Modeling conformational flexibility of kinases in inactive states

Abstract Kinase structures in the inactive “DFG‐out” state provide a wealth of druggable binding site variants. The conformational plasticity of this state can be mainly described by different conformations of binding site‐forming elements such as DFG motif, A‐loop, P‐loop, and αC‐helix. Compared to DFG‐in structures, DFG‐out structures are largely underrepresented in the Protein Data Bank (PDB). Thus, structure‐based drug design efforts for DFG‐out inhibitors may benefit from an efficient approach to generate an ensemble of DFG‐out structures. Accordingly, the presented modeling pipeline systematically generates homology models of kinases in several DFG‐out conformations based on a sophisticated creation of template structures that represent the major states of the flexible structural elements. Eighteen template classes were initially selected from all available kinase structures in the PDB and subsequently employed for modeling the entire kinome in different DFG‐out variants by fusing individual structural elements to multiple chimeric template structures. Molecular dynamics simulations revealed that conformational transitions between the different DFG‐out states generally do not occur within trajectories of a few hundred nanoseconds length. This underlines the benefits of the presented homology modeling pipeline to generate relevant conformations of “DFG‐out” kinase structures for subsequent in silico screening or binding site analysis studies.

Receptor and ligand preparation. One PDB structure per ligand and target pair was chosen as a reference (see Table S3), waters were removed from all crystal structures and hydrogens added to all oxygen and nitrogen atoms. All ABL1 complexes (and homology models) were aligned to 2HYY_A and all KDR complexes (and homology models) to 3WZE_A. All receptor structures and ligands were prepared with Open Babel 2.4.1. [2] into pdbqt format. Note that the KDR homology model 9 was not generated by the homology modelling program and that the reference structure of 0LI into ABL1 (PDB 3OXZ) is from mus musculus. Docking calculations. Docking was performed with Smina [3] which is based on AutoDock Vina 1.1.2. [4]. As homology models were generated without the presence of any ligand, the '-
The numbering of psi and xi refers to the glycine-rich motif (IDs 50-55). psi G-motif-1 and psi G-motif+1 are backbone dihedrals just before and after the GxGxPhiG motif, xi G-motif{+1, +2} is a pseudo-torsional angle between the Cα atoms of the four residues following Phi (IDs 55-58), and the Cα-Cα distance is between the Phi residue (ID 54) and HRD+4 (ID 170).

Figure S1: Distribution of DFG-out structures in PDB with respect to varying A-loop, P-Loop, and αC-helix classes.
Coloured are kinases that have at least one DFG-out structure in the respective class, while those without assigned classes, either due to structural incompleteness (i.e. in the case of the Ploop) or due to differing geometrical criteria, are coloured in white. All kinome tree figures were generated via KinMap (www.kinhub.org/kinmap/). [6] (A) Distribution of A-loop classes 'closed type 2' (green), 'open DFG-out' (orange) or 'closed A-under-P' (blue). 'Closed type 2' conformations are only detected so far for the TK, CAMK, and CMGC groups. However, a profiling study of type II inhibitors [7] showed that these inhibitors can target more than 200 kinases, suggesting that most kinases can form a 'closed type 2' or 'closed A-under-P' A-loop. Thus, we decided to include both conformational classes as templates of the modelling procedure of the entire kinome. 'Open DFG-out' conformations are relatively even distributed amongst the kinome and only two kinase groups lack an existing PDB structure of this conformation. Hence, a kinome-wide distribution can be also assumed.

(B)
Distribution of P-loop classes 'collapsed' (orange) and 'stretched' (green). The 'stretched' P-loop conformation occurs in every kinase group, whereas the number of 'collapsed' conformations in DFG-out structures are overall lower. Since collapsed P-loop conformations in DFG-in structures occur in every kinase group (data not shown), we assumed this structural class to be potentially also sampled in the entire kinome in the DFG-out state.

(C)
Distribution of αC-helix classes 'aC-out' (orange), 'aC-inter' (blue), and 'aC-in' (green). The classification was obtained by employing the Brooijmans αC-helix classifier. [8] All three conformations are present in most kinase groups. Hence, a kinome-wide distribution of all three αC-helix classes can be assumed and all three classes were included as templates of the modelling procedure.   The P-loop conformations can be mainly differentiated into 'stretched' (green) and 'collapsed' (blue). The analysis of the derived feature importance of a trained random forest classifier revealed the five plotted features in the figure to be of high potential value for P-loop classification. The feature d3, describing Cα-Cα distances, was subsequently removed due to the high correlation with feature d2. The initial 40 structural features of the P-loop residues included: phi and psi backbone dihedrals of 13 P-loop residues (IDs 47-59), 10 pseudo-torsional angles of the same 13 residues and 4 Cα-Cα distances between either the first Gly of the GxGxPhiG motif (ID 50) or the Phi (Tyr/Phe) residue of the same motif (ID 54) and either the HRD+4 residue (ID 170) or the third residue following the hinge's end (ID 127).