FLT3 inhibitor design using molecular docking based virtual screening for acute myeloid leukemia

Acute Myeloid Leukaemia (AML) is a blood cancer, which affects the red blood cells in the bone marrow. Of the possible proteins that are affected in AML, fms-like tyrosine kinase 3 (FLT3) has long been recognized as a potential therapeutic target as it affects the other signaling pathways and leads to a cascade of events. First-generation inhibitors sorafenib and midostaurin, as well as secondgeneration agents such as quizartinib and crenolanib are known. It is of interest to identify new compounds against FLT3 with improved activity using molecular docking and virtual screening. Molecular docking of existing inhibitors selected a top scoring bestestablished candidate Quizartinib having PubChem CID: 24889392. Similarity searching resulted in compound XGIQBUNWFCCMASUHFFFAOYSA-NPubChemCID: 44598530 which shows higher affinity scores. A comparative study of both the compounds using a drug-drug comparison, ADMET studies, boiled egg plot and pharmacophore parameters and properties confirmed the result and predicted the ligand to be an efficient inhibitor of FLT3.

The patho-physiology of Acute Myeloid Leukemia is such that it arises in patients with an underlying hematological disorder, or as a serious side effect of prior therapy, for example, exposure to topo-isomerases II, alkylating agents or radiation being the key factors. But, moreover, we see that in most of the cases it appears as a de novo or new malignancy in previously healthy individuals [1]. The pathogenesis of AML involves the abnormal proliferation and differentiation of a clonal population of myeloid stem cells. It progresses rapidly, with myeloid cells ©Biomedical Informatics (2019) interfering with the production of normal white blood cells, red blood cells, and platelets. This also results in the accumulation of poorly differentiated myeloid cells.
In addition to large chromosomal rearrangements, molecular changes have also added to the development and prognosis of AML. We see that genetic mutations are identified in more than 97% of cases, often in the absence of any large chromosomal abnormality. There are 3 classes of mutations in acute myeloid leukemia. [2] The class I mutations result in the activation of proproliferative pathways. The class II mutations impair normal hematopoietic differentiation in order for leukemia to develop. These 2 classes occur in conjunction with each other. Common class I mutations, such as FLT3 (internal tandem duplications, ITD, and tyrosine kinase domain mutations, TKD), K/NRAS, TP53, and c-KIT are found in 28,12,8 and 4% of cases, respectively [3]. Mutations in receptor tyrosine kinases, for example, FLT3 duplications are seen up to 50% of AML cases and statistically denote a worse prognosis. Alterations in genes involved in epigenetic regulation have recently emerged as a third class of mutations, with downstream effects on both cellular differentiation and proliferation that is, the class I and class II type mutations [4]. The identification of recurrent genetic mutations is considered in FLT3. It's prognostic impact of FLT3-ITD complex interaction may depend on the presence of biallelic mutations [5]. FLT3 tyrosine kinase receptor is located on chromosome 13 (13q12) of the human genome.
Fatigue, anorexia and weight loss, bleeding, bruising, infection, red spots on the skin, or shortness of breath are symptoms of acute myeloid leukemia. If AML is left untreated, it will lead to immediate death within months of diagnosis. Young adults have undergone multicenter treatment trials [6] and the survival rates are calculated [7]. Acute Myeloid Leukemia has 6 major classifications namely, AML with recurrent genetic abnormalities, AML with myelodysplasia-related changes, therapy-related myeloid neoplasms, myeloid sarcoma, myeloid proliferations related to Down syndrome and AML not otherwise specified. These can be broken down to have 11 subtypes collectively.

Materials and Methodology: Selection of FLT3 inhibitors:
Literature findings were conducted to find out the pre-established inhibitors of FLT3 for acute myeloid leukemia (Table 1). These inhibitors were selectively sorted out on the basis of its binding capacity to the protein, in turn deciding the behavior of the complex formed. The total numbers of inhibitors selected for further analysis were 47. 3D Structures of these inhibitors were obtained. Most of these structures are available in the PubChem database from where it was directly downloaded. Some of the unavailable structures were built using molecular editor software, MarvinSketch. All of them were saved in 3D.sdf format.

Protein and ligand Preparation:
The crystal structure of the target protein was obtained from the Protein Data Bank with PDB ID: 4RT7 [37]. In a similar way, compiling all the 3D structures of ligands using the LigPrep module of Schrodinger suite, 2013 (Schrodinger did the ligand preparation. LLC, New York, NY) where, these were optimized through OPLS 2005 force field algorithm [38][39][40][41][42]. Both the structure files of the protein and the ligands were saved for the ease of processing for docking analysis. All the ligand structures were saved under a single file using a .sdf extension [43-47].

Molecular Docking:
Molecular docking analysis was carried out using Molegro Virtual Docker (MVD), which unifies high potential Piece-Wise Linear Potential (PLP) and MolDock scoring function [48][49][50][51][52][53][54]. The preexisting ligand was removed from the complex structure of the FLT3 protein structure and all its cavities were prepared using Molegro Virtual Docker [55][56][57]. The cavity one, having the largest volume was selected, verifiably the one in which the previous ligand was docked. The preferred orientation of one molecule with the other to form a stable complex was found after docking the .sdf file of the set of ligands with the target protein. The docking procedure holding parameters were a maximum iteration of 1500, grid solution 0.2 having a binding affinity and a maximum population size 50. The protein and ligands were assessed on the subsequent conformation of the Internal Electrostatic interaction (Internal ES), sp2-sp2 torsions, and internal hydrogen bond interaction [58][59][60][61][62]. Post-docking, energy minimization and H-bond optimization were carried out. Placing of Simplex Evolution at max steps 300 and neighbor distance faster 1.00. This confirms that the predominant molecule obtained after docking forms a stable complex. After docking to minimize the complex energy of ligandreceptor interaction the Nelder Mead Simplex Minimization (using non-grid force field and H-bond directionality) was used [63-66].
©Biomedical Informatics (2019) The best pose of the drug was tallied from the result generated and was then imported. The resultant structure generated was saved as the best-posed drug and was stored in PDB format, for that particular compound [74-76]. These steps were repeated for the second docking file containing all the information of the virtual screened compound and the protein cavity. The best fit was studied and an excel sheet was organized to check and compare all the affinities, hydrogen interaction, steric energy and high re-rank score to draw out a comparison between the two drugs.

ADMET studies:
The admetSAR database is an open and free interface, which takes into account the absorption, distribution, metabolism, elimination, and toxicity (ADMET) of the drug molecules [38 -39]. These tell us about the pH, solubility and the overall physicochemical properties of drugs to study the drug metabolism and drug-drug and drug-body interactions. AdmetSAR database is available at http://lmmd.ecust.edu.cn:8000/ for analysis of a particular drug. The properties and parameters help inproviding us with essential information related to the development and discovery of drugs. The admetSAR database mostly consists of 22 qualitative classifications and 5 quantitative regression models, which gives us a high precision based predicted outcome. Hence, the estimation of the properties of the compounds was done using admetSAR. The best docked established compound Quizartinib having PubChem CID-24889392 and the best virtual screened compound with PubChem CID-44598530were considered and the bioactivity properties and toxicity were predicted by using admetSAR [77-79].

Softwares, Suites and Web servers Used:
The 3D chemical structures were retrieved from NCBI's PubChem database in 3DSDF format. Some compounds that lack PubChem ID or the 3D structure was unavailable in PubChem were drawn with the help of MarvinSketch5. 6 The drug to be studied is also classified as the P-gp substrate (PGP+; blue dots) and P-gpnonsubstrate (PGP-; red dots), which provide the absorption and distribution of the drug in the central nervous system (CNS). Thus, we get to know the distribution rate and localized accumulation of the drug for passive absorption (inside/outside the white), passive brain access (inside/outside the yellow) and active efflux from CNS to the gastrointestinal tract (GI). In order to get the functional properties of the drug for target protein FLT3, the mapping of the parameters of GI and BBB and optimization of the BOILED-Egg plot was done by using the rerank affinity of the docking results of the pre-established drug and the virtual screened compound.

Virtual Screening Results:
Virtual screening of the best-established compound against PubChem database ensued in a total number of 109 compound structures that showed a similarity percentage of =>95. Table 3 enlists top 10 compounds that display the highest affinity to the target protein based on their re-rank scores. The compound with PubChem CID 44598530 was observed to have the lowest re-rank score, which made it the best virtual screened compound. Physical properties of this compound include a molecular weight of 651.801 g/mol, a hydrogen bond donor count of 2, a hydrogen bond acceptor count of 10 and a logP value of 5.4. The re-rank score of this compound stands at -190.091 and the H-bond interaction score at -5.93428. Table 4 provides a comparative account of value and re-rank scores of the best-established drug and the best virtual screened drug. It brings about the estimate between the similarities and dissimilarities of the 2 best drugs when they are docked in the first cavity of the target protein structure. It is very clear from the table given below that the best virtual screened compound binds with higher affinity to the target protein than their best-established compound on the basis of the re-rank score. Apart from rerankscore, we take into consideration the other parameters such as External ligand interactions, protein-ligand interactions, and hydrogen bonds. We see the best virtual screened compound to have lower values for all these parameters than the best established compound, which indicates a better affinity to FLT3 protein structure. Steric values computed by both PLP as well as LJ12-6 methods are again lower for the best virtual screened drug, so is the hydrogen bond value with no directionality. Pharmacophore mapping: Pharmacophore mapping takes into account the molecular features of a ligand for the recognizable arrangement to the target protein. It brings in the optimal spatial systematic topographies of molecular interaction with the target protein receptor. A precise pharmacophore model will help bind a novel structurally diverse ligand to the same receptor site, assisted by annotations and characterized by the aligned poses of the molecule. Figure 1 gives the target protein structure. Binding of the target protein FLT3 with best virtual screened drug PubChem CID 44598530 is realized to be effective and efficient and hence, pharmacophore studies were conducted to improve the understanding of the varying interactions observed in the complex so formed. The interaction studies were carried out only for the purpose of including hydrogen bond interactions, van der Walls interaction, electrostatic interactions, aromatic interactions, and Ligand-binding Interactions.   Figure 3 presents the interacting residues of FLT3 protein structure with compound PubChem CID: 44598530 embedded in its cavity. The residues represented by pink circles exhibit electrostatic interactions whereas those in green exhibit van der Waals (vdW) interactions. Green, as well as blue dotted arrows between the interacting species, denotes hydrogen bonds. Hence, Tyr 693, Lys 614, Lys 644 act as hydrogen bond acceptors and the other 2 hydrogen interactions act as hydrogen bond donors. Also, there is a formation of a sigma-pi bond between the inhibitor and Phe 105. Leu 616 forms a sigma bond with the drug and shows a direct van der Waals (vdW) interaction with the drug.  ADMET Profile: Table 5 provides a comparative account of ADMET property prediction of the best-established compound Quizartinib with PubChem CID: 24889392 and best virtual screened compound having PubChem CID: 44598530. Looking at individual properties it is seen that both the compounds display a positive Blood-Brain Barrier, with the virtual screened drug encasing a lower probability value compared to the established drug. Human Intestinal Absorption (HIA), which provides the prediction of absorption of the compound in the intestine, shows us that the established drug only has a slightly higher increment probability of 0.9953 than the virtual screened drug probability of 0.9709. This shows us that both the drugs can be used in varying dosages and the probability of absorption of either of them is relatively close.

Drug-drug comparative study:
The P-glycoprotein Substrate and P-glycoprotein Inhibitor predictions of both the compounds highlight higher probability values associated with the virtual screened drug when compared to the values associated with the established drug. A difference is observed in the probability of the drug distributed over its centers and its sub-cellular localization in the mitochondria. They have a common distribution center over which the established drug has a 0.5208 probability and the virtual screened drug has a 0.3781 probability. Metabolism predictions varies in points like CYP450 2C9 Substrate, CYP450 2D6 Substrate, CYP450 3A4 Substrate, CYP450 1A2 Inhibitor, CYP450 2D6 Inhibitor, CYP450 2C19 Inhibitor, CYP450 3A4 Inhibitor, CYP Inhibitory Promiscuity with both the compounds acting as non-substrates as well as noninhibitors in all the cases, except in case of CYP450 3A4 Substrate ©Biomedical Informatics (2019) were the established drug and the virtual screened drug is predicted to act as a substrate. CYP450 2C9 Inhibitor is another exception in which the established drug acts as an inhibitor and the virtual screened drug acts like a non-inhibitor.
Toxicity studies infer that both the compounds are non-carcinogens and are not readily biodegradable although the virtual screened drug shows a better result as per the table in these two categories. Both the compounds are also non-AMES toxic but the toxicity value of the established drug is relatively same when compared to that of the virtual screened drug. It can be summarized that the best virtual screened compound displays slightly more preferable probabilities when compared to the best-established compound. Table 6 summarizes the regression prediction comparison of ADMET analysis of the two compounds namely the bestestablished drug and the best virtual screened drug. The regression model highlights that the Rat Acute Toxicity level and the Fish Toxicity has a marginal difference in the readings comparison between the established compound and the virtual screened compound, as given in the table as per the LD50 and pLC50 values respectively. Tetrahymena Pyriformis Toxicity in the virtual screened compound has a lower toxic endpoint level than the established compound.   Comparative ADMET profile study of the compounds and the control A relative ADMET profile comparison was carried out for selected inhibitors by taking the predicted probability values of parameters such as Blood-Brain Barrier (BBB), Human Intestinal Absorption (HIA), AMES Toxicity, and LD50 rat toxicity ( Table 7). The two best established compound results from docking were considered of PubChem CID: 124889392 and PubChem CID: 24826799 respectively against the two best virtual screened compound having PubChem CID: 44598530 and PubChem CID: 52934143 respectively. These four compounds were utilized for this study and graphically represented using R-programming as shown in Figure 5. The parameters, BBB, HIA, AMES Toxicity, and LD50 acquired from the admetSAR database were used to tabulate the comparative results according to their estimated values. The best virtual screened compound displays the lowest probability for Blood-Brain Barrier (BBB) and nearby values for AMES toxicity, HIA and LD50 values of all the compounds. The overall profile study is more favorable for the virtual screened compound than the established compound.

Boiled-egg Plot
The Boiled--egg model delivers a rapid, easily producible yet robust method to predict the passive gastrointestinal absorption and brain access of small molecules useful for drug discovery and development. The results of the Boiled-egg plot graphically represent the three out of four compounds in (Figure 6). Four drugs were analyzed for the Boiled-egg plot analysis. Two pre-established drugs having PubChemCID: 24889392 and PubChem CID: 24826799 were considered along with two virtual screened drugs having PubChem CID: 44598530 and PubChem CID: 52934143. The selected drugs are tabulated below in Table 8. The result is conclusive of the outside grey region stands for molecules with properties implying predicted low absorption and limited brain penetration which holds good for our best virtual screened drug. It is favorable to the virtual screened drug as it's not Blood-Brain barrier permeant (yellow region). One of the molecules is out of range as its TPSA and WlogP values exceed beyond the graph and hence it's not shown.

Conclusion:
The best inhibition effect given from the set of established inhibitors was by Quizartinib, which has PubChem CID: 24889392. This compound was tallied and searched against the database to obtain an entirely new ligand, which is the best virtual screened product, with PubChem CID: 44598530. A drugdrug comparative study yielded lower values for External ligand interactions, protein-ligand interactions, hydrogen bonds and rerank score parameters than the established compound pertaining to the higher affinity of the compound to the protein. The stable overall arrangement of the virtual screened ligand-protein complex is done by taking into account all of their features and structure to give us a good result of five H-bond forming amino acids. The positive conclusive result from the boiled-egg plot shows that it does not cross the blood-brain barrier (BBB) and hence, is effective for treating acute myeloid leukemia. This virtual screened compound is better than the other drugs.