The Role of Select Subtype Polymorphisms on HIV-1 Protease Conformational Sampling and Dynamics

Background: HIV-1 protease is an essential enzyme for HIV maturation.

Results: Select and naturally occurring polymorphisms alter the conformational sampling and backbone dynamics of HIV-1 protease.

Conclusion: These mutations lead to an alternative flap ensemble that we suspect is a curled flap orientation.

Significance: The mechanism of distal mutations on drug resistance is unclear, but altered dynamics and conformational equilibria likely play key roles.

Abstract

HIV-1 protease is an essential enzyme for viral particle maturation and is a target in the fight against HIV-1 infection worldwide. Several natural polymorphisms are also associated with drug resistance. Here, we utilized both pulsed electron double resonance, also called double electron-electron resonance, and NMR ¹⁵N relaxation measurements to characterize equilibrium conformational sampling and backbone dynamics of an HIV-1 protease construct containing four specific natural polymorphisms commonly found in subtypes A, F, and CRF_01 A/E. Results show enhanced backbone dynamics, particularly in the flap region, and the persistence of a novel conformational ensemble that we hypothesize is an alternative flap orientation of a curled open state or an asymmetric configuration when interacting with inhibitors.

Introduction

HIV-1 protease (HIV-1 PR)³ is an enzyme crucial for cleaving Gag and Gag-Pol polyprotein precursors in the process of the generation of mature infectious viral particles. Consequently, it serves as a therapeutic target in successful antiretroviral therapy. The discovery of protease inhibitors (PIs) has helped to significantly prolong the life span of patients by competitively binding with HIV-1 PR to slow down the virus replication process. However, drug resistance develops very rapidly during the PI antiretroviral treatment. “Major” mutations that evolve decrease the affinity of ligand binding (including both PIs and the polyprotein substrates) (1, 2). These changes are followed by “secondary” mutations that often act to restore retroviral fitness (3–8). HIV-1 PR exists as a C₂ homodimer with each subunit consisting of 99 amino acids. There are two highly flexible β-hairpins, termed the flaps that work together as a gate to control the access to the active pocket (Fig. 1). Mutations occurring in the active site region decrease inhibitor binding and are termed “major or primary mutations”; mutations occurring in the periphery of the enzyme are termed “secondary mutations” and can lead to inhibitor cross-resistance and modulate kinetic efficiency.

FIGURE 1.

Ribbon diagram of HIV-1 PR conformations. Wide open (left, MD simulation (72)), semi-open (middle, Protein Data Bank code 1HHP), and closed (right, Protein Data Bank code 4DQB).

HIV-1 is categorized into different groups, subtypes, and circulating recombinant forms (CRF) (8, 9). Subtype B is the most prevalent subtype in North America and Western Europe, whereas most infected people in Southeast Asia, such as Cambodia, Thailand, and Malaysia, carry CRF_01 A/E (10–12). Although subtype B accounts for only 11% of the worldwide infection of HIV-1 (13), current FDA-approved PIs were designed and optimized for targeting subtype B. Naturally occurring polymorphisms have been shown to alter HIV-1 PR conformational sampling ensembles (14) and dynamics (15) and to decrease binding affinity by 2–7-fold (8, 9). When concurrent with drug pressure-induced primary mutations, the presence of these polymorphisms has been shown to have an even more dramatic decrease in binding affinity (16–18). Thus, an understanding of the mechanism of how natural polymorphisms elicit enhanced drug resistance is important for the design of next generation PIs to tailor-fit particular subtypes or circulating recombinant forms.

HIV-1 PR is known to undergo significant conformational changes during the catalytic cycle. The flaps must open to allow substrate entry and likely the release of catalytic product (19, 20). Molecular dynamics (MD) simulations have shown nominally three states in this process as follows: namely closed, semi-open, and wide open conformations (20, 21). MD and x-ray crystallography indicate that the apo-form of HIV-1 PR favors the semi-open conformational ensemble, whereas inhibitor binding induces a closed flap conformation (Fig. 1). NMR investigations demonstrate that the backbone dynamics decrease from the nanosecond to microsecond to millisecond regime upon inhibitor binding (22). In recent years, our laboratory has pioneered the application of site-directed spin labeling (SDSL) with pulsed electron double resonance, also called double electron-electron resonance (DEER) to characterize flap conformational ensembles in HIV-1 PR (14, 23–28). Other groups have also utilized this methodology to investigate the catalytic mechanism of HIV-1 PR (29, 30). In this method, a nitroxide radical is incorporated into the protein sequence via chemical modification of a cysteine side chain by a thiol-reactive spin label (Fig. 2). Then the dipole couplings between the two electrons are interrogated via pulsed EPR spectroscopy, and the resultant DEER distance profile is analyzed in terms of a conformational ensemble for HIV-1 PR. Our previous studies have shown that this method is successful in characterizing ligand-induced shifts in the conformational ensembles (23, 26), that natural polymorphisms and drug pressure-selected mutations alter the fractional occupancy of the conformational ensemble (14, 24, 25, 27), and that there are correlations of fractional occupancy with enzyme kinetics and inhibitor K_i values (25–27). Taken together, these findings indicate that changes in dynamics and conformational sampling likely play important roles in lowering inhibitor effects seen with drug pressure-selected mutations.

FIGURE 2.

A, MTSL label scheme and the ribbon diagram of the symmetric homodimer subtype B HIV-1 protease (Protein Data Bank code 1HHP). The mutation sites incorporated in the PR5 constructs compared with subtype B are shown as red spheres. The active site residues, Asp-25 and Asp-25′, are rendered as green sticks. MTSL spin probes (K55R1 and K55′R1) are incorporated in silico via MMM 2011.2 and rendered as blue sticks. Distance between two MTSL spin probes were measured and used as an indicator for HIV-1 PR flap conformation. B, sequence comparison of subtypes. The indicated amino acid substitutions relative to subtype B LAI (68) that emerge in subtype Bi, PR5a, and PR5i are natural polymorphism mutations. In italics are the locations of amino acid changes that were introduced to stabilize the protease from proteolytic cleavage (56) or substitutions of native cysteine residues to alanine. The latter mutations ensure site-specific labeling at K55C in DEER experiments and prevent nonspecific disulfide linkages. The inactivate substitution D25N is underlined.

Of particular interest to us has been the appearance of a fourth distinct distance population in our DEER distance profiles, which has a distance between spin labels that cannot readily be assigned to the nominal closed, semi-open, and wide open conformations. This “peak” in the distance profile has been most prevalent in constructs containing select secondary mutations (14, 26, 27) and in the circulating recombinant form CRF_01 A/E (25). It has been assigned to a curled flap conformation that is speculated as an alternative “open” flap conformation not readily seen in native subtype B HIV-1PR conformational sampling, which may have an impact on inhibitor binding and drug resistance. An alternative explanation based on MD simulations of others may also be an asymmetric flap curling upon interaction with inhibitors (31). To gain insight regarding the origin of this additional curled flap conformation, a combination of four naturally occurring polymorphisms common in the sequences of subtypes A, F, and C were introduced into subtype B. Shown in Fig. 2 are the locations of these mutations, which cluster in the secondary nonactive site region of the hinge of the flap. A comparison of HIV-1 PR sequences is also provided in Fig. 2.

Fig. 3 summarizes the prevalence of these mutations within various subtypes (Stanford database). I15V is commonly found in subtypes C and F. E35D occurs >80% of the time in subtypes A, F, and CRF_01 A/E. R41K is prevalent in all other subtypes. R57K is usually seen in subtypes A and F. The “back” mutation of S37N was incorporated because our subtype B construct contains Ser-37, which is found in only 7% of sequences deposited into the Stanford Database. Hence, Asn-37 is the amino acid found 73–96% of the time in the other subtypes (our mutated sample is called PR5-I15V/E35D/S37N/R41K/R57K).

FIGURE 3.

Prevalence of selected mutations in HIV-1 PR subtypes. Data were downloaded from Stanford database.

The results within indicate that the fourth distance population, speculated as a curled open conformation, also exists in PR5. Additionally, NMR relaxation measurements show that PR5 and CRF_01 A/E have an increase in their overall backbone dynamics as compared with subtype B. A close look at various x-ray crystal structures in the Protein Data Bank reveals that the mutation E35D, which exists in both CRF_01 A/E and PR5, removes the salt bridge that normally forms between Glu-35 and Arg-57 in subtype B. We hypothesize that the removal of this salt bridge is likely responsible for destabilizing the compact fold of the flap, thus inducing an alternative flap conformation that may be a curled open or asymmetric conformation; concurrently removal of this salt bridge interaction also leads to increased backbone dynamics.

EXPERIMENTAL PROCEDURES

Cloning and Site-directed Mutagenesis

DNA that encodes Escherichia coli codon-optimized subtype B, CRF_01 A/E, and PR5 HIV-1 PR were purchased from DNA 2.0 (Menlo Park, CA). Each construct was cloned into pET-23a vector (Novagen, Madison, WI) under the control of the T7 promoter. Stabilized (Q7K, L33I, and L63I) inactive (D25N) constructs of subtype B (Bi), CRF_01 A/E (AEi), and PR5 (PR5i) and active constructs (Asp-25) of subtype B (Ba) and PR5 (PR5a) with and without incorporated labeling sites (K55C), where italic letters i and a refer to the inactive (D25N) and active (Asp-25) constructs, respectively, were made using the site-directed mutagenesis kit (Stratagene). Additionally, mutation of the natural cysteine residues to alanine residues (C67A and C95A) were generated in each construct via site-directed mutagenesis to ensure site-specific labeling and prevent nonspecific disulfide bond formation. These amino acid changes are highlighted in Fig. 2. Note that this procedure renders all mutations symmetrically applied to both subunits of the homodimer. The fidelity of the HIV-1 PR genes was confirmed by Sanger DNA sequencing (ICBR Genomics Facility, University of Florida). The complete amino acid sequences of the variants utilized in this study are given in the supplemental material.

Protein Expression, Purification, and Spin Labeling for DEER Experiments

Constructs containing the spin-labeling site (K55C) were used in DEER experiments. Protein expression, purification, and spin labeling were carried out as described previously (14, 32) with the following modification; the inclusion body resuspension buffer pH used for anion exchange depends upon the isoelectric point (pI) of a given construct. The isoelectric point buffer pH for B, CRF_01 A/E, and PR5 were adjusted to 9.30, 9.20, and 9.48, respectively. For active (Asp-25) constructs, inhibitors were added immediately after the refolding into formic acid buffer to prevent protease self-cleavage (33). (1-Oxyl-2,2,5,5-tetramethyl-Δ3-pyrroline-3-methyl) methanesulfonate label (MTSL) (purchased from Toronto Research Chemicals, North York, Ontario, Canada) was added in 3–4-fold molar excess to 8 μm HIV-1 PR homodimer in 10 mm Tris-HCl, pH 6.9, for both inactive and active constructs. The reaction is allowed to proceed in the dark for 12 h at room temperature. Excess free spin label is removed by buffer exchange into 2 mm NaOAc, pH 5.0, using the HiPrep 26/10 desalting column.

Sample Preparation, DEER Data Collection, and Data Analysis

Protein samples were made with 100 μm HIV-1 PR homodimer in 20 mm d₃-NaOAc/D₂O, pH 5.0, buffer using desalting column. 30% v/v d₈-glycerol was added to the sample. Inhibitor was presented as 3-fold excess to HIV-1 PR in the final concentration, and the solution was allowed to equilibrate at room temperature for 30–45 min. Samples were then transferred to a 4-mm quartz EPR tube and flash frozen in liquid nitrogen before inserting the tube into the resonator. All pulsed EPR data were collected with a Bruker EleXsys E580 spectrometer equipped with the ER 4118X-MD-5 dielectric ring resonator at 65 K using a four-pulse DEER sequence (34), described in detail previously (24). The DEER dipolar modulation curves were background-corrected, high pass-filtered, and converted to distance distribution profiles via Tikhonov regularization using DeerAnalysis2008 (35, 36). The correct background correction level was determined using a self-consistent analysis procedure, where a series of Gaussian-shaped populations representing the nominal conformations of HIV-1 PR (14, 23) with estimated relative percentage, full width at half-maximum, and most probable distance were summed to reconstruct the distance profile via DeerSim. DeerSim is a MATLAB-based program that our laboratory developed and is available upon request. Using this software, the dipolar evolution curve is regenerated from the summed Gaussian profile for comparison with the experimental background-corrected data and Tikhonov regularization fit (23). The optimal regularization parameter (37) was selected to guarantee conversion accuracy from the dipolar modulation curve to a Tikhonov regularization distance profile as described previously (14, 23, 24). To exclude artifacts as well as to validate the populations, theoretical echo curves were generated by suppressing each population individually or in all possible combinations, and compared with experimental data echo curves (shown in the supplemental Figs. 7–18). Details about DEER data analysis and population validation are provided in the supplemental Figs. 7–18.

Protein Expression and NMR Sample Preparation

DNA encoding E. coli codon-optimized HIV-1 PR amino acid sequence Bi_, AEi, and PR5i lacking the K55C substitution (i.e. Lys-55) were cloned into pET-23a vector (Novagen, Madison, WI) under the control of T7 promoter. The vector was transformed in BL21*(DE3)pLysS E. coli cells (Invitrogen) and grown in modified minimal media with either ¹⁵NH₄Cl (Sigma) or ¹⁵NH₄Cl and d-[¹³C₆]glucose. Overexpression of HIV-1 PR was induced when optical density of the culture was 0.8 (measured as absorbance at 600 nm), by adding isopropyl β-d-thiogalactoside to a final media concentration of 1 mm. Induction was allowed to proceed at 37 °C for 5–6 h. HIV-1 PR was purified from inclusion bodies as described previously (14, 23, 24). The NMR sample contained 0.1–0.15 mm ¹⁵N- or ¹³C/¹⁵N-labeled HIV-1 PR in 2 mm D₃-NaOAc buffer at pH 5.0 with 10% D₂O and 0.1 mm 4,4-dimethyl-4-silapentane-1-sulfonic acid as internal reference.

NMR Spectroscopy and Data Analysis

All spectra were obtained at 293 K. All 600 MHz data were collected on a Bruker Avance spectrometer with a 5-mm TXI cryoprobe (AMRIS Facility, University of Florida). All 700 MHz data were collected on a Bruker Avance system with a 5-mm TCI 700S4 h-C/N-D-05Z cryoprobe (National High Magnetic Field Laboratory, Department of Chemistry and Biochemistry, Florida State University). All 800 MHz data were collected on a Bruker Avance system equipped with a 5-mm TCI cryoprobe (Department of Chemistry and Biochemistry, University of Notre Dame). NMRPipe (38) and Sparky (Goddard and Kneller, Sparky 3, University of California at San Francisco) were used for processing and analysis of all NMR data.

Backbone resonance assignments for PR5 inactive (PR5i) were carried out using three-dimensional HNCACB (39), CBCA(CO)NH (40), HNCA (41), and HN(CO)CA (42) experiments at 600 MHz with a uniformly ¹³C/¹⁵N-labeled sample. Assignment for PR5i has been deposited in the BMRB database at Madison, WI, with accession number 19072. Backbone chemical shift assignments for inactive subtype B (Bi) and inactive CRF_01 A/E (AEi) were determined and reported in a previous publication (43).

Relaxation measurements were performed at two different magnetic fields as follows: Bi (600 and 700 MHz), AEi (600 and 800 MHz), and PR5i (600 and 700 MHz), with ¹⁵N-labeled samples. ¹⁵N longitudinal rate R₁ (with relaxation delay values of 0.016, 0.064, 0.128, 0.256, 0.384, 0.512, 0.64, 0.768, and 0.896 s) values were measured using hsqct1etf3gpsi (R₁) with recycle delays of 3 s. Transverse relaxation rates R₂ were measured (with relaxation delay values of 0.008, 0.0173, 0.0346, 0.0519, 0.0692, 0.0864, 0.104, 0.121 s) with the CPMG pulse train using hsqct2etf3gpsi (R₂) with recycle delays of 2 s. Peak heights were fitted to a single exponential curve (I(t) = B·exp(−t·R_1,2)) to get R₁ and R₂ values using relaxGUI (44–46). Errors of extracted parameters were estimated using 500 Monte Carlo simulations. (¹H)-¹⁵N NOEs were measured in an interleaved manner using the pulse sequence hsqcnoef3gpsi with recycle delay of 5 s. NOE values were obtained by taking the ratio of resonance intensities from experiments performed with and without ¹H presaturation. NOE experiments were recorded twice and averaged to generate resonance intensities and respective errors. Experimental errors for NOE values were evaluated by error propagation using resonance intensity errors. Consistency of measurements performed at different magnetic field strengths was confirmed as proposed by Morin and Gagne (47). Model-free analysis was performed using relaxGUI (44–46).

Differential Scanning Calorimetry (DSC) Experiments and Sample Preparation

Active and inactive constructs lacking the K55C (i.e. Lys-55) substitution were used in DSC experiments. The HIV-1 PR samples were prepared as described previously (14, 23, 24) and gave a final protein concentration of 16-20 μm as dimer in 10 mm sodium acetate, pH 5.0 buffer, in the presence of 2% DMSO total volume. The inhibitors were present at 2:1 final concentration to dimeric protein. All experiments were performed under the same conditions. The protease (dimer) concentrations for all samples were 16 μm, except for Bi samples, which were 20 μm. Scanning rate was 60 °C/h. The experiments were carried out on a Micro-Cal VP-DSC microcalorimeter, and raw data were analyzed using Origin7.0 for DSC (MicroCal Software, Inc., Northampton, MA). The transition midpoint T_m was determined from the peak maximum.

RESULTS

Naturally Occurring Polymorphism Effects on Protein and Protease Inhibitor Complex Stability

The PR5 construct contains four natural occurring polymorphisms, I15V, E35D, R41K, and R57K, which are specific polymorphisms found in subtypes C, F, or CRF_01 A/E along with the back mutation S37N. Additionally, these substitutions are commonly seen as secondary mutations, which are believed to restore enzyme catalytic function after drug-resistant primary mutations develop (48, 49). As shown in Fig. 2A, positions 35, 37, and 41 lie within the elbow/hinge region. For EPR and NMR investigations, we and others (50, 51), often inactivate HIV-1PR through the D25N active site mutation, which limits self-proteolysis. The DSC measurements serve to demonstrate how protein stability and protein complex stability are altered by the PR5 mutations and D25N-inactivating mutation.

DSC is a thermo-analytical technique that measures the rate of heat evolution or absorption of a specimen that is undergoing a programmed temperature change. For HIV-1 PR, the coupled unfolding and dimer stability of the protein or protease-inhibitor complex is reported by the transition midpoint T_m (51, 52). Fig. 4A shows the high sensitivity DSC measurement results of protease structural stabilities for apo subtype B and apo-PR5 (Bi, Ba, PR5i, and PR5a), where i means inactive enzyme with the D25N mutation and a means active Asp-25. As shown in Fig. 4A, the transition midpoint, T_m, for apo-inactive proteins are similar as follows: 62.4 ± 0.3 and 63.2 ± 0.3 °C, as are values for apo-active proteins, 67.1 ± 0.3 and 67.3 ± 0.3 °C, for B and PR5, respectively. These results indicate that incorporation of the five mutations has minimal effect on protein stability. For both inactive proteases, the D25N substitution decreases T_m by roughly 5 °C compared with the active enzyme, hence slightly destabilizing the dimer. This result is consistent with the finding that the D25N substitution has been shown to increase the equilibrium dimer dissociation constant by a factor greater than 100-fold relative to active consensus subtype B PRs (49–52).

FIGURE 4.

Structural stability of Bi, PR5i, Ba, and PR5a HIV-1 PR, as determined by DSC. A, DSC curves for the four apo-HIV-1 PR constructs. Transition temperature T_m values (°C) are given in parentheses. B and C, plots of the change in transition temperature, ΔT_m = T_m (protein + inhibitor) − T_m (free protein) for the inactive and active HIV-1 PR constructs. DRV, darunavir; IDV, indinavir; LPV, lopinavir; NFV, nelfinavir; RTV, ritonavir.

DSC is also a useful method to probe stability changes upon binding of proteins to small molecules, drugs, and other proteins (53–55). When a small molecule or ligand preferentially binds to (the native form of) protein, it can stabilize the protein leading to a T_m value for the protein-ligand complex that is higher than that of the protein in the absence of ligand. Thus, ΔT_m = T_m (protein + inhibitor) − T_m (free protein) can be used to evaluate the effect that inhibitor binding has on complex thermal stability. Values of T_m and ΔT_m for the inactive and active subtype B and PR5 constructs with nine FDA approved inhibitors are given in Table 1. The corresponding ΔT_m values for inhibited and apo-PR are plotted in Fig. 4, B and C. The results show that for both the active and inactive constructs with all inhibitors except amprenavir (APV), ΔT_m is less for PR5 than for subtype B, indicating a slightly weaker stabilization of the PR5-inhibitor complexes. This is not surprising given that inhibitor designs were optimized against subtype B, and binding affinities to non-B subtypes, which have been investigated via isothermal titration calorimetry, are usually weaker than that of subtype B (8, 9). A second general observation is that inhibitor binding to the active constructs reveals a larger degree of protein stabilization, indicated by a larger value of ΔT_m and consistent with a higher binding affinity. ΔT_m values are typically ≤5 °C for inactive constructs (except TPV) but >10 °C for active constructs. These differences are consistent with published results showing that D25N decreases the binding affinity in other HIV-1 PR variants by a factor of >1000-fold (51). Interestingly, TPV has a ΔT_m value above 20 °C for both active and inactive constructs, which is substantially greater than the effects of other inhibitors. This unique response of TPV may be explained by the fact that TPV forms seven direct hydrogen bonds with the active consensus subtype B protease, much more than other inhibitors (57). Perhaps even in the D25N constructs and PR5, TPV maintains strong interactions with the flap region via non-water-mediated bonds and thus has greater adaptability for binding cavity changes caused by mutations (57).

TABLE 1.

Transition temperature (T_m) of inactive and active subtype B and PR5 HIV-1 PR constructs measured by DSC

Inhibitor	Bi		PR5i		Ba		PR5a
	Tm	ΔTm	Tm	ΔTm	Tm	ΔTm	Tm	ΔTm
	°C	°C	°C	°C	°C	°C	°C	°C
None	62.4 ± 0.4		63.22 ± 0.01		67.1 ± 0.2		67.3 ± 0.3
+APV	62.46 ± 0.34	0.1 ± 0.6	63.05 ± 0.30	−0.2 ± 0.4	79.4 ± 0.3	12.2 ± 0.4	79.8 ± 0.10	12.5 ± 0.3
+ATV	63.42 ± 0.32	1.0 ± 0.6	63.64 ± 0.09	0.42 ± 0.09	84.0 ± 0.3	16.9 ± 0.4	81.9 ± 0.4	14.6 ± 0.5
+DRV	63.90 ± 0.32	1.5 ± 0.6	63.8 ± 0.3	0.6 ± 0.3	85.7 ± 0.3	18.6 ± 0.4	84.04 ± 0.09	16.7 ± 0.3
+IDV	62.53 ± 0.14	0.1 ± 0.5	62.9 ± 0.3	−0.5 ± 0.3	80.7 ± 0.3	13.6 ± 0.4	79.5 ± 0.3	12.1 ± 0.4
+LPV	67.52 ± 0.60	5.1 ± 0.8	67.34 ± 0.24	4.12 ± 0.24	87.0 ± 0.3	19.9 ± 0.4	84.7 ± 0.3	17.4 ± 0.4
+NFV	63.84 ± 0.1	1.4 ± 0.5	63.32 ± 0.16	0.10 ± 0.16	82.4 ± 0.3	15.3 ± 0.4	79.9 ± 0.3	12.6 ± 0.4
+RTV	64.34 ± 0.25	1.9 ± 0.6	64.13 ± 0.20	0.91 ± 0.20	82.3 ± 0.3	15.2 ± 0.4	79.4 ± 0.3	12.1 ± 0.4
+SQV	65.16 ± 0.34	2.7 ± 0.6	65.49 ± 0.18	2.27 ± 0.18	82.6 ± 0.3	15.4 ± 0.4	81.8 ± 0.4	14.4 ± 0.5
+TPV	88.38 ± 0.13	25.9 ± 0.5	85.06 ± 0.14	21.84 ± 0.14	94.7 ± 0.3	27.6 ± 0.4	89.5 ± 0.4	22.2 ± 0.5

Overall, the DSC results indicate that the PR5 construct is as stable as subtype B and that the effects of the D25N mutation destabilize both constructs similarly. As expected, based upon published isothermal titration calorimetry results of inhibitors with other subtypes compared with subtype B (8, 9), the stability of the PR5-inhibitor complex is slightly weaker than those with subtype B. Therefore, it is unlikely that the fourth population seen in the DEER data of PR5 (described within) arises from changes in protein stability.

PR5 HIV-1 PR Has an Additional Curled Flap Conformation Compared with Subtype B

The DEER data for both apo-PR5i and Bi HIV-1 PR have a most probable distance near 36 Å corresponding to a predominant semi-open conformation (14, 23, 25). Fig. 5 shows DEER distance profiles and corresponding population analyses for PR5i and Bi HIV-1 PR with select inhibitors; specifically, data are shown for saquinavir (SQV), APV, and atazanavir (ATV). In general, the inhibitor-induced conformational shifts for PR5i are similar to those for Bi, except that for PR5i the distinct presence of a peak ∼27 Å persists. Fig. 5 also shows results from population analyses of the DEER distance profiles (full details of data analysis are given in supplemental material), which proceeds by assuming the distance profile is an ensemble of all conformational states and that this ensemble can be modeled as a linear combination of Gaussian-shaped functions, where each Gaussian-shaped peak (sometimes combined peaks) is taken to correspond to a conformational ensemble of protein/spin label (23–28). For our HIV-1 PR DEER data, typically four final functions, i.e. populations, are necessary to regenerate the experimentally determined distance profile. These four populations are assigned to the curled open, closed, semi-open, and wide open. From previous x-ray structures and molecular dynamics simulation studies (21), we assign distances of 26–30, 33, 36 and 37, and 40–45 Å to the curled open, closed, semi-open, and wide open conformation, respectively. The curled open conformation we hypothesize to be an “open-like” conformation where the flaps curl or tuck open, possibly in an asymmetric manner, having weaker interactions with inhibitor or allowing for inhibitor escape (27). Others have described a curled flap conformation in MD simulation studies (14, 25, 58). The curled open conformation is not statistically detectable in our Bi DEER data, but in addition to the data shown here, a fourth conformation has been observed in other constructs that we have studied by DEER spectroscopy, such as subtype C, CRF_01 A/E, and subtype B with specific mutations (14, 23, 25, 27).

FIGURE 5.

A, distance distribution profiles for inhibitor-bound Bi (red solid line) and PR5i (blue dashed line); B, relative percentage of each conformational population for inhibitor-bound protease Bi (red bar) and PR5i (blue hatched bar); ATV = atazanavir; APV = amprenavir; SQV = saquinavir. Background-corrected DEER dipolar echo curves and full data analyses are given in the supplemental material.

Fig. 6 shows the percentage occupancy of the curled open and closed states for the apo state and with all nine FDA-approved inhibitors, for PR5i and AEi, which possess the polymorphisms E35D, S37N, and R41K in common with PR5. The raw experimental DEER data and analyses have been reported elsewhere (25, 59, 60). The bar graphs in Figs. 5B and 6A, clearly reveal that the curled population (marked by shaded area in Figs. 5A and 6A) includes ∼10–20% of the conformational ensemble for PR5i and AEi, even in the presence of inhibitors. Interestingly, the presence of the curled open conformation is removed by ATV binding to AEi and TPV binding to PR5i (denoted by * in Fig. 6A). It was unexpected, based upon our earlier studies of subtype B (23) and MDR769 (26), that Ca-P2 binding does not completely remove fractional occupancy of this new state.

FIGURE 6.

DEER results showing effects of inhibitor binding on fractional occupancy of the curled open conformation (A) and the closed conformation (B) for PR5 and CRF_01 A/E with nine FDA-approved inhibitors and a nonhydrolyzable form of the substrate analog Ca-P2. The shaded area in A reflects the degree of uncertainty of the population of curled open state in the apo construct. The errors are largely due to the relatively smooth distance profile for the open construct. Asterisks designate inhibitors that remove the curled open population from the ensemble. C, plot of inhibitor effects on Δc, defined as the difference in the closed population in the presence and absence of inhibitor. The solid lines denote the subjective choices of where to define weak (below 20%), moderate (between 20 and 50%), and strong (greater than 50%) inhibitor interactions on conformational shifts.

Previously, we showed that the interaction of substrates and inhibitors with protease containing the D25N mutation can be interrogated by DEER and NMR spectroscopy (23, 25, 26). FDA inhibitors with active HIV-1 PR have association constants of >10⁹ nm⁻¹. The D25N mutation lowers this affinity >1000-fold, making EPR and NMR investigations sensitive to differences in protein-inhibitor interactions as a function of amino acid changes (25). Specifically, we have shown that inhibitors vary in their ability to shift the conformational sampling to a closed state, with stronger protein-inhibitor interactions correlating with slower exchange dynamics having the greatest effect on flap closure for inactive HIV-1 PR (25). In our earlier works, we define the degree of protease inhibitor interaction to be “weak,” “moderate,” or “strong,” depending on the degree to which a ligand changes the relative population of the closed state, a parameter defined as Δc. Where weak implies Δc <20% and “strong” describes cases where Δc >50%, moderate is the percentage in between. Table 2 summarizes the categorizing of inhibitors for PR5 and those previously described for subtype B and CRF_01 A/E. Overall, the inhibitor shifts for PR5 resemble those obtained for subtype B with two differences. Surprising, indinavir appears to interact stronger with PR5 than B or AE, whereas darunavir appears to have slightly diminished interactions, Δc = 48 ± 8% for PR5 compared with subtype B Δc ∼80%. Although darunavir is considered strong for CRF_01 A/E, the inhibitor-induced conformation shift to the closed dates, Δc = 53 ± 5%, is more similar to PR5 than subtype B Δc >80% (25, 27). For PR5, the largest shift was obtained for TPV, with Δc = 71 ± 8%.

TABLE 2.

Summary of induce flap closure ability for inhibitors on different HIV-1 PR constructs based on the DEER percentage change in flap closed conformation

The abbreviations used are as follows: DRV, darunavir; IDV, indinavir; LPV, lopinavir; NFV, nelfinavir; RTV, ritonavir.

HIV-1 PR constructs	DEER data
	Weak	Moderate	Strong
Subtype Ba	IDV, NFV	ATV	APV, TPV, LPV, RTV, DRV, SQV
CRF_01 A/Eb	IDV, NFV	ATV, APV, RTV	TPV, LPV, DRV, SQV
PR5	NFV	IDV, ATV, DRV	APV, TPV, LPD, RTV, SQV

^a Data were taken from Ref. 23.

^b Data were taken from Ref. 25.

To investigate whether the presence of the fourth conformation arises as an artifact of the D25N mutation, DEER data were collected for active (Asp-25) PR5a, in the presence of three inhibitors (ATV, APV, and SQV). The rationale for choosing these inhibitors was based upon their moderate mode of interaction with inactive protease, thus providing modest changes in the DEER data when comparing active/inactive constructs with these inhibitors. Fig. 7 summarizes the results, which clearly show that the hypothesized curled open conformation peak at 27 Å remains, possibly suggesting that this conformation either does not interact with inhibitors or alternatively may be stabilized upon interaction with inhibitors. For the active construct, the distance profile for ATV is now narrowed compared with the inactive construct; this effect likely arises from a tighter interaction of ATV with the protease due to the presence of the aspartic acid residues in the active site. The DEER distance profiles for SQV and APV are similar for the active and inactive PR5 constructs. The bar graphs in Fig. 7B show that except for ATV, within error, there is no change in the relative populations of the four conformational states between active and inactive enzyme for strong binding inhibitors. Additionally, the population of the curled open state does not change within the error of our measurements. Taken together, these findings indicate that the fourth population does not arise because of the D25N mutation in the active site of PR5i, and it remains even in the presence of the “tight” binding inhibitors.

FIGURE 7.

A, distance distribution profiles for inhibitor-bound active (Asp-25) PR5a (red solid line) and inactive (D25N) PR5i (blue dashed line). B, relative percentage of each conformational population for inhibitor-bound active (Asp-25) PR5a (red) and inactive (D25N) PR5i (blue hatched); ATV = atazanavir; APV = amprenavir; SQV = saquinavir.

Mutant HIV-1 PRs Have Increased Backbone Dynamics

Although the above DEER spectroscopy data give insights into how polymorphisms can alter conformational sampling, no insights into site-specific changes of protein dynamics can be gained from these measurements. To understand the dynamic difference between the variants, high resolution NMR spectroscopy was used to measure the transverse relaxation rate (R₂), the longitudinal relaxation rate (R₁), and the (¹H)-¹⁵N heteronuclear NOE for peptide nitrogen of the different protease constructs, Bi, AEi, and PR5i. Relaxation data are shown in supplemental Figs. S1–S6. These relaxation data contain dynamic information on the picosecond to nanosecond time scale for each backbone ¹⁵N-¹H vector. Model-free analysis (61–63) can be used to translate NMR relaxation data into interpretable dynamic parameters such as order parameter (S²), which characterizes the amplitude of the motion. Details of data fitting to obtain order parameters are given in the supplemental material. The theoretical value for S² ranges from 0 to 1, where the higher the S² value the more rigid the residue.

Fig. 8 shows plots of S² values determined for apo-inactive constructs Bi, PR5i, and AEi The results of Bi are similar to those obtained previously for a related subtype B construct called PMPR (19, 21, 31, 59, 64). Consistent with earlier findings, S² values plotted as a function of residue number (Fig. 8A) for all three HIV-1PR constructs investigated here reveal faster dynamics of the hinge and flap regions, which include residues 37–42 and 47–55, respectively, compared with other regions of the protein. Fig. 8B shows color-coded ribbon diagrams for S² analyses of HIV-1 PR backbone dynamics for each construct investigated here. In general, it is found that the trend of flexibility proceeds as Bi < AEi < PR5i; indicating that the polymorphisms of the AE and PR5 constructs induce more flexibility into the protein backbone, compared with subtype B.

FIGURE 8.

A, backbone order parameter S² of Bi (black square), AEi (blue triangle); and PR5i (red dot). B, color-coded HIV-1 PR ribbon protein structure based on order parameter S² value. Missing values are labeled as gray. The polymorphisms compare with Bi are labeled as spheres.

DISCUSSION

Common Polymorphisms May Be Responsible for the Increased Backbone Dynamics and Altered Conformational Sampling

As shown here and in our previous work (14, 25), both PR5 and CRF_01 A/E have a 10–20% fractional occupancy of the hypothesized curled open flap conformation and have increased backbone dynamics compared with subtype B. Amino acid sequence comparisons show that PR5 and CRF_01 A/E share the same polymorphisms E35D, S37N, and R41K, which likely account for the differences in conformational sampling and dynamics. A previous molecular dynamics simulation study of others reported that the single mutation E35D increased the backbone dynamics of HIV-1 protease (65). These combined changes in conformational sampling and increased backbone dynamics may play a role in the slightly lowered inhibitor efficiency observed for CRF_01 A/E (8). These mutations are also found in the sequences of subtype F (66) and subtype A (67).

HIV-1 Protease Variants Have Different Salt Bridge Formation Patterns

To provide molecular level structural evidence to support our hypothesis that the population in DEER distance profiles ∼27 Å is a novel conformer rarely seen in molecular dynamics simulations of subtype B, and to explore how this mutation may structurally influence HIV-1 PR, we scrutinized several x-ray crystal structures, including subtype B (51, 66, 68), C (69, 70), subtype F (66), CRF01_A/E (8), and other mutants with (49, 71) and without the E35D mutation. We found, as shown in Fig. 9A, that in subtype B a salt bridge interaction exists between Glu-35 and Arg-57. However, as observed in the x-ray structure of CRF_01 A/E, mutation of the Glu to an Asp (E35D) results in a shorter side chain that eliminates the salt bridge interaction with Arg-57 in the crystal. Moreover, the hinge loop, where residue 35 is located, is observed to move away from Arg-57, making the Asp-35 and Arg-57 distance too large to form a salt bridge (Fig. 9B). A similar absence of the salt bridge pattern and turned away hinge loop pattern is seen in the recently reported x-ray structure of a clinically isolated multidrug-resistant construct PR20, which contains E35D (49). It is likely that the Glu-35/Arg-57 salt bridge may hold this loop into a more structured conformation in subtype B, where loss of this interaction may result in an increase in backbone dynamics and lead to a more flexible mode of flap motion, resulting in the novel fourth curled open or asymmetric flap conformation.

FIGURE 9.

Comparison of residues 35–57 salt bridge formation pattern for subtype B (Protein Data Bank code 3BVB) (A) and CRF_01 A/E (Protein Data Bank code 3LZS) (B).

Table 3 summarizes a comparative analysis of crystal structure sequences and the relationship between the distance difference between residues 35/57, the presence of the curled open conformation in DEER distance profiles, and the ability of inhibitors to “remove” the fractional occupancy of this fourth conformational state with a shift to the closed state. For subtype C, where x-ray structures show the existence of the salt bridge and where DEER data have a population with distances between 25 and 30 Å, inhibitor binding (almost all inhibitors) was found to remove the fractional occupancy of this fourth state with a shift to the closed state conformation (25). This result is in agreement with the suggestion that mutations in subtype C weaken the salt bridge between Glu-35 and Arg-57 but do not completely remove it (70). However, for CRF_01 A/E, which shows the absence of the salt bridge (8), inhibitor addition, except for ATV, did not remove the population observed at 25–30 Å. There is currently no crystal structure for PR5, but given that the inhibitor binding, except for TPV, does not remove the population of this fourth state and that this population was seen in active PR5 constructs with strong binding inhibitors, we propose the absence of the salt bridge with a turned away hinge loop pattern may be the molecular basis of this previously undefined peak. We are currently performing crystallization trials to test our hypothesis and plan to generate a series of altered salt-bridge constructs to further investigate our ideas.

TABLE 3.

Summary of x-ray structures containing salt bridge interactions between residues 35 and 57 and appearance of curled open conformation in HIV-1 PR variants

Subtypes	35-Position residue	57- Position residue	35–57 O···HN distance	Protein Data Bank code	Salt bridge (Yes/No)	Curled opena (Yes/No)	Inhibitor removal of C/T (Yes/No)
			Å
B	Glu	Arg	2.7	1HHP	Yes	Noa	NAb
	Glu	Arg	2.7	3BVB
	Glu	Arg	3.0	2P3A
C	Glu	Arg	1.9	3U71	Yes but weak	Yesa	Yes
	Glu	Arg	3.0	2R5P
	Glu	Arg	2.8	2R5Q
CRF_01 A/E	Asp	Arg	15	3LZS	No	Yesa	No
	Asp	Arg	15	3D3T
F	Asp	Lys	10	2P3C	No	NA	NA
M36V/I84V	Glu	Arg	2.7	1RV7	Yes	NA	NA
PR20	Glu	Arg	15	3UF3	No	NA	NA
E35D + 14 others	Asp	Arg	15	2FDD	No	NA	NA

^a Data were taken from SDSL DEER analyses reported in previous publications (23, 25, 27).

^b NA means we have not collected DEER spectroscopy data.

Hence, it is possible that the fourth population for K55R1-K55′R1 distance within the 25–30 Å range in DEER distance profiles of nonsubtype B constructs and drug-resistant strains of subtype B correspond to different structural ensembles that are stabilized by various molecular interactions. Detailed molecular dynamics simulations and crystallization screens are underway to further investigate the nature of these altered conformational states. We speculate that the novel conformation that results from the loss of the Glu-35–Arg-57 salt bridge brings about a conformation that has difficulty interacting with current FDA inhibitors, which opens the door for new avenues to target optimized inhibitors for HIV-1 PR.