HIV-1 Protease Inhibitor Docking

HIV protease inhibitors (PIs) have been an important and successful model for structure-assisted drug design.  At least ten PIs have been approved by the FDA since 1995.  However because of HIV-1’s high mutation rate, PI therapies often succumb to drug resistance mutations.  Computational methods that predict how a putative inhibitor will interact with various strains of HIV-1 protease will speed the development of new inhibitors that avoid resistance mutation.

HIV-1 protease mutants exhibit structural diversity both outside of the binding site and within it.  Many docking applications are now able to predict small structural changes arising from mutations within the binding site.  However current computational methods are unable to predict how mutations outside the binding site affect the structure of the HIV-1 protease binding site.

We hypothesized that these conformational changes could be predicted by using ensembles of HIV-1 protease starting structures as templates for docking studies.  We predicted binding free energies on a total of 176 complexes between 34 mutants of HIV-1 protease in complex with 10 different inhibitors. All complexes were modeled using RosettaLigand and a library of 171 HIV-1 protease crystal structures. Predicted binding free energies (ΔΔG_pred) were compared to experimentally determined values (ΔΔG_exp) obtained from the Binding Database (www.bindingdb.org). We find that the correlation coefficient improves from R=0.16 to R=0.62 when conformational selection is used. This result supports the hypothesis that the crystal structure ensemble reflects the conformational flexibility of the backbone of the HIV-1 protease fold. Escape mutants distant from the binding site shift the distribution of conformers and thus affect inhibitor binding free energy.