Discovery, optimization and biological evaluation for novel c-Met kinase inhibitors

Article abstract of DOI:10.1016/j.ejmech.2017.11.073

The c-Met kinase has emerged as an attractive target for developing antitumor agents because of its close relationship with the development of many human cancers, poor clinical outcomes and even drug resistance. A series of novel c-Met kinase inhibitors h

Full text of DOI:10.1016/j.ejmech.2017.11.073

Accepted Manuscript
Discovery, optimization and biological evaluation for novel c-Met kinase inhibitors
Haoliang Yuan, Qiufeng Liu, Li Zhang, Shihe Hu, Tiantian Chen, Huifang Li, Yadong
Chen, Yechun Xu, Tao Lu
PII:
S0223-5234(17)30976-5
10.1016/j.ejmech.2017.11.073
EJMECH 9948
DOI:
Reference:
To appear in: European Journal of Medicinal Chemistry
Received Date: 23 September 2017
Revised Date: 23 November 2017
Accepted Date: 26 November 2017
Please cite this article as: H. Yuan, Q. Liu, L. Zhang, S. Hu, T. Chen, H. Li, Y. Chen, Y. Xu, T. Lu,
Discovery, optimization and biological evaluation for novel c-Met kinase inhibitors, European Journal of
Medicinal Chemistry (2017), doi: 10.1016/j.ejmech.2017.11.073.
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to
our customers we are providing this early version of the manuscript. The manuscript will undergo
copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please
note that during the production process errors may be discovered which could affect the content, and all
legal disclaimers that apply to the journal pertain.

ACCEPTED MANUSCRIPT
Discovery, Optimization and Biological Evaluation for Novel c-Met kinase
Inhibitors
Haoliang Yuan,ab# Qiufeng Liu,cde# Li Zhang,a# Shihe Hu,a Tiantian Chen,c Huifang Li,f Yadong Chen,a
Yechun Xuc* and Tao Luab*
a Laboratory of molecular design and drug discovery, School of Science, b State Key Laboratory of Natural
Medicines, Jiangsu Key Laboratory of Drug Discovery for Metabolic Disease, Institute of Pharmaceutical
Research, China Pharmaceutical University, 24 Tongjiaxiang, Nanjing 210009, China.
c
CAS Key Laboratory of Receptor Research, Drug Discovery and Design Center, Shanghai Institute of
Materia Medica, Chinese Academy of Sciences (CAS), Shanghai 201203, China.
d School of Life Science and Technology, ShanghaiTech University, Shanghai, 200031, China
e University of Chinese Academy of Sciences, Beijing, 100049, China
f
Vancouver Prostate Centre, University of British Columbia, 2660 Oak Street, Vancouver, British
Columbia V6H 3Z6, Canada
# Authors contributed equally to this work.
Corresponding Author
*Yechun Xu
Phone:86-21-50801267; Fax: 86-21-50801267; E-mail: ycxu@simm.ac.cn.
*Tao Lu
Phone: 86-25-86185086; Fax: 86-25-83302827; E-mail: lutao@cpu.edu.cn.
Notes
The authors declare no competing interest.

ACCEPTED MANUSCRIPT
ABSTRACT
The c-Met kinase has emerged as an attractive target for developing antitumor agents because of its close
relationship with the development of many human cancers, poor clinical outcomes and even drug
resistance. A series of novel c-Met kinase inhibitors have been identified with multiple workflow in this
work, including virtual screening, X-ray crystallography, biological evaluation and structural optimization.
The experimentally determined crystal structure of the best hit compound HL-11 in c-Met kinase domain
was highly consistent with the computational prediction. Comparison of the hit compounds with different
c-Met kinase inhibitory activity by molecular dynamics simulations suggested the key protein-ligand
interactions for structural optimization. Based on these, structural optimization produced compound 11e
with better c-Met kinase inhibitory activity and improved anti-proliferative activity. These experimental
findings proved the reliability and efficiency of our in silico methods. This strategy will facilitate further
lead discovery and optimization for novel c-Met kinase inhibitors.
Keywords
c-Met, Virtual Screening, Pharmacophore, X-ray Crystallography, Structural Optimization

ACCEPTED MANUSCRIPT
1. Introduction
The c-Met kinase is a receptor tyrosine kinase that expressed in endothelial and epithelial cells.
Hepatocyte growth factor (HGF) is the natural ligand of c-Met, also known as a scatter factor. The
HGF/c-Met signaling pathway plays important roles in normal development, organogenesis, and
homeostasis [1]. The activation of c-Met by HGF causes receptor dimerization and autophosphorylation of
tyrosine 1234 and 1235. Phosphorylated c-Met further triggers the activation of downstream signaling
pathways such as the Ras/MAPK, c-Src, and PI3K/Akt pathways. Meanwhile, c-Met induces an invasive
program consisting of cell proliferation, migration, invasion, and survival that is essential for physiological
events during normal processes such as morphogenesis, liver regeneration, and wound healing. Aberrant
HGF/c-Met signaling through constitutive activation, gene amplification, mutations, and activation of an
autocrine loop occurs in virtually all types of solid tumors, and as such is implicated in multiple tumor
oncogenic processes, such as mitogenesis, survival, angiogenesis, and invasive growth, especially in the
metastatic process [2-5]. Furthermore, overexpression of c-Met and HGF was demonstrated to correlate
with poor prognosis or metastatic progression in a number of major human cancers, including lung,
prostate, renal, ovarian, gastric, and liver cancers [6]. In this sense, c-Met has recently attracted
considerable interest as a therapeutic target for various cancers [7].
The c-Met kinase inhibitors are grouped according to their binding mode in the binding site. Compounds
with a U-shaped binding mode in the ATP binding site are Type I c-Met kinase inhibitors. In general,
compounds of this type exhibit relatively good cellular selectivity profiles with a limited number of
off-target kinase hits, such as Crizotinib and PF-04217903. They compete with ATP to bind in the ATP
binding site and mimic the interactions between ATP and the binding site, including the hydrogen bond
interactions with the residues in the hinge region (Pro1158, Tyr1159 and Met1160). Interaction with the
activation loop is also important to kinase inhibitors, including DFG motif and Tyr1230 in c-Met kinase.
The DFG motif (Asp1222, Phe1223 and Gly1224) is conserved in kinase, interactions with which are
beneficial for improving kinase inhibitory activity. Furthermore, Tyr1230 in c-Met kinase domain is unique
so that compounds got interactions with it showed good selectivity. Type II c-Met kinase inhibitors are
multi-targeted c-Met kinase inhibitors that pass the gatekeeper and occupy the deep hydrophobic back
pocket. A significant movement of the A-loop is required for entering the c-Met back pocket which is
compensated by additional interactions with the hydrophobic back pocket, often leading to compounds with
high molecular weights and high lipophilicities. But hydrogen bond interactions with the hinge region are

ACCEPTED MANUSCRIPT
the shared protein-ligand interactions of Type I and II c-Met kinase inhibitors. The representative c-Met
inhibitors of Type I, II and other types in different clinical phase studies have been shown in Fig. 1.
Especially, Crizotinib (c-Met/ALK) and Cabozantinib (VEGFR2/c-Met) have been approved for the
treatment of non-small cell lung carcinoma [6]. Besides, some other c-Met kinase inhibitors have also been
put into clinical studies, including PF-04217903 and JNJ-38877605 of Type I, BMS-777607, GSK1363089
and MGCD-265 of Type II.
Type I
O
N
NH
N
HO
N
Cl
O
N
N
N
N
HN
F
N
F F
N
N
N
N
N
N
Cl
N
S
N
N
N
N
N
N
N
N
N
H2N
N
AMG-208
Phase I
JNJ-38877605
Phase I
Crizotinib/PF-02341066
Launched
PF-04217903
Phase I
Type II
H
H
O
N
N
H
N
F
N
N
F
O
N
O
O
O
N
F
O
O
NH
O
O
O
O
N
F
N
H2N
O
N
S
N
Cl
F
MGCD-265
Phase I/II
BMS-777607
Phase I/II
Cabozantinib/XL-184
Launched
Other Types
H
H
O
F
N
N
O
O
O
O
H
O
F
N
N
N
N
S
O
N
O
O
O
HN
N
N
O
O
N
O
N
H
XL-880/GSK1363089
Phase II
MK-2461
Phase I/II
ARQ-197/tivantinib
Phase I
Fig. 1 The representative c-Met kinase inhibitors of different structural types.
Encouraged by these findings, a multiple workflow, including virtual screening, biological evaluation,
X-ray crystallography, molecular dynamics simulation [8] and structural optimization, has been applied to
identify novel c-Met inhibitors in this study. Complex-based pharmacophore model and molecular docking
were used as the structure-based approaches, while the ligand-based approaches were ligand-based
pharmacophore we have developed [9], 2D chemical similarity, and 3D-shape. These two types of
approaches focused on different aspects and complemented each other. A series of novel and potent c-Met
inhibitors have been identified after biological evaluation. The experimentally resolved crystal structure of

ACCEPTED MANUSCRIPT
the best hit compound HL-11 in complex with c-Met kinase domain validated the computational
prediction. Structural optimization yielded compound 11e with better c-Met kinase inhibitory activity and
significantly improved anti-proliferative activity.
2. Materials and Methods
2.1 Structure-Based Virtual Screening
Molecular Docking. The available programs, Glide, Gold, Surflex and CDocker, were used for docking.
As a test set, 11 ligands in the crystal structures of the kinase domain of c-Met were extracted and
re-docked into the binding site. The PDB IDs for the 11 complex structures were 2RFS [10], 2WD1 [11],
2WGJ [12], 2WKM [13], 3A4P [14], 3CCN [15], 3CD8 [15], 3DKF [16], 3DKG [16], 3F66 [17], and 3I5N
[18]. The root mean square deviation (RMSD) between the docked and crystal conformations of the ligand
was a criteria to assess the docking accuracy.
Glide. The protein in the crystal structures was prepared using the Protein Preparation Wizard workflow.
The generated receptor grid was centered on the ligand in the crystal structure, which was defined as the
ligand-binding site search region. The compounds to be docked was confirmed by an enclosing box that
was similar in size to the crystal ligand. The compounds were prepared with LigPrep and docked into the
binding site using Glide standard/extra precision (SP/XP) mode. Default settings were used for all the other
parameters. The best pose was determined by the protein-ligand interactions and Glide score. All these
procedures were completed in Schrödinger 2009.
Gold. Gold uses a genetic algorithm to explore the binding site for bioactive ligands. Protein structures
were prepared by adding hydrogen atoms and deleting water molecules as well as the ligand in Sybyl 6.9. A
10 Å radius surrounding the ligand was defined as the ligand binding site. Only the best conformation of
the docked ligand was included in the output and other settings in the genetic algorithm were left default.
Surflex. An empirical scoring function was used to dock ligands into the binding site. The active site was
defined as protomol, an idealized ligand that makes every potential interaction with the binding site. It was
used to align the ligand fragments and also provide the direction to grow the fragments. All the preparations
were completed in Sybyl 6.9.
CDocker. CDocker is a grid-based molecular docking method that employs CHARMm. During the
docking procedure, the receptor keeps rigid while ligand is flexible. Protein and ligands were prepared in
DS 2.5 (Accelrys Discovery Studio 2.5). The binding site was defined based on the volume occupied by the
bound ligand. Only the best conformation of the docked ligand was output.

ACCEPTED MANUSCRIPT
Generation of Complex-based Pharmacophore Models. We focused on developing a complex-based
pharmacophore model for type I c-Met inhibitors. At the outset of this study, only 11 crystal structures of
c-Met kinase domain in complex with type I inhibitors were available in PDB. All these complex structures
were aligned by Cα atoms so that the active sites were thus aligned. Water molecules were retained in order
to include the possible water-mediated hydrogen bonds.
Pharmacophore model was generated for each complex based on the aligned structures. The
protein-ligand interactions were converted to pharmacophoric features, including hydrophobic centers,
aromatic rings, and hydrogen bond donors and acceptors along with their direction vectors. Excluded
volume spheres representing different spatial positions occupied by adjacent residues were automatically
added to the final model. All features in the 11 pharmacophore models were clustered according to
different types of interactions with the active site. Excluded volume spheres used as steric constraints were
clustered based on the specific residues they represented. The cluster centers were identified using Edit and
Cluster Pharmacophore tool in DS 2.5. Finally, the pharmacophore model was complemented with
3D-shape constraints to improve the selectivity of the c-Met inhibitors.
Validation of Complex-based Pharmacophore Models. Retrospective discovery of known c-Met
inhibitors was used to validate the pharmacophores’ capability of distinguishing active inhibitors from
inactive compounds. A database included other 86 reported type I c-Met inhibitors and 538 randomly
selected decoy compounds was constructed. Decoys were collected from FDA-approved small molecule
drug without reported biological activities on c-Met in Drugbank (https://www.drugbank.ca/). Structures of
all these molecules were prepared using SciTegic Pipeline Pilot 7.5. Multi-conformations of each
compound were generated using the FAST conformational search protocol implemented in DS 2.5 with an
energy threshold of 20 kcal/mol and a maximum of 255 conformers. A multi-conformation database
included all the conformers of 624 compounds was resulted. All screening experiments were performed by
using Best Flexible Search algorithm. Enrichment Factor (EF), Goodness of Hit (GH) and receiver
operating characteristic (ROC) scores were used to evaluate the performance of pharmacophore screening.
2.2 Ligand-Based Virtual Screening
Ligand-based Pharmacophore Modeling. A HypoGen ligand-based pharmacophore model has been
developed and rigorously validated in our previous study [9]. The pharmacophore was generated using 23
known c-Met inhibitors of diverse structures. It consisted of two Hydrogen Bond Acceptors (HBA), one
Hydrophobic center (HY) and one Ring Aromatic (RA). The capability of it in identifying active

ACCEPTED MANUSCRIPT
compounds from inactive ones was illustrated with an enrichment factor of 21.07 and a ROC score of 0.896.
It was suggested that such a phamacophore model could serve as a reliable tool for the discovery of new
c-Met inhibitors.
2D Chemical Similarity and 3D-shape. Molecular similarity was calculated using SciTegic Pipeline Pilot
7.5. The c-Met inhibitors collected from literatures were used as the reference compounds. Tanimoto
coefficient between the ChemDiv and reference compounds was calculated. 3D-shape was applied in
combination with the complex-based pharmacophore model.
2.3 Crystal Structure Selection and Database Preparation
Crystal Structure Selection. Cross-docking was performed to select the most suitable crystal structure for
the structure-based virtual screening. The complex structures were firstly superimposed onto the reference
structure (PDB ID: 2WGJ). The co-crystal ligands were extracted and prepared before they were re-docked
using the selected docking programs. RMSD values between the crystal and docked conformations were
considered as the criteria for crystal structure selection. Because the docking accuracy may be affected by
the type of the bound inhibitor, a representative structure for each structural type of inhibitor was used in
the structure-based virtual screening.
Database Preparation. The ChemDiv compound library containing 0.7 million compounds was used for
virtual screening in this study. For molecular docking, all the structures in the compound database were
prepared using SciTegic Pipeline Pilot 7.5. They were protonated, added with partial charges, and then
minimized with the MMFF94x force field to a gradient of 0.0001 kcal/mol Å. For the
pharmacophore-based virtual screening, all the structures were subjected to conformation modeling using
the Catalyst module of DS 2.5. Multi-conformations of each compound were generated using the FAST
conformational search protocol implemented in DS 2.5 with an energy threshold of 20 kcal/mol and a
maximum of 255 conformers. These compounds were converted into a searchable multi-conformation
database.
2.4 Further Scoring
The compounds passed through the parallel virtual screening lines and cascade molecular docking were
then further scored. First, they were energy minimized in MOE 2009 (Molecular Operating Environment,
version 2009.10) to obtain their LigX_pKi binding affinity, which is mainly presented by the energy of
hydrogen bonds and hydrophobic interactions. Second, they were scored using molecular mechanics, the
generalized Born model, and the solvent accessibility (MM-GB/SA) method with OPLS_2005 and GB/SA

ACCEPTED MANUSCRIPT
in MacroModel of Schrödinger 2009 to calculate the free energies (E_Macromodel) of the optimal
chemical conformations. Third, another module implemented in Schrödinger 2009, Prime MM-GBSA
which calculates ligand binding energies and ligand strain energies between a set of ligands and a single
receptor, was also applied to calculate the binding free energies between the compounds and kinase domain
of c-Met. Including Glide Gscore, four different scoring methods in total were applied to select the
potential compounds from the docking results.
2.5 Kinase inhibitory and anti-proliferative assays
Chemical Source. The selected compounds were all purchased from Topscience
(http://www.tsbiochem.com/). The structure and purity of all chemicals were confirmed by mass
spectroscopy (MS) and liquid chromatography-tandem mass spectrometry (LC-MS/MS).
Biochemical kinase Assays. The c-Met kinase inhibitory activity was determined using Hot-SpotSM
kinase assay by Reaction Biology Corp. (Malvern PA, USA). After the substrate was prepared in freshly
prepared reaction buffer (20 mM Hepes pH 7.5, 10 mM MgCl2, 1 mM EGTA, 0.02% Brij35, 0.02 mg/mL
BSA, 0.1 mM Na3VO4, 2 mM DTT, 1% DMSO), 5 nM of human GST-tagged target kinase was delivered
into the substrate solution and mixed gently. The testing compounds were dissolved in 100% DMSO to
specific concentration and added into the kinase reaction mixture by Acoustic technology (Echo550;
33
nanoliter range). The reaction mixture was incubated for 20 min at room temperature. P-ATP (Specific
activity 10 µCi/µL) was delivered into the reaction mixture to initiate the reaction and incubated at room
temperature for 2 h. The kinase activities were detected by filter-binding method. IC50 values were obtained
using GraphPad Prism (San Diego, CA).
AlamarBlue® Cell proliferation Assays. The human tumor cell lines used in preliminary
anti-proliferation assays were A549 (lung), HCT116 (colon), MCF-7 (breast), PC3 (prostate) and
SGC-7901 (gastric) initially, and EBC-1 (lung), MKN45 (gastric), NCI-H460 (lung), PC3, Colo-205
(colon) were used for anti-proliferation assays of new compounds obtained after structural optimization. All
experiments were done under standard conditions (37 °C and 5% CO2). Cells were seeded in 384-well
plates in medium supplemented with 10% fetal bovine serum (FBS). After 24 hours, the wells were
changed with serum-free medium [with 0.04% bovine serum albumin (BSA)]. Cells were treated with test
compounds for 72 hours. Add 4 µl alamarBlue® reagent directly to cells in 40 µl culture medium. Incubate
for 1 to 4 hours at 37 °C in a cell culture incubator, protected from direct light. Detect the fluorescence
intensity at excitation wavelength of 540-570 nm (peak excitation is 570 nm) and fluorescence emission at

ACCEPTED MANUSCRIPT
580-610 nm (peak emission is 585 nm) with the Safire microplate Reader (Tecan, Switzerland). The IC50
values were calculated by fitting the data to the equation for competitive inhibition using nonlinear
regression method (GraphPad Prism, San Diego, CA).
2.6 Crystal Structure Determination
Protein Purification and Crystallization. Production of the kinase domain (1038-1346 aa) of
recombinant human c-Met followed the protocols of Wang [19]. with certain modifications. The cDNA
fragment was cloned into the vector pET28a and the protein was co-expressed with catYopH subcloned in
pET15b (164-468AA) [20]. The expressed c-Met kinase domain was passed through a Ni-NTA column
(Qiagen) and further purified by QHP ion exchange column (GE) which eluted with 25 mM Tris pH 8.5,
100 mM NaCl, 10% glycerol, 1mM DTT. The protein was concentrated to ~10 mg/mL for further
crystallization.
Cocrystallization of the c-Met kinase domain with compound HL-11 was carried out by mixing a
solution of the protein-ligand complex with an equal volume of precipitant solution (0.1M Tris pH7.5, 15%
glycerol, 12% MPD,5% isopropanol, 15% PEG5KMME). The protein-ligand complex was prepared by
adding the compound to the protein solution to a final concentration of 1 mM of HL-11. Cocrystallization
utilized the vapour-diffusion method in hanging drops. Crystals were flash frozen in liquid nitrogen in the
presence of well solution supplemented with 25% glycerol.
Structure Determination and Refinement. Data were collected at 100 K on beamline BL17U at the
Shanghai Synchrotron Radiation Facility (SSRF), and were processed with the XDS [21] software
packages. The structure was solved by molecular replacement, using the program PHASER [13] with the
search model of PDB ID 4GG5 [22]. The structure was refined with PHENIX [23]. With the aid of the
program Coot [24], compound HL-11, water molecules were fitted into to the initial Fo-Fc map. The
complete statistics, as well as the quality of the solved structures, are shown in Table S1.
2.7 Molecular Dynamics Simulation
Molecular dynamics simulations were performed using AMBER 9.0 software package with the ff99SB
force field to simulate c-Met in complex with its inhibitors and calculate their binding free energy. The
co-complex structures were obtained by molecular docking. The ligands were firstly fully minimized by the
AM1 method and electrostatic potentials computed at the HF/6-31G* level in the Gaussian 09 program.
The RESP fitting technique in AMBER was used to determine the partial charges. The force-field
parameters for the ligand were generated with the general AMBER force field (GAFF) by Antechamber.

ACCEPTED MANUSCRIPT
Hydrogen atoms were assigned using the LEaP module, which sets ionisable residues to their default
protonation states at a neutral pH value. Each complex was immersed in a cubic box of TIP3P water model
with a 10 Å minimum solute-wall distance. A total of five Cl¯ ions were added to neutralize each
protein-ligand complex system. Energy was minimized in the solvated system employing 1000 steps of the
steepest descent algorithm and 2000 steps of the conjugate gradient algorithm with a nonbonded cutoff of
10Å. The protocol for molecular dynamics simulation consisting of gradual heating, density equilibration,
equilibration and production procedures in an isothermal isobaric ensemble (NPT, P = 1 atm and T = 298K)
MD. The system was gradually heated from 0 to 298K in 50ps, followed by density equilibration at 298K
for 500ps, and then constant equilibration at 298K for 500ps. Then each protein-ligand complex system
underwent a process of equilibration procedure until the system achieved a continuous stable status, i.e.
production stage. The time step was set to 2 fs while the snapshots were taken every 10 ps to record the
conformation trajectory during production MD. The nonbonded interactions were treated with a 10Å cutoff.
SHAKE algorithm was applied to constrain all bonds involving hydrogen atoms to their equilibrium length.
The conformations of different systems were collected every 50ps after the system achieved their
equilibrium status. The collected snapshots were used for structural and energetic analysis of each complex
system.
Binding free energy calculations were also performed to investigate binding affinity between different
ligands and the binding pocket. The methods like MMPBSA [25] and MM/GBSA [26] are usually used for
investigating the energetic contribution of protein-ligand binding affinities. For different protein-ligand
systems, the process of converged status was used for the binding free energy calculation. The solute and
solvent dielectric constants were set as 1.0 and 80.0, respectively. The binding free energy of different
ligands to the protein was calculated as follows:
∆ꢀbinding=Gcomplex-[Gprotein+Gligand
]
For different protein-ligand systems, the process for binding free energy calculations were also applied
for hydrogen bond occupancy calculations. The hydrogen bond distance was set as 3.5Å and angle was
120.0°. Other parameters were kept default.
2.8 Chemistry
1H NMR and 13C NMR spectra were obtained at room temperature using a Bruker Avance 300
spectrometer. DMSO-d6 or CDCl3 containing 0.03% tetramethylsilane (TMS) (99.8% D, Adamas, Inc.) was
1
used as a solvent for NMR measurements. Chemical shifts (δ) for H NMR are given in parts per million

13
ACCEPTED MANUSCRIPT
(ppm) relative to residual DMSO (δ 2.50 ppm) or H2O (δ 3.33 ppm). Chemical shifts (δ) for C NMR are
given in ppm relative to DMSO (δ 39.50 ppm). Splitting patterns are designated as follows: s = singlet, d =
doublet, t = triplet, q = quartet, m = multiplet, br = broad. High resolution mass spectrometry (HRMS) was
obtained on a Q-TOF micro spectrometer. Melting points were determined with a Micro melting point
apparatus. TLC plates were visualized by exposure to ultraviolet light.
2.8.1 4-((4-amino-5-mercapto-4H-1,2,4-triazol-3-yl)methyl)phenol (1)
2-(4-hydroxyphenyl)acetic acid (5.02 g, 33 mmol), hydrazinecarbothiohydrazide (3.80 g, 36 mmol) and
CH3SO3H (6.91 g, 72mmol) were dissolved in 8 mL sulfolane and 6 mL H2O. The solution was stirred at
o
90 C for 24 h. After cooled to room temperature, the mixture was poured into 50 mL ice water. Saturated
aqueous sodium carbonate solution was added to adjust the PH to 7-8. The mixture was filtered and dried in
vacuo to give 3.60 g of 1 as blue solid: 50% yield. 1H NMR (300 MHz, DMSO-d6) δ 13.52 (s, 1H), 9.35 (s,
1H), 7.07 (d, J = 8.0 Hz, 2H), 6.69 (d, J = 7.3 Hz, 2H), 5.54 (s, 2H), 3.89 (s, 2H); m.p.: 206-208oC; ESI-MS
m/z: 223.2 [M-H]- .
2.8.2 4-((6-(1H-indol-2-yl)-[1,2,4]triazolo[3,4-b][1,3,4]thiadiazol-3-yl) methyl)phenyl 1H-indole-2-
carboxylate (2a)
1 (1.00 g, 4.50 mmol), 1H-indole-2-carboxylic acid (0.60 g, 5 mmol) were dissolved in 3 mL sulfolane, and
o
12 mL POCl3 was added. The reaction mixture was stirred at 85 C for 18 h, then cooled to room
temperature, poured into 200 mL ice water. NaOH was added to adjust the PH to 7. The mixture was
filtered and dried in vacuo to give 0.60 g of 2a as white solid: 27% yield. 1H NMR (300 MHz, DMSO-d6) δ
9.35 (s, 1H), 8.60 (s, 1H), 8.09 (s, 1H), 7.14 (d, J = 7.9 Hz, 2H), 6.75 – 6.66 (m, 2H), 4.26 (s, 2H), 3.93 (s,
3H); m.p.: 280-281oC; ESI-MS m/z: 491.3 [M+H]+.
2.8.3 4-((6-(1-methyl-1H-pyrazol-4-yl)-[1,2,4]triazolo[3,4-b][1,3,4] thiadiazol-3-yl)methyl)phenyl
1-methyl-1H-pyrazole-4-carboxylate (2b)
1
Synthetic procedure was followed as described for compound 2a to give 2b as white solid: 27% yield. H
NMR (300 MHz, DMSO-d6) δ 8.61 (s, 1H), 8.53 (s, 1H), 8.10 (s, 1H), 8.01 (s, 1H), 7.42 (d, J = 8.2 Hz, 2H),
7.18 (d, J = 8.0 Hz, 2H), 4.45 (s, 2H), 3.93 (s, 3H), 3.92 (s, 3H); m.p.: 167-168 oC; ESI-MS m/z: 421.2
[M+H]+.
2.8.4 4-((6-(1H-indol-2-yl)-[1,2,4]triazolo[3,4-b][1,3,4] thiadiazol-3-yl)methyl)phenol (3a)
2a (0.49 g, 1.00mmol) was dissolved in 15 mL THF and NaOH (0.2 g, 5 mmol) was added. The mixture
was stirred at room temperature for 24 h. 1N HCl was added to adjust the PH to 7 and the solution was

ACCEPTED MANUSCRIPT
extracted with ethyl acetate (20 mL x 2). The organic layer was washed with brine (30 mL), dried over
Na2SO4 and concentrated. The residue was purified by flash column chromatography (SiO2, chloroform /
MeOH) to give 0.32 g of 3a as gray solid: 91% yield. 1H NMR (300 MHz, DMSO-d6) δ 12.28 (s, 1H), 7.66
(d, J = 8.1 Hz, 1H), 7.46 (dd, J = 12.9, 8.3 Hz, 3H), 7.28 (dd, J = 22.1, 13.2 Hz, 4H), 7.11 (t, J = 7.6 Hz,
1H), 4.44 (s, 2H); m.p.: 191-193 oC; ESI-MS m/z: 348.4 [M+H]+.
2.8.5 4-((6-(1-methyl-1H-pyrazol-4-yl)-[1,2,4]triazolo[3,4-b][1,3,4] thiadiazol-3-yl)methyl)phenol (3b)
1
Synthetic procedure was followed as described for compound 3a to give 3b as white solid: 85% yield. H
NMR (300 MHz, DMSO-d6) δ 9.35 (s, 1H), 8.60 (s, 1H), 8.09 (s, 1H), 7.14 (d, J = 7.9 Hz, 2H), 6.75 – 6.66
(m, 2H), 4.26 (s, 2H), 3.93 (s, 3H); m.p.: 243 oC decomposed); ESI-MS m/z: 313.1 [M+H]+.
(
2.8.6 2-(4-((6-(1H-indol-2-yl)-[1,2,4]triazolo[3,4-b][1,3,4]thiadiazol-3-yl)methyl)phenoxy)-N,N-
dimethylacetamide (11a)
3a (94 mg, 0.30 mmol) and 2-chloro-N,N-dimethylacetamide (44 mg, 0.36 mmol) was dissolved in 10 mL
DMF and Cs2CO3 (0.32 g, 0.90 mmol) was added. The reaction mixture was stirred at 80 oC for 3 h. After
cooled to room temperature, 30 mL water was added to the reaction mixture, and the mixture was extracted
with ethyl acetate (30 mL x 2). The organic layer was washed with brine (50 mL), dried over Na2SO4 and
concentrated. The residue was purified by flash column chromatography (SiO2, chloroform / MeOH) to give
30 mg 11-1 as white solid: 23% yield. 1H NMR (300 MHz, DMSO-d6) δ 8.59 (s, 1H), 8.08 (d, J = 0.8 Hz, 1H),
7.25 (d, J = 8.3 Hz, 2H), 6.88 (d, J = 8.6 Hz, 2H), 4.79 (s, 2H), 4.32 (s, 2H), 3.92 (s, 3H), 3.57 (s, 4H), 3.43 (s,
4H); 13C NMR (75 MHz, DMSO-d6) δ 167.06, 158.76, 157.10, 138.12, 129.56 (2C), 127.84, 126.32, 125.01,
121.54, 120.59, 114.74 (2C), 112.36, 107.37, 65.79, 38.71, 35.55, 34.89, 29.53; m.p.: 225-227oC; HRMS
(ESI+) m/z 455.1244 (455.1261 calcd for C22H20N6NaO2S+, [M+Na]+).
2.8.7 2-(4-((6-(1-methyl-1H-pyrazol-4-yl)-[1,2,4]triazolo[3,4-b][1,3,4] thiadiazol-3-yl)methyl)phenoxy)
-1-morpholinoethan-1-one (11b)
Synthetic procedure was followed as described for compound 11-1 to give 11-2 as white solid: 40% yield. 1H
NMR (300 MHz, DMSO-d6) δ 8.59 (s, 1H), 8.08 (d, J = 0.8 Hz, 1H), 7.25 (d, J = 8.3 Hz, 2H), 6.88 (d, J = 8.6
Hz, 2H), 4.79 (s, 2H), 4.32 (s, 2H), 3.92 (s, 3H), 3.51-3.61 (brs, 4H), 3.48-3.40 (s, 4H); 13C NMR (75 MHz,
DMSO-d6) δ 165.96, 146.54, 137.66, 131.74, 129.61(2C), 127.93, 114.72(2C), 65.99(2C), 65.83, 44.73,
41.55, 29.50; m.p.: 164-166oC; HRMS (ESI+) m/z 462.1307 (462.1319 calcd for C20H21N7NaO3S+,
[M+Na]+).
2.8.8 2-((7-methoxyquinolin-4-yl)oxy)acetic acid (4)

o
ACCEPTED MANUSCRIPT
2-hydroxyacetic acid (5.00 g, 65 mmol) and KOH (6.00 g, 90 mmol) were mixed and heated to 170 C.
4-chloro-7-methoxyquinoline (5.00 g, 26 mmol) dissolved in DMSO (20 mL) was added dropwise. The
reaction mixture was stirred at 170 oC for 2.5 h. After cooled to room temperature, the solution was poured
to 50 mL ice water. Saturated aqueous sodium carbonate solution was added to adjust the PH to 7-8. The
mixture was filtered and dried in vacuo to give 3.70 g of 4 as brown solid: 62% yield. m.p.: 223-225oC;
+
HRMS (ESI+) m/z 234.0763 (234.0761 calcd for C12H12NO4 , [M+H]+).
2.8.9 4-amino-5-(((7-methoxyquinolin-4-yl)oxy)methyl) -4H-1,2,4-triazole-3-thiol (5)
4 (2.00 g, 8.60 mmol), hydrazinecarbothiohydrazide (1.00 g, 9.50 mmol) and CH3SO3H (1.80 g, 19 mmol)
were dissolved in 3 mL sulfolane and 3 mL H2O. The solution was stirred at 90 oC for 24 h. After cooled to
room temperature, the mixture was poured into 30 mL ice water. Saturated aqueous sodium carbonate
solution was added to adjust the PH to 7-8. The mixture was filtered and dried in vacuo to give 1.80 g of 5
as brown solid: 69% yield. HRMS (ESI+) m/z 304.0864 (304.0863 calcd for C13H14N5O2S+, [M+H]+).
2.8.10 6-(1H-imidazol-2-yl)-3-(((7-methoxyquinolin-4-yl)oxy)methyl) -[1,2,4]triazolo[3,4-b][1,3,4]
thiadiazole (11c)
5 (1.00 g, 3.30 mmol ) and 1H-imidazole-2-carboxylic acid (0.40 g, 3.60 mmol) were dissolved in 3 mL
sulfolane, and 12 mL POCl3 was added. The reaction mixture was stirred at 85 oC for 18 h, then cooled to
room temperature, poured into 200 mL ice water. NaOH was added to adjust the PH to 7. The mixture was
filtered and dried in vacuo to give 92 mg of 11-3 as gray solid: 7% yield. 1H NMR (300 MHz, DMSO-d6) δ
13.83 (s, 1H), 8.77 (d, J = 5.4 Hz, 1H), 8.01 (d, J = 9.0 Hz, 1H), 7.52 (s, 1H), 7.36 (d, J = 2.5 Hz, 1H), 7.28
(d, J = 5.2 Hz, 2H), 7.18 (dd, J = 9.2, 2.6 Hz, 1H), 5.82 (s, 2H), 3.90 (s, 3H); m.p.: 205-207oC; HRMS
(ESI+) m/z 380.0914 (380.0924 calcd for C17H14N7O2S+, [M+H]+).
2.8.11 3-(((7-methoxyquinolin-4-yl)oxy)methyl)-6-(thiazol-4-yl) -[1,2,4]triazolo[3,4-b][1,3,4]
thiadiazole (11d)
Synthetic procedure was followed as described for compound 11-3 to give 92 mg of 11-4 as yellow solid: 7%
yield. 1H NMR (300 MHz, Chloroform-d) δ 8.94 (d, J = 2.0 Hz, 1H), 8.74 (d, J = 5.4 Hz, 1H), 8.29 – 8.07
(m, 2H), 7.41 (d, J = 2.6 Hz, 1H), 7.18 – 7.08 (m, 1H), 7.04 (d, J = 5.4 Hz, 1H), 5.78 (s, 2H), 3.94 (s, 3H);
+
m.p.: 211-213oC; HRMS (ESI+) m/z 397.0528 (380.0536 calcd for C17H13N6O2S2 , [M+H]+).
2.8.12 6-(1H-indol-2-yl)-3-(((7-methoxyquinolin-4-yl)oxy)methyl) -[1,2,4]triazolo[3,4-b][1,3,4]
thiadiazole (11e)
Synthetic procedure was followed as described for compound 11-3 to give 113 mg of 11-5 as light brown
solid: 8% yield. 1H NMR (300 MHz, DMSO-d6) δ 12.33 (s, 1H), 8.75 (d, J = 5.2 Hz, 1H), 8.02 (d, J = 9.2

ACCEPTED MANUSCRIPT
Hz, 1H), 7.67 (d, J = 8.0 Hz, 1H), 7.52 – 7.23 (m, 5H), 7.20 – 7.00 (m, 2H), 5.83 (s, 2H), 3.89 (s, 3H); m.p.:
258-259oC; HRMS (ESI+) m/z 429.1125 (429.1128 calcd for C22H16N6O2S+, [M+H]+).
2.8.13 6-(1H-benzo[d]imidazol-5-yl)-3-(((7-methoxyquinolin-4-yl)oxy) methyl)-[1,2,4]triazolo[3,4-b]
[1,3,4]thiadiazole (11f)
Synthetic procedure was followed as described for compound 11-3 to give 156 mg of 11-6 as gray solid: 11%
yield. 1H NMR (300 MHz, DMSO-d6) δ 13.12 (s, 1H), 8.79 (d, J = 5.5 Hz, 1H), 8.44 (s, 1H), 8.20 (s, 1H),
8.10 (d, J = 9.1 Hz, 1H), 7.78 (s, 2H), 7.40 – 7.28 (m, 2H), 7.21 (dd, J = 9.3, 2.6 Hz, 1H), 5.93 (s, 2H), 3.91
(s, 3H); m.p.: 249-251oC; HRMS (ESI+) m/z 430.1069 (380.0536 calcd for C21H16N7O2S+, [M+H]+).
3. RESULTS AND DISCUSSIONS
3.1 Virtual Screening
3.1.1 Selection of Docking Programs and PDBs
Table 1. RMSD values between the crystal and docked conformations of inhibitors a
Glide(XP) b
PDB ID
CDocker
0.49
0.85
2.70
2.12
0.42
0.36
0.50
1.11
0.51
0.75
0.48
0.93
0.74
Gold Surflex
2RFS
0.34
0.28
0.64
1.59
0.15
0.15
0.18
0.28
0.31
0.86
0.25
0.46
0.42
1.52
0.63
0.72
1.03
1.93
0.24
0.64
0.28
0.42
0.72
0.40
0.78
0.50
1.33
0.73
0.48
2.54
0.79
0.43
0.20
0.21
0.42
0.32
0.31
0.71
0.66
2WD1
2WGJ
2WKM
3A4P
3CCN
3CD8
3DKF
3DKG
3F66
3I5N
Mean (Å)
Std. dev. (Å)
0Å<RMSD<0.5Å 0.5Å<RMSD<1Å 1Å<RMSD<1.5Å 1.5Å<RMSD<2Å 2Å<RMSD
aA color code bar is given according to the RMSD range. bGlide extra docking precision mode.
Four docking programs, including CDocker, Glide, Gold, and Surflex, were evaluated by re-generating
the ligand binding conformation in the 11 crystal structures. The resulted RMSDs (Table 1) between the
docked and crystal conformations of inhibitors suggested Glide and Gold were preferred for molecular
docking. Glide outperformed the other programs as only one RMSD value was over 1.00 Å (1.59Å for
2WKM) and the average RMSD value was also the smallest. Gold was also considered as a suitable
program since the docking results were all in the acceptable range. Therefore, Gold and Glide were selected
for the molecular docking in this study.

ACCEPTED MANUSCRIPT
Only limited number of crystal structures of c-Met kinase domain in complex with type I inhibitors were
available at the outset of this study. All these available crystal structures of c-Met kinase domain in
complex with type I inhibitors were evaluated. Considering both of the cross-docking accuracy and
structure types of the ligands, crystal structures (PDB ID: 2WD1 [11], 2WGJ [12], 2WKM [12], 3A4P [14],
and 3QTI [27]) were selected for cascade molecular docking (supplementary Table S2).
3.1.2 Complex-based Pharmacophore Generation and Validation
Complex-based pharmacophore models were generated based on the eleven crystal structures (Table 1)
with different type I c-Met inhibitors (supplementary Fig. S1) to cover comprehensive pharmacophoric
features. Structural alignment and feature cluster (Fig. 2) yielded different groups of pharmacophoric
features, including hydrogen bond donors (HBD), hydrogen bond acceptors (HBA), hydrophobic center
(HC) and aromatic ring (RA). HBD was not selected because of its small number and dispersed spatial
distribution. Two representative HBA features reflected hydrogen bond interactions between the ligand and
Met1160, Asp1222 were kept. Two main HC groups were found. One was located in the adenine binding
pocket, and the other was between the hydrophobic residues Met1211 and Tyr1230. RA cluster center
covered right the spatial position of the second HC. It was found that most of the chemical groups
represented by the second HC were aromatic. Therefore, RA was more feasible to represent the chemical
groups required in this region. Considering these, the first HC and the RA were kept in the pharmacophore
model. Thus, the initial pharmacophore model (Hypo1) consisted of two HBA, one HC and one RA
(supplementary Fig. S2.A).
Fig. 2 (A) The 11 complex structures were well aligned in a common reference structure (PDB ID: 2WGJ).
(B) The pharmacophoric features generated based on the aligned complex structures: HBA, green; HBD,
magenta; HC, cyan; RA, orange.

ACCEPTED MANUSCRIPT
To improve selectivity, the initial model Hypo1 was optimized by adding excluded volume spheres,
which resembled sterically inaccessible regions within the binding site. Twelve excluded volume features
were found and clustered in the ATP-binding site, whose spatial positions were respectively occupied by
residues Lys1161, Met1160, Leu1157, Ala1108, Ile1084, Met1211, Leu1140, Gly1085, Tyr1230, Lys1110,
Asn1209, and Asp1222. The resultant pharmacophore model was Hypo2 (supplementary Fig. S2.B). It was
complemented with shape constraint based on the crystal binding conformation of Crizotinib to obtain a
more restrict model, Hypo3 (supplementary Fig. S2.C).
It was suggested that ligand pharmacophore mapping module could generate low-energy conformations
similar to bioactive conformations of active inhibitors in the binding pocket. With “Best Fit” selected,
PF-02341066 was well mapped onto Hypo3. The RMSD value between the crystal and mapping
conformation is 1.5689Å (supplementary Fig. S3) suggested acceptable predictive ability of this model for
binding conformation.
Table 2. Statistics of pharmacophore model validation.
Parameter
Total molecules in database (D)
Total number of actives in database (A)
Total Hits (Ht)
Hypo1
624
86
Hypo2
624
Hypo3
624
86
86
394
78
287
12
Active Hits (Ha)
74
12
% Yield of actives [(Ha/Ht) × 100]
% Ratio of actives [(Ha/A) × 100]
Enrichment factor (EF) [(Ha × D)/(Ht × A)]
False Negatives [A - Ha]
19.80
90.70
1.44
8
25.78
86.05
1.87
12
1
13.95
7.26
0
False Positives [Ht - Ha]
Goodness of Hit Score (GH)a
316
0.15
213
0
0.24
0.78
a [(Ha/4HtA) (3A + Ht)) × (1 - ((Ht - Ha)/(D -A))]; GH score of >0.6 indicates a functioning model.
Enrichment factor (EF), Goodness of Hit Score (GH) and ROC scores were applied as the criteria to
assess the performance of these three models by retrospective discovery of active c-Met inhibitors from
inactive ones. Most of the active compounds could be identified by Hypo1 and Hypo2 (Table 2), but also
many false positives. For Hypo3, only 12 compounds were retrieved, but all active ones, indicating strong
reliability of this model in rejecting inactive compounds. Hypo3 was more statistically significant than
Hypo1 and Hypo2 (Table 2). The GH score was 0.78 while the accepted value of GH must ≧ 0.5 for a
screening model to be considered as well defined. ROC score was 0.807 (supplementary Fig. S4),
indicating that in eight out of ten cases, one randomly selected active c-Met inhibitors is better ranked than
an inactive one. Considering all of these, Hypo3 was the most suitable model for virtual screening.

ACCEPTED MANUSCRIPT
3.1.3 Virtual Screening Workflow
Fig. 3 The discovery process for novel c-Met inhibitors.
ChemDiv compound database was used for virtual screening (Fig. 3). The first parallel line was the
cascade molecular docking using Gold and Glide. Four known c-Met inhibitors (IC50: 2, 20, 100, and 640
nM) were used as positive control. Their lowest Gold Fitness 55.00 and highest Glide Gscore -7.3 were set
as the threshold values to discard the compounds poorly scored in the Gold and Glide HTVS docking
procedures. The first 10% of the Glide SP docking results were then put into the cascade docking using the
selected PDBs. The second parallel line was performed by Glide independently. The first 10% ranked
compounds of former mode were put into the next docking mode. The third parallel line was the
complex-based pharmacophore model. The first 10% compounds mapped well on Hypo3 were put into
cascade docking. Ligand-based pharmacophore model was used as another query for screening the
compound database. The first 10% compounds mapped well on it were put into cascade docking. The other
parallel line is 2D chemical similarity. Compounds with Tanimoto coefficient less than 0.5 were kept. After
all the parallel lines, cascade molecular docking using the selected crystal structures was applied to all the
retrieved compounds. The compounds with preferred binding pose in all the selected crystal structures were
considered to be potential hits. Through visual inspection, 480 novel compounds were obtained. After
filtered by further scoring (Fig. 3) and drug-likeness, the first 64 compounds of each structural type by
different scoring methods were selected and purchased from vendors in sufficient purity and quantity.
3.2 In Vitro Biological Evaluation of Virtual Hits
After biological evaluation, fourteen hit compounds were found, with Crizotinib as the positive control.
Among them, compounds HL-9, HL-10, and HL-11 sharing similar scaffold got the IC50 values of 11.19,
3.31 and 0.19 µM (Table 3), respectively. The other hit compounds (data not shown) were in micromolar

ACCEPTED MANUSCRIPT
range, which were in the process of further structural optimization [28]. However, the three hit compounds
did not show significant inhibitory activity against the selected cancer cell lines (Table 3), indicating further
structural optimizations were needed. Compound HL-11 showed high selectivity against KDR and B-raf
(supplementary Table S3). It was supposed that this specificity might be closely related with the ‘U’ shaped
binding mode of type I inhibitors. But this hypothesis should be validated by experimental approaches like
X-ray crystallography to disclose binding mode of HL-11.
Table 3. Structures and activities of the hit compounds.
Kinase Inhibition Tumor Cell lines Inhibition Rate (%) under 10µM
ChemDiv
Comp.
Structure
IDNUMBER
c-Met
11.19
A549
HCT116 MCF-7 PC3 SGC-7901
HL-9
D727-0159
D727-0656
D727-0806
─
10.42
8.2
35.7
57.74
15.53
12.37
HL-10
HL-11
3.31
0.19
20.11
6.08
4.73
0.64
45.41 19.43
14.94
55.1
12.26
Crizotinib
0.001053
0.1343 a 0.2536 a 0.045 a ND b
0.3213 a
a IC50 values in µM; bnot determine
3.3 X-ray Crystallographic study
Fig. 4 A) Interactions of HL-11 with the kinase domain of c-Met revealed by the crystal structure (PDB ID: 5YA5). The
compound was shown by cyan sticks. Hydrogen bonds were shown as green dash line. The residues hydrogen bonding to
HL-11 were coloured in green while other surrounding residues were colored in yellow. B) A (Fo-Fc) difference
electron-density map contoured at 3.0 σ for HL-11 in its complex structure with c-Met kinase determined by the X-ray
diffraction. C) The crystal-structure determined (cyan) and docked (magenta) conformations of HL-11 were aligned in the
binding site.
The crystal structure of c-Met kinase domain in complex with HL-11 has been determined at 1.89 Å
resolution (PDB ID: 5YA5; supplementary Table S1). It was found that HL-11 utilized a ‘U’ shaped crystal

ACCEPTED MANUSCRIPT
binding pose (Fig. 4A). Unlike Crizotinib, although it took such a conformation, the protein-ligand
interactions mainly focused on the side of Arg1208, Asp1222 and Tyr1230. Hydrogen bonds were
generated between HL-11 and the two residues, Arg1208 and Asp1222. Hydrophobic and π-π stacking
interactions between the scaffold of HL-11 and the side-chain of Tyr1230 contributed greatly to the
binding. The hydrogen bond between HL-11 and the main-chain of Met1160 was considered to be critical
for c-Met kinase inhibitors. The electrostatic density of HL-11 (Fig. 4B) showed that it was possible for the
methyl to rotate in different directions, but not disturb this hydrogen bond. Therefore, HL-11 perfectly
fitted into the binding cavity of c-Met kinase.
Importantly, the crystal binding conformation of HL-11 turned out to be very similar to the
computational prediction (RMSD = 0.4147Å; Fig. 4C), proving the accuracy of molecular docking. Based
on this experimental validation, HL-9 and HL-10 were supposed to take similar binding mode of HL-11
due to their similar structures and docking conformations. Molecular dynamics simulations were applied to
find their differences for further structural optimization.
3.4 Molecular Dynamics Simulation
Molecular dynamics simulations were performed to elucidate the differences of protein-ligand
interactions between compound HL-10 and HL-11. Both of the simulations were performed for 10 ns (Fig.
5). For HL-11, the protein-ligand system kept equilibrium all through the process while it converged after
7.3 ns for HL-10. The predicted root mean square fluctuation (RMSF) values (supplementary Fig. S5)
illustrated that the protein kept stable in the process of modeling, especially the conservative residues, like
the hinge region, Asp1222 and Tyr1230. Both of the binding free energy values for HL-11 and HL-10
calculated with MM/GBSA and MM/PBSA methods illustrated that it was more stable for protein-ligand
binding of HL-11 than HL-10, consistent with their c-Met kinase inhibitory activity. The individual energy
terms (Table 4) illustrated that van der waals energy contributions and polar desolvation energy were the
major reasons for the distinct binding affinity of HL-11 and HL-10. Hydrogen bond analysis found that
both of these two compounds formed conserved hydrogen bonds with the amino acid residues Met1160 and
Asp1222 (Table 5). The occupancies of the hydrogen bond with Met1160 were 98.51% and 92.66%,
respectively. But the occupancies of the other hydrogen bond with Asp1222 were 32.18% and 74%,
indicating that it was more stable for the hydrogen bond with HL-11 than HL-10. This was supposed to be
one of the reasons that HL-10 was less potent than HL-11. Besides, this also illustrated that hydrogen bond
between the ligand and Asp1222 was very important for their activity.

ACCEPTED MANUSCRIPT
Fig. 5 RMSD of protein-ligand complex in 10 ns, red: backbone atoms of c-Met kinase domain; blue: heavy atoms of the
ligand. A) HL-10. B) HL-11.
Table 4. Predicted binding free energies and individual energy terms of HL-10 and HL-11 in complex with
c-Met kinase domain (kcal/mol-1), calculated by MM/PBSA and MM/GBSA.
HL-10
HL-11
Contribution
Mean/(kcal/mol) Mean/(kcal/mol) Mean/(kcal/mol) Std.
△E ele
-6.50
-47.42
0.00
2.67
2.95
0.00
4.05
0.14
3.16
3.45
3.93
1.97
1.66
3.00
-16.91
-49.19
0.00
2.71
2.26
0.00
2.93
0.12
2.06
2.38
2.62
1.96
1.53
2.22
△E vdw
△E intra
△E gas
△G np
-53.92
-6.00
-66.10
-5.70
△G pb
34.98
28.48
-24.94
22.46
15.96
-37.46
36.47
19.57
-35.33
27.75
10.84
-44.05
△G pbele
△G pb,tot
△G gb
△G gbele
△G gb,tot
Eele: electrostatic energy term; Evd: van der waals energy term; Eintra: internal energy contributions generated because of
conformation changes; Egas = Eele + Evdw + Eintra; Gnp: non-polar desolvation free energy term; Gpb: polar desolvation free
energy calculated using MM/PBSA; Gpbele = Eele + Gpb; G pb,tot = Egas + Gpb + Gnp; Ggb: polar desolvation free energy
calculated using MM/GBSA; Ggbele = Eele + Ggb; Ggb,tot = Egas + Ggb + Gnp.
Table 5. Statistics of hydrogen bonds formation between compound HL-11, HL-10 and the corresponding
residues of c-Met.
Compd.
Hbond
Occupied (%)
98.51
Distance (Å)
Angle (°)
O9@MOL
N14@MOL
O21@MOL
N7@MOL
3.095 (±0.17) 2.966(±0.14) 23.77(±11.69)
3.263 (±0.15) 3.206 (± 0.17) 47.59 (±8.98)
2.966(±0.14) 3.095 (±0.17) 22.98 (±12.15)
3.206 (± 0.17) 3.263 (±0.15) 40.88 (±11.14)
HL-10
32.18
92.66
HL-11
74.00
Occupied (%): Percentage of snapshots with H-bond formation during MD simulations. The larger the occupied value, the
more stable the hydrogen bonds are. Percentages lower than 50% mean the hydrogen bonds are not stable.
3.5 Structural Optimization
3.5.1 Molecular Design and Chemical Synthesis
Further structural optimization was performed to develop more potent c-Met kinase inhibitors based on
the experimental and computational results (Fig. 6) [8, 9, 28, 29]. The hydrogen bonds with Met1160 and

1
ACCEPTED MANUSCRIPT
Asp1222 were retained. For R , we proposed that it should be extended to the solvent accessible region to
get hydrophilic interactions (compound 11a and 11b). Besides, we also supposed that R1 should be bulky
and hydrophobic groups to fully occupy the adenine binding pocket (compound 11c~11f). Linker group
was used to keep the ‘U’ shaped binding conformation of the entire ligand. We selected the commonly used
linker groups to replace the original one (compound 11c~11f). The hydrogen bond between R2 and
Arg1208 was not stable in the process of molecular dynamics simulation. Compounds with different R2
groups were designed to investigate whether it was important or not. The hydrophobicity of R2 was also
investigated. These newly designed compounds were listed in Table 6. Their general synthetic routes for
compound 11a~11f were outlined in Scheme 1 and 2. The procedures and characterization were detailed in
section 2.8, and the NMR spectra was in the supplementary material.
Fig. 6 Binding mode for structural optimization.
Docking conformations of all these new compounds in the binding site of c-Met kinase domain were
shown in Fig. S6 in the supplementary material. They kept similar binding mode with the hit compound
HL-11 and the extending of R1 and R2 in the binding site was consistent with our design. Hydrogen bond
interactions were found between all these new compound and Asp122, Met1160. Furthermore, compound
11a, 11c and 11e kept hydrogen bond interactions with the residue Arg1208 as expected.
Table 6. The new designed compounds and their in vitro c-Met kinase inhibitory activity.
Comp.
R1
Linker
R2
c-Met (IC50: µM)
11a
-CH2-
6.51
11b
-CH2-
5.49

ACCEPTED MANUSCRIPT
11c
11d
11e
11f
-OCH2-
1.15
-OCH2-
-OCH2-
-OCH2-
0.154
0.0793
0.0139
R2
O
HO
S
OH
R2
R2
=
O
2a:
2b:
N
H2N
NH2
N
N
H
R2COOH
b
H
H
H2N
N
N
N
N
S
a
N
N
N
R2
N
=
COOH
N
SH
1
R1
HO
O
R1-Cl
d
c
N
N
S
N
N
R2
N
R2
N
N
S
N
N
O
N
, R2
=
R2
R2
=
=
R1
R1
=
3a:
3b:
11a:
11b:
N
N
H
H
N
N
N
N
, R2
=
O
=
O
Scheme 1. Reagents and conditions: a) CH3SO3H, sulfolane, H2O, 90°C; b)POCl3, sulfolane, 85 °C; c) NaOH, THF, rt; d) CS2CO3, DMF, 80°C.
3.5.2 In Vitro Biological Evaluation of 11a~11f
From Table 6, the optimization to extend R1 into the solvent accessible region gave negative results.
Compound 11a and 11b only got micromolar inhibitory activity. On the opposite, the bulky and
hydrophobic R1 group (compound 11c-11f) gave the expected results. Comparison between the c-Met

1
ACCEPTED MANUSCRIPT
kinase inhibitory activity of compound 11a and 11e validated the two hypotheses for R optimization. For
the linker, it didn’t make significant differences because both of them could keep the ‘U’ shaped binding
conformations. But they could be used for expanding the structural diversity. The different R2 groups
showed certain hydrophobicity and aromatic property was preferred to strengthen the protein-ligand
interactions, including π-π interaction with Tyr1230. From the results of compound 11d and 11e, hydrogen
bond between the ligand and Arg1208 should be favored for improving c-Met inhibitory activity.
Furthermore, compound 11e and 11f with the best c-Met kinase inhibitory activity also got improved
anti-proliferative activity (Table 7) compared with the starting point. Compound 11e showed comparable
inhibitory activity against NCI-H460, PC3 and Colo-205 to Crizotinib. But their anti-proliferative activity
against EBC-1 and MKN45 were still needed to be improved.
Table 7. Anti-proliferative activity (IC50: µM) of 11e and 11f.
Comp.
11e
EBC-1 MKN45 NCI-H460 PC3 Colo-205
0.506
3.17
1.171
1.02
2.055
NDa
2.970
NDa
2.242
NDa
11f
Crizotinib 0.013
a not determined
0.022
2.244
9.787
2.449
4. CONCLUSION
A new series of novel Type I c-Met inhibitors have been identified in this study with the multi
workflows, including virtual screening, biological evaluations, X-ray crystallography, molecular dynamics
simulation and structural optimization. The crystal binding conformation of the best hit compound HL-11
was highly consistent with the computational result. Molecular dynamics simulations elucidated the main
detailed differences between compound HL-10 and HL-11. Based on this starting point, hypothesis for
structural optimization was proposed and novel analogues of compound HL-11 were designed. Compound
11d~11f showed better in vitro c-Met inhibitory activity and anti-proliferative activity. Compound 11e
showed significantly improved anti-proliferative activity against several cancer cell lines. The experimental
results illustrated that bulky hydrophobic R1 groups that got hydrogen bonds with Met1160 were beneficial
for better c-Met inhibitory activity. Aromatic and hydrophobic R2 groups were preferred to strengthen the
protein-ligand interactions. It was also favoured if additional stable hydrogen bond between the ligand and
Arg1208. All these experimental findings proved the reliability and efficiency of our in silico methods.
Acknowledgments
This work was financially supported by the National Science Foundation of China (81473078 and
81703367 ), the scientific research funds for new teacher of China Pharmaceutical University

ACCEPTED MANUSCRIPT
(3014070086) and the Fundamental Research Funds for the Central Universities (2632017PY12). This
work was also supported by the “111 Project” from the Ministry of Education of China, the State
Administration of Foreign Expert Affairs of China (No. 111-2-07).
Appendix A. Supplementary data
Supplementary data related to this article can be found at
References
[1] M.F. Di Renzo, R.P. Narsimhan, M. Olivero, S. Bretti, S. Giordano, E. Medico, P. Gaglia, P. Zara, P.M. Comoglio,
Expression of the Met/HGF receptor in normal and neoplastic human tissues, Oncogene 6 (1991) 1997-2003.
[2] E. Gherardi, W. Birchmeier, C. Birchmeier, G. Vande Woude, Targeting MET in cancer: rationale and progress, Nat.
Rev. Cancer 12 (2012) 89-103.
[3] C. Birchmeier, W. Birchmeier, E. Gherardi, G.F. Vande Woude, Met, metastasis, motility and more, Nat. Rev. Mol.
Cell Biol. 4 (2003) 915-925.
[4] C. Boccaccio, P.M. Comoglio, Invasive growth: a MET-driven genetic programme for cancer and stem cells, Nat. Rev.
Cancer 6 (2006) 637-645.
[5] P.C. Ma, M.S. Tretiakova, A.C. MacKinnon, N. Ramnath, C. Johnson, S. Dietrich, T. Seiwert, J.G. Christensen, R.
Jagadeeswaran, T. Krausz, E.E. Vokes, A.N. Husain, R. Salgia, Expression and mutational analysis of MET in human
solid cancers, Genes Chromosomes Cancer 47 (2008) 1025-1037.
[6] J.J. Cui, Targeting receptor tyrosine kinase MET in cancer: small molecule inhibitors and clinical progress, J. Med.
Chem. 57 (2014) 4427-4453.
[7] S. Peters, A.A. Adjei, MET: a promising anticancer therapeutic target, Nat. Rev. Clin. Oncol. 9 (2012) 314-326.
[8] H.L. Yuan, J. Zhuang, S.H. Hu, H.C. Li, J.X. Xu, Y.N. Hu, X. Xiong, Y.D. Chen, T. Lu, Molecular modeling of
exquisitely selective c-Met inhibitors through 3D-QSAR and molecular dynamics simulations, J. Chem. Inf. Model.
54 (2014) 2544-2554.
[9] W. Tai, T. Lu, H. Yuan, F. Wang, H. Liu, S. Lu, Y. Leng, W. Zhang, Y. Jiang, Y. Chen, Pharmacophore modeling and
virtual screening studies to identify new c-Met inhibitors, J. Mol. Model. 18 (2012) 3087-3100.
[10] S.F. Bellon, P. Kaplan-Lefko, Y. Yang, Y. Zhang, J. Moriguchi, K. Rex, C.W. Johnson, P.E. Rose, A.M. Long, A.B.
O'Connor, Y. Gu, A. Coxon, T.S. Kim, A. Tasker, T.L. Burgess, I. Dussault, c-Met inhibitors with novel binding
mode show activity against several hereditary papillary renal cell carcinoma-related mutations, J. Biol. Chem. 283
(2008) 2675-2683.
[11] J. Porter, S. Lumb, R.J. Franklin, J.M. Gascon-Simorte, M. Calmiano, K.L. Riche, B. Lallemand, J. Keyaerts, H.
Edwards, A. Maloney, J. Delgado, L. King, A. Foley, F. Lecomte, J. Reuberson, C. Meier, M. Batchelor, Discovery
of 4-azaindoles as novel inhibitors of c-Met kinase, Bioorg. Med. Chem. Lett. 19 (2009) 2780-2784.
[12] J.J. Cui, M. Tran-Dube, H. Shen, M. Nambu, P.P. Kung, M. Pairish, L. Jia, J. Meng, L. Funk, I. Botrous, M. McTigue,
N. Grodsky, K. Ryan, E. Padrique, G. Alton, S. Timofeevski, S. Yamazaki, Q. Li, H. Zou, J. Christensen, B.
Mroczkowski, S. Bender, R.S. Kania, M.P. Edwards, Structure based drug design of crizotinib (PF-02341066), a
potent and selective dual inhibitor of mesenchymal-epithelial transition factor (c-MET) kinase and anaplastic
lymphoma kinase (ALK), J. Med. Chem. 54 (2011) 6342-6363.
[13] Afinogenov, II, [History of the pharmacy of the Main Hospital], Voen. Med. Zh. 318 (1997) 57-60.
[14] H. Nishii, T. Chiba, K. Morikami, T.A. Fukami, H. Sakamoto, K. Ko, H. Koyano, Discovery of
6-benzyloxyquinolines as c-Met selective kinase inhibitors, Bioorg. Med. Chem. Lett. 20 (2010) 1405-1409.
[15] B.K. Albrecht, J.C. Harmange, D. Bauer, L. Berry, C. Bode, A.A. Boezio, A. Chen, D. Choquette, I. Dussault, C.
Fridrich, S. Hirai, D. Hoffman, J.F. Larrow, P. Kaplan-Lefko, J. Lin, J. Lohman, A.M. Long, J. Moriguchi, A.

O'Connor, M.H. Potashman, M. Reese, K. Rex, A. Siegmund, K. Shah, R. Shimanovich, S.K. Springer, Y. Teffera, Y.
ACCEPTED MANUSCRIPT
Yang, Y. Zhang, S.F. Bellon, Discovery and optimization of triazolopyridazines as potent and selective inhibitors of
the c-Met kinase, J. Med. Chem. 51 (2008) 2879-2882.
[16] S.G. Buchanan, J. Hendle, P.S. Lee, C.R. Smith, P.Y. Bounaud, K.A. Jessen, C.M. Tang, N.H. Huser, J.D. Felce, K.J.
Froning, M.C. Peterman, B.E. Aubol, S.F. Gessert, J.M. Sauder, K.D. Schwinn, M. Russell, I.A. Rooney, J. Adams,
B.C. Leon, T.H. Do, J.M. Blaney, P.A. Sprengeler, D.A. Thompson, L. Smyth, L.A. Pelletier, S. Atwell, K. Holme,
S.R. Wasserman, S. Emtage, S.K. Burley, S.H. Reich, SGX523 is an exquisitely selective, ATP-competitive inhibitor
of the MET receptor tyrosine kinase with antitumor activity in vivo, Mol. Cancer Ther. 8 (2009) 3181-3190.
[17] J. Porter, S. Lumb, F. Lecomte, J. Reuberson, A. Foley, M. Calmiano, K. le Riche, H. Edwards, J. Delgado, R.J.
Franklin, J.M. Gascon-Simorte, A. Maloney, C. Meier, M. Batchelor, Discovery of a novel series of quinoxalines as
inhibitors of c-Met kinase, Bioorg. Med. Chem. Lett. 19 (2009) 397-400.
[18] A.A. Boezio, L. Berry, B.K. Albrecht, D. Bauer, S.F. Bellon, C. Bode, A. Chen, D. Choquette, I. Dussault, M. Fang,
S. Hirai, P. Kaplan-Lefko, J.F. Larrow, M.H. Lin, J. Lohman, M.H. Potashman, Y. Qu, K. Rex, M. Santostefano, K.
Shah, R. Shimanovich, S.K. Springer, Y. Teffera, Y. Yang, Y. Zhang, J.C. Harmange, Discovery and optimization of
potent and selective triazolopyridazine series of c-Met inhibitors, Bioorg. Med. Chem. Lett. 19 (2009) 6307-6312.
[19] W. Wang, A. Marimuthu, J. Tsai, A. Kumar, H.I. Krupka, C. Zhang, B. Powell, Y. Suzuki, H. Nguyen, M. Tabrizizad,
C. Luu, B.L. West, Structural characterization of autoinhibited c-Met kinase produced by coexpression in bacteria
with phosphatase, Proc. Natl. Acad. Sci. U S A 103 (2006) 3563-3568.
[20] M.A. Seeliger, M. Young, M.N. Henderson, P. Pellicena, D.S. King, A.M. Falick, J. Kuriyan, High yield bacterial
expression of active c-Abl and c-Src tyrosine kinases, Protein Sci. 14 (2005) 3135-3139.
[21] W. Kabsch, Xds, Acta Crystallogr. D Biol. Crystallogr. 66 (2010) 125-132.
[22] K. Wu, J. Ai, Q. Liu, T. Chen, A. Zhao, X. Peng, Y. Wang, Y. Ji, Q. Yao, Y. Xu, M. Geng, A. Zhang,
Multisubstituted quinoxalines and pyrido[2,3-d]pyrimidines: synthesis and SAR study as tyrosine kinase c-Met
inhibitors, Bioorg. Med. Chem. Lett. 22 (2012) 6368-6372.
[23] P.D. Adams, P.V. Afonine, G. Bunkoczi, V.B. Chen, I.W. Davis, N. Echols, J.J. Headd, L.W. Hung, G.J. Kapral,
R.W. Grosse-Kunstleve, A.J. McCoy, N.W. Moriarty, R. Oeffner, R.J. Read, D.C. Richardson, J.S. Richardson, T.C.
Terwilliger, P.H. Zwart, PHENIX: a comprehensive Python-based system for macromolecular structure solution, Acta
Crystallogr. D Biol. Crystallogr. 66 (2010) 213-221.
[24] P. Emsley, B. Lohkamp, W.G. Scott, K. Cowtan, Features and development of Coot, Acta Crystallogr. D Biol.
Crystallogr. 66 (2010) 486-501.
[25] A. Onufriev, D. Bashford, D.A. Case, Exploring protein native states and large-scale conformational changes with a
modified generalized born model, Proteins 55 (2004) 383-394.
[26] P.A. Kollman, I. Massova, C. Reyes, B. Kuhn, S. Huo, L. Chong, M. Lee, T. Lee, Y. Duan, W. Wang, O. Donini, P.
Cieplak, J. Srinivasan, D.A. Case, T.E. Cheatham, 3rd, Calculating structures and free energies of complex
molecules: combining molecular mechanics and continuum models, Acc. Chem. Res. 33 (2000) 889-897.
[27] R. Tiedt, E. Degenkolbe, P. Furet, B.A. Appleton, S. Wagner, J. Schoepfer, E. Buck, D.A. Ruddy, J.E. Monahan,
M.D. Jones, J. Blank, D. Haasen, P. Drueckes, M. Wartmann, C. McCarthy, W.R. Sellers, F. Hofmann, A drug
resistance screen using a selective MET inhibitor reveals a spectrum of mutations that partially overlap with
activating mutations found in cancer patients, Cancer Res. 71 (2011) 5255-5264.
[28] L. Zhang, B. Zhang, J. Zhao, Y. Zhi, L. Wang, T. Lu, Y. Chen, Structure-based design, synthesis, and evaluation of
4,5,6,7-tetrahydro-1H-pyrazolo[4,3-c]pyridine derivatives as novel c-Met inhibitors, Eur. J. Med. Chem. 138 (2017)
942-951.
[29] H.L. Yuan, T. Lu, T. Ran, H.C. Liu, S. Lu, W.T. Tai, Y. Leng, W.W. Zhang, J. Wang, Y.D. Chen, Novel strategy for
three-dimensional fragment-based lead discovery, J. Chem. Inf. Model. 51 (2011) 959-974.

ACCEPTED MANUSCRIPT
HIGHLIGHTS
ò The best hit compound HL-11 identified with the reliable virtual screening workflow exhibited potent
c-Met kinase inhibitory activity.
ò Comparison of the hit compounds with different c-Met inhibitory activity by molecular dynamics
simulations suggested the key protein-ligand interactions for structural optimization.
ò The experimentally resolved crystal binding mode of compound HL-11 was highly consistent with the
computational prediction.
ò Structural optimization of compound HL-11 resulted in 11e with better c-Met kinase inhibitory activity
and improved anti-proliferative activity.