Discovery of potent DNMT1 inhibitors against sickle cell disease using structural-based virtual screening, MM-GBSA and molecular dynamics simulation-based approaches

Abstract Sickle cell disease (SCD) is an autosomal recessive genetic disorder affecting millions of people worldwide. A reversible and selective DNMT1 inhibitor, GSK3482364, has been known to decrease the overall methylation activity of DNMT1, resulting in the increase of HbF levels and percentage of HbF-expressing erythrocytes in an in vitro and in vivo model. In this study, a structure-based virtual screening was done with GSK3685032, a co-crystalized ligand of DNMT1 (PDB ID: 6X9K) with an IC50 value of 0.036 μM and identified 3988 compounds from three databases (ChEMBL, PubChem and Drug Bank). Using this screening method, we identified around 15 compounds with XP docking scores greater than −8 kcal/mol. Further, prime MM-GBSA calculations have been performed and found compound SCHEMBL19716714 with the highest binding free energy of −83.31 kcal/mol. Finally, four compounds were identified based on glide energy and ΔG bind scores that have the most binding with DG7, DG19, DG20 bases and Lys1535, His1507, Trp1510, Ser1230, which were required for the target enzyme inhibition. Furthermore, molecular dynamics simulation studies of top ligands validate the stability of the docked complexes by examining root mean square deviations, root mean square fluctuations, solvent accessible surface area, and radius of gyration graphs from simulation trajectories. These findings suggest that the top four hit compounds may be capable of inhibiting DNMT1 and that additional in vitro and in vivo studies will be essential to prove the clinical effectiveness of the selected lead compounds. Communicated by Ramaswamy H. Sarma


Introduction
Sickle cell disease (SCD) is an autosomal recessive genetic disorder that occurs due to a point mutation of glutamic acid with valine at the sixth position of the b-chain of hemoglobin (Inusa et al., 2019).According to WHO, >0.6 million newborn babies are affected yearly by these b-hemoglobinopathies. Central Africa, Nigeria and India account for 57% of neonates with these defective genes worldwide (Mukherjee et al., 2022).In addition, SCD frequency increased in developed countries due to the migration of people from the country with higher SCD cases (Gardner, 2018).It is characterized by painful vaso-occlusive crises (VOC), hemolysis, pulmonary hypertension, acute chest syndrome and many other disorders that shorten life expectancy (Gbotosho et al., 2023;Kato et al., 2017).
During embryonic development, mammalians express fetal hemoglobin (HbF) composed of 2a and 2c-globin chains.Within six months after birth, the c-globin gene gets epigenetically silenced and HbF is switched to adult hemoglobin (HbA), containing 2a and 2b-globin chains.The pathophysiology of SCD is dependent on the polymerization of deoxy sickle hemoglobin.Patients with mutant hemoglobin (HbS) lose their flexibility in the red blood cells (RBCs) and polymerize upon prolonged deoxygenation.Since HbF does not interact with HbS, it delays the polymerization of HbS within the RBC.Therefore, induction of HbF levels forms an effective therapeutic strategy for the management of b-haemoglobinopathies like SCD and b-thalassemia (Lanaro et al., 2018).
Hydroxyurea reduces both polymerization of HbS and pain in SCD patients (Ata et al., 2023).However, the molecular mechanism that underlies the induction of HbF remains unclear.Large-scale clinical trials of HU in SCD patients revealed a noticeable decrease in the frequency of painful episodes and mitigation of symptomatic and organ damage (Lohani et al., 2018).According to a recent trial, the antioxidant properties of L-glutamine are believed to be the critical therapeutic mechanism as it reduces pain episodes in SCD patients (Ilboudo et al., 2021).Another FDA-approved molecule, voxelotor, has been shown to inhibit HbS polymerization, enhance hemoglobin oxygen affinity and lowers sickling of RBCs in SCD patients (Howard et al., 2021).Crizanlizumab therapy is a monoclonal antibody medication that significantly reduces the rate of sickle cell-related pain crises in patients (Stevens et al., 2021).
The current work focuses on DNMT enzymes that catalyze methyl group transfer from the co-factor S-adenosyl-L-methionine (SAM) to the fifth carbon of cytosine, predominantly in CpG dinucleotides.DNA methylation is one of the most extensively studied epigenetic modifications, as they are much involved in gene expression during the development of mammalian cells (Wong et al., 2019).The DNMT enzymes are categorized into catalytically active (DNMT1, 3 A and 3B) and inactive (DNMT2 and 3 L).Among these, DNMT1 is associated with hereditary sensory neuropathy and various cancers.It is also known as maintenance DNMT as it converts the hemimethylated DNA to full-methylated ones after replication (Chattopadhyaya & Ghosal, 2022;Medina-Franco et al., 2022).In the case of SCD, DNMT1 causes repression of c-globin in adult RBCs.Therefore, DNMT1 inhibitors help reactivate c-globin genes that induce the HbF levels and act as one of the effective therapies for the treatment of SCD (Starlard-Davenport et al., 2022).
DNMTi are categorized into two types, namely nucleoside and non-nucleoside analogs.The former type contains drugs such as 5-azacytidine (2004) and decitabine (2006), approved against acute myeloid leukemia, chronic myelomonocytic leukemia and myelodysplastic syndrome (Feng & De Carvalho, 2022).However, these irreversible, covalent inhibitors have strong secondary effects like low specificity, renal toxicity, chemical instability and poor bioavailability (Akone et al., 2020).Due to these effects of nucleoside analogs, the quest for non-nucleoside drugs is carried out globally.Various approaches, such as virtual screening, drug repurposing and DNMT focused library screening, including natural and synthetic molecules, were carried out to obtain potent DNMT inhibitors (Li et al., 2022;San Jos� e-En� eriz et al., 2017;Zhang et al., 2022).However, none of these drugs are selective for DNMT1 with clear translation from in vitro to in vivo efficacy.Currently, research focuses on developing selective DNMT1 inhibitors effective for cancer treatment.The studies include the development of several carbazole derivatives (Li et al., 2022), quinazolines (Medina-Franco et al., 2022), and dicyanopyridine-based molecules (Horton et al., 2022).Among these, the dicyanopyridine compound developed by GlaxoSmithKline (GSK3685032) showed a significant IC 50 value of 36 nM against DNMT1 (Pappalardi et al., 2021).Another inhibitor, GSK3482364, contributed to the reduction of DNMT1 activity, which resulted in the reactivation of the globin genes HBG1 and HBG2.Thus it increased HbF expression and the percentage of HbF containing erythrocytes in both in vitro and in vivo transgenic murine models of SCD.Therefore, a thorough literature survey provides the rationale for identifying reversible, selective DNMT1 inhibitors against SCD.
In this study, we have used the Tanimoto similarity search to identify compounds with a similar structure to GSK3685032.To these molecules, ensemble docking, binding free energy calculations, and molecular dynamics simulations were carried out to identify potential DNMT1 inhibitors (Figure 1).

Protein preparation and receptor grid generation
The structure of human DNMT1 (DNA methyl transferase 1) bound to inhibitor GSK3685032A was retrieved from the protein data bank, RCSB (PDB ID: 6X9K, 2.65 Å) (Pappalardi et al., 2021).The downloaded protein was prepared using the Protein Preparation Wizard module of Schrodinger.Firstly, the protein structure was pre-processed by adding hydrogen atoms, assigning bond orders and atomic charges, adding missed fractions, and eliminating water molecules beyond 3 Å.Secondly, the preprocessed protein was optimized with PROPKA (pH:7) and minimized with optimized potentials using the OPLS4 force field (Tokalı et al., 2022).The grid box was generated based on native ligand coordinates (X¼ À 44.58, Y¼ À 11.72, Z ¼ 28.91) in the workspace of Receptor Grid Generation Wizard.The bounding box size (inner box -10, 10, 10 and outer box -16, 16, 16) was set similarly to the workspace ligand and no constraints were imposed on any of the binding pocket atoms of the selected protein (Sahayarayan et al., 2021).

Ligand preparation
The potent, selective DNMT1 inhibitor (GSK3685032) underwent a Tanimoto similarity search in three databases, PubChem (Kim et al., 2021) with � 90% Tanimoto threshold (664 molecules), Drug bank (Wishart et al., 2018) with �0.5 similarity threshold (6 molecules), ChEMBL (Gaulton et al., 2017) with �0.5 Tanimoto coefficient (3318 molecules) and downloaded 3988 molecules in SDF format.It was taken as input for ligand preparation using the Schrodinger LigPrep module.Then the ionization states were neutralized by adding or removing protons from the ligand, and stereoisomers were generated from the 3D structures.At last, the minimization of ligands using the OPLS4 force field was performed to complete the ligand preparation process (Guzelj et al., 2022).

Structure-based virtual screening
Molecular docking studies were carried out using a glide module of Schrodinger 2022-2 suite, Xenon W3565 processor, and Ubuntu 18.04 OS (Mak et al., 2022).For HTVS, all the prepared ligands of 5179 (3988 downloaded compounds þ 1191 conformers obtained after ligand preparation) were docked at the binding site of 6X9K protein.The resultant compounds with a docking score better than the cocrystal ligand had passed to the SP filter (149 molecules), which aids in reducing the number of spurious conformers throughout the docking process.At last, the top 30% of SP ligands (about 50 molecules) had subjected to XP docking (Bhowmick et al., 2021).

Molecular mechanics -generalized born surface area (MM-GBSA)
The prime MM-GBSA calculations were performed for the top 30% XP results using the default VSGB solvent model and OPLS4 force field to generate energy properties.The energies obtained from the prime MM-GBSA reports help calculate ligand and receptor strain and binding energies (E.Wang et al., 2019).The binding free energies were calculated using the following equation: DG bind is the free energy for binding the ligand (L) to the protein (P) to form the PL complex.

In-silico ADMET
The in-silico absorption, distribution, metabolism, excretion and toxicity properties of the top four compounds were analyzed using the Schrodinger QikProp module and OSIRIS Data Warrior tools (Sander et al., 2015).

Dynamic behavior of the docked complex
Molecular dynamics simulations were performed using the academic version of the Desmond module, D.E Shaw research group (Academic License, Version 2020-4).The docked complex was soaked into a TIP3P solvent model within a 10 Å boundary size box (Aksoydan & Durdagi, 2022).Then, counter ions (31 Na þ ions) and salts were added to neutralize the charges and maintain physiological conditions.Then the complex was minimized and equilibrated in an NPT ensemble at 310 k, pressure (bar) 1.01 and using OPLS4 force field runs for 100 ns.At last, a simulation interaction diagram (SID) was generated, which includes various parameters of the dynamics studies, such as RMSD and RMSF of the protein-ligand complex, radius of gyration of the ligand, and protein-ligand interaction data (Mahmud et al., 2021).

Molecular docking and prime MM-GBSA calculations
The molecular docking results revealed that four hit compounds might be potential inhibitors of DNMT1 as they exhibit interactions with its catalytic site residues.Based on the reported potent molecule GSK3685032, the molecules from the selected three databases were prepared using LigPrep, resulting in 5179 compounds.Next, Structure-based drug design (SBDD) was employed, one of the computeraided drug design (CADD) approaches used to identify novel active small molecules against a chosen therapeutic target.It leading to significant advances in pharmaceutical research and development (X.Wang et al., 2018), (Macalino et al., 2015).It includes structure-based virtual screening (SBVS), which helps to predict or identify the active compounds from chemical databases towards a particular drug target at the early stages of drug discovery.It utilizes three dimensional (3D) structure of biological targets to dock the molecules of chemical libraries and forecast the binding affinity for further biological evaluation (Toledo Warshaviak et al., 2014).As a part of SBVS, we performed three modes of docking, namely high throughput virtual screening (HTVS), standard precision (SP) and extra precision (XP), using the downloaded libraries of compounds.The initial HTVS study resulted in 149 ligands which were then processed to SP followed by XP docking to ensure accurate screening (Figure 2).The Glide ligand efficiency, XP Gscore, Glide evdw, Glide ecoul, Glide energy and Glide model results were considered for screening ligands from libraries (Table 1).The higher negative score derived from these attributes were utilized to rank the top 15 molecules from the databases.Furthermore, these top fifteen molecules from XP docking were post-processed to prime MM-GBSA calculation studies, followed by the final selection of ligands.The prime MM-GBSA reports the fundamental free energies of ligands, receptors and complexes (E.Wang et al., 2019).It revealed that the binding free energies of selected top molecules range from À 41.4 to À 83.3 kcal/mol.In contrast, the reference compound (cocrystal ligand) and approved drugs have binding free energies of  À 61.43, À 35.73 and À 34.31 kcal/mol, respectively.Except for SCHEMBL19716714, all other compounds had binding energies that were less negative than À 60 kcal/mol, implying modest binding of ligands with the enzyme compared to the cocrystal ligand.The MM-GBSA binding energies of the top fifteen molecules were tabulated in Table 1.Next, the calculations were done to observe the binding energy (DG bind) at the active site of DNMT1.Only four compounds were selected based on superior glide energy and binding free energy of top molecules when compared to the potent molecule.Among all the molecules, SCHEMBL19716714 had the highest XP Gscore (À 10.73 kcal/mol) and binding energy (À 83.31 kcal/mol) towards the binding pocket of DNMT1.
The validation was done by docking the native ligand and found the RMSD of 0.83 Å, as shown in Figure 3.The binding score and glide energy of the cocrystal ligand was found to be À 7.1 and À 67.641 kcal/mol, respectively.Similarly, the docking of approved drugs (azacytidine and decitabine) was also carried out and found the docking score of À 4.67 and À 3.78 kcal/mol.Firstly, the amine group of mercaptoacetamide of the cocrystal ligand forms two hydrogen bond interactions with DG19 and DG20 at a distance of 2.41 Å and 1.92 Å, respectively.The pyridine ring of the ligand forms p-p stacking contact with DG7 (4 Å) and DG19 (3.95 Å) at the major groove of DNA (Table 2).
Secondly, the hydroxy groups of approved drugs actively participated in forming hydrogen bond linkages with DG19 and DG20 bases.Further, the triazine ring of both drugs was involved in p-p stacking interactions with the same bases at the active site of DNMT1.Additionally, amino acid residues like His1507, Trp1510, Lys1535 and Ser1230 are also in contact with cocrystal ligand and approved drugs at the catalytic site via Van Der Waal interactions (supplementary material, Figure 1).
The dicyanopyridine ring present in all four hits shows p-p stacking interactions with DG7 and DG19 bases similar to that of cocrystal ligand.Despite the presence of phenylacetamide moiety in all the titled compounds, only  SCHEMBL19716714 and cocrystal ligand were involved in the interaction with DG19 and DG20 via hydrogen bonds.Interestingly, substituting 3-amino azetidine for 4-amino piperidine at the sixth position of the crystal ligand reduced the negative binding free energy in 132248943 (À 50.08 kcal/mol).On the contrary, the same position had been replaced with 1-methyl diazepane, which resulted in higher negative binding free energy in SCHEMBL19716906 (À 60.17 kcal/mol).Additionally, the Comp.code -compound code; Donor HB -hydrogen bond donor (0-6); Acceptor HB -hydrogen bond acceptor (0-20); log Po/w -partition coefficient (octanol/water) (À 2.0 to 6.50); log S -solubility compound (À 6.5 to 0.5 mol dm À 3 ); P Caco -Caco-2 cell permeability in nm s À 1 (<25 poor, >500 good); log BBbrain/blood partition coefficient (À 3.0 to 1.2); % human oral absorption -percentage human oral intestinal absorption (up to 100); Druglikeness -Values were positive for 80% of the marketed drugs.cyclopropane substitution in the fourth position of SCHEMBL19716714 resulted in a significant drift in the negative binding energy (À 83.31 kcal/mol).However, modifying the cyclic diazepane to acyclic diamine significantly decreased the negative binding free score in SCHEMBL19716460 (À 42.19 kcal/mol) (Figure 4).Based on a critical examination of these chemical features, cyclic and bulky substitutions at fourth and sixth position of dicyanopyridine based inhibitors could have superior binding profiles compared to acyclic and compact inhibitors.Furthermore, except SCHEMBL19716906, compounds such as SCHEMBL19716714, SCHEMBL19716460 and 132248943 had van der Waal contacts with Lys1535 residue.This might be the reason of top hits with greater docking scores along with other common contacts.Similarly, the binding interactions of the top four molecules were illustrated in (supplementary material, Figure 2).Table 2 lists the residues of DNMT1 that interacted with ligand molecules at the active site of 6X9K.

In-silico predicted ADME and toxicity parameters
The QikProp module of Schrodinger analyzed the physiochemical descriptors of the top four molecules and the results were compared with cocrystal ligand results.All the examined molecules followed the Lipinski rule of five (Mol.wt, donor Hb, acceptor Hb and logP) and were all within the acceptable range of marketed drugs.The solvent accessible surface area, aqueous solubility, blood/brain partition coefficient and the number of rotatory bonds were satisfactory and within the recommended range of QikProp (supplementary material, Table 1).The descriptors like P caco and percentage oral absorption of investigated molecules were out of recommended thresholds.However, the top hit compounds had comparable values to reference molecule with slight variations in Caco-2 cell permeability and human oral absorption values.The toxicity results demonstrated the non-toxicity of the top hit compounds.In addition, the cancer-causing properties, including mutagenicity and tumorigenicity, were nil.Other effects, such as reproductive effects and irritation, were also found to be none.Druglikeness values of the four hit compounds ranged from À 1.35 to 0.85, with positive values for SCHEMBL19716714 and SCHEMBL19716460 and negative values for SCHEMBL19716906 and 132248943.However, the druglikeness value of the reference ligand was also in negative for further consideration of top hits (Table 3).Therefore, these compounds can be further explored with these favorable ADMET results.

Molecular dynamics (MD) simulation analysis
The molecular dynamics simulation analysis was performed for the top four hit molecules to check the stability, conformational changes, and intermolecular interactions at the active site of DNMT. Figure 5 depicts the RMSD of proteins and ligands.The protein RMSD of all complexes and apoenzyme were found in the range of 1.5 to 4 Å.Among all the complexes, SCHEMBL19716460 showed a slight drift after 65 ns to the rest of the MD simulation.Comparatively, the RMSD graphs of all complexes showed results similar to those of the apoenzyme with minor variations during the 100 ns simulation.The ligand RMSD of the top four hit molecules was found in the range of 2 to 7 Å, slightly higher than the cocrystal ligand RMSD.The SCHEMBL19716906 exhibited a substantially higher deviation than other top ligands due to p-cation interaction with the DG7, DC21 and DC5 residues and low contacts with active residues DG19, DG20, His1507 and Trp1510.SCHEMBL19716714 could replicate similar docking contacts with minimal variations lasting up to 30 ns before achieving stability for the rest of the simulation.Further, the remaining compounds, 132248943 and SCHEMBL19716460, were revealed to be entirely in contact with the protein backbone, with few fluctuations at the beginning of the simulation (Figure 5).
Additionally, RMSF peaks depict the flexibility of the amino acid residues during the simulation.From Figure 5, all the complexes had RMSF values less than 4 Å except for a few amino acid residues that fluctuated up to 7 Å during the simulation.However, the interacting residues of all complexes showed lower RMSF values of less than 2 Å, which were comparable with those of apoenzyme.Solvent accessible surface area (SASA) trends indicate the volume of protein is in contact with its solvent during the simulation.The SASA trends demonstrate that all the complexes were stable without significant conformational changes in the protein structure throughout the simulation period.Among all the complexes, SCHEMBL19716714 had higher SASA values, indicating that most complex residues are associated with solvent molecules.While complex SCHEMBL19716906 has a low SASA value and other remaining complexes have similar SASA values to that of the cocrystal complex with little fluctuations with the complex SCHEMBL19716460 after 40 ns to the rest of the simulation trajectory.
Moreover, the radius of gyration (Rg) graph describes the extendedness of the enzyme-ligand complex.The Rgs of all the complexes were in the range of 3.8 to 4.5 nm and the results were comparable with the reference complex Rg value.The Rg results of all complexes indicate that each enzyme substrate showed stable compactness throughout the simulation time.
According to the H-bond plot, all hit molecules fall between 1 and 7 H-bond contacts with the target.Among them, SCHEMBL19716714 had maximum H-bond connections, followed by SCHEMBL19716906 and SCHEMBL19716460 throughout the simulation.Further, 132248943 displayed a similar H-bond pattern to a crystal ligand with one or two contacts throughout the 100 ns of simulation (Figure 5).
All the ligands continuously interacted with the active site and displayed stable hydrogen bonds on the active residues DG7, DG19, DG20, Ser1230, His1507, Trp1510, Gln1227 and Lys1535 throughout the simulation time.However, the results of the H-bond plot reveal that hit molecules exhibited strong contacts when compared with the cocrystal ligand at the common binding site.As a result, it reflects that the activity of these ligand molecules is most likely comparable to that of an active cocrystal ligand.
The binding pattern of the top four best compounds and crystallized ligand is depicted in Figure 6.All the ligands interacted with DNA guanine bases during most of the simulation duration and the actively involved bases include DG7, DG19 and DG20.Weak interactions were also observed with Ser1230, His1507 and Lys1535 at the binding pocket of DNMT1.At first, the cocrystal ligand retained the docked pose interactions after MD simulations.The native ligand predominately forms DNA contacts with bases DG7 (p-stacking), DG19 (H-bond) and DG20 (H-bond) and the complex was stabilized with water-mediated interactions with Lys1535, Ser1230 and Gln1536 during the simulation trajectory.Further, the compound SCHEMBL19716714 demonstrated that the amide group of 2-(pyridin-4-yl)-acetamide forms two hydrogen bonds with DG19 and DG20.A p-stacking interaction was also observed between the dicyanopyridine ring and DG7.Moreover, the complex was strengthened by the presence of Ser1230, His1507, Lys1535 and Trp1510 at the catalytic site throughout the simulation.
The compound SCHEMBL19716460 was primarily bound by water-bridged hydrogen contact with Ser1230, His1507, Gly1228 and Gln1536 for about 10%, 5%, 4% and 2% of the simulation time.Further, the dicyanopyridine ring interacts with DG7 through hydrophobic interactions, strengthening the complex.SCHEMBL19716906 also forms water-mediated interactions, followed by hydrogen bonds and hydrophobic interactions with the active site residues.The oxygen of the acetamide group made hydrogen contact with Ser1230 about 50% of the simulation period.The ligand forms p-cation and p-stacking links with DG7 and DG19 bases through diazepine and dicyanopyridine moieties that help in the stabilization of the complex.Similarly, the amine group of 3amino azetidine in compound 132248943 bound to Ser1230 through water-assisted interaction contributes 50% of the simulation trajectory.In addition, it was observed that the guanine bases DG7, DG19, DG20 and amino acid residues Lys1535 and Trp1510 coordinated with the ligand through van der Waals interactions at the binding pocket of DNMT1.
Upon critical examination, the best four hits' compounds revealed steadiness with DNA guanine bases and amino acid residues in the system.Low fluctuations have been observed between ligands with their complexed proteins.Our results suggest that the DNA guanine bases DG7, DG19, DG20 and amino acid residues Ser1230, His1507, Trp1510, and Lys1535 contribute to the PL interactions (Figure 7).Notably, the dicyanopyridine ring of all ligands readily formed p-stacking interaction with the DNA guanine bases (mostly DG7) and contributed to stability during the simulation time.Finally, time-frame snapshots have been taken to determine the conformational changes of the best four compounds throughout the MD simulation trajectory and illustrated in Figure 7. Every 25 ns time interval was analyzed to observe the variation of ligands from their binding pose.The results suggest that all four complexes studied contribute to the high stability with minimal conformational changes at the active site of DNMT1.
Therefore, our findings from the analysis of CADD tools reveal that four lead compounds may disrupt the DNMT1 activity and may increase the HbF expression.However, additional laboratory testing is required to determine these compounds' effectiveness and inhibitory capability under in vitro conditions.

Conclusion
In the current work, We performed a similarity search on three databases using a potent co-crystallized ligand in a complex with DNMT1 enzyme (PDB ID: 6X9K).Then, the compounds were subjected to structure-based virtual screening and based on the glide XP score, 15 top hits were selected.All the hits showed a better docking score with a minimum value of À 8.05 kcal/mol and better than the native ligand.Moreover, the docked poses were subjected to MM-GBSA calculations and found the binding free energies of top hits towards the DNMT1 protein.The best four compounds were identified and among these, SCHEMBL19716714 was found to have the highest XP Gscore (À 10.73 kcal/mol) and binding energy (À 83.31 kcal/mol) towards DNMT1.Based on glide energy and MM-GBSA results, the top four hits were selected and evaluated for molecular dynamics simulations.However, the predicted ADME and toxicity profile of the selected top hits showed superior drug-likeness properties combined with zero toxicity possibilities.
We also confirmed the presence of the binding residues of DNMT1, such as DG19, DG 20, DG7, Lys1535, His1507, Trp1510, and Ser1230, and their existence in the post-MD structures.Since this work is mainly based on simulation analysis and various computational tools, it needs further in vitro evaluation.Finally, the four most promising compounds may increase HbF levels by reducing the overall DNMT1 methylation activity in a way similar to the reference molecule and could aid in developing potent candidates against SCD.

Figure 1 .
Figure 1.Drugs against SCD that have been approved by the US Food and Drug Administration (A), as well as nucleoside (B) and non-nucleoside analogs (C) of DNMT inhibitors.

Figure 2 .
Figure 2. The structure-based virtual screening workflow using the glide module of Schrodinger and identification of hit molecules (4).

Figure 4 .
Figure 4.Chemical feature comparison of cocrystal ligand at the center and top four hit compounds in the 2D interaction diagram.

Figure 5 .
Figure 5.The molecular dynamics analysis of the top four compounds along with native ligand, here a, b, c, d, e and f, indicates the RMSD of protein, RMSD of ligands, RMSF, SASA, Rgs and HB contacts plots, respectively.

Figure 6 .
Figure 6.The 2D interaction diagram (post-MD) and Protein-ligand contacts of DNMT1 with the best four compounds and reference ligand throughout the MD trajectory.

Figure 7 .
Figure 7.The surface view of DNMT1 with the top four hit compounds and their snapshots were taken at 0 (docked pose), 25, 50, 75 and 100 ns during the progress of the MD simulation.

Table 1 .
Docking results (kcal/mol) and binding free energies (DG bind) of the top 15 compounds.

Table 2 .
Non-covalent interactions of the top four hit molecules and cocrystal ligand with the target protein (PDB: 6X9K).

Table 3 .
In-silico predicted ADMET properties of top four hit molecules and cocrystal ligand.