2021 年 44 巻 12 号 p. 1872-1877
FMS-like tyrosine kinase 3 (FLT3) plays a very important role in regulating the proliferation, differentiation and survival of normal hematopoietic stem cells. Internal tandem duplications of the FLT3 gene (FLT3-ITD) mutations are present in 25% of all acute myeloid leukemia (AML) patients and are frequently associated with adverse clinical outcomes. Therefore, FLT3-ITD is a promising target for the treatment of AML. The use of covalent virtual screenings has shown that efficient rational approaches for the rapid discovery of new drugs scaffold. Herein, we report a hybrid virtual screening strategy that led to the discovery of FLT3 inhibitors. Using the combination of non-covalent docking and covalent docking, 8 compounds were found to inhibit FLT3, and G856-8335, S346-0154 are also effective against mutant FLT3. These two compounds also show selectivity to receptor tyrosine kinase (C-KIT), which has the potential for optimization. And this work can be extended to the screening of other covalent inhibitors.
FMS-like tyrosine kinase 3 (FLT3) is a transmembrane tyrosine receptor and belongs to the type III receptor tyrosine kinase family.1,2) It plays a very important role in regulating the proliferation, differentiation and survival of normal hematopoietic stem cells. FLT3 is expressed by early bone marrow hematopoietic progenitor cells, and its structure is composed of an Extracellular domain (five immunoglobulins), a transmembrane domain, a Juxtamembrane domain, two tyrosine kinase domains and a C-terminal.3) Wild-type FLT3 is activated by endogenous ligands and the activated FLT3 receptor dimerizes, changes the conformation of the intracellular kinase domain, and exposes the phosphorylation binding site.4) Activate multiple downstream signaling pathways to drive cell growth and survival by binding to the corresponding protein.
About 30% of adult patients with acute myeloid leukemia (AML) have FLT3 mutations, which is the most common and important cause associated with poor prognosis of AML.5) Current studies have found that there are two main forms of FLT3 mutations, one is the internal tandem duplication (ITD) in the proximal membrane region, which accounts for about 23% of AML patients; the other is the activation loop of the kinase region, and the main point mutation in the tyrosine kinase domain (TKD) is a D835Y point mutation, and there are also less common types of point mutations at other sites.6,7) The difference from the wild-type FLT3 is that the FLT3 receptor with insertion mutation and point mutation destroys the auto-inhibitory resting conformation of the receptor, induces the constitutive activation of FLT3 and does not depend on the ligand FL for cell growth.8,9) Activate downstream mitogen-activated protein kinase (MAPK), Janus kinase (JAK)-signal transducer and activator of transcription (STAT), phosphatidylinositol 3-kinase (PI3K) and other signaling pathways, and ultimately induce abnormal proliferation and growth of hematopoietic cells, leading to the occurrence of cancer.10,11)
At present, a series of FLT3 inhibitors have been approved by the U.S. Food and Drug Administration (FDA) or are in the clinical research phase as shown in Fig. 1. According to the different binding sites of FLT3 inhibitors and FLT3 and the mechanism of action, they have been divided into type I inhibitors and type II inhibitors.12,13) Type I inhibitors, such as Sunitinib, Midostaurin and Gilteritinib, can inhibit cells with ITD mutations or TKD point mutations.12) FLT3 signaling pathway; however, type II inhibitors, such as Sorafenib, Ponatinib, and Quizartinib can only inhibit the FLT3 signaling pathway in cells with ITD mutations and have weak inhibitory effects on cells with TKD point mutations.14) However, it is worth noting that many responding patients are still in incomplete remission and have not achieved complete remission with the restoration of normal hematopoietic function. In addition, it is reported that in 8 FLT3-ITD patients who achieved complete remission with Quizartinib monotherapy, another mutation in the kinase domain (D835Y, D835V, D835F, or F691L) was additionally acquired at the time of relapse.15) These clinical results indicate that selective and sustained FLT3 inhibition can provide patients with complete remission, but FLT3 inhibitors need to be effective against FLT3-ITD and FLT3-TKD mutations.16–18)
To overcome the problem of drug resistance caused by mutations, covalent inhibitors may be a good solution due to their excellent selectivity and strong affinity.19,20) FF-10101, an FLT3 covalent inhibitor, showed sustained FLT3 inhibition in patients with refractory or relapsed AML in phase I clinical study, including those who are resistant to Gilteritinib.21) In this study, we found that the ATP binding pocket of FLT3 has two covalent binding sites: Cys 695 and Cys 828 (Fig. 2). Through covalent virtual screening, covalent inhibitors with novel scaffolds acting on the two sites respectively were obtained. Further biological studies have found that some of these compounds can inhibit FLT3 activity, including D835Y mutant protein.
The structure of the commercial small molecule library ChemDiv containing 1.5 million small-molecule ligands and the covalent inhibitor libraries Covalent-inh-generic, Covalent-inh-smart and Covalent-inh-sp3-b-lactams were filtered using the Structure Filtering module of Schrödinger for substructure search.
Non-covalent Virtual ScreeningThe crystal structure of FLT3 protein (PDB ID: 5X02) was obtained from the RCSB website database. During the non-covalent virtual screening, the key amino acid Cys 695 and Cys 828 was mutated to glycine (Gly) using Mutate residues module to expose the pocket. The mutant FLT3 protein was prepared using the receptor module in the MGLTools 1.5.6 toolkit such as hydrogenation, protonation, and energy minimization. Autogrid4 module was used to prepare the receptor pocket file with a pocket diameter of 20 Å. The covalent compound library was performed pretreatments such as hydrogenation, protonation, and energy minimization using the Ligand Preparation module. Finally, the Autodock GPU v1.2 was used for preliminary non-covalent virtual screening, the nrun parameter was set to 5, and the rest were the default parameters. The GPU NVIDIA GeForce RTX 2080Ti was driven by OpenCL.
The default parameters of the LigPrep module of Schrödinger were used to process small covalent molecules selected by Autodock GPU. Hydrogenation, protonation and energy minimization of the molecules were carried out under the OPLS3e force field. The active pocket, with a diameter of 20 Å, of the receptor based on the original ligand was construct using the Receptor Grid Generation. The virtual screening was run under the standard prediction (SP) mode using the Ligand Docking function to give 5 docking conformations for each ligand. Perform Pose filtering on the top-scored molecules.
Covalent Virtual ScreeningThe Docking Pose filter module of Schrödinger was used to screen the ligands of the Michael acceptor and the key cysteine residues Cys 695 or Cys 828 of the receptor and the small ligand molecules with a Contact maximum distance of 5 Å.
CovDock was used for covalent virtual screening of compounds obtained by the Pose filter. The compounds were hydrogenation, protonation and energy minimization using the LigPrep module under the force field of OPLS3e and the remaining parameters were set as default. The protein does not undergo mutation treatment during covalent virtual screening, and the default parameters of the Protein Preparation Wizard module were used to preprocess the protein structure, optimize hydrogen atoms and minimize energy directly.
Use the CovDock module to specify the receptor active amino acid residues (Cys 695 or Cys 828) and reaction type (Michael addition), specify the interfaced pocket based on the original ligand position of the crystal structure, and perform covalent virtual screening in Virtual Screening (Fast) mode. Then, the molecules with the highest docking scores were selected for covalent docking in Pose Prediction (Thorough) mode, and each ligand outputs 5 docking conformations.
Kinase Inhibition AssaysThe substrate was added to the mixture of the compound and enzyme buffer, and the reaction was started in a 384 reaction plate (ProxiPlateTM-384 Plus, PerkinElmer, Inc., U.S.A.). After incubating for one hour at room temperature, XL665 and antibody were added to incubate for one hour. Finally, detect the fluorescent signal at 665 and 620 nm. At the same time, a blank control group without enzyme, a solvent control group with dimethyl sulfoxide (DMSO) instead of the compound, and a Gilteritinib positive control group were set up. The final volume of the reaction was 10 µL, and the specific reaction system was 2% DMSO, 0.5 ng/µL FLT3, 1 µM TK-s, 2 µM ATP, 5 mM MgCl2, 1 mM MnCl2, 1 mM dithiothreitol (DTT), 1× kinase buffer. There were 3 replicate wells for each sample and each concentration.
The preparation of protein and ligands is critical in any docking process. Tyrosine kinases all contain a circular activation region with a DFG (aspartic acid (Asp)-phenylalanine (Phe)-Gly) motif at the start and an APE (alanine (Ala)-Pro-glutamic acid (Glu)) motif at the end. When the Asp and Phe in the DFG motif are facing the ATP binding site (DFG-in), it is the activation conformation of the kinase, and vice versa, it is DFG-out. We choose the FLT3 protein in the DFG-out conformation for virtual screening. The co-crystal structure PDB ID: 5X02 was chosen as the protein because it is the only crystal structure that has FLT3 DFG-out conformation and is combined with the covalent inhibitor FF-10101 in Cys 695. A covalent compound library containing 124721 compounds was established according to the method previously mentioned.
To test the reliability of the docking method, CovDock, a docking method with a low error rate reported in the literature, was used to re-docking the original ligand (FF-10101) and protein of the crystal complex of FLT3 (PDB ID: 5X02). The CovDock module was carried out after specifying the receptor active amino acid residue Cys 695, select the reaction type was Michael addition based on the OPLS3e force field. Then the covalent docking was performed in the Pose Prediction (Thorough) mode, and the root-mean-square deviation (RMSD) value of the docking conformation of FF-10101 and the bonding conformation in the original crystal structure was calculated. The re-docking test found that it can reproduce the real docking conformation very well, and the top-scoring ligand poses were predicted to have an RMSD within 2.09 Å from the experimental binding mode22) (Fig. 3).
The virtual screening protocol employed to discover FLT3 inhibitors was illustrated in Fig. 4. The covalently bound residues on the protein crystal complex (Cys 695 and Cys 828) were mutated to Gly. Then the chemical library was filtered in the form of non-covalent docking ranked by Autodock GPU scoring function. The binding force and selectivity of the compound and protein were fully optimized by common effects such as hydrogen bonding, hydrophobic interaction and van der Waals force. After filtering by Autodock GPU, the first 8100 compounds with a score of <−9.5 were selected for the Glide_SP, which act as a bridge between non-covalent docking and covalent docking. Then 6435 compounds, with a Glide score of <−7, perform Pose filtering with amino acid residues Cys 695 and Cys 828, respectively. The Pose filter showed that 4315 compounds may be covalently bound to Cys 695 and 1413 compounds may be bound to Cys 828. Finally, the compounds were filtered in the form of covalent docking ranked by CovDock_Fast and CovDock_Thorough scoring function. Sixty-eight compounds achieved scored <−5 from 4315 compounds using CovDock_Fast virtual screening covalently docked with Cys 695; 14 compounds achieved scored <−5 from 1413 compounds covalently docked with Cys 828 using the same condition. A more accurate covalent docking method, CovDock_Thorough, was used to filter molecules to improve the hit rate. Three high-scoring compounds were considered to be a reasonable binding mode and bound to Cys 695; 5 compounds covalently bound to Cys 828 were selected for the biological activity test (Fig. 5). The docking scores of selected compounds were shown in Table 1.
| Key residue | Compound | Docking Score | ||
|---|---|---|---|---|
| Autodock GPU | Glide | CovDock | ||
| Cys695 | k781-1280 | −9.83 | −7.296 | −6.251 |
| G856-8335 | −10.84 | −7.398 | −5.855 | |
| G856-8800 | −11.31 | −7.967 | −5.05 | |
| Cys828 | P218-0398 | −9.60 | −8.203 | −6.413 |
| F537-0060 | −9.79 | −7.264 | −7.422 | |
| 3236-0134 | −9.79 | −7.654 | −6.878 | |
| S551-0611 | −9.81 | −8.365 | −6.567 | |
| S346-0154 | −10.11 | −7.549 | −6.581 | |
These 8 compounds were purchased and tested for FLT3-WT, FLT3-D835Y and C-KIT activities, and Gilteritinib was used as a positive control (Table 2). C-KIT is a kinase of the same family as the FLT3 kinase, and inhibition of both FLT3 and C-KIT can produce myelosuppressive toxicity. Among the 3 compounds selected for potential Michael addition to Cys 695, although K781-1280 showed strong FLT3-WT inhibitory ability, it had a weaker effect on the mutant FLT3-D835Y. G856-8800 showed no selectivity to C-KIT. G856-8335 performed best, with an IC50 of about 1 µM for wild-type; about 3 µM for mutant type; and selectivity for C-KIT. Compounds P218-0398 and S551-0611 showed the strongest inhibitory activity against wild-type FLT3, with IC50 of 0.39 and 0.43 µM, respectively, in the selected compounds with potential Michael addition to Cys 828. However, they had weaker activity against mutant FLT3-D835Y, and there is no selectivity to C-KIT. S346-0154 showed similar strong activity to wild type FLT3 (IC50 = 1.5 µM) and mutant type FLT3 (IC50 = 1.3 µM), as well as good C-KIT selectivity (IC50 > 10 µM).
| Key residue | Compound | IC50 (nM) | ||
|---|---|---|---|---|
| FLT3-WT | FLT3-D835Y | C-KIT | ||
| Cys695 | k781-1280 | 55 ± 0 | >10 µM | >10 µM |
| G856-8335 | 964 ± 314 | 3462 ± 797 | >10 µM | |
| G856-8800 | 694 ± 111 | 6308 ± 569 | 80 ± 0 | |
| Cys828 | P218-0398 | 390 ± 50 | 2459 ± 315 | 5391 ± 842 |
| F537-0060 | 6216 ± 366 | >10 µM | >10 µM | |
| 3236-0134 | 1629 ± 544 | 4598 ± 915 | 9257 ± 1108 | |
| S551-0611 | 426 ± 33 | >10 µM | 65 ± 30 | |
| S346-0154 | 1536 ± 7 | 1327 ± 55 | >10 µM | |
| Gilteritinib | 6 ± 1 | 2 ± 0 | 1969 ± 523 | |
The binding mode analysis of compound G856-8335 showed that Cinnamamide undergone a Michael addition reaction with the amino acid residue Cys 695, as well as formed a hydrogen bond with the Cys 695. The fluorine on the benzene ring extending to the inside of the kinase formed a hydrogen bond with the amino acid residue Cys 828. Benzene ring and thiazolo[3,2-b][1,2,4]triazole ring generating π interaction with the surrounding residues including Leu 616, Val 621, Ala 642, Leu 818 and Phe 830. However, compound G856-8335 does not form hydrogen bonds with amino acid residues Cys 694 and Glu 692 in the hinge region of FLT3, well the positive control Gilteritinib does (Fig. 6). Further structural optimization strategies can focus on the benzene ring that extends inside the kinase. The hydrogen bond donor and the hydrogen bond acceptor can be introduced on the benzene ring to form hydrogen bonds with the amino acid residues Cys 694 and Glu 692 in the hinge region to enhance the affinity and improve the activity.
The binding mode analysis of compound S346-0154 showed that Cinnamamide reached deep into the kinase pocket and undergone a Michael addition reaction with the amino acid residue Cys 828. Compound S346-0154 also failed to form a hydrogen bond with the amino acid residue Cys 694 and Glu 692 in the hinge region for there is no hydrogen bond donor and/or hydrogen bond acceptor on the benzene ring in the middle of the molecule (Fig. 7). It can form a hydrogen bond with the hinge region through scaffold hopping or the introduction of functional groups. The 5,6,7,8-tetrahydro-[1,2,4]triazolo[4,3-a]pyridine extend to the solvent zone, can also be modified to improve the physicochemical properties and in vivo processes of compounds.
In conclusion, FLT3 represents a promising target for the treatment of cancer. In this study, discriminative covalent virtual screening was employed to identify potent FLT3 inhibitors. CovDock was proven to reproduce the experimental binding conformation very well through re-docking. By filtering the established covalent compound library through a combination of non-covalent screening and covalent screening, potential covalent inhibitors can be efficiently obtained. The identification of 8 compounds was further subjected to an in vitro inhibitory activity assay. All of them exhibited good-to-moderate FLT3-WT inhibitory effects with IC50 values of 55 ± 0 to 6217 ± 367 nM. Moreover, compounds G856-8335 and S346-0154 displayed strong activity against mutant type FLT3 proteins and selectivity against C-KIT. These two compounds can be optimized as lead compounds. And this work can be extended to the screening of other covalent inhibitors.
This work was supported by Grant from the China Postdoctoral Science Foundation (2020M671768). Thanks to the activity test provided by the National Center for Drug Screening.
The authors declare no conflict of interest.
