Phosphorylation-dependent activity-based conformational changes in P21-activated kinase family members and screening of novel ATP competitive inhibitors

P21-activated kinases (PAKs) are serine/threonine protein kinases that are subdivided into two groups on the basis of their domain architecture: group-I (PAK1–3) and group-II (PAK4–6). PAKs are considered as attractive drug targets that play vital role in cell proliferation, survival, motility, angiogenesis and cytoskeletal dynamics. In current study, molecular dynamics simulation-based comparative residual contributions and differential transitions were monitored in both active and inactive states of human PAK homologs for therapeutic intervention. Due to their involvement in cancer, infectious diseases, and neurological disorders, it is inevitable to develop novel therapeutic strategies that specifically target PAKs on the basis of their activity pattern. In order to isolate novel inhibitors that are able to bind at the active sites of PAK1 and PAK4, high throughput structure-based virtual screening was performed. Multiple lead compounds were proposed on the basis of their binding potential and targeting region either phosphorylated (active) or unphosphorylated PAK isoform (inactive). Thus, ATP-competitive inhibitors may prove ideal therapeutic choice against PAK family members. The detailed conformational readjustements occurring in the PAKs upon phosphorylation-dephosphorylation events may serve as starting point for devising novel drug molecules that are able to target on activity basis. Overall, the observations of current study may add valuable contribution in the inventory of novel inhibitors that may serve as attractive lead compounds for targeting PAK family members on the basis of activity-based conformational changes.


Introduction
Phosphorylation is the most prevalent type of post-translational modification that is involved in multiple cellular processes including metabolism, differentiation, growth, motility, membrane transport, muscle contraction and immunity [1,2]. Considering importance of protein kinases in signal transduction pathways, they are considered as one of the largest gene families in eukaryotes contributing about~2% of the genome [1,3,4] PAKs have been involved in a variety of diseases including ovarian, breast, bladder, and other cancers [31], impaired synaptic plasticity, defects in learning, memory and heart defect [32]. Over the past few years, numerous selective inhibitors have been reported for PAK family members. For example, among group-II PAK (PAK4-6) inhibitors that are based on benzimidazole core, group-I PAK selective series based on a pyrido [2,3-d]pyrimidine-7-one core and an allosteric dibenzodiazepine-based PAK1 inhibitor series, only single inhibitor named as PF-3578309 has been selected for initial clinical trials, however, it failed beyond this step [31,33].
In current study, we applied various in silico approaches to evaluate the comparative conformational changes in the active and inactive states of group-I and II PAKs due to phosphorylation. Our findings facilitate in exploring the synergistic ATP binding profiles of PAK family members by evaluating the Lys-Glu residual relationship and monitoring the open and close kinase conformations due to the influence of glycine-loop. Subsequently, structure-based virtual screening of PAK1 and PAK4 was performed to explore conformation-specific inhibitors. These findings will largely help in devising novel therapeutic strategies against PAK family members.

Structural studies
X-ray structure of PAK1 Arg299Lys (PDB ID: 3FXZ; resolution 1.64Å) was utilized to model PAK1 structure through ModellerV9.14. In the absence of a well-defined structure of PAK2, its FASTA sequence was retrieved through UniProtKB/Swiss-Port database (http://www. uniprot.org) and subjected to Basic Local Alignment Search Tool (BLAST) (https://blast.ncbi. nlm.nih.gov/Blast.cgi). PAK1 structure was utilized as a template (100% query coverage and 90.57% identity) for PAK2 structure modeling. ModellerV9.14 modeled PAK2 structure with a calculated RMSD value of 0.25Å between template and target structure. X-ray structure of PAK3 Asp537Ala (PDB ID: 6FD3; resolution 1.52Å) was utilized to model PAK3 structure through ModellerV9.14. An RMSD value of 0.182Å was observed between template and target PAK3 structure. These structures were validated by MolProbity [36] and Verify3D [37]. Win-Coot [38] was used for the geometry optimization and UCSF Chimera 1.11 [39] was employed for phosphorylation. PAK1, PAK2 and PAK3 structures were phosphorylated at Thr423, Thr402 and Thr436 residues, respectively. In contrast, in PAK4, PAK5 and PAK6 structures, activation segments are phosphorylated at Ser474, Ser602 and Ser560 positions, where phosphate groups were removed by UCSF Chimera 1.11 [39], respectively. Structure minimization was performed using GROMACS 5.1.4 [40].

Molecular dynamics simulation assay
In order to measure conformational changes, stability and dynamic behaviour of active and inactive PAKs, Molecular Dynamics (MD) simulation assays were performed through GRO-MACS 5.1.4. GROMOS96 43a1 extended phosphorylated force field was employed for the simulation of all PAKs members [40]. Briefly, SPC216 water model was used in a periodic box to solvate the system, trailed by addition of Na + and Clcounter ions for system neutralization. In order to remove initial steric clashes, energy minimization (steepest descent algorithm for 500 steps) was accomplished by a tolerance of 1000 kJ/molÅ 2 . The energy-minimized systems were equilibrated for 1000 ps under constant temperature and pressure. MD simulation runs were performed under constant pressure (1 atm) and temperature (300 K) for 100 ns time scale using the Berendsen thermostat and barostat. Long-range electrostatic interactions were analysed with a cut off of 1 nm for the direct interaction through fast smooth Particle-Mesh Ewald (PME) summation [41]. Snapshots were gathered for each system throughout MD simulation and PDBs were retrieved at 10 ns time interval to explore the stability profile, timedependent behaviour and residual fluctuations. Periodic box dimensions for group-I and group-II PAKs were in the range of 8.50 x 8.50 x 8.50Å.

Virtual screening
Virtual Screening (VS) is generally described as a series of screening methodologies to scrutinize a set of compounds to be verified for biological activity against the proposed drug target. For VS, minimized 3D structures of PAK1 Tpo423 , PAK1, PAK4 and PAK4 Sep474 were subjected to docking analysis against Chemical library of Korea Chemical Bank in Korea Research Institute of Chemical Technology [10], through AutoDock Vina. AutoDock Vina required PDBQT file format generated by AutoDock tool (ADT) (http://vina.scripps.edu/manual.html). ADT assigns polar hydrogen, united atom Kollman charges, solvation parameters and fragmental volumes to the protein. PDBQT files were generated for PAK1 Tpo423 , PAK1, PAK4 and PAK4 Sep474 respectively. AutoGrid was used for the grid map preparation through a grid box. For PAK1, the grid size was set to 66 ×64×62Å (xyz points) with grid a spacing of 1.0Å and grid center was designated at x, y, and z dimensions: 38 [42,43]. In all cases, protein was kept as rigid, while ligand was flexible. For PAK1, a known inhibitor G-5555 (PDB: 5DEY) [33], for PAK1 Tpo423 , control inhibitor was compound 17 (PDB: 4ZY5) [44], for PAK4, there is no known inhibitor available in RCSB PDB, while for PAK4 Sep474 , KY-04031(4NJD) [10] was utilized as a control. For interaction mapping, Molecular Operating Environment (MOE) [45] tool was used. MOE is very useful tool for analysis and visualization of protein-ligand complex.

Sequence and structural analysis
At sequence level, group-I and group-II PAK family members were highly conserved among individual groups (Fig 1). In group-I, the conserved Gly-loop motif (GQGASG) was replaced by GEGSTG in group-II. ATP binding region AIK in PAK1-3 was modified into AVK in PAK4-6. Similarly, in group-I PAKs, a conserved Thr residue located in the activation or Tloop was converted to Ser residue in case of group-II activation loops. These residues are required for phosphorylation. These differences may play specific role in the binding specificities of PAK family members with other proteins. In order to ascribe these modifications at structural level, group-I and group-II PAK structures were recruited for comparative analysis (Fig 1).

Moleculer dynamics simulation analysis
Another improtant aspect of this study is to explore the comparative structural details leading to active states through phosphorylation at the T-loops of PAK family members. In order to accomplish these tasks, active and inactive PAK structures were subjected to MD simulation runs for 100 ns. Dynamic behavior of individual simulated system was carefully explored to gauge the overall stability and conformational changes by plotting the RMSD (Root mean square deviation), RMSF (Root mean square fluctuation) and distance calculation values.
PDB files were generated for all simulated systems at regular time interval (10, 20, 30, 40, 50, 60, 70, 80, 90 and 100 ns) in order to explore the significant conformational changes occurring in the active states of PAK family members. The main structural differences were observed in the lengths of Gly-loop and activation loops due to inter convertion of β-strands and loop regions. In order to maintain the conformational plasticity in the active states of PAKs, both activation and Gly-loops remained well-ordered.
In PAK1 Tpo423 , phosphorylation induced the conversion of a proximal loop into β-conformation (β8), resulting in the movement of adjacent loop to the N-lobe. Consequently, a salt bridge formation between β5-specific Glu345 and β8-specific Lys404 residue resulted in the narrowing (4.1 to 3.6Å) of N-and C-lobes of PAK1 Tpo423 (Fig 3A and 3B). In case of inactive PAK1, absence of β1-strand facilitated in the intrinsic flexibility of Gly-loop. Similarly, β8 strand were not visible in PAK1, rather activation loop was disordered (Figs 3A, 4A and 4C). In the active state, Gly-loop and activation segment of PAK1 Tpo423 attained a close conformation due to phoshporylation. Helix α-C position was shifted more towards Gly-loop (Figs 3B, 4B and 4D). In PAK2 Tpo402 , an atypcial conversion of loop into extended conformation Targeting of P21-activated kinase (β8-strand) induced more stability through salt bridge formation between Lys383 and Glu324 residues. Gly-loop was integral with helix α-C but in opposite orientation with respect to each other, while the activation loop was intact due to narrowing of N-and C-lobe (Fig 3C and 3D). The upward position of Gly-loop was maintained due to the formation of β1-strand that was missing in the inactive form of PAK2 (Fig 4E-4H).
PAK3 Tpo436 activation pattern was quite similar to PAK1 Tpo423 and PAK2 Tpo402 , due to the influence of β8 and β1 strands. The residual distance was reduced (6.2 to 3.5Å) between Lys417 and Glu358 residues resulting in the formation of a salt bridge that induced conformational stability in the PAK3 Tpo436 structure (Fig 3E and 3F). Overall, the combinatorial movement of α-C and α-A towards Gly-loop and orderness of activation segment helped in attaining the closed conformation (Fig 4I-4L).
In case of PAK4 Sep474 , an independent movement of helix α-C was observed beside a reduction in its overall size. In contrast, helix α-A demonstrated no conformational change. The ion-pair formation between Lys455 and Glu396 resulted in the narrowing of N-and C-terminal lobes from 3.9 to 2Å (Fig 6A and 6B). Another difference was observed in the intrinsic Targeting of P21-activated kinase movement of Gly-loop with respect to helical axis of α-C and activation segment, that achieved orderness due to positioning of β8-strand (Figs 6A, 6B and 7A-7D).
In PAK5 Sep602 and PAK6 Sep560 , conformational stability of activation segment was observed due to phosphorylation. The conserved Glu524 and Glu428 resdiues located at the N-terminal lobes of PAK5 Sep602 and PAK6 Sep560 participated in the salt bridge formation with Lys583 and Lys541 residues of C-terminal lobes and stabilized the closed conformation of these kinases (Fig 6C-6F). Gly-loop further facilitated in the narrowing of C-and N-terminal lobes. The movement of N-terminal helix α-C to the active site was independent of α-A (Fig 7E-7L).

Discussion
P21 activated kinases (PAKs) play crucial role in the junctional signaling through phosphorylating multiple proteins that are involved in the regulation of cell shape and polarity [46]. Other functional implications of PAK signaling have been reported in oncogenesis [47], viral pathogenesis [48]. cardiovascular [49] and neurological disorders [50]. PAKs are classified into group-I (PAK1-3) and group-II (PAK4-6) PAKs that differ in their kinase regulation, intracellular localization and binding partners [51]. Upon unfolding, PAKs undergo autophosphorylation at Thr (group-I) or Ser (group-II) residue of the activation loop [14]. In PAK1, T423E substitution results in the constitutive activity of kinase [52], while S474E substitution in PAK4 has no effect on the activity [14]. These observations necessitate the exploration of Targeting of P21-activated kinase concerted conformational rearrangements in PAK subdomains through computational means.
In this study, comparative structural features involved in the switching of kinase active and inactive states were explored through comparing atomistic MD trajectories of group-I and group-II PAK-specific kinase domains. As described earlier, conformational transitions of PAK N-and C-lobes were mediated by involving similar structural components [21]; however, phosphorylation/dephosphoryaltion paradigm of the activation segement residues of PAKs may induce distinct conformational changes that result in their activation. In group-I PAKs, no conservation was detected in the conformational transitions at secondary structure level (Table D in S1 File). In contrast, group-II PAKs shared more similarity in the regions bearing conformational changes (Table D in S1 File). Evidently, activation loop-specific conserved residues (Val469, Val597 and Val555, respectively) of PAK4 Sep474 , PAK5 Sep602 and PAK6 Sep560 attained more stabilization due to addition of phosphate group (Fig 10). These findings reveal that conformational changes leading to the phosphyrlation-dependent regulation differ among group-I and group-II PAK family members. Such differences may largely help in devising novel therapeutic strategies based on the active or inactive kinase conformations in cell type specific manner. Through understanding the dynamic role of individual residual conformations involving in kinase activity, these enzymes may be targeted more efficiently.
The elaboration of kinase inactive to active state-specific conformational changes at residual level is a challenging task for designing better and more potent inhibitors. Recently, type-II inhibitors have been developed that are able to bind exclusively at DFG (Asp-Phe-Gly)-out conformation as compared to type-I inhibitors that preferentially bind to DFG-in conformation [53]. Quite interestingly, in all PAKs, highly conserved DFGFCAQ motif is located at the N-terminus of activation segment, where F residue of tripeptide DFG motif is known to regulate the phosphoacceptor selectivity for Ser residue. In the active state, F side-chain is pointed outward to the ATP-binding cleft, while D side-chain is resided at the outer region of pocket (DFG-in or active conformation). Comparatively, our results of DFG motif analysis suggest that in the active state, D and G residues of DFG motif attain more closeness as compared to inactive PAKs.
The residual contributions and underlying conformational changes specified in this study (Table D in S1 File) may largely help in the isolation of novel inhibitors that may efficiently target PAK family members based on their active or inactive states. Through extensive VS, we proposed benzamide, phthalazinone and methanone derivatives that may prove to be effective therapeutic options against ATP binding cleft in the active PAK1 without interrupting the  Targeting of P21-activated kinase inactive PAK1. These compounds share common binding attributes as described for type I inhibitor [53]. To date, only single aminopyrazole-based pan-PAK1 inhibitor PF-3578309 has been progressed into clinical trial-I that failed beyond this point [54]. PAK1 binding was observed with pyrrolidinedione, furamide and ethanone derivatives at the region reported for PAKs are now known to be potential regulators of intracellular activity, cytoskeleton remodeling, cell survival, transformation, cell cycle and gene transcription pathways [57]. PAK2 activation by Rac, cdc42 cleavage, caspases or caspase-like proteases correlates with the programmed cell death. Thus, PAK2 is unique among the PAK isoforms due to its involvement in the stimulation of cell survival and cell death events depending on the activation mechanism [58]. Group II PAK signaling pathways have been observed downstream to membrane receptors and multiple potential regulators of intracellular activity. The three family members appear to have distinct and overlapping cellular functions and interact with an array of downstream effectors to elicit their cellular responses. EGF, epidermal growth factor; HGF, hepatocyte growth factor; PP1B, protein phosphatase 1B [21]. https://doi.org/10.1371/journal.pone.0225132.g010 Targeting of P21-activated kinase type I inhibitors [55]. PAK4 Sep474 binding was observed with piperidinecarbonitrile, acetamide and urea derivatives at ATP-binding groove. In contrast, inactive PAK4 binding was prominent with pyrimidinone, pyrazinecarboxamide and acrylamide derivatives [56]. Thus ATPcompetitive inhibitors may prove ideal therapeutic choice for PAK family members. The detailed conformational readjustements occurring in PAKs during phosphorylation-dephosphorylation transition may serve as a starting point for devising novel drug molecules that may target these anzymes. Overall, the observations of current study may add valuable contribution in the inventory of novel inhibitors that may serve as attractive lead compounds for targeting PAK family members on the basis of activity-based conformational changes.