HIV-1 protease inhibitors from inverse design in the substrate envelope exhibit subnanomolar binding to drug-resistant variants

Abstract

The acquisition of drug-resistant mutations by infectious pathogens remains a pressing health concern, and the development of strategies to combat this threat is a priority. Here we have applied a general strategy, inverse design using the substrate envelope, to develop inhibitors of HIV-1 protease. Structure-based computation was used to design inhibitors predicted to stay within a consensus substrate volume in the binding site. Two rounds of design, synthesis, experimental testing, and structural analysis were carried out, resulting in a total of 51 compounds. Improvements in design methodology led to a roughly 1000-fold affinity enhancement to a wild-type protease for the best binders, from a Ki of 30-50 nM in round one to below 100 pM in round two. Crystal structures of a subset of complexes revealed a binding mode similar to each design that respected the substrate envelope in nearly all cases. All four best binders from round one exhibited broad specificity against a clinically relevant panel of drug-resistant HIV-1 protease variants, losing no more than 6-13-fold affinity relative to wild type. Testing a subset of second-round compounds against the panel of resistant variants revealed three classes of inhibitors: robust binders (maximum affinity loss of 14-16-fold), moderate binders (35-80-fold), and susceptible binders (greater than 100-fold). Although for especially high-affinity inhibitors additional factors may also be important, overall, these results suggest that designing inhibitors using the substrate envelope may be a useful strategy in the development of therapeutics with low susceptibility to resistance.

Figure 1

Illustration of the substrate envelope hypothesis. In the wild-type drug target, the traditional inhibitor (A, top) occupies more of the binding site and makes more contacts than a substrate (A, bottom). In B, the drug target has mutated to expand the active site in a region that only contacts the inhibitor (star). The inhibitor (B, top) loses contacts and consequently binding affinity, while the substrate (B, bottom) loses negligible affinity as it never contacted the mutable residue. If the inhibitor had been designed to only make interactions made by the substrate, this resistance mutation might have little effect on its binding affinity.

Figure 2

Reaction scheme for the synthesis of the first-round protease inhibitor library. Reagents and conditions: (a) EtOH or iPrOH, 80 °C, 2–3 h; (b) aq. Na₂CO₃, CH₂Cl₂, 0 °C to rt, 4–8 h; (c) Et₃N, CH₂Cl₂, 0 °C to rt, 4–8 h; (d) TFA, CH₂Cl₂, rt, 1 h; (e) EDCI, HOBt, DIPEA, DMF–CH₂Cl₂ (1:1), 0 °C to rt, overnight.

Figure 3

Reaction scheme for the synthesis of the second-round protease inhibitor library. Reagents and conditions: (a) EtOH or iPrOH, 80 °C, 2–3 h; (b) aq. Na₂CO₃, CH₂Cl₂, 0 °C to rt, 4–8 h; (c) TFA, CH₂Cl₂, rt, 1 h; (d) EDCI, HOBt, DIPEA, DMF–CH₂Cl₂ (1:1), 0 °C to rt, overnight (e) EDCI, HOBt, H₂O–CH₂Cl₂ (1:1), 0 °C, 24 h.

Figure 4

Crystallographic contacts with mutable residues for pairs of similar designed compounds that exhibited different resistance profiles. Compound 12h (cyan carbons) contacts the M1 mutable residue Gly48, and makes minimal contacts with the residue Ile84, mutated to Val in the M3 variant. Compound 12j (purple carbons) makes close contacts with the M3 mutable residue Ile84, while avoiding Gly48 (A). Compound 30a (cyan carbons) and 28a (purple carbons) differ by one methyl group, in the proximity of Val82, which is mutated to Ala in the M1 variant (B). Compound 28a is sensitive to the M1 mutations, even though it makes less contact with Val82. Both 30a (cyan carbons) and 30d (purple carbons) do not contact Ile84, even though the latter is M3 sensitive (C).

Figure 5

Comparisons between predicted and experimental binding affinities. (A) The round one designed compounds (green), as well as the clinical inhibitor amprenavir (APV) (black), were generated and scored using a substrate envelope inside a protease structure derived from a substrate complex. (B) Round one (green) and round two compounds (blue), in addition to the clinical inhibitors amprenavir (APV) and darunavir (DRV) (black), were designed and scored inside a “maximal” envelope that completely fills the active site of a HIV-1 protease structure derived from a complex with darunavir. Generating and scoring compounds without the substrate envelope constraint and inside a structure bound to a similar scaffold improves the ability to differentiate between tighter and weaker binders. Experimental K_i values were converted to binding energies by assuming that K_i = K_d and that ΔG_bind = +RT lnK_d.

Figure 6

Comparison between predicted and experimentally determined binding modes for nine designed inhibitors. Predicted structures were derived from inverse design calculations without the substrate envelope constraint, and are drawn in atom colors (cyan carbons) while the crystal structure is in purple. Green atoms are fluorine. For clarity, hydrogen atoms have been omitted and only one of the two crystallographically determined orientations for compound 29b is shown. Although the protease has not been shown for clarity, the alignment of predicted and crystallographic structures were prepared by aligning all Cα atoms of the protease and not the inhibitor structures.

Figure 7

Superposition of the substrate envelope (transparent orange) on the crystallographic or predicted structures of selected designed inhibitors. In the crystal structure of 12h, the thiophene ring exceeded the substrate envelope (A), while the predicted structure for 12h within the substrate envelope (B) adopted a different conformation for the R2 substituent that was reminiscent of the substrate-bound structures. The terminal methyl group of the R1 substituent of compound 29b exceeded the envelope in one of the two inhibitor geometries observed in the crystal structure (C). In the alternative experimentally observed conformation for 29b, the R1 substituent adopted a different conformation that remained inside the substrate envelope, similar to the structure originally predicted by computational design (D). Structures that protruded (A, C) are drawn using a space-filling model to highlight regions that exceed the substrate envelope.