As explained in my previous post [https://openlabnotebooks.org/2714-2/], my goal is to test whether structure-based approaches can guide the design of PROTACs. As a first step, I am evaluating whether protein-protein docking tools can accurately predict the interface between an E3 ligase and its target, a question recently explore by Drummond and Williams.1 This step is necessary to define the relative orientation of the chemical moiety binding the E3 ligase and the chemical moiety binding the target protein. Once we have this information, the second step will be to design PROTACs that are compatible with the relative orientation of these 2 chemical moieties.
The 3 protein-protein docking tools that I am testing are listed in the table below, and all performed among the best at past CAPRI protein docking competitions (http://www.capri-docking.org/)
| Sr. No | Docking Tool | Description | Method/Approach |
| 1 | Rosetta | Global/Local protein-protein docking | Monte Carlo-based algorithm |
| 2 | HADDOCK | High ambiguity driven biomolecular docking | Flexible docking |
| 3 | ICM | Global protein-protein docking | Fast Fourier Transform (FFT) correlation techniques |
I evaluated how accurately the Haddock, Rosetta and ICM docking protocols detailed above could predict the following complex crystal structures: CRBN-BRD4BD1 [6boy, 6bnb], VHL-BRD4BD2 [5t35] & VHL-SMARCA2BD [6hay]. In my hands, HADDOCK is performing best, as summarized in the Figure and table below, and as detailed in Zenodo. (“Cα-RMSD” is the RMSD between the predicted and experimental binding poses of the protein target when the E3 ligase structures are superimposed. “Ligand RMSD” is the RMSD between the predicted and experimental positions of the small molecule inhibitor binding the target when the E3 ligase structures are superimposed.)
Table 1: Comparison of Haddock, Rosetta and ICM docking results on the 3 protein complexes tested here.
In summary, I find that all 3 methods generate a diverse array of predicted structures that includes the experimental one, but none of the methods accurately rank the experimental structure at the top (though Haddock results are more enriched in experimental structures). If the experimental structure is the only acceptable E3-target interface, this is a problem, as using the wrong E3-target complex for subsequent PROTAC design is bound to fail. However, Nowak et al have shown that different PROTACs can lead to difference protein-protein interfaces for the same E3-target pair.2 It is therefore possible that the best scoring poses (even if different from experimental poses) are indeed useful for PROTAC design. Something I will test in the next step.
Reference:
- Drummond et al. In Silico modeling of PROTAC-medicated ternary complexes: Validation and application. (2019) J. Chem. Inf. Model. 59:1634-1644
- Nowak et al. Plasticity in binding confers selectivity in ligand-induced protein degradation. (2018) Nat. Chem. Biol.14:706–714