Background:  Mutations in HIV-1 drug targets lead to resistance and consequent therapeutic failure.  Rules-based interpretations may not accurately predict the effects of multiple mutations, while phenotypic assays are time consuming and costly.  To improve resistance prediction, we developed a structural phenotyping procedure that models HIV-1 protease inhibitor complexes and rapidly evaluates binding affinities of protease inhibitors (PI) to mutant variants.

Methods: Models of wild type Pr-PI complexes were produced by energy optimization of published structures.  Amino acid substitutions were introduced and the models were refined by energy minimization.  Calculated changes in the PI binding energy (DE_bind) of mutant vs wild type complexes were correlated with published estimates of the changes in E_bind or the ratios of phenotypic IC₅₀s. Complexes of Pr variants from clinical isolates were modeled with 5 PIs (saquinavir, indinavir, ritonavir, nelfinavir, amprenavir) andE_bind cutoffs corresponding to a 5x increase in IC₅₀ (Virologic PhenoSense) used to define a structural phenotype of susceptible, resistant or equivocal.

Results: Significant correlations (r² = 0.7 - 0.8, p < 0.01) between calculated binding energies and empiric estimates obtained from published K_is were found for saquinavir, indinavir, ritonavir and amprenavir.  Resistance-associated mutations markedly increased (³ 1.5 kcal/mol) E_bind. For 48 Pr variants, calculated E_bind for 4/5 PIs showed significant correlations (r² = 0.7-0.8, p < 0.001) with fold changes in IC₅₀s (Virco Antivirogram).  For a second group of 46 variants, E_bind for all 5 PIs showed good correlations (r² = 0.75-0.85, p < 0.001) with fold changes in IC₅₀s (Virologic PhenoSense).  The overall concordance of structurally predicted resistance with phenotype exceeded 80% for both data sets.

Conclusions: A novel computational method for structure-based phenotyping of HIV-1 protease was developed that correlates the binding energies of 5 PIs to HIV-1 Pr and changes in inhibitor IC₅₀s as measured by phenotyping. This method rapidly predicts drug resistance of clinical HIV-1 Pr variants, with an accuracy approaching that of cell-based phenotypic assays.