Clinically approved inhibitors of the HIV-1 protease function via a competitive mechanism. A particular vulnerability of competitive inhibitors is their sensitivity to increases in substrate concentration, as may occur during virion assembly, budding and processing into a mature infectious viral particle. Advances in chemical synthesis have led to the development of new high-diversity chemical libraries using rapid in-solution syntheses. These libraries have been shown previously to be effective at disrupting protein–protein and protein–nucleic acid interfaces. We have screened 44000 compounds from such a library to identify inhibitors of the HIV-1 protease. One compound was identified that inhibits wild-type protease, as well as a drug-resistant protease with six mutations. Moreover, analysis of this compound suggests an allosteric non-competitive mechanism of inhibition and may represent a starting point for an additional strategy for anti-retroviral therapy.
HIV infection continues to be a worldwide health crisis, with over 33 million infected people worldwide . Despite improvements in anti-retroviral therapeutic development, drug resistance remains a major obstacle to the effective control of infection in HIV-infected patients. Numerous advances and improvements have been made in drugs targeting the viral protease, which is required for maturation of virions into infectious particles . However, common mutations associated with protease–inhibitor drug resistance appear in drug-experienced patients and, with certain subtypes of the virus, in drug-naïve patients as well , often leading to virologic failure and onset of disease progression. Drug resistance complicates the use of therapeutics in the treatment of HIV infection, necessitating an ongoing search for novel therapeutics targeting the viral protease.
The HIV-1 protease is a 22-kDa homodimeric aspartic protease consisting of two 99-residue polypeptide chains that self-assemble to form the enzymatically active dimer. Currently, all U.S.A. FDA (Food and Drug Administration)-approved PIs (protease inhibitors) are in the same mechanistic class, i.e. they are competitive inhibitors that bind to the active site of the protease, preventing the association of the protease with substrates and resulting in disruption of virion maturation . One drawback of competitive inhibitors is that similar active-site mutations can deleteriously affect small molecule binding in the active site, leading to an increased risk of cross-resistance to other competitive inhibitors. An additional potential pitfall is that competitive inhibitors are sensitive to substrate concentrations . An alternative to competitive inhibitors has been the identification of inhibitors that target non-active-site regions of the protease, such as the dimer interface [5,6], flaps [7,8] or other non-substrate active-site regions . Moreover, inhibitors that utilize non-competitive, uncompetitive or mixed-mode mechanisms have also been identified [5–8,10,11]. A potential advantage of a non-competitive mechanism will be the insensitivity to substrate concentrations, which may better maintain a therapeutic threshold in the substrate-rich virion. The improved therapeutic threshold and potential insensitivity to current resistance mutations may mean such inhibitors are more efficacious at inhibiting viral replication and the emergence of drug resistance.
To identify inhibitors that may target other features of the protease structure, to inhibit its function, we screened a library of compounds shown previously to include inhibitors of protein–protein and protein–nucleic acid interactions [12,13]. One compound, compound (1), was found to inhibit wild-type protease from the NL4-3 strain of HIV-1 in the low micromolar range. Moreover, compound (1) also inhibited a MDR (multidrug resistant) protease containing six mutations associated with PI resistance . The kinetics of the wild-type protease demonstrated a mechanism of inhibition consistent with non-competitive inhibition and cross-competitive inhibition studies, with compound (1) and pepstatin A, a competitive inhibitor, implicated a non-active-site binding effect. Taken together, these findings suggest that compound (1) functions as a non-competitive allosteric PI.
MATERIALS AND METHODS
Enzyme activity assays
HIV-1 protease enzymatic activity was assayed as described previously , using the fluorescently labelled anthranilyl protease substrate Abz (aminobenzoyl)-Thr-Ile-Nle-p-nitro-Phe-Gln-Arg-NH2 (H-2992; Bachem) . In brief, bacterially purified HIV-1 protease was mixed with inhibitor compounds in a reaction buffer containing 25 mM Mes, pH 5.6, 200 mM NaCl, 5% (v/v) DMSO, 5% (v/v) glycerol, 0.0002% Triton X-100 and 1 mM dithiothreitol in a pre-warmed 96-well plate. All clones used for protease bacterial expression were generated from the NL4-3 wild-type or a MDR protease containing six mutations (L24I/M46I/F53L/L63P/V77I/V82A), termed 6X, associated with resistance to saquinavir, nelfinavir, ritonavir and TL3 . The enzyme–substrate reaction was started by the addition of fluorescently labelled substrate and the reaction progress was measured by fluorescence intensity using an FLx-800 fluorescence plate reader (BioTek). For IC50 determinations, the final reaction concentrations were 25 nM protease, 30 μM substrate (the approx. Km) and 0.001–600 μM inhibitor.
Boger et al.  have established and reported previously on a collection of chemical libraries consisting of approx. 66000 compounds, which was prepared by using solution-phase technology with liquid–liquid acid–base extraction purification, evaluated for composition and purity, and stored for further assessment [18,19]. Figure 1 shows a representative diagram of chemical scaffolds and substitutions used to generate the library. From the original library, 44000 compounds were evaluated in the present study. The lead compounds (1) and (2) (Figure 2) identified from the screening were synthesized, then evaluated for composition and purity (below) before use [18,19].
Representative group of chemical substituents that are used in the reaction in the context of the compound scaffold
Compounds used in the present study
Compound (1): 1H-NMR (400 MHz, DMSO-d6, 25 °C) d 11.84 (s, 1H), 10.74 (s, 1H), 9.86 (s, 1H), 8.39 (d, J=1.9, 1H), 8.14 (dd, J = 8.3, 1.6 Hz, 2H), 8.08 (s, 1H), 7.92 (d, J = 8.8, 1H), 7.83 (dd, J = 8.9, 1.9, 1H), 7.39 (s, 1H), 3.84 (s, 3H), 1.49 (s, 9H); MS-ESI (electrospray ionization) (m/z) calculated for [C24H22N4O7S2+Cl]− 577.1; found: 577.1.
Compound (2): 1H-NMR (400 MHz, DMSO-d6, 25 °C) d 11.81 (s, 1H), 10.85 (s, 1H), 8.41 (d, J = 1.8 Hz, 1H), 8.08 (dd, J = 15.1, 1.4 Hz, 2H), 8.03 (s, 1H), 7.94 (d, J = 8.6 Hz, 1H), 7.84 (dd, J = 8.7, 1.6 Hz, 1H), 7.69 (m, 2H), 7.55 (m, 1H); MS-ESI (m/z) calculated for [C18H12N4O5S2+H]+ 429.0; found: 429.0.
Michaelis–Menten kinetic measurements
For Michaelis–Menten kinetic measurements, the protease substrate was titrated from 1 to 200 μM. To assay for promiscuous inhibition, 0.001–0.01% Triton X-100 was added to the reaction. To assess inhibitor specificity horseradish peroxidase (Sigma–Aldrich) activity was assayed in 14 mM potassium phosphate, pH 6.0, and 0.5% (v/v) hydrogen peroxide with an enzyme concentration of 5 nM. The reaction was initiated with the addition of the substrate O-phenylenediamine at concentrations from 20 μM to 100 μM in 100 mM sodium phosphate and 50 mM sodium citrate, pH 5.0. Kinetic constants were determined by non-linear regression of initial reaction velocities as a function of inhibitor concentration using Prism 5.0c (GraphPad Software).
IC50 values were fitted with the following equation:
Michaelis–Menten kinetic constants were fitted to the following equation:
Ki constants for compounds (1) and (2) were fitted to the following equation:
Cross-competitive inhibitor measurements
A variation of Yonetani and Theorell  analysis was used to evaluate the binding mode of compound (1) [20,21]. The use of the variation of Yonetani and Theorell analysis, as discussed by Martinez-Irujo et al. , takes into account the binding interactions, on an enzyme, of competitive and non-competitive inhibitors. The cross-competitive inhibitor assessment was accomplished by using various concentrations of Pepstatin A (Roche), a competitive inhibitor, with a fixed concentration of compound (1), a non-competitive inhibitor, while keeping the substrate and protease concentrations constant. The experimental conditions for assessing protease function were identical to those used to determine IC50. Pepstatin A was used at concentrations from 0.6 to 3.0 μM, whereas compound (1) was held constant at 45, 30, 20 or 0 μM. The determination of the interaction term γ, which defines the degree to which binding of one inhibitor influences the binding of the second inhibitor, was determined utilizing Prism 5.0c using the following equation :
Docking studies of compound (1)
The docked conformation of compound (1) with the HIV protease was generated using AutoDock Vina 1.02 . A high-resolution HIV-1 protease structure (PDB code 2HS1) was chosen as the receptor. Two overlapping search spaces were used, each measuring 25×32×40 Å (1 Å=0.1 nm), which together spanned chain A of the structure. The darunavir molecule bound in the active site was preserved. In each docking run, nine conformations were reported and only the most favourable is detailed below. Three-dimensional co-ordinates for the ligand were determined using Corina . Other docking parameters were kept to their default values.
RESULTS AND DISCUSSION
We have screened a library of compounds shown previously to inhibit protein–protein and protein–nucleic acid interactions [12,13,17,24]. A chemically diverse library of 44000 compounds was synthesized using a solution-phase combinatorial synthesis as described previously [13,18,19,25]. To facilitate synthesis and screening, some compounds were synthesized as part of a screened mixture, with some mixtures containing up to ten related, but distinct, compounds. A representative group of compounds from which the lead compounds emerged is shown in Figure 1.
Compounds were screened initially for the ability to inhibit the wild-type HIV-1 protease, obtained from the NL4-3 virus, in a real-time kinetics assay using a fluorigenic substrate at a concentration equal to the Km. Compounds that had significant affects on the baseline fluorescent signal, which was designated as 10% above baseline independently of the substrate peptide, were excluded from further screening. Assay conditions were chosen to reduce the possibility of false positives resulting from promiscuous inhibitors, including minimizing compound aggregate formation by the inclusion of detergent and reducing compound incubation time [26,27]. Compound groups that showed greater than 50% inhibition of the wild-type protease at a compound concentration of 20 μM were then tested against the 6X protease, a MDR protease containing six mutations (L24I/M46I/F53L/L63P/V77I/V82A) associated with resistance to saquinavir, nelfinavir, ritonavir and TL3 .
Compound families showing greater than 50% inhibition against both wild-type and 6X proteases were then deconvoluted and synthesized as individual compounds. These compounds were then tested individually against the wild-type and MDR 6X proteases. Individual compounds again showing greater than 50% inhibition at a concentration of 20 μM were selected for more detailed kinetics analyses. One compound, compound (1) (Figure 2), was found to inhibit the wild-type protease in the low micromolar range (Figure 3). Furthermore, compound (1) also showed low micromolar inhibition of the 6X protease. The half maximal inhibitory concentrations (IC50 values) were determined against the wild-type and 6X proteases and found to be 17 μM against the wild-type protease and 11 μM against the 6X protease (Figure 3). Thus compound (1) is effective in inhibiting both wild-type and an MDR 6X protease at a similar IC50.
IC50 titration of compound (1) against wild-type and the 6X MDR proteases
To address whether compound (1) was a general enzymatic inhibitor we evaluated whether the compound altered horseradish peroxidase function at various concentrations. No measurable effect on the Michaelis–Menten kinetics of the reaction was observed at any of the compound concentrations evaluated, suggesting that compound (1) is not a general enzymatic inhibitor (results not shown). Furthermore, the inhibitory activity of compound (1) on HIV-1 protease was not abrogated by the addition of non-ionic detergents, strengthening further the case that compound (1) is not a promiscuous inhibitor (results not shown) [26,27].
In order to identify the minimal chemical moieties necessary for PI activity, we synthesized a derivative library based on compound (1) and screened each fragment against the wild-type protease independently. Whereas the majority of the derivatives showed significantly reduced inhibition of protease activity, with IC50 values ranging from 20 to greater than 1000 μM, one derivative, compound (2) (Figure 2), showed more potent inhibitory activity. Compound (2) is similar to compound (1), but with the Boc (t-butoxycarbonyl) and methyl groups removed. When the inhibitory activity of compound (2) was compared with compound (1), it showed a slight decrease in both the IC50 and Ki values as determined with the fluorigenic protease substrate assay (Figures 3 and 4). Therefore both compounds were active against the wild-type protease and compound (1) demonstrated activity against the MDR 6X protease mutant.
Michaelis–Menten kinetics of compounds (1) and (2) against wild-type protease
We next determined the effect of compounds (1) and (2) on the Km and Vmax of the wild-type protease with reactions performed within a range of substrate concentrations from 1 to 200 μM, centred around the Km of 30 μM, at several inhibitor concentrations, from 0 to 30 μM. The values for the initial velocities were then fitted to a Michaelis–Menten model using non-linear regression to determine the dose-dependent effects of the compounds on the Km and Vmax for the HIV protease, as shown in Figures 4(A) and 4(B). When we measured protease activity as a function of both substrate concentration and inhibitor concentration, and used non-linear regression to fit the resulting initial velocities to a Michaelis–Menten model, we observed a curvi-linear response in Vmax as a function of increasing concentration of both compounds (Figure 4C). The results are consistent with a non-competitive mechanism of inhibition.
To glean further insights into the underlying molecular process of protease inhibition by compound (1), we utilized a variation of Yonetani and Theorell analysis [20,21] to evaluate the binding mode. As the inhibition by compound (1) is consistent with a non-competitive mechanism, which might predict that the substrate may still bind to the protease active site when compound (1) is bound, pepstatin A was used as a cross-competitive inhibitor for the analysis. This method of inhibitor cross-competitive analysis allows determination of the degree to which the binding of compound (1) to the protease influences the binding of the second inhibitor to the active site [4,20,28]. The choice of pepstatin A was based on the ability to inhibit the protease , the well-established biochemical and structural reports of its binding location in the active site , that it has a competitive inhibition mechanism [31,32] and the reported use of acetyl-pepstatin A for inhibitor cross-competitive studies for non-active-site inhibitors . The graphical findings from a representative cross-competitive study utilizing compound (1) and pepstatin A is shown in Figure 5. In each case, non-parallel lines were obtained, which converged at the x-axis, consistent with the interpretation that compound (1) and pepstatin A may bind independent sites [4,20,21]. The interaction term γ, which defines the degree to which the binding of one inhibitor to the enzyme influences the binding of the second inhibitor, can be determined through interpolation of the x-intercept or can be calculated [4,20,21]. A small γ value (<1) signifies a synergistic interaction between the inhibitors, whereas a large γ value (>1) indicates mutual antagonism and in the case that γ=1 the inhibitors bind to the enzyme in an independent manner. Calculation of γ yielded approximately 1, consistent with compound (1) binding to a protease site independent of that of pepstatin A, binding in the active site. These findings are consistent with compound (1) binding and providing inhibition through a site independent of the active site.
Yonetani and Theorell plot of v/vi against concentration of pepstatin A and compound (1)
Given our findings from the inhibitor cross-competitive study, indicating that compound (1) was not binding in the active site, we investigated whether compound (1) functions as a dimerization inhibitor. A number of compounds have been reported to promote inhibition through disruption of protease dimerization [5,6]. To address whether compound (1) disrupts dimerization we utilized a tethered homodimeric protease, formed by a direct repeat of protease monomers linked by a five-residue amino acid sequence . The IC50 of compound (1) was found to be similar for both the non-covalent wild-type protease dimer and the covalently tethered dimer protease, as shown in Figure 6. As compound (1) was active against the protease-tethered dimer, this implies that dimerization disruption is not required for inhibitory activity.
IC50 titrations of compound (1) against wild-type and tethered dimer proteases
As compound (1) was shown to have a distinct binding location compared with pepstatin A and not to promote dimer interface disruption, possible binding modes were explored using molecular docking. The search was focussed on the outside surface of the protein and a low-energy conformation was discovered that placed compound (1) in a long solvent-exposed cleft, termed the exo site , as shown in Figure 7. The exo site is composed of distinct regions that include the elbow, cantilever and fulcrum components of the protease. Molecular dynamic simulations of protease flap movement relative to the exo site has indicated that the exo site is compressed when the flaps are open and is extended when the flaps are closed . Moreover, the exo site has been shown, via a fragment-based screen, to accommodate small molecules . The predicted binding energy from the compound (1) docking simulation [−7.2 kcal/mol (1 kcal≈4.184 kJ)] corresponds to a Ki of 5.2 μM, very close to the experimentally observed Ki of 6.1 μM. Together with the biochemical findings from the present study, the docked compound (1) conformation supports a plausible allosteric binding mechanism, which is consistent with structural data . It is tempting to speculate that binding of compound (1) to the exo site influences flap dynamics, perhaps by locking the flaps closed and rendering the protease unable to bind substrate. A number of recent reports have implicated novel compounds that disrupt flap movement, thereby altering enzymatic function [36,37].
Docked conformation showing compound (1) bound outside of the HIV-1 protease (2HS1) active site
Currently, all approved PIs are competitive inhibitors, which target the active site. Given the rise in PI-resistant HIVs, new inhibitors with novel inhibitory mechanisms are needed. Non-active-site allosteric inhibitors may avoid the selective pressure associated with active-site inhibitors, which results in drug resistance mutations. The identification of compound (1) from a novel library of diverse compounds was found to inhibit both wild-type and a MDR protease through a non-competitive allosteric mechanism. This compound provides a rationale starting point from which to chemically investigate novel inhibitory mechanisms that may provide another avenue of viral suppression.
Rolf Muller, Jeremiah Savage and Ying Lin screened the combinatorial chemical library for HIV-1 protease inhibitory activity. Ying Lin and Jeremiah Savage produced the protease for the biochemical assays and protease screening. Sukwon Hong, Wei Jin and Landon Whitby synthesized, determined the composition and purity of the library, and deconvoluted the library. Michael Giffin and Max Chang designed and performed all biochemical analyses on the selected PIs. Max Chang performed all the docking studies. John Elder, Dale Boger and Bruce Torbett were involved in the design and interpretation of the results. Michael Giffin, Max Chang and Bruce Torbett were primarily involved in writing the manuscript. Bruce Torbett and Max Chang edited the manuscript.
This work was supported by the National Institutes of Health [grant numbers 5T32AI007354 (to M.J.G.), 5T32NSO412119 (to M.W.C.), Ge83658, GM48870 and AI40882 (to B.E.T. and J.H.E.) CA78045 (to D.L.B.)]; and by the Center for AIDS Research [grant number 3 P30 AI036214-13S1]. This is publication MEM 20132 from The Scripps Research Institute.
These authors contributed equally to this work.