Synthesis and biological evaluation of novel (E)-N'-benzylidene hydrazides as novel c-Met inhibitors through fragment based virtual screening

Abstract C-Met plays a crucial role in the development and progression of neoplastic disease. Type II c-Met inhibitors recognise the inactive DFG-out conformation of the kinase, result in better anti-tumour effects due to synergistic effect against the other kinases. According to our previous works, an (E)-N'-benzylidene group was selected as the initial fragment. Two series of (E)-N'-benzylidene hydrazides were designed by fragment growth method. The inhibitory activities were in vitro investigated against c-Met and VEGFR-2. Compound 10b exhibited the most potent inhibitory activity against the c-Met inhibitor (IC50 = 0.37 nM). Compound 11b exhibited multi-target c-Met kinase inhibitory activity as a potential type II c-Met inhibitor (IC50 = 3.41 nM against c-Met; 25.34 nM against VEGFR-2). The two compounds also demonstrate the feasibility of fragment-based virtual screening method for drug discovery.


Introduction
Hepatocyte growth factor receptor (HGFR/c-Met) is a transmembrane heterodimer comprising two disulphide-linked chains, including an outer a-chain (50 kDa) and a transmembrane b-chain (140 kDa) 1 , encoded by c-Met gene 2 . Under normal physiological conditions, it regulates important cellular processes, such as differentiation, proliferation, cell cycle, motility, and apoptosis. The intracellular portion of c-Met comprises a catalytic tyrosine kinase domain, which contains multifunctional docking sites 3 . Many downstream pathways such as PI3K, MAPK, and STAT3 will be activated, when the HGF binds to the extracellular domains of c-Met, followed by autophosphorylation of tyrosine kinase residues in the catalytic domain 4 . Abnormal activation in the c-Met pathway induces excessive cell proliferation and is related to the development and progression of the neoplastic disease 5 .
With the development of research on the mechanism in drug resistance to cancer cells, the crossing talk between various membrane receptors and c-Met is monitored in tumours and takes part in the resistance to EGFR-tyrosine kinase inhibitors including gefitinib and erlotinib 6,7 . Continuous angiogenesis is a key mechanism of tumour growth. Therefore, anti-angiogenesis plays an important role in cancer therapy. VEGF has a strong stimulating effect on the proliferation of endothelial cells through increasing vascular permeability and promoting the formation of tumour blood vessels 8 . It specifically binds to the extracellular domain of VEGFR-2 and is involved in endothelial cell sprouting, migration, vascular permeability, and tumour cell survival. However, VEGFR-2 inhibitors can lead to c-Met-dependent invasion and metastasis of tumour cells by increasing the degree of hypoxia when used alone 9 . While the combination of sunitinib and c-Met inhibitor PF-04217903 inhibits both VEGFR and c-Met pathways, significantly decreases tumour invasion and metastasis 10,11 . Studies on the multi-target c-Met kinase inhibitors show that it is different from the type I inhibitor which possesses high selectivity. The multitarget c-Met kinase inhibitor binds to the inactive conformation of the kinase and belongs to the type II c-Met kinase inhibitor 12,13 , which binds to not only c-Met but also VEGFR, FGFR, ALK, EGFR, MAT1R. All these studies suggest that simultaneous inhibition of the two tyrosine kinase receptors results in better anti-tumour effects 14,15 .
The number of multi-target c-Met inhibitors that entering the clinical stage is small and the skeleton of them is lacking in diversity at present, it is easy to cause over-fitting when building virtual screening models according to the ligand 16 . We established a fragment-based multistage screening method for discovering VEGFR-2 inhibitors in previous work 17,18 , obtained a novel molecule with a high antineoplastic effect. Not affected by the ligand conditions, the fragment-based theory held that the active pockets of drug targets were made up of multiple sub-pockets, and the fragments are units that combine with these sub-pockets. Finding these fragments and linking them together often led to higher active compounds 19,20,21 . In this study, we intended to use this method to establish a virtual screening model for discovering multi-target c-Met inhibitors, increasing the structural diversity of these inhibitors.
inactive pocket of c-Met is generally divided into a back pocket behind the gatekeeper (c-Met: Phe1223), and a front pocket within the solvent-accessible area (Figure 1(a)). The back pocket is the unique pocket of the inactive conformation of tyrosine kinase, namely the DFG-out site. The individual pockets are constructed by several amino acids, the surrounding amino acids of each pocket impart distinct properties. The back pocket in c-Met is characterised as a hydrophobic pocket due to the surrounding hydrophobic amino acids: Phe1134, Leu1195, and Phe1200 (Figure 1(b)). Meanwhile, the front-pocket is located in a solventaccessible area and is surrounded by Ile1084, Lys1161 and His1162 (Figure 1(b)). Analysis of protein-ligand interaction of c-Met inhibitors (exemplified by GSK1363089 in c-Met) showed that the common interaction patterns include: a conserved hydrogen bond with the backbone of Asp1222, an arene-arene stacking interaction with the side chain of Phe1223.  For non-competitive inhibitors combining with the DFG-out site of  tyrosine kinases, an ASP and a Lys around gatekeeper are key  residues 24,25 . In our previous work, we found that the transacylhydrazine can also form hydrogen bonds with Asp1110 and Lys1222 of c-Met. Given that the hydrophobic residues were in the back pocket, the aromatic acyl hydrazine was used as the initial fragment (Figure 2(a)). Then the fragment growth method in MOE was used to generate novel molecules. The newly generated small molecules were then filtered by the screening model of VEGFR-2. Finally, a novel skeleton that linked a phenyl group to the aromatic acyl hydrazine with a triazole fragment was achieved. It was worth noting that a series of small molecules with large substituents on the phenyl group were abandoned by the second model, which  indicated that large substituents were not beneficial for increasing the inhibitory activity of small molecules against VEGFR-2.

Docking
The hit compound was docked into the inactive conformation of c-Met (Figure 2(b)). The halogenated aromatic ring was in the hydrophobic region at the bottom of the pocket and the transacylhydrazine fragment formed two hydrogen bonds with Asp1110 and Lys1222, respectively. On the basis of the new skeleton, the halogen atom at the tail of the novel compound was further changed, exploring the approach to increase the activity of this small molecule. Owing to the large substituents on the phenyl group seem to play an opposite role in the c-Met and VEGFR-2 inhibitory activities of the small molecule, an acetylamino group was taken into account as a compromise.
In the previously designed route, we used intermediate 5 reacted with hydrazine hydrate and then form the target compounds with benzaldehyde derivatives. We found an interesting phenomenon during the previous synthetic route exploration. There will be cis-trans isomers generated in this process. So we improved the synthetic route to get the pure target compounds (Scheme 1).
The difficulties in the synthesis of the target compounds (10a-10k, 11a-11k) lie in the cis-trans isomerisation of the acyl hydrazine fragment. To synthesise a pure single conformation is the focus of synthesis. It was found that acyl hydrazine fragment synthesised by click chemistry link to another fragment results in the appearance of cis-trans isomers due to the direction of the linkage. The mixture is difficult to separate according to the previous synthesis studies conducted by our group. The yields of the cis and trans configurations were equal and were confirmed in nuclear magnetics. This was also verified in the study of cis-trans isomers by Jian Wu 26 , Mohammad Sayed Alam 27 . In response to the above problems, we improved the synthetic route, the benzaldehyde derivatives were used to form the (E)-benzylidenehydrazine derivatives fragment with hydrating hydrazine. For another fragment, phenylacetylene and 3-ethynylaniline were used as raw materials and obtained by click chemistry and hydrolysis. The new synthetic method solved the problem of generating cis-trans isomers, simplified the synthetic method, reduced the synthetic steps and has great reference value for the synthesis and purification of cis-trans isomers.

QSAR study
The capacity of the novel compounds for inhibiting c-Met activity was evaluated using a cell-free assay. The evaluation results are summarised in Table 1.

PCA-based small molecule descriptor selection
To find out the molecules with the optimal structure, a 3D-QSAR study on the derivatives was carried out using the molecular descriptors method in Molecular Operation Environment System (MOE; Chemical Computing Group Inc., Montreal, QC, Canada). Based on the molecular model, which showed a significant correlation with the experimental IC 50 activity data, further improvement of this correlation was analysed by molecular descriptors. The available molecular descriptors implemented in MOE were tested. As a result, the molecular descriptors were used to analyse the structure-activity relationship of each series. To reduce the model noise and eliminate the interaction between descriptors, the PCA-based small molecule descriptor method was adopted. According to the significant correlation with IC 50 , 10 descriptors were screened out from 379 descriptors by the PCA method (RMSE ¼ 22124.5 R 2 ¼ 0.746558). Target compounds correlation coefficients are shown in Table 2. After calculation, the normalised linear model displayed significant relation to descriptors. The following equation showed the generated QSAR model that included eleven descriptors (AM1 E , AM1 Eele , MNDO E , PM3 E , PM3 Eele , pmi, pmi2, pmi3, pmiX, pmiY) to predict IC 50 /SD values of c-Met inhibitors.
In the QSAR model above, AM1 E , MNDO E , and PM3 E were used to describe total energy (molecular dynamic kinetic energy in space, the energy of the electrons of a molecule, intramolecular energy, vibrational energy between atoms in a molecule, the energy of the molecule rotating around the centre of mass); AM1 Eele and PM3 Eele represented electronic energy; pmi, pmi2, pmi3, pmiX and pmiY showed principal moment of inertia (measurement of inertia of a rigid body rotating around an axis). The magnitude of the value indicated its correlation with IC 50 . According to the QSAR model, the activity of the compounds was mainly related to electronic energy (AM1 Eele and PM3 Eele ).

QSAR suggestion
After analysis of the contribution of the substituents to the activities by MOE, a QSAR suggestion was given as a set of compounds with more potential activities. The compounds in the QSAR suggestion set combined the advantages of the synthesised compounds and discarded their disadvantages. The results of the QSAR study are shown in Figure 3. The activities of the compounds in series 10 were better than that of series 11. For the compounds in series 11, the substituents on the benzene ring preferred for halogen atoms, such as bromine atom. This could be explained by high total energy and electronic energy of compounds that bromine atom brought. While the introduction of hydroxy group reduced the activities of the compounds. As a result of the principal moment of inertia, the activities of the  meta-substituted compounds were better than that of the orthosubstituted and para-substituted compounds. However, the activity of the compound 11h was superior to 11g, prompting that it could be dominated by the increased total energy rather than the principal moment of inertia. Therefore, the QSAR model suggested meta-chloro substitution (11b). For the compounds in series 10, the contribution of ortho-substituents was uncertain. However, the combination of hydroxyl and chlorine atoms could make the molecule have better physicochemical properties since the QSAR suggestion gave 10f, 11f, and 11i. The compound 10i, 10j, and 10k substituted by methoxy showed different inhibitory effects. It may due to the different position and number of substituents brought the change of compounds' energy and moment of inertia. The virtual screening model of VEGFR-2 was used to screen the QSAR Suggestion, five molecules were screened out. Followed by synthesising and detecting their kinase activities against c-Met and VEGFR-2, respectively, 11b and 10f showed better inhibitory activity than the others as shown in Table 3. It was consistent with the QSAR predicted results. Unfortunately, compound 10b did not exhibited expected activity against VEGFR-2, which may be related to acetyl substitution. It was not accidental that molecules with activity against c-Met can also inhibit VEGFR-2, the multi-target inhibitors against c-Met and VEGFR-2 have a physiological basis. Both of the proteins belong to the family of tyrosine kinase, which catalyse the phosphorylation of tyrosine. The similar functions make them possess similar active sites. While subtle differences may make small molecules show different activities against the two kinases. So we superimposed the two kinases in three dimensions to explore the reason.
The structures of c-Met and VEGFR-2 were superposed in 3D space as shown in Figure 4(a). The RMSD results of the superposition were 1.63 Å, indicating that the similarity of the two conformations was very high. Moreover, the back pockets in the two kinases, named the DFG out site, possessed higher similarity (Figure 4(c), the residues marked with red circle). This also provides the basis for the discovery of dual target drugs. However, when the active site was amplified, the difference between the two pockets was detected. The opening of c-Met pocket was wide, which can accommodate small molecules. So the inhibitor of c-Met was longer. While the opening of VEGFR-2 pocket was more narrow that it could not accommodate any group (Figure 4(b), the ribbons marked with red arrows). So the inhibitor of VEGFR-2 was not as long as that of c-Met. So it was suggested that the opening of VEGFR-2 pocket could not accommodate the acetylamino group, causing compound 10b to show promising inhibitory activity against c-Met but no activity against VEGFR-2.

Synthesis and characterisation of novel compounds
For all target compounds, 1 H-NMR and 13 C-NMR spectra were recorded on a Varian NMR spectrometer operating at 600 MHz for 1 H, and 150 MHz for 13 C. All chemical shifts were measured in DMSO-d 6 as solvents. All chemicals were purchased from Sinoreagent Chemical Reagent (Beijing, China) and were used as received, unless stated otherwise. Analytical TLC is performed on silica gel 60 F254 plates (Qingdao Haiyang Chemical Company, Qingdao, Shandong, China) and visualised by UV. Flash column chromatography is performed on gel 60 (40-63 mm) (Qingdao Haiyang Chemical Company, Qingdao, Shandong, China). Melting points were determined with an Electro thermal melting point apparatus, are uncorrected. 1 H-NMR and 13 C-NMR of all the target compounds are described in Supplementary material.
Compound 5 was prepared with the same method using ethynylbenzene (3), to give pure compound as a pale yellow oily liquid.
Compound 7 was prepared with the same method by using corresponding raw materials 5.
General procedure for the synthesis of the (E)-benzylidenehydrazine derivatives (9a-9k): A solution of 80% hydrazine hydrate (8.9 g, 177.85 mmol) was added dropwise to 4-chlorobenzaldehyde (8a) (5.0 g, 35.57 mmol) in 10 ml water. The reaction mixture was stirred at room temperature for 5 h and monitored by TLC. After completion of the reaction, the reaction mixture was poured into 50 ml water and stirred for 20 min. The reaction mixture was filtered off. Filter cake was washed with water, and dried to give the compound 9a as a white solid, yield: 81.27%.
Compounds 9b to 9k were prepared with the same method by using corresponding raw materials 8b to 8k.

Fragment-based drug design
Fragment based drug design was performed using the Add Group to Ligand module in Molecular Operating Environment package (MOE; Chemical Computing Group Inc., Montreal, QC, Canada). The aromatic acyl hydrazine was used as the initial fragment, the hydrogen atom at the head of the acetylhydrazine group was selected as the connection point. During the generation, a filter of Lipinski's rule of five was toggled ON. The novel small molecules were generated based on the shape of the active pocket.

QSAR study
The target compounds database was prepared using compute module in MOE. The preparation involved introduction of every single target compounds using ChemDraw, converted structure to 3D and saved as mol2 formats in Chem3D. Further preparation comprised compounds database generation with importing all mol2 formats and energy minimise in molecule module in compute module using default settings. The IC 50 activities data were added in another field. Summaries of the relevant datasets employed for generating the QSARs relating the various molecular descriptors to the compounds 10a-10k and 11a-11k.
The Regression analysis was performed in QuaSAR module. The QSAR model was performed for compounds 10a-10k and 11a-11k using IC 50 as activity field and PCR method. The RMSE and R 2 values were derived from Quasar fit. The estimated linear model which contained all molecular descriptors was established. The QSAR model was constructed for the 10a-10k and 11a-11k compounds with relatively important descriptors.
The database contained compounds structures, IC 50 values and relatively important descriptors were established. The basic parent structure molecule was input as a guide and scaffolds were calculated as preset. The IC 50 style was changed into KI/IC 50 (nM).

Superposing
The PDB files of the two proteins were imported into MOE, and operated in the sequence editing panel. First, the sequences of the two proteins were aligned to ensure the similarity. Then the 3D structures of the two proteins were superposed with each other, which was visible in the main panel of MOE. Finally, the RMSD result was obtained in the protein Superpose RMSD plot panel.

Conclusion
The difficulty in the synthesis of the target compound (10a-10k, 11a-11k) lies in the cis-transisomerizm. After improving the synthetic route, two series of (E)-N'-benzylidene derivatives were synthesised and evaluated for their inhibitory activities against c-Met. The QSAR results showed that halogens and acetyl of aromatic rings could improve the activity of the compounds against c-Met. However, the compounds with acetyl groups did not perform well in the activity test of VEGFR-2. Nevertheless, we have obtained a highly active type I c-Met inhibitor 10b and a multitarget c-Met kinase inhibitor 11b. They could be potential antitumor agents that deserve further research.

Disclosure statement
No potential conflict of interest was reported by the authors.

Funding
This work was supported by National Natural Science Foundation of China [81274182 and 81573687].