Anticancer compound XL765 as PI3K/mTOR dual inhibitor: A structural insight into the inhibitory mechanism using computational approaches

The PI3K-AKT-mTOR pathway is often a commonly disrupted pathway in human cancer and, therefore, it is widely exploited for cancer therapy. The inhibitors for the important proteins of the pathway including PI3K and mTOR have been increasingly designed. The dual inhibitors targeting PI3K and mTOR both have proven to be more effective than those targeting single protein only. An orally-active compound XL765 is well established as PI3K/mTOR dual inhibitor and have shown in vitro and in vivo anticancer activity against a variety of cancer types and is undergoing clinical trials. The present study explored the exact binding pose and the the interactive forces holding XL765 within the active sites of PI3Kγ and mTOR using molecular docking analyses. The XL765 interacting residues of both the proteins were delineated and the degree of participation in binding was estimated by various methods. In the process, among the interacting residues of PI3Kγ, the Lys-890 and the Met-953 were recognized as the key residues involved in XL765 binding. While, in mTOR case, the Trp-2239 was recognized as the key residue playing role in the XL765 binding. In order to explore the better inhibitors, the study also generated combinatorial chemical library by modifying the scaffold considered from XL765. The virtual screening of the generated compound library led to identification of six novel promising compounds proposed as PI3K/mTOR dual inhibitors. Thus, the present work will through light on the drug inhibitory mechanism of XL765 for PI3K and mTOR, and will also assist in designing novel efficacious drug candidates.


Introduction
Cancer is world-wide deadly disease, and in 2012 alone, it is bringing about 14.1 million new cancer occurrences and 8.2 million deaths. It is expected that these figures will rise to whopping 22  isoforms α, β, γ, and δ) and for mTOR is 157 nM [45]. The XL765 shows antitumor activity alone or in combination with temozolomide in variety of diverse xenografts and animal models [46][47][48][49]. This drug is undergoing clinical trials alone and in combination with other drugs for variety of cancer types [44,50]. A few recent studies involving multiple inhibitors along with XL765 includes mTOR consensus docking [51] and PI3K isoform specific docking [52] were performed and conclusions were drawn from consensus docking poses of multiple inhibitors. However, in the present work, the attempts are made to explore the exact binding pose, interacting residues, molecular interactions, and the key interacting residues of PI3Kγ and mTOR using XL765 docking. Further, in order to explore the better inhibitors, study also generated a combinatorial chemical library by modification of scaffold considered from XL765 followed by its virtual screening against PI3Kγ and mTOR.

Data retrieval
The three dimensional coordinates of XL765 was obtained from PubChem compound database (CID: 49867926), while the structures of proteins were retrieved from Protein Data Bank (PDB): PI3Kγ p110 with PDB Id, 3L54 and mTOR with PDB Id, 4JT6 respectively. The kinase domain of mTOR (residues ranging from 1867 to 2436) was considered in the study and used for all analyses. Both of the retrieved structures were co-complex structures with bound ligands (PI3Kγ with bound LXX, mTOR with bound PI-103), and these bound ligands were used as clues for catalytic site grid generation in molecular docking.

Molecular docking and modification of chemical compound
All the molecular docking simulations were performed by Dock v.6.5 [53]. The pre-processing of proteins and ligands, called structure preparation, required as input for docking was performed by Chimera v.1.6.2 [54]. The chemical compounds were modified using MarvinSketch v.18.4, ChemAxon (http://www.chemaxon.com/products/marvin).

Analyses of docked protein-ligand complex
The docked complexes of protein and ligand were visually analyzed by PyMOL v.1.3 [55] and illustrations were prepared. The molecular interactions between the proteins and ligands were analyzed and illustrated by Ligplot+ v.1.4.3 program [56][57]. For a residue, the degree of taking part in binding was evaluated by loss in accessible surface area (ASA). A residue is said to be taking part in ligand binding if it loses more than 10 Å 2 ASA due to binding [58]. All the ASA calculations of the protein-ligand complexes and the unbound proteins were performed by Naccess v.2.1.1 [59]. To check the binding strength of the proteins towards the ligand, the binding energies and dissociation constants were calculated by X-Score v.1.2.11 [60][61].

Drug-likeness and pharmacokinetic predictions
The "pkCSM-pharmacokinetics" online web-server (http://biosig.unimelb.edu.au/pkcsm/) was used for predictions of drug-likeness and pharmacokinetic properties absorption, distribution, metabolism, excretion, and toxicity [62]. This method uses graph based signature for a chemical compound containing all sets of distance patterns between atoms. These signatures of compounds were used to predict regression and classification models for multiple pharmacokinetic properties.

Enrichment evaluation for molecular docking
Enrichment procedure is used for molecular docking evaluation and it measures how active compounds rank versus a background of decoys. Decoys act as negative controls and should not actually bind. Directory of useful decoys (DUD, http://dude.docking.org), Shoichet Laboratory in the Department of Pharmaceutical Chemistry at the University of California, San Francisco (UCSF), USA was used for generating decoys for the proposed inhibitors [63]. The DUD decoys are similar to known ligands physically but dissimilar topologically. DUD uses 2-D similarity fingerprints to minimize the topological similarity between decoys and ligands. The enrichment factor (EF) is defined as the ratio between the percentage of active compounds in the selected upper subset and the percentage in the entire database [64].

EF ¼ HitsðsampleÞ=NðsampleÞ HitsðdatabaseÞ=NðdatabaseÞ
where Hits(sample) is the number of target-specific active compounds picked by docking in sample dataset (at a specific % level of ranked database, let's say 20%); N(sample) is the total number of compounds in sample dataset (upper 20% of the ranked database); Hits(database) is the total number of target-specific active compounds in the database; N(database) is the total number of compounds in the database.

Molecular docking study of XL765 with PI3Kγ
The molecular docking study showed that the XL765 binding pocket was lined by the residues Lys-802, Met-804, Trp-812, Ile-831, Val-882, Ala-885, Thr-886, Lys-890, Met-953, and Ile-963. (Fig 2, Table 1). These 10 residues exerted 28 non-bonding interactions on the drug and stabilized the drug-protein complex ( Table 1). The high absolute values of dock score, binding energy, and dissociation constant showed good quality binding ( Table 2). The residue Lys-890 showed maximum ΔASA (59.19 Å 2 ) due to binding which meant it had great involvement in drug binding (Table 1). Another important residue Met-953 was involved in maximum of eight non-bonding interactions with the drug suggesting its critical role in stabilizing the the drug-protein complex (Table 1). These residues Lys-890 and Met-953 were identified as the key residues required for XL765 binding in the present study.
On comparing the binding of the docked XL765 with that of native ligand LXX, six residues Met-804, Trp-812, Ile-831, Val-882, Met-953, and Ile-963 were found common for both the ligands (Fig 2B and 2C). Of the common residues, Met-953 was showing maximum nonbonding interactions. The Val-882 formed hydrogen bond with the bound ligand LXX, however, in the present work, it was participating in non-bonding interactions and showed adequate ΔASA (14.1 Å 2 ) due to binding. The rest four residues showed their importance in binding for their involvement in non-bonding interactions and ample ΔASA due to binding. From these findings, the relevance of the common residues became obvious and this also showed that XL765 was also engaging the residues common to that of the native ligand and thus inhibiting the protein [44][45].

Molecular docking study of XL765 with mTOR
Molecular docking study of XL765 with mTOR showed that the drug in the catalytic site was found rapped by 11 residues including Ile-2163, Leu-2185, Trp-2239, Val-2240, Asp-2244, Thr-2245, Ala-2248, Arg-2251, Asp-2252, Met-2345, and Ile-2356 (Fig 3, Table 3). The residue Asp-2251 was engaged in hydrogen bonding using N-amino atom of guanidium group to one of the N-atom of quinoxaline moiety of XL765 and all the interacting residues exerted 37 nonbonding interactions making the drug-protein complex stabilized ( Fig 3B, Table 2). The strength of XL765 binding was evident from high absolute values of the dock score, binding energy, and dissociation constant ( Table 2). The residue Trp-2239 was pinpointed as the key interacting residue as it observed maximum ΔASA (56.39 Å 2 ) and was also participating in maximum number of 10 non-bonding interactions. Further, the Trp-2239 was also seeming to be involved in aromatic interaction with the terminal benzene ring of bulkier group attached to quinoxaline moiety of XL765.
On comparing the docked XL765 with the native ligand PI-103, five residues Ile-2163, Leu-2185, Trp-2239, Val-2240, and Ile-2356 were identified as common interacting residues for both the ligands (Fig 3B and 3C). Of these common interacting residues, the residue Trp-2239 was marked as the key residue in XL765 binding. Another common interacting residue, Val-2240 formed a hydrogen bond with the native ligand, however in case of XL765, it was involved in non-bonding interaction only. These findings showed the relevance of the common interacting residues in binding and thus, XL765 engaged the important interacting residues in binding like that of the native ligand and inhibited mTOR in the similar way [44][45].

Combinatorial library generation for XL765 scaffold
Keeping the scaffold intact from XL765 and replacement of R 1 and R 2 with substituents ( Fig 4) generated a library of compounds. The nine substituents were considered for R 1 replacement  from the previous known PI3K/mTOR inhibitors ( Fig 5) [65]. Whereas 5 small substituents including '-O-CH 3 ', '-OH', '-NH 2 ', '-F', and '-Br' were used for R 2 replacement. These R 1 and R 2 replacements generate a library of total 45 compounds used in this study for virtual screening against PI3Kγ and mTOR. The compounds generated were named as sequential numbers from 1 to 45 in the order of systematic substitution of R 1 and R 2 groups (S1 Table).

Virtual screening of generated combinatorial library
The combinatorial library of all the 45 compounds used for virtual screening and their dock scores for PI3Kγ and mTOR is provided as S1 Table. The six compounds were found common among top 20 scoring compounds of PI3Kγ and mTOR (Fig 1B-1G, Table 2). In general, these six compounds have better dock score, binding energy, and dissociation constant than XL765. These compounds were proposed as potential dual PI3K/mTOR inhibitors and further checked for drug-likeness and pharmacokinetic properties prediction and enrichment evaluation analysis for docking accuracy.

Drug-likeness and pharmacokinetic properties
The values for all 5 conditions (rule of 5) for the proposed inhibitors were better than that of XL765 (Table 4). XL765 was having undesired molecular weight (> 500), while all the proposed inhibitors have desired molecular weight less than 500 dalton except for two compounds 38 & 9. The lipophilicity (LogP) value of XL765 has undesired value (>5), while all the proposed inhibitors have desired LogP values within the range of five. While comparing pharmacokinetic properties (absorption, distribution, metabolism, excretion, and toxicity), the proposed inhibitors passed most of the tests and were comparable to XL765 (Table 5). Therefore, this study proposed the six proposed inhibitors as safe drug candidates for treatment in humans.

Enrichment evaluation analysis
For enrichment analysis, 100 decoys were generated for each compound and thus, making the count to 600 for the six proposed inhibitors. These 600 decoys along with six proposed inhibitors were screened using molecular docking against molecular targets PI3Kγ and mTOR. The enrichment analysis of the screening is shown in Fig 6. The complete random selection of proposed inhibitors will yield EF = 1. An EF = 5 means proposed inhibitors were observed five times more in the top 20% of the ranked database than observed in random 20% sampling of the database. When the sample subsetting level is set at 20%, the theoretical maximum that the enrichment factor achieve is 20. In the current study, the EF values at 20% subsetting level were 6.67 and 16.67 for PI3Kγ and mTOR respectively. The enrichment analysis for the target PI3Kγ revealed that the five of the six proposed inhibitors were picked in the top 55% of the screened database of decoys and proposed inhibitors. While for the target mTOR, all the six proposed inhibitors were picked in the top 23% of the screened database. Thus, the enrichment analysis showed good enrichment for mTOR, while modest enrichment for PI3K. However,  the current work explored dual PI3K/mTOR inhibitors and thus sought common compounds in top 20 scoring for both the targets, and thus may compensate for modest docking accuracy of PI3K. Overall, the enrichment analysis suggest that the six proposed inhibitors were selectively picked over the decoys dataset by virtual screening procedure for both the targets. Molecular interactions of dual inhibitor XL765 and human oncoproteins PI3K and mTOR
Summing up, all the six compounds showed similar binding pattern to that of XL765 and the binding scores were also comparable. The two residues Asp-964 and Ser-806 were commonly found as interacting residues for all the six compounds and one residue Asp-964 was consistently appearing as interacting residue in all the six compounds and the native ligand.

Conclusions
The current study explored the binding pose and the molecular interactions of dual inhibitor XL765 with PI3Kγ and mTOR using molecular docking analyses. The binding pose of XL765 with various interacting residues were determined and characterized. Among XL765 interacting residues of PI3Kγ, Lys-890 and Met-953 were pinpointed as the key residues required for binding. Whereas in case of mTOR, the Trp-2239 was pinpointed as the key interacting residue and another residue Asp-2251 contributed a hydrogen-bonding interaction using Namino atom of guanidium group to one of the N-atom of quinoxaline moiety of XL765. The virtual screening of combinatorial library generated by modification of scaffold considered from XL765 led to identification of six novel compounds. The compounds passes most of the tests in drug-likeness and pharmacokinetic properties evaluation, which suggest that the six novel compounds can be used as safe drug candidates for treatment in humans. In addition to the better binding scores, the enrichment analyses also prove the selective and quality binding Table 7. Interacting residues of mTOR for the compounds XL765, 28, 18, 38, 9, 10, 19, and native ligand. Each column has interacting residues of mTOR for a particular compound whose name is mentioned in bold at the top in the first row. The interacting residues common with those of XL765 are shown in bold and italics. 28  18  38  9  10  19  Native   Ile-2163  -Ile-2163  --Ile-2163  Ile-2163  Ile-2163   ----Thr-  to the targets PI3Kγ and mTOR. The detailed and comparative analyses with XL765 indicated these six novel compounds as better dual PI3K/mTOR inhibitors than the starting compound XL765. Thus, the present docking analyses of dual inhibitor XL765 with PI3Kγ and mTOR will provide an excellent model for studying molecular interactions of drug-protein complex where the drug is targeting multiple proteins and will also help in designing the novel and efficacious drugs.

XL765
Supporting information S1 Table. The compounds generated are named as sequential numbers from 1 to 45 in the order of systematic substitution of R1 and R2 groups. The compounds are provided with structure of varying R 1 and R 2 groups, and dock scores for PI3Kγ and mTOR docking.