Computational drug repurposing of Akt-1 allosteric inhibitors for non-small cell lung cancer

Non-small cell lung carcinomas (NSCLC) are the predominant form of lung malignancy and the reason for the highest number of cancer-related deaths. Widespread deregulation of Akt, a serine/threonine kinase, has been reported in NSCLC. Allosteric Akt inhibitors bind in the space separating the Pleckstrin homology (PH) and catalytic domains, typically with tryptophan residue (Trp-80). This could decrease the regulatory site phosphorylation by stabilizing the PH-in conformation. Hence, in this study, a computational investigation was undertaken to identify allosteric Akt-1 inhibitors from FDA-approved drugs. The molecules were docked at standard precision (SP) and extra-precision (XP), followed by Prime molecular mechanics—generalized Born surface area (MM-GBSA), and molecular dynamics (MD) simulations on selected hits. Post XP-docking, fourteen best hits were identified from a library of 2115 optimized FDA-approved compounds, demonstrating several beneficial interactions such as pi–pi stacking, pi-cation, direct, and water-bridged hydrogen bonds with the crucial residues (Trp-80 and Tyr-272) and several amino acid residues in the allosteric ligand-binding pocket of Akt-1. Subsequent MD simulations to verify the stability of chosen drugs to the Akt-1 allosteric site showed valganciclovir, dasatinib, indacaterol, and novobiocin to have high stability. Further, predictions for possible biological interactions were performed using computational tools such as ProTox-II, CLC-Pred, and PASSOnline. The shortlisted drugs open a new class of allosteric Akt-1 inhibitors for the therapy of NSCLC.

Selection of target protein. The target protein considered for this computational study was Akt-1, which was in an autoinhibited form, co-scrystallized with an allosteric inhibitor-12j/0R4. The crystal structures of Akt-1 in the auto-inhibited form were searched in the Protein Data Bank (PDB) (http:// www. rcsb. org/) 22 . In the RCSB database, we could find three PDB IDs, 3O96, 4EJN and 5KCV, as protein structures wherein Akt-1 was co-scrystallized with different allosteric inhibitors. Among them, PDB ID: 4EJN, co-crystallized with a ligand 12j/0R4, was selected for this computational study. The same crystal structure was being used for another computational study that involved the identification of imidazopyridine analogues as Akt-1 allosteric inhibitors by molecular docking and MD simulations by Gu et al. 23 .
Protein preparation. The PDB-ID 4EJN of Akt-1 bound to co-crystal ligand (12j/0R4, IC 50 = 5 nM) with a resolution of 2.19 Å was downloaded and imported to the Maestro interface 24 . The procedure followed by Gu et al., 2021 for screening and establishment of imidazopyridine analogues Akt1 allosteric inhibitors using 3D-QSAR, molecular docking and molecular dynamics simulations 23 was followed by us for protein preparations. An initial setup of protein preparation was performed using the protein preparation wizard (PPW) as the downloaded protein cannot be used directly for computational calculations 25 . In the initial step, pre-processing of the protein structure was done by assigning bond orders, adding hydrogen, creating disulfide bonds, and filling the missing side chains and loops with the Prime tool. The termini were capped and water molecules beyond 5 Å were removed from the hetero group to rule out the hindrance. The different hetero atoms and unwanted crystalographic water molecules were removed from the crystal structure to refine the chain. The protonation states of ligands and residues was set using the PROPKA tool so that it helps to simulate the exact experimental conditions. Under the OPLS3e force field, the refined protonated structure minimization generated a lower energy protein 26,27 . Ligand preparation. The DrugBank database (http:// www. drugb ank. ca) was used to identify 2115 US FDA-approved drugs which form the ligands in the docking study. The ligand structures were optimized using the LigPrep tool 28 . The ligands were prepared at pH 7.0 ± 2.0 by the Epik module, retaining the specific chiralities, desalting and generating the tautomers. Ultimately, the 2D structures were converted to 3D, which were geometry minimized under the OPLS3e forcefield for producing the lowest energy state 3D conformation of compounds with all the required corrections 29 . The optimized ligands were used in the docking calculations.
Search for the binding site and molecular docking. The receptor grid generation tool specified the docking site for ligands in the protein structure in the Glide module surrounding the co-crystal ligand. In generating the receptor grid, the co-scrystallized ligand was picked from the minimized protein structure to eliminate it so that it would not be a part of the ligand-receptor docking. The receptor grid was generated using the default www.nature.com/scientificreports/ values of Vander Waal's radius scaling factor of 1 Å with a partial charge cut-off of 0.25 Å. Flexible docking was conducted using the Glide module, wherein the prepared ligands were to fit into the receptor grid at Standard Precision (SP) docking. The output of SP docking was put forward in extra precision (XP) to save the single best pose for each molecule 30 . Flexible docking algorithms generated the docking poses with docking scores and interaction with catalytic and other active site residues 31 . The shortlisted ligands that showed a better affinity towards the Akt-1 allosteric site were identified based on the molecular interactions with the ligands, the target protein, and the docking score.

Validation of the docked poses.
An extra precision (XP) Glide docking approach was used to assess the correctness of docking process. The co-crystallized ligand as well as other previously reported allosteric Akt-1 inhibitors were redocked to the binding site was docked with Glide XP docking mode. The bonding contacts identified after redocking and also the associations documented in the literature were utilised to confirm the docking results 32 .
Free ligand binding energy calculations. Prime Molecular Mechanics/Generalized Born Surface Area (MM-GBSA) method was used to calculate the binding free energy of the protein-ligand complexes as an indication of the stability of the ligand-target complex. This attempt refined and rescored the docking results by incorporating molecular mechanics (MM) force fields combined with a generalized Born and surface area continuum (implicit) solvation model. The free binding energy calculation of the shortlisted molecules/hits was executed by incorporating a variable dielectric VSGB 2.0 solvation model,OPLS3e force field and the energy svisualizer in the prime module 33,34 .

MD simulation.
The eight best ligand-protein complexes shortlisted were subjected to MD simulation using Desmond module 35 . The XP-docked ligand-protein complex was subjected to an MD simulation for 100 ns with periodic boundary conditions in the Desmond module. The system builder tool created an orthorhombic box to submerge the protein-ligand complex in TIP3P explicit water model. Every atom in the ligand-protein complex was located at a minimum distance of 10 Å fixed between the complex and the box wall. A fair number of counter ions (sodium and chloride) been added to neutralize the system, and physiological or isosmotic conditions were emulated by adding 150 mM NaCl to the simulation box. The system energy was sminimized to align the protein structure within the simulation boundaries comfortably. The whole system was minimized with 2000 iterations with convergence criteria of 1 kcal/mol/Å using an OPLS3e force field. A predefined equilibration protocol was run before the production run of the simulation. MD simulation was performed under NPT conditions at a temperature of 300 K and 1.013 bar pressure. The temperature and pressure were maintained using a Nose-Hoover-Chain thermostat and Martyna-Tobias-Klein barostat. The energy and structure were recorded and saved in the trajectory every 10 ps, with 1000 frames saved to the trajectory. After the MD simulation, the stability of the ligand-protein complex was determined from the Simulation Interaction Diagram (SID) generated from the simulation trajectory. Other parameters described in the SID like the RMSD, measured as an average change in the displacement of selected atoms of a defined frame to the reference frame and several other parameters were calculated.
Prediction of toxicity, cell line cytotoxicity and biological activity prediction. The SMILES representation of compounds abstracted from the PubChem database (https:// pubch em. ncbi. nlm. nih. gov/) submitted to the Prediction of Toxicity of Chemicals (ProTox-II). This web-based tool relies on molecular similarity, pharmacophores, fragment propensities, and machine-learning predictive models. A freely accessible toxicity estimation software tool (TEST) was used to predict small molecules relative in silico toxicity. Various toxicological endpoints, including hepatoxicity, carcinogenicity, immunotoxicity, mutagenicity, and cytotoxicity, were predicted from this tool 36 . A naive Bayes approach to predict cytotoxicity against the tumour and non-tumour cell lines was made from Cell-Line Cytotoxicity Predictor (CLC-Pred) (http:// www. way2d rug. com/ Cell-line/) webserver at the pharmacological activity (Pa) and pharmacological inactivity (Pi), with the criteria Pa > Pi (suggested to be active) 37 . Prediction of Activity Spectra for Substances Online (PASSOnline), (http:// www. way2d rug. com/ PASSO nline), a web-based server, uses Bayesian analysis to predict biological activity, including pharmacological effects, mechanisms of action based on the structural formula of the chemical 38 .
Ethics declaration. The work does not include human, animal or biologicals.

Results and discussion
Protein structure. PI3K-Akt-mTOR signalling is activated aberrantly in many cancers, including lung cancers. Akt is expressed frequently in lung cancer, resulting in an abnormal increase in phosphorylated level. This has contributed to tumour progression through increased cell growth, migration, invasion, and drug resistance development by the phosphorylation of its downstream substrates, which makes Akt an appealing cancer target 39 . An inhibition of Akt activity in cells induced cell death. Allosteric or ATP-competitive inhibitors are the ways to achieve Akt inhibition, and many molecules are in clinical development. Recent research shows that allosteric Akt inhibitors engage at the PH and kinase domain interface with remarkable pharmacological selectivity, limiting non-catalytic Akt activity like acetate excretion 40 .
The crystal structure coordinate 4EJN, downloaded from the PDB, was used for this computational study. The selection was based solely on parameters like resolution (2.19 Å), source organism (homo sapiens), receptor comprehensiveness, expression system (Spodoptera frugiperda), R-value free (0.276), R-factor (0.237), and had www.nature.com/scientificreports/ a co-crystal ligand (12j/ 0R4, IC 50 = 5 nM). The chain A of crystal structure 4EJN had a sequence length of 446 amino acids. The co-crystallized ligand was identified near the junction between PH and kinase domains, about 10 Å away ATP binding pocket, making several polar and non-polar contacts. The core of 12j made a hydrophobic contact with conserved tryptophan (Trp-80) residue. The ring-nitrogen interacted via a water-mediated hydrogen bond network to the aspartic acid residue of the DFG-motif (Asp-292). The phenyl ring of the ligand interacted directly hydrophobically with the conserved Tyr-272 residue. A hydrophobic interaction was observed with the aromatic core and the Val-270 residue. The contact between the PH and Akt-1 kinase domains rendered the kinase catalytically inactive and closed. The lipid binding site appeared occluded by the catalytic domain. As a result, we assumed that Akt-1 would make an appealing drug target, and we used docking and MD simulations to find allosteric Akt-1 inhibitors.
Ligand docking. Computational studies such as molecular or ligand docking had been performed to identify potent hits against Akt-1. Using computational technologies, in silico drug identification reduces millions of chemicals to a few possible ones against a target quickly and effectively 41 . Computational studies allow the prediction or study of the intermolecular interactions as the significant factors that could significantly impact the affinity of a ligand for a receptor. Hence, a chemical library of 2115 FDA-approved molecules, whose structures were soptimized using LigPrep, was used for this study. After identifying the binding site was defined at default dimensions around the co-crystal ligand on the linkage of the PH domain and the kinase domain interface employing the receptor grid generation tool. After introducing the optimized ligands to the virtual space, SP-docking was performed to shortlist top hits. Based on their docking score, the top hits underwent a flexible docking approach at XP to rank the ligand-protein complexes. Glide SP could determine the binding ability by adding the interaction energies from several compounds contributing to the scoring matrix 30,31 . With Glide, XP scoring seems to be the most precise since it incorporates more extensive assessment and filtering than SP. It either rewards or penalizes hydrogen bonds, hydrophobic contacts, and π-cations. The docking score could provide only a statistically valid and semiquantitative rating of a ligand's ability to bind to a specific receptor geometry. It also help in removing false-positive results via a statistically valid and semiquantitative rating 42 .
Docking the molecules using SP and XP revealed both binding affinity and orientations of ligand-protein associations in space that could impede a protein function. Some molecules were shortlisted to perform further studies. All the shortlisted hits showed a significant interaction with the conserved residues Trp-80, or Tyr-272 and other critical amino acid residues in the kinase hinge region that could contribute to the allosteric inhibition. The molecules with the lowest glide score value were considered the best docked and potent lead molecules. The XP-docking score of the shortlisted ligand-protein complex varied from −12.121 to −6.911 (Table 1). Table 2 shows the docking score, glide score, Glide Van der Waal (G evdw ), Glide energy (G energy ), Glide model (G model ), Glide ECoulomb (G ecoulomb ) and Glide eInternal (G einternal ) of the top-14 ligands, namely vilazodone, indacaterol, pitavastatin, nomegestrol, raltitrexed, novobiocin, ezetimibe, ditazole, nebivolol floxuridine, delorazepam, valganciclovir, dasatinib, and lorazepam.
Upon visual observations of molecular interactions and MD simulations, the 3D structures of top-4 docked ligand-protein complexes and their molecular interactions with amino acids are represented in Fig. 1. The remaining details of ten molecules of MD simulations are included in the supplementary-1 materials (Fig. 1S).
Allosteric inhibitors interact with and engage outside ATP binding pocket, resulting in a closed activation loop conformation. Studies have demonstrated that allosteric inhibitors of Akt bind and interact with the tryptophan residue (Trp-80) of the PH domain 14,43 . Computational docking analyses investigating the binding model by MK-2206 (an allosteric Ak1/2 inhibitor in clinical investigations), on Akt-1 revealed that the compound interacted with the residues Trp-80 and Tyr-272 along with other residues like Asn-53, Gln-59, Leu-78, Val-201, Leu-264, and Val-270 of Akt-1 44 . Generally, ligand-protein complexes sstabilize if more hydrophobic and polar contacts are exhibited in the docking. The similarity exhibited among interacting residues for the coscrystallized and docked ligands, a high negative docking score, implies that they bind similarly and can interfere with and disrupt the target protein function in the same way as observed with the co-scrystallized ligand. Thus, the docked complexes could favour the PH-in conformation and inactive or autoinhibited state of Akt-1, thereby preventing phosphorylation-mediated kinase activation. Our findings, therefore, provide evidence that identified compounds might suppress Akt-1 by functioning as an allosteric inhibitor by maintaining the favoured closed conformation necessary for its activity.

Validation of molecular docking
The docking protocol employed was docking the co-crystallized ligand, and other reported allosteric Akt-1/2 inhibitors (MK-2206, ARQ-02 (Miransertib), TAS-117 and BAY 1125976) to the earlier generated receptor grid. The binding mode of the above mentioned ligands is reported in supplementary-2 file. The molecules were docked at XP-mode using the Glide module to the generated receptor grid. Interpretation of the results demonstrated interactions between amino acid residues. The interactions with Trp-80, Val-270, Val-271, Tyr-272 and Asp-292 appear common among the docked complexes.
Free ligand binding energy calculations. The docking studies experiments generated false-positive findings for binding affinities among a protein and a ligand, which might explain why the ligand and the protein complexes. Poisson-Boltzmann surface area (MM-GBSA), binding free energy estimates of different ligand-protein complexes to represent the associations for distinct protein and ligand are being used to represent binding free energies (ΔG) as molecular mechanics estimates 45 . From the results of Glide docking and investigation of their 2D and 3D binding mode, interactions made with residues in the allosteric site, 14 docked ligand-protein www.nature.com/scientificreports/ MD simulation. The SP and XP docking had many limitations, such as flexibility and complexity with the ligand flexibility, entropic effects, solvation/desolvation and influence of water molecules (and ions) during binding 46 . Hence, MD simulations were performed on the solvated system of the respective docked complexes using Desmond to sanalyze the binding stability as fluctuations in RMSD, followed by the persistence of proteinligand interaction profiling during the 100 ns within the Akt-1 allosteric site mimicking biological conditions. Furthermore, MD simulations seek to simulate atom motions over some time. The RMSD for the protein is shown on the left y-axis, while the ligand RMSD profile matched on the protein structure backbone is given on the right x-axis. RMSD parameter is used to calculate the system's dynamical stability. The ligand, was bound in the binding pocket of a biomolecule, it is thought to impact the protein residue atoms by inducing flexibility. As a result, it represents a full measure of structural variations during MD simulation. The plot of the protein structure backbone counted throughout the MD simulation was matched the original structure to the enumeration of RMSD. A RMSD shift of 1-3 Å is acceptable during MD simulations for small protein complexes attached to ligands. However, a significantly high number implies a major conformational shift of the protein, and the system is not stable. The RMSD graphs from the MD simulations revealed considerable transformations for protein, ligand, and stability in contacts throughout the MD simulation, as seen in Fig. 2.
Valganciclovir was bound to the allosteric site of Akt-1 and demonstrated both hydrophilic and hydrophobic contacts during MD simulation. After a preliminary fluctuation due to the equilibration for 35 ns, the protein structures RMSD varied between 0.7 and 3.4 Å until the simulation culminated ( Fig. 2A). The protein structure oscillations between 2.7 Å indicated a stable protein structure where the complex had not suffered substantial changes in conformation. Similarly, the ligand RMSD varied from 1.6 to 2.9 Å up to the end of the simulation. The ligand structure fluctuations remained between 1.3 Å, demonstrating that the ligand is steadily bound to the kinase allosteric site and has not significantly diffused from the bound position. Figure 3 depicts the ligand-protein interactions and the protein-ligand contacts recorded throughout the MD simulation. Interactions that last more than 30% of the simulation time in the selected trajectory were considered. Various protein-ligand contacts such as H-bonds, Hydrophobic, Ionic, and Water bridges were observed during the simulation. The purine ring in the valganciclovir made π-π interactions with the conserved tryptophan (Trp-80) of Akt-1. This interaction persisted for more than 83% of the simulation time. Dasatinib, a tyrosine kinase inhibitor for treating lymphoblastic or chronic myeloid leukaemia, was bound to the Akt-1 allosteric site, demonstrating several hydrophilic and hydrophobic interactions during the MD simulation. After the initial fluctuation due to the equilibration for 35 ns, the protein structures RMSD varied between 0.9 and 3.2 Å till the end of the simulation (Fig. 2B). The protein structure fluctuations between 2.3 Å indicated a stable protein structure where the complex had not undergone significant conformational changes. Similarly, the ligand structures RMSD varied between 2.0 and 3.2 Å till the end of the simulation. The ligand structure fluctuations remained between 1.2 Å, indicating that the ligand is stably bound to the kinase allosteric site and has not diffused significantly from the bound position. Figure 4 depicts the ligand-protein interactions and the protein-ligand contacts recorded throughout the MD simulation, wherein interactions that last more than 30% of the simulation time were considered. Several molecular interactions as H-bonds, hydrophobic, ionic, and water bridges, were observed as protein-ligand contacts during the simulation. The aromatic benzene ring with the 2-chloro-6-methyl and thiazole functional groups demonstrated the conserved residues Trp-80 and Tyr-272, respectively. Both of these interactions were proven te an essential for allosteric inhibition of molecules targeting the allosteric site of Akt-1. This interaction persisted for 58 and 44% of the simulation time.  Indacaterol is a β2 adrenergic agonist used to treat COPD and moderate to severe asthma. After the initial fluctuation due to the equilibration for 35 ns, the protein structures RMSD varied between 0.7 and 2.4 Å till the end of the simulation (Fig. 2C). The protein structure fluctuations between 1.7 Å indicated a stable protein structure where the complex had not undergone significant conformational changes. Similarly, the ligand structures RMSD varied between 1.8 and 2.7 Å till the end of the simulation. The ligand structure fluctuations remained between 0.9 Å, indicating that the ligand is stably bound to the kinase allosteric site and has not diffused significantly from the bound position. Figure 5 depicts the ligand-protein interactions and the protein-ligand contacts recorded throughout the MD simulation, wherein interactions that last more than 30% of the simulation time were considered. Several molecular interactions as H-bonds, hydrophobic, ionic, and water bridges, were observed as protein-ligand contacts during the simulation. The amino functional group near the hydroxyethyl linker demonstrated two direct hydrogen bonding interactions with residues, Gln-79 and Asp-292, Table 2. Demonstrates the results of molecular docking as glide score, Glide Van der Waal (G evdw ), Glide energy (G energy ), Glide model (G model ), Glide Ecoulomb (Ge coulomb ) and Glide eInternal (G einternal ) of the top-14 molecules.

Molecules
Glide g-score    www.nature.com/scientificreports/ of the simulation time. The carbonyl (ketone) functionality in the quinoline-2-one ring demonstrated a waterbridged hydrogen bonding interaction that persisted for 52% of the interactions. Novobiocin: Novobiocin is an antibiotic compound. After the initial fluctuation due to the equilibration for 35 ns, the protein structures RMSD varied between 0.7 to 2.8 Å till the end of the simulation (Fig. 2D). The protein structure fluctuations between 2.1 Å indicated a stable protein structure where the complex had not undergone significant conformational changes. Similarly, the ligand structures RMSD varied between 2.3 and 3.6 Å till the end of the simulation. The ligand structure fluctuations remained between 1.3 Å, indicating that the ligand is stably bound to the kinase allosteric site and has not diffused significantly from the bound position. Figure 6 depicts the ligand-protein interactions as H-bonds, hydrophobic, ionic, and water bridges were observed as protein-ligand contacts recorded throughout the MD simulation over 30% of the simulation time. The hydroxyl group attached to the benzene ring and 3-methylbut-2-en-1-yl ring demonstrated a direct hydrogen bonding interaction with the Val-271 residue that persisted for 56% of the simulation. The aromatic benzene ring demonstrated a π-π interaction with the residue Tyr-272 that persisted for 64% of the simula- was not for a specific type of cancer and specific target. In this work, we have specified the target Akt-1 for drug repurposing in NSCLC. In summary, docking the FDA-approved drugs to the allosteric site of Akt-1, followed by MD simulations, allowed us to shortlist the top-14 molecules. The literature on these top-14 molecules suggests that using these in the treatment has added benefits in lung cancer. Vilazodone is developed as an antidepressant drug 48 which might reduce anxiety for patients with lung cancer. In addition, recently, it has been shown to inhibit the human colon cancer cells HCT116 49 . Vilazodone was srecognized as an inositol polyphosphate multikinase (IPMK) antagonist in a structure-based virtual screening of licenced drugs. Vilazodone inhibited IPMK and Akt phosphorylation 49 , hence, it can be repurposed in lung cancer. The drug indacaterol is a longacting β 2 -agonists used to treat asthma and COPD 50 . It was recently studied on human lung cancer cells (NSCLC) with EGFR T790M mutation and was found to induce apoptosis in lung cancer cells via downregulation of activated Akt 51 . In this study, we propose that indacaterol inhibits Akt-1, a newer mechanism of action as an anti-cancer agent. Pitavastatin is a lipid-lowering agent with multifaceted pharmacological properties 52 . This drug is currently the centre of attraction for development as an anti-cancer agent for many different human cancers, and has demonstrated anInhibition of prenylation-dependent Ras/Raf/MEK and PI3K/Akt/mTOR signalling in lung cancer cells and human lung tumour-associated endothelial cells 53 . Moreover, arecent clinical study found it beneficial in combination with neoadjuvant therapy 54 . Nomegestrol is developed as a progestin contraceptive and postmenopausal hormone replacement therapy. However, it is also reported to be anti-cancer in human endometrial RL95-2 cancer cells, both in vitro and in vivo xenograft model 55 . Studies suggest that synthetic progestin medroxyprogesterone acetate (MPA), connected to nomegesterol, can offset oestrogen's positive impact on endothelial function. MPA reduced nitric oxide (NO) production by vascular endothelium in human endothelial cells via inhibition of Akt and ERK signalling cascade 56 . Most recent studies also reveal that progesterone receptor membrane component 1 (PGRMC1) is the prospective therapeutic target in lung cancer. Hence nomegestrol can be repurposed 57 . Raltitrexed was developed as a novel thymidylate synthase inhibitor to treat cancer. It was also reported to be active in human lung cancer cells A549 and human colon cancer cells HCT-116 in preclinical studies 58 . Raltitrexed improved the anti-cancer outcomes of lapatinib on human oesophageal squamous carcinoma cells by decreasing the phosphorylation of Akt and Erk 59 . Our present study demonstrated that it also possesses allosteric Akt-1 inhibition and is shypothesized to inhibit Akt phosphorylation. Novobiocin was developed and approved as an antibiotic targeting DNA gyrase to treat severe infections due to Staphylococcus aureus. However, because of safety or effectiveness concerns, the drug was withdrawn as antibiotics in 2009. Novobiocin was found to hinder the interaction between the CH1 region of p300/CBP and HIF-1α C-TAD, which www.nature.com/scientificreports/ www.nature.com/scientificreports/ resulted in down-regulated expression of genes regulated by HIF-1α, especially CA9, resulting in an inhibitory effect on tumour cell proliferation, that eventually decreased mRNA expression of AKT1 and mTOR in A549 and MCF-7 cells 60 . Further, novobiocin has been tried in several cancers and highly explored by medicinal chemists as derivatives and found beneficial in multiple human cancers 61 . Ezetimibe is a lipid-lowering drug currently used in therapy and was found to be inhibiting the growth of gastrointestinal cancer 62 . Ditazole was developed as an analgesic and anti-inflammatory drug similar to NSAIDs with additional antiplatelet properties 63 . Bone is among the most frequent site of metastasis for all malignancies. Bone metastases occur in up to 75% of patients with advanced breast and prostate cancer and up to 40% of patients with advanced lung cancer 64 . Palliative radiation or narcotic analgesic medications are prescribed for bone pain. Hypercalcemia, nerve compression, pathological fracture, spinal cord compression, and orthopaedic surgical complications could arise from these can harm a person's quality of life (QoL) 65 . These properties will have additional benefits in relieving cancerinduced pain/ inflammation. Nebivolol is a cardiovascular drug, β-blocker and recently found inhibition of mitochondrial activity, thereby inhibiting angiogenesis and arresting human tumour growth 66 . According to recent research, prior β-blocker usage is related to prolonged time-to-discontinuation (TTD) and survival rates (OS) in treatment-naive patients who have advanced lung cancer following first-line EGFR-TKIs 67 . Floxuridine is also known as 5-fluorodeoxyuridine, a pyrimidine analogue. This molecule was developed as an antimetabolite/ anti-cancer agent used to treat colorectal cancer 68 . However, its ability to bind Akt-1 allosteric site was not being www.nature.com/scientificreports/ discovered to date. Delorazepam is a metabolite of diazepam and was developed as an antidepressant drug to reduce anxiety also therapeutically used in bipolar disorders 69 . The reduction of anxiety is necessary for patients with cancer. Valganciclovir is a purine metabolite developed as an antiviral agent to treat human cytomegalovirus.
Valganciclovir is a l-valyl ester of ganciclovir, a prodrug whose concentrations after oral administration are low and transient, t-half is 30 min 70,71 . It was recently found to be an anti-cancer agent. Several clinical studies have found it beneficial in chemotherapy in combination, especially for treating glioblastoma 72 . Hence, targeting the drug to the lung as an appropriate pharmaceutical formulation technology could help repurpising and developing valganciclovir as a drug in NSCLC. The drug dasatinib was developed as a tyrosine kinase inhibitor (TKIs) and used to treat different human cancer. Song et al. 73 demonstrated that dasatinib exerted growth inhibitory activity in apoptosis in gefitinib-sensitive EGFR mutant lung cancer cells in vitro via down-regulation of activated survival proteins, namely Akt and STAT3. Recently proven, its effect in lung cancer cells 74 . Here we are adding the novel mechanism of action for dasatinib as an allosteric Akt-1 inhibitor. Lorazepam is an anti-anxiety drug similar to diazepam. This drug is found beneficial in chemotherapy-induced emesis and anxiety 75 . Thus repurposing these drugs could have added advantages in lung cancer. www.nature.com/scientificreports/ The protein-ligand RMSD plot generated after the MD simulation and various other aspects like protein-ligand contacts, ligand-protein contacts, various ligand properties like ligand RMSD radius of gyration (rgyr), molecular surface area (MolSA), solvent accessible surface area (SASA), and polar surface area (PSA). Analysis of these parameters also demonstrated a stable binding for the dasatinib, valganciclovir, indacaterol and novobiocin within the allosteric binding pocket of Akt-1. The floxuridine, delorazepam, ezetimibe and pitavastatin had moderate binding stability, and the other drugs had higher than 3.0 Å fluctuations. Hence, only valganciclovir, dasatinib, indacaterol and novobiocin are further studied for computational efficacy and toxicity predictions.
Prediction of toxicity, cell line cytotoxicity and biological activity prediction. ProTox-II server classifies the chemicals into different classes based on the predicted LD50. ProTox-II classifies chemicals into toxicity classes sper the globally harmonized system of classification of labelling of chemicals (GHS). Class I belongs to fatal if swallowed (LD50 ≤ 5), and Class II is fatal if swallowed with a higher LD50 range (5 < LD50 ≤ 50). Similarly, Class III is toxic if swallowed (50 < LD50 ≤ 300), Class IV is harmful if swallowed (300 < LD50 ≤ 2000), Class V may be harmful if swallowed (2000 < LD50 ≤ 5000), and Class VI is said to be non-toxic (LD50 > 5000). The results in our studies from the ProTox-II server, valganciclovir,were categorized as class V. In contrast, dasatinib, indacaterol and novobiocin belonged to class IV, with LD50 range from 369 to 5000 mg/kg. Indacaterol belonged to class IV and was harmful if swallowed. Further, valganciclovir and dasatinib predicted a high risk of carcinogenicity, and not with indacaterol and novobiocin. Further, dasatinib, indacaterol and novobiocin were predicted to be immunotoxic. The results are represented in Table 4.
CLC-Pred is a predictor tool for cell line cytotoxicity and reduces the time and cost of the experimental in vivo screening, especially for anti-cancer activity 37 . In this tool the, cytotoxicity prediction was done based on Naive Bayes learning algorithms. The structure-cell line toxicity relationship was assessed using particular training sets with a leave-one-out cross-validation procedure 76 . The prediction of cytotoxicity was made in terms of Pa (probable activity) and Pi (probable inactivity), which ranged from 0.000 to 1.000. By default in PASS, the Pa = Pi value is considered a threshold; those with Pa > Pi are proposed to be cytotoxic or active, wherein www.nature.com/scientificreports/ www.nature.com/scientificreports/ Pa > 0.5 is considered highly cytotoxic and highly active, Pa > 0.3 is believed to have intermediate activity, and Pa < 0.3 is deemed to be the lowest activity 37,77,78 . The in silico prediction of cytotoxic activity using CLC-Pred for four compounds, valganciclovir, dasatinib, indacaterol and novobiocin, with Pa > Pi for various lung cancer cell lines, are provided in Table 5. The results indicate that valganciclovir is expected to exhibit more cytotoxic potential against NSCLC cells NCI-H1299 (Pa = 0.254, and Pi = 0.115). However, this molecule was predicted to be cytotoxic against non-tumorigenic lung fibroblasts, i.e. MRC5 cells (Pa = 0.388, Pi = 0.025). Prediction from the PASS online server using the 2D structure based on their mechanisms of action depending on their interaction with targets and general effects such as antineoplastic, antihypertensive, antiepileptic etc. The biological activity spectrum refers to the list of biological activities a chemical substance exhibits due to its interaction with different biological entities 38 . In drug repurposing, re-tasking, or re-profiling, new uses for existing or sauthorized medications approved or clinically unrelated or original medicinal indication are established 79 . The anti-carcinogenic activity was predicted for molecules valganciclovir (Pa = 0.509, Pi = 0.018), indacaterol (Pa = 0.342, Pi = 0.025), and novobiocin (Pa = 0.52, Pi = 0.017). The different predicted results are illustrated in Table 6.

Conclusion
In this computational study, we identified eight best hits that could bind to the Akt-1 allosteric site, capable of treating NSCLC. The top best hits are valganciclovir, dasatinib, indacaterol, novobiocin, ezetimibe, delerozepam, pitavastatin and nebivolol. These compounds could be repurposed as potent Akt-1 allosteric inhibitors capable of binding directly to the site where the co-crystal ligand, in a similar binding mode as standard ligand 12j or 0R4, thus it could stabilize the 'PH-in' form of the Akt-1. Further, these compounds are predicted to be cytotoxic on human lung cancer cells, and valganciclovir was the best anti-cancer agent among the tested compounds. Thus, the shortlisted molecules could represent novel treatment options and form the basis for further experimental validation and step to rational drug design of a new class of allosteric Akt-1 inhibitors. Table 5. Cytotoxicity prediction of the shortlisted compounds on tumour and non-tumour cell lines of the lung using CLC-Pred at Pa > Pi. The data are predicted values and are obtained from the CLC-Pred web-based prediction tool.

Data availability
Most of the data used is mentioned in the manuscript. The data is also supplemented with the manuscript.  Table 6. The predicted biological effect by the four shortlisted compounds using PASSOnline webserver at Pa > Pi. Data presented are predicted values from the web-server-based tool PASSOnline.