the results demonstrated
that compound A4 could effectively inhibit angiogenesis and block
the VEGFR2 signaling in intact cells.
3.4. Prospective study 2: the optimization of SYK inhibitor R406
Spleen tyrosine kinase (SYK), a non-receptor tyrosine kinase, is a
key integrator of intracellular signals triggered by activated im-
mune receptors, including B cell receptors (BCR) and Fc receptors,
which are important for the development and function of lymphoid
cells [41]. SYK has been thought an attractive drug target for the
treatment of autoimmune disorders, such as rheumatoid arthritis,
allergic asthma, and multiple sclerosis [42,43]. Though some SYK
inhibitors have been reported, there is no SYK inhibitor used in
clinic so far. R406 is a typical SYK inhibitor, which is the active
moiety of R788, an SYK inhibitor that is in clinical trials [43].
Nevertheless, it just has moderate activity against SYK with IC50 of
64 nM. Here we shall use LEADOPT to optimize its potency.
The structure of R406-SYK complex was taken from the Protein
Data Bank (PDB code: 3FQS) [44]. The 5-fluoropyrimidine-2,4-
diamine motif in R406 was defined as the core scaffold (Fig. 6).
After LEADOPT running, 82,690 molecules were generated. Among
them, 67,975 molecules have an LE value larger than 0.2940 (an LE
value of R406) and an ADMET-score value larger than 9. We
selected four compounds (B1eB4), which have an LE value larger
than 0.2940, to synthesize. Again, these compounds were selected
because they have never been reported before and can be very
easily synthesized without too much synthetic efforts in our cur-
rent experimental conditions.
Fig. 7. The binding mode of compound B4 with SYK.

536
G.-B. Li et al. / European Journal of Medicinal Chemistry 93 (2015) 523e538
Fig. 8. In vitro growth inhibitory profile of compound B4 against (A) Ramos B cells and (B) MV4-11 cells using the MTT assay. (C) The inhibitory effects of compound B4 on
phosphorylation and downstream signals of SYK in MV4-11 cells was examined using the western blot analysis.
3.4.1. Chemical synthesis for compounds B1eB4
64 nM). Fig. 7A depicts the interaction mode between compound
B4 and SYK. Compound B4 perfectly interacts with the residues in
the binding site of SYK. Apparently, the assembled fragment, 2-
methylquinolin-4-amine, forms a strong hydrogenebond interac-
tion with Asp512 and aromatic interactions with Phe382, which
reflect that this fragment is very suitable to the hydrophobic pocket
surrounded by residues of Phe382, Val433 (see Fig. 7B).
The general synthetic route chosen for the synthesis of com-
pounds B1eB4 is depicted in Scheme 2. Compounds B1eB4 were
all synthesized from the initial compound, 2,4-dichloro-5-
fluoropyrimidine, which contains two regioselective chlorine
atoms. Firstly, commercially available or readily synthesized
amines (F4eF6, 1.0 equiv) were reacted regioselectively with the 4-
chlorine of 2,4-dichloro-5-fluoropyrimidine (1.2 equiv) at
90 ꢂCe100 ꢂC in the presence of diisopropylethylamine (DIPEA) to
afford the corresponding 2-chloro-5-fluoro-N-substitued-pyr-
imidin-4-amine intermediates. Subsequently, the nucleophilic
displacement of 2-chlorine of the intermediates by 3,4,5-
trimethoxyaniline (1.0 equiv) or 4-(4-methylpiperazin-1-yl)aniline
(1.0 equiv) led to the target compounds B1eB4 with the yields
ranging from 32% to 95%, which required conditions of a temper-
ature of 90 ꢂCor 125 ꢂC and the catalysis of acid (hydrochloric acid
for 4-(4-methylpiperazin-1-yl)aniline) or base (DIPEA for 3,4,5-
trimethoxyaniline). Details about compounds B1eB4 are given in
Supporting Information.
3.4.3. Functional evaluations of compound B4
We first assessed the inhibitory effect of compound B4 on tumor
cell viability by MTT assays. Here, two tumor cell lines, Ramos B [45]
and MV4-11 [46], were used; the two tumor cell lines were chosen
because their viabilities have been reported to be regulated by SYK
[45,46]. Compound B4 displayed a good anti-viability potency
against both Ramos B and MV4-11 cell lines with IC50 values of
0.51 mM and 0.003 mM, respectively (Fig. 8A and B). For both cell
lines, compound B4 showed a higher potency than R406. Then, we
used western blot analysis to examine the effect of compound B4
on the SYK signaling in intact MV4-11 cells. As shown in Fig. 8C,
compound B4 significantly inhibited the phosphorylation of FLT3
and ERK1/2, which are SYK downstream signaling proteins. All of
these demonstrated that compound B4 has a good functional
activity.
3.4.2. Inhibitory activities of compounds B1eB4 against SYK
The inhibitory activities of compounds B1eB4 are present in
Fig. 6. The four compounds all exhibited potency against SYK with
IC50 values in the range of 2958 nMe13 nM. However only com-
Though a more potent SYK inhibitor was obtained here, the
performance of LEADOPT was apparently poorer than that in the
pound B4 showed a higher potency (IC50: 13 nM) than R406 (IC50
:

G.-B. Li et al. / European Journal of Medicinal Chemistry 93 (2015) 523e538
537
[7] U. Rester, From virtuality to reality e virtual screening in lead discovery and
lead optimization: a medicinal chemistry perspective, Curr. Opin. Drug Discov.
Devel. 11 (2008) 559e568.
[8] V.M. Popov, W.A. Yee, A.C. Anderson, Towards in silico lead optimization:
scores from ensembles of protein/ligand conformations reliably correlate with
biological activity, Proteins 66 (2007) 375e387.
case study of VEGFR2 inhibitor optimization. In fact, because we
just chose a very limited number of simple compounds to synthe-
size and measured their potency in the two case studies, it is hard
to make a performance comparison from the results. Of course, we
also acknowledge the possibility that LEADOPT has a different
performance on different targets.
[9] W.L. Jorgensen, Efficient drug lead discovery and optimization, Acc. Chem.
Res. 42 (2009) 724e733.
[10] L.-L. Yang, G.-B. Li, H.-X. Yan, Q.-Z. Sun, S. Ma, P. Ji, Z.-R. Wang, S. Feng, J. Zou,
S.-Y. Yang, Discovery of N6-phenyl-1H-pyrazolo[3,4-d]pyrimidine-3,6-
diamine derivatives as novel CK1 inhibitors using common-feature pharma-
cophore model based virtual screening and hit-to-lead optimization, Eur. J.
Med. Chem. 56 (2012) 30e38.
[11] R.D. Cramer, D.E. Patterson, J.D. Bunce, Comparative molecular field analysis
(CoMFA). 1. Effect of shape on binding of steroids to carrier proteins, J. Am.
Chem. Soc. 110 (1988) 5959e5967.
[12] G. Klebe, U. Abraham, T. Mietzner, Molecular similarity indices in a compar-
ative analysis (CoMSIA) of drug molecules to correlate and predict their bio-
logical activity, J. Med. Chem. 37 (1994) 4130e4146.
[13] C.L. Kuo, H. Assefa, S. Kamath, Z. Brzozowski, J. Slawinski, F. Saczewski,
J.K. Buolamwini, N. Neamati, Application of CoMFA and CoMSIA 3D-QSAR and
docking studies in optimization of mercaptobenzenesulfonamides as HIV-1
integrase inhibitors, J. Med. Chem. 47 (2004) 385e399.
4. Concluding remarks
In this investigation, we developed an automatic tool for
structure-based lead optimization, termed LEADOPT. LEADOPT
starts its work from an input structure of ligandereceptor complex,
which can come from either x-ray crystal structure or molecular
docking. The structural modifications in LEADOPT mainly include
two operations: fragment growing and fragment replacing, which
are restricted to carry out in the active pocket of target protein with
the core scaffold structure of ligand kept unchanged. LE rather than
scoring function is used as a measure to sort the newly generated
molecules. Twelve important pharmacokinetic and toxic properties
are also evaluated using SCADMET. LEADOPT was evaluated with
two retrospective cases. The results showed that LEADOPT could
generate derivatives that are very similar with known potent
compounds. LEADOPT was finally applied to structural optimiza-
tions of VEGFR2 inhibitor sorafenib, and SYK inhibitor R406.
Though just several compounds were synthesized, we have ob-
tained some compounds that are more potent than sorafenib and
R406. All of these have validated, at least to some extent, the
effectiveness of LEADOPT.
Finally, we have to mention that the binding affinities of the
generated molecules are estimated directly by ID-Score without
further sophisticated molecular docking because the main purpose
of LEADOPT is to provide a rapid automatic tool for lead optimi-
zation. In addition, to make the generated molecules easy to syn-
thesize, atom mutations are not allowed in LEADOPT. It is also
necessary to point out that, although pharmacokinetic and toxic
properties are evaluated in LEADOPT, we did not validate these
issues in this investigation mainly due to our very limited experi-
mental resource. Even so, it does not hamper LEADOPT as a useful
tool for experimental medicinal chemists.
[14] B.M. Baron, R.J. Cregge, R.A. Farr, D. Friedrich, R.S. Gross, B.L. Harrison,
D.A. Janowick, D. Matthews, T.C. McCloskey, S. Meikrantz, P.L. Nyce, R. Vaz,
W.A. Metz, CoMFA, synthesis, and pharmacological evaluation of (E)-3-(2-
carboxy-2-arylvinyl)-4,6-dichloro-1H-indole-2-carboxylic acids: 3-[2-(3-
aminophenyl)-2-carboxyvinyl]-4,6-dichloro-1H-indole-2-carboxylic acid,
a
potent selective glycine-site NMDA receptor antagonist, J. Med. Chem. 48
(2005) 995e1018.
[15] A. Politi, S. Durdagi, P. Moutevelis-Minakakis, G. Kokotos, M.G. Papadopoulos,
T. Mavromoustakos, Application of 3D QSAR CoMFA/CoMSIA and in silico
docking studies on novel renin inhibitors against cardiovascular diseases, Eur.
J. Med. Chem. 44 (2009) 3703e3711.
[16] B. Wendt, M. Mulbaier, S. Wawro, C. Schultes, J. Alonso, B. Janssen, J. Lewis,
Toluidinesulfonamide hypoxia-induced factor 1 inhibitors: alleviating drug-
drug interactions through use of PubChem data and comparative molecular
field analysis guided synthesis, J. Med. Chem. 54 (2011) 3982e3986.
[17] S. Sartini, V. Coviello, A. Bruno, V. La Pietra, L. Marinelli, F. Simorini, S. Taliani,
S. Salerno, A.M. Marini, A. Fioravanti, P. Orlandi, A. Antonelli, F. Da Settimo,
E. Novellino, G. Bocci, C. La Motta, Structure-based optimization of tyrosine
kinase inhibitor CLM3. Design, synthesis, functional evaluation and molecular
modeling studies, J. Med. Chem. 57 (2014) 1225e1235.
[18] M. Mori, A. Massaro, V. Calderone, M. Fragai, C. Luchinat, A. Mordini, Discovery
of a new class of potent MMP inhibitors by structure-based optimization of
the Arylsulfonamide scaffold, ACS Med. Chem. Lett. 4 (2013) 565e569.
[19] J. Jacobs, V. Grum-Tokars, Y. Zhou, M. Turlington, S.A. Saldanha, P. Chase,
A. Eggler, E.S. Dawson, Y.M. Baez-Santos, S. Tomar, A.M. Mielech, S.C. Baker,
C.W. Lindsley, P. Hodder, A. Mesecar, S.R. Stauffer, Discovery, synthesis, and
structure-based optimization of a series of n-(tert-Butyl)-2-(n-arylamido)-2-
(pyridin-3-yl) Acetamides (ML188) as potent Noncovalent small molecule
inhibitors of the Severe Acute Respiratory Syndrome Coronavirus (SARS-CoV)
3CL Protease, J. Med. Chem. 56 (2013) 534e546.
Acknowledgments
[20] J.D. Bauman, D. Patel, S.F. Baker, R.S.K. Vijayan, A. Xiang, A.K. Parhi,
L. Martinez-Sobrido, E.J. LaVoie, K. Das, E. Arnold, Crystallographic fragment
screening and structure-based optimization yields a new class of influenza
endonuclease inhibitors, ACS Chem. Biol. 8 (2013) 2501e2508.
[21] O. Nicolotti, I. Giangreco, T.F. Miscioscia, A. Carotti, Improving quantitative
structure-activity relationships through multiobjective optimization, J. Chem.
Inf. Model. 49 (2009) 2290e2302.
This work was supported by the 973 Program (2013CB967204),
the 863 Hi-Tech Programs (2012AA020301, 2012AA020308), the
National Natural Science Funds for Distinguished Young Scholar
(81325021), the National S&T Major Project (2012ZX09501001-
003), and partly by PCSIRT (NO: IRT13031).
[22] C. Abad-Zapatero, J.T. Metz, Ligand efficiency indices as guideposts for drug
discovery, Drug Discov. Today 10 (2005) 464e469.
[23] C. Abad-Zapatero, Ligand efficiency indices for effective drug discovery,
Expert. Opin. Drug Discov. 2 (2007) 469e488.
[24] A. Gaulton, L.J. Bellis, A.P. Bento, J. Chambers, M. Davies, A. Hersey, Y. Light,
S. McGlinchey, D. Michalovich, B. Al-Lazikani, J.P. Overington, ChEMBL: a
large-scale bioactivity database for drug discovery, Nucleic Acids Res. 40
(2012) D1100eD1107.
[25] D.S. Wishart, C. Knox, A.C. Guo, D. Cheng, S. Shrivastava, D. Tzur, B. Gautam,
M. Hassanali, DrugBank: a knowledgebase for drugs, drug actions and drug
targets, Nucleic Acids Res. 36 (2008) D901eD906.
Appendix A. Supplementary data
Supplementary data related to this article can be found at http://
dx.doi.org/10.1016/j.ejmech.2015.02.019.
References
[26] X.Q. Lewell, D.B. Judd, S.P. Watson, M.M. Hann, RECAP-retrosynthetic
combinatorial analysis procedure: a powerful new technique for identifying
privileged molecular fragments with useful applications in combinatorial
chemistry, J. Chem. Inf. Comput. Sci. 38 (1998) 511e522.
[27] G.B. Li, L.L. Yang, W.J. Wang, L.L. Li, S.Y. Yang, ID-Score: a new empirical
scoring function based on a comprehensive set of descriptors related to
protein-ligand interactions, J. Chem. Inf. Model. 53 (2013) 592e600.
[28] I.D. Kuntz, K. Chen, K.A. Sharp, P.A. Kollman, The maximal affinity of ligands,
Proc. Natl. Acad. Sci. USA 96 (1999) 9997e10002.
[1] T. Kenakin, Predicting therapeutic value in the lead optimization phase of drug
discovery, Nat. Rev. Drug Discov. 2 (2003) 429e438.
[2] G.M. Keseru, G.M. Makara, The influence of lead discovery strategies on the
properties of drug candidates, Nat. Rev. Drug Discov. 8 (2009) 203e212.
[3] C.W. Murray, D.C. Rees, The rise of fragment-based drug discovery, Nat. Chem.
1 (2009) 187e192.
[4] L.M. Mayr, D. Bojanic, Novel trends in high-throughput screening, Curr. Opin.
Pharmacol. 9 (2009) 580e588.
[29] C.H. Reynolds, B.A. Tounge, S.D. Bembenek, Ligand binding efficiency: trends,
physical basis, and implications, J. Med. Chem. 51 (2008) 2432e2438.
[30] S.Y. Yang, Q. Huang, L.L. Li, C.Y. Ma, H. Zhang, R. Bai, Q.Z. Teng, M.L. Xiang,
Y.Q. Wei, An integrated scheme for feature selection and parameter setting in
the support vector machine modeling and its application to the prediction of
[5] J.P. Kennedy, L. Williams, T.M. Bridges, R.N. Daniels, D. Weaver, C.W. Lindsley,
Application of combinatorial chemistry science on modern drug discovery,
J. Comb. Chem. 10 (2008) 345e354.
[6] C. McInnes, Virtual screening strategies in drug discovery, Curr. Opin. Chem.
Biol. 11 (2007) 494e502.

538
G.-B. Li et al. / European Journal of Medicinal Chemistry 93 (2015) 523e538
pharmacokinetic properties of drugs, Artif. Intell. Med. 46 (2009) 155e163.
[39] S. Wilhelm, C. Carter, M. Lynch, T. Lowinger, J. Dumas, R.A. Smith, B. Schwartz,
R. Simantov, S. Kelley, Discovery and development of sorafenib: a multikinase
inhibitor for treating cancer, Nat. Rev. Drug Discov. 5 (2006) 835e844.
[40] L.L. Yang, G.B. Li, S. Ma, C. Zou, S. Zhou, Q.Z. Sun, C. Cheng, X. Chen, L.J. Wang,
S. Feng, L.L. Li, S.Y. Yang, Structure-activity relationship studies of pyrazolo
[31] H. Zhang, Q.Y. Chen, M.L. Xiang, C.Y. Ma, Q. Huang, S.Y. Yang, In silico pre-
diction of mitochondrial toxicity by using GA-CG-SVM approach, Toxicol.
Vitro 23 (2009) 134e140.
[32] H. Zhang, M.L. Xiang, C.Y. Ma, Q. Huang, W. Li, Y. Xie, Y.Q. Wei, S.Y. Yang,
Three-class classification models of logS and logP derived by using GA-CG-
SVM approach, Mol. Divers. 13 (2009) 261e268.
[33] C.Y. Ma, S.Y. Yang, H. Zhang, M.L. Xiang, Q. Huang, Y.Q. Wei, Prediction models
of human plasma protein binding rate and oral bioavailability derived by
using GA-CG-SVM method, J. Pharm. Biomed. Anal. 47 (2008) 677e682.
[34] T.C. Tran, B. Sneed, J. Haider, D. Blavo, A. White, T. Aiyejorun, T.C. Baranowski,
A.L. Rubinstein, T.N. Doan, R. Dingledine, E.M. Sandberg, Automated, quanti-
tative screening assay for Antiangiogenic compounds using transgenic
zebrafish, Cancer Res. 67 (2007) 11386e11392.
[3,4-d]pyrimidine derivatives leading to the discovery of
a novel multi-
kinase inhibitor that Potently inhibits FLT3 and VEGFR2 and evaluation of its
activity against acute myeloid leukemia in vitro and in vivo, J. Med. Chem. 56
(2013) 1641e1655.
[41] A. Mocsai, J. Ruland, V.L.J. Tybulewicz, The SYK tyrosine kinase: a crucial
player in diverse biological functions, Nat. Rev. Immunol. 10 (2010) 387e402.
[42] M. Ulanova, F. Duta, L. Puttagunta, A.D. Schreiber, A.D. Befus, Spleen tyrosine
kinase (Syk) as a novel target for allergic asthma and rhinitis, Expert Opin.
Ther. Targets 9 (2005) 901e921.
[35] M. Zak, C.A. Hurley, S.I. Ward, P. Bergeron, K. Barrett, M. Balazs, W.S. Blair,
R. Bull, P. Chakravarty, C. Chang, Identification of C-2 hydroxyethyl imidazo-
pyrrolopyridines as potent JAK1 inhibitors with favorable physicochemical
properties and high selectivity over JAK2, J. Med. Chem. 56 (2013)
4764e4785.
[36] S. Naud, I.M. Westwood, A. Faisal, P. Sheldrake, V. Bavetsias, B. Atrash,
K.M. Cheung, M. Liu, A. Hayes, J. Schmitt, A. Wood, V. Choi, K. Boxall, G. Mak,
M. Gurden, M. Valenti, A. de Haven Brandon, A. Henley, R. Baker,
C. McAndrew, B. Matijssen, R. Burke, S. Hoelder, S.A. Eccles, F.I. Raynaud,
S. Linardopoulos, R.L. van Montfort, J. Blagg, Structure-based design of orally
bioavailable 1H-pyrrolo[3,2-c]pyridine inhibitors of mitotic kinase monopolar
spindle 1 (MPS1), J. Med. Chem. 56 (2013) 10045e10065.
[37] N. Ferrara, R.S. Kerbel, Angiogenesis as a therapeutic target, Nature 438 (2005)
967e974.
[38] M. McTigue, B.W. Murray, J.H. Chen, Y.L. Deng, J. Solowiej, R.S. Kania, Mo-
lecular conformations, interactions, and properties associated with drug ef-
ficiency and clinical performance among VEGFR TK inhibitors, Proc. Natl.
Acad. Sci. USA 109 (2012) 18281e18289.
[43] M.E. Weinblatt, A. Kavanaugh, M.C. Genovese, T.K. Musser, E.B. Grossbard,
D.B. Magilavy, An oral spleen tyrosine kinase (Syk) inhibitor for rheumatoid
arthritis, New. Engl. J. Med. 363 (2010) 1303e1312.
[44] A.G. Villasenor, R. Kondru, H. Ho, S. Wang, E. Papp, D. Shaw, J.W. Barnett,
M.F. Browner, A. Kuglstatter, Structural insights for design of potent spleen
tyrosine kinase inhibitors from crystallographic analysis of three inhibitor
complexes, Chem. Biol. Drug Des. 73 (2009) 466e470.
[45] F. Padilla, N. Bhagirath, S. Chen, E. Chiao, D.M. Goldstein, J.C. Hermann, J. Hsu,
J.J. Kennedy-Smith, A. Kuglstatter, C. Liao, W. Liu, L.E. Lowrie, K.C. Luk,
S.M. Lynch, J. Menke, L. Niu, T.D. Owens, C. O-Yang, A. Railkar, R.C. Schoenfeld,
~
M. Slade, S. Steiner, Y.-C. Tan, A.G. Villasenor, C. Wang, J. Wanner, W. Xie,
D. Xu, X. Zhang, M. Zhou, M.C. Lucas, Pyrrolopyrazines as selective spleen
tyrosine kinase inhibitors, J. Med. Chem. 56 (2013) 1677e1692.
[46] A. Puissant, N. Fenouille, G. Alexe, Y. Pikman, C.F. Bassil, S. Mehta, J. Du,
J.U. Kazi, F. Luciano, L. Ronnstrand, A.L. Kung, J.C. Aster, I. Galinsky, R.M. Stone,
D.J. DeAngelo, M.T. Hemann, K. Stegmaier, SYK is a critical regulator of FLT3 in
acute myeloid leukemia, Cancer Cell 25 (2014) 226e242.