Modelling Conformational Flexibility of Kinases in Inactive States

I would like to shamelessly advertise my master thesis project which just got published in Proteins. Keep on reading if you are interested in kinases and/or systematic modelling of protein families.

Protein kinases are players in intracellular signalling and popular drug targets of the pharmaceutical industry. Active sites of protein kinases are highly conserved and therefore difficult to target selectively. Identifying selectivity-determining features is fairly difficult and our methodology may help by systematic modelling of kinase conformations in the so called ‘DFG-out’ state. The DFG motif (one-letter code for aspartic acid, phenylalanine, glycine) is highly conserved amongst kinases and its orientation is crucial for the kinase’s catalytic activity (‘DFG-out’ is generally inactive). Other kinase features have distinct conformations as well, namely the activation loop (A-loop) that binds the kinase’s protein substrate, the P-loop which stacks above the nucleotide substrate and the αC-helix which contains an important glutamic acid.

You might ask yourself now, why ‘DFG-out’ and not ‘DFG-in’? The ‘DFG-out’ state is inactive and thus structurally less conserved. People suspected to find more selectivity-determining features than in the structurally conserved active state, but this could not be shown yet. The actual reason is therefore the potential for new intellectual property (IP) as ‘DFG-in’ inhibitors had been the focus of pharmaceutical research previously. This is also reflected in the number of PDB structures; compared to ‘DFG-in’ structures, ‘DFG-out’ structures are largely underrepresented (also, ‘DFG-in’ states are easier to crystallise).

Our approach uses distinct classes of the three flexible kinase features (see Figure 1) and systematically combines them to generate 18 different homology models for each kinase (in the ‘DFG-out’ state). To achieve this, we classified all human kinases in the PDB and selected template structures that were in the ‘DFG-out’ state and also contained a certain flexible feature in a certain class. We then structurally aligned them and stitched together 18 chimeric templates, each representing a different combination of A-loop, P-loop and αC-helix. Using this approach generated model ensembles for over 95% of the human kinome.