TL;DR
This paper presents a novel homology modeling pipeline to generate diverse "DFG-out" conformations for all human kinases, addressing the underrepresentation of these crucial inactive states in the Protein Data Bank. This approach enables more comprehensive in silico screening and binding site analysis for developing targeted kinase inhibitors.
What Problem Does This Paper Address?
Protein kinases are critical regulators of cellular processes, and their dysregulation is implicated in numerous diseases, particularly cancer. Consequently, they are major targets for drug development. Kinases exist in different conformational states, with the "DFG-in" state typically being catalytically active and the "DFG-out" state being inactive. The DFG-out conformation is particularly interesting for drug design because it exposes unique binding site variations that can be exploited by "type II" inhibitors. However, DFG-out structures are significantly underrepresented in the Protein Data Bank (PDB) compared to their DFG-in counterparts. This scarcity hinders structure-based drug design efforts, as it limits the availability of diverse structural models for exploring potential drug targets. Existing homology modeling methods often focus on single template structures or specific flexible elements, failing to capture the full conformational plasticity of the DFG-out state. This paper aims to bridge this gap by developing an efficient method to generate a comprehensive ensemble of DFG-out conformations for the entire human kinome.
Methods
The core of this work is a sophisticated homology modeling pipeline designed to generate diverse DFG-out kinase structures. The process begins with a thorough classification of existing DFG-out structures in the PDB based on the conformations of three key flexible elements: the DFG motif itself, the A-loop, and the αC-helix. Additionally, the P-loop's conformation is also considered.
The pipeline then constructs "chimeric template structures" by combining rigid C-lobes from DFG-in structures with representative N-lobes and A-loops from the classified DFG-out structures. Specifically, they identified:
- Three major A-loop conformations: "closed type 2," "open DFG-out," and "closed A-under-P."
- Two major P-loop conformations: "collapsed" and "stretched."
- Three major αC-helix conformations: "αC-in," "αC-inter," and "αC-out."
By combining these, they created 18 distinct chimeric template structures. These templates were then used to generate homology models for the entire human kinome, assuming that all kinases can adopt these various DFG-out conformations. To validate the stability and relevance of these generated models, molecular dynamics (MD) simulations of 200 ns were performed on selected ABL1 homology models.
Datasets
The primary dataset used in this study is the Protein Data Bank (PDB), specifically focusing on human kinase domains. The authors performed a classification of available DFG-out structures within the PDB to identify representative conformations of the flexible elements. They also utilized a manually curated multiple sequence alignment of nearly 500 kinases, provided by Möbitz, for target-to-template alignment. For testing the utility of their generated models, they selected FDA-approved type II inhibitors for two kinases, ABL1 and KDR, and their corresponding crystal structures.
Key Insights and Ideas
- DFG-out plasticity is key: The inactive DFG-out state isn't a single conformation but a spectrum of possibilities driven by the interplay of the DFG motif, A-loop, P-loop, and αC-helix.
- Underrepresentation is a bottleneck: The scarcity of DFG-out structures in the PDB is a significant hurdle for structure-based drug design, necessitating methods to generate these missing conformations.
- Chimeric templates unlock diversity: By systematically combining different flexible element conformations into "chimeric templates," the authors can generate a much broader range of DFG-out models than previously possible.
- MD simulations confirm stability: The generated homology models, particularly for the A-loop and DFG motif, appear stable over short MD simulation timescales (200 ns), suggesting they represent relevant, albeit not necessarily the most dynamic, states. Larger conformational changes, like those involving the A-loop, are expected to occur on much longer timescales.
- Ensembles are valuable for docking: The generated DFG-out ensembles show promise for docking studies, with successful identification of near-crystal-like poses for known type II inhibitors.
Key Results
- The pipeline successfully generated 18 distinct chimeric template structures representing major DFG-out conformations.
- Homology models were generated for almost all typical human kinases, with Z-scores indicating good to satisfactory quality for most kinase groups.
- MD simulations on ABL1 models showed that while the C-helix can exhibit some movement, the overall structure and the key DFG-out conformations remained stable within the 200 ns simulation time.
- Docking studies using the generated ABL1 and KDR DFG-out ensembles with FDA-approved type II inhibitors yielded encouraging results, with good docking poses found within the top predictions for four out of five cases.
Limitations and Future Work
- Limited MD timescale: The MD simulations were limited to 200 ns, which is insufficient to capture the microsecond to millisecond timescale dynamics of larger conformational changes like those in the A-loop. This means the generated models represent snapshots of potential DFG-out states rather than fully dynamic ensembles.
- Assumption of universal accessibility: The pipeline assumes that all kinases can adopt all 18 combinations of structural features. While this is a reasonable starting point, it's acknowledged that not all combinations might be energetically accessible for every kinase.
- Potential for improvement in template selection: The authors suggest that differentiating template selection based on kinase groups (e.g., tyrosine kinases vs. serine/threonine kinases) could be an improvement, especially as more PDB structures become available.
- Future work: The authors envision their generated ensembles being valuable for in silico screening, binding site analysis, and understanding selectivity-determining features, particularly for those states that are not well-represented in experimental data. Further exploration of more dynamic simulations and kinase-group-specific template selection are potential future directions.
Summary updated