Binding of Clinical Inhibitors to a Model Precursor of a Rationally Selected Multidrug Resistant HIV-1 Protease Is Significantly Weaker Than That to the Released Mature Enzyme

Abstract

We have systematically validated the activity and inhibition of a HIV-1 protease (PR) variant bearing 17 mutations (PRS17), selected to represent high resistance by machine learning on genotype-phenotype data. Three of five mutations in PRS17 correlating with major drug resistance, M46L, G48V and V82S, and five of eleven natural variations, differ from two clinically derived extreme mutants, PR20 and PR22 bearing 19 and 22 mutations, respectively. PRS17, which forms a stable dimer (<10 nM), is ~10- and 2-fold less efficient in processing the Gag polyprotein relative to the wild-type and PR20, respectively, but maintains the same cleavage order. Isolation of a model precursor of PRS17 flanked by the 56 amino acid transframe region (TFP-p6pol) at its N-terminus, which is impossible when expressing an analogous PR20 precursor, allowed systematic comparison of inhibition of TFP-p6pol-PRS17 and mature PRS17. Resistance of PRS17 to 8 protease inhibitors (PIs) relative to PR ranges from 1.5 to 5 orders of magnitude increase in Ki from 0.01 to 8.4 μM. Amprenavir, darunavir, atazanavir and lopinavir, the most effective of the 8 PIs, inhibit precursor autoprocessing at the p6pol/PR site with IC50 ranging from ~7.5 to 60 μM. Thus this process, crucial for stable dimer formation, shows ~200 to 800-fold weaker inhibition than the mature PRS17. TFP/p6pol cleavage, which occurs faster, is inhibited even more weakly by all PIs except darunavir (IC50 of 15 μM); amprenavir shows a 2-fold increase in IC50 (~15 μM), and atazanavir and lopinavir show increased IC50 of >42 μM and >70 μM, respectively.

Structural Studies of a Rationally Selected Multi-Drug Resistant HIV-1 Protease Reveal Synergistic Effect of Distal Mutations on Flap Dynamics

“…The sequence of PR S17 was selected from genotype-phenotype data by a new method for predicting drug resistance using machine learning with a unified encoding of the sequence and 3-D structure [17–18]. This method accurately classified genotype data to predict drug resistance.…”

“…Mean shift clustering with regression analysis identified mutants, such as PR S17 , with high levels of resistance to multiple drugs that were representative of wide classes of drug-resistant proteins [20]. We demonstrated that purified mature PR S17 exhibits extreme resistance to all 8 potent PIs tested [17]. It forms a stable dimer and is ~10- and 2-fold less efficient in processing the Gag polyprotein relative to the wild-type and PR20, respectively [17].…”

Analysis of drug resistance in HIV protease

“…The inhibitors bind to an active site that is under selective pressure to still recognize its biological substrate. Many of the most highly resistant strains demonstrate multi-drug resistance [6, 13, 14, 24]. Therefore, we expect some level cross-prediction and this work quantifies it.…”

“…The existence of cross-resistance is well known and our lab has used similar approaches to identify interesting multi-drug resistant mutants for structural study [6, 13, 14]. …”

“…The validity of this approach was verified by experimental studies [13, 14]. Machine learning was used to rigorously select representative highly resistant PR sequences for biochemical and structural characterization.…”