Effects of Hinge-region Natural Polymorphisms on Human Immunodeficiency Virus-Type 1 Protease Structure, Dynamics, and Drug Pressure Evolution*

Multidrug resistance to current Food and Drug Administration-approved HIV-1 protease (PR) inhibitors drives the need to understand the fundamental mechanisms of how drug pressure-selected mutations, which are oftentimes natural polymorphisms, elicit their effect on enzyme function and resistance. Here, the impacts of the hinge-region natural polymorphism at residue 35, glutamate to aspartate (E35D), alone and in conjunction with residue 57, arginine to lysine (R57K), are characterized with the goal of understanding how altered salt bridge interactions between the hinge and flap regions are associated with changes in structure, motional dynamics, conformational sampling, kinetic parameters, and inhibitor affinity. The combined results reveal that the single E35D substitution leads to diminished salt bridge interactions between residues 35 and 57 and gives rise to the stabilization of open-like conformational states with overall increased backbone dynamics. In HIV-1 PR constructs where sites 35 and 57 are both mutated (e.g. E35D and R57K), x-ray structures reveal an altered network of interactions that replace the salt bridge thus stabilizing the structural integrity between the flap and hinge regions. Despite the altered conformational sampling and dynamics when the salt bridge is disrupted, enzyme kinetic parameters and inhibition constants are similar to those obtained for subtype B PR. Results demonstrate that these hinge-region natural polymorphisms, which may arise as drug pressure secondary mutations, alter protein dynamics and the conformational landscape, which are important thermodynamic parameters to consider for development of inhibitors that target for non-subtype B PR.

HIV-1 is the causative agent of acquired immunodeficiency syndrome (AIDS), which has been a worldwide epidemic since the 1980s. Although the success of antiretroviral therapy has led to a decrease in the mortality rate from HIV-1 infection, in 2015 there were still over 2.1 million new infections and 1.1 million deaths worldwide (1)(2)(3). HIV-1 protease (PR) 6 is a major drug target in the battle against HIV-1 infection, where inhibition of HIV-1 PR prevents viral maturation and results in immature noninfectious virions (4 -6). Given the effectiveness of PR antiretroviral treatment, studies on the biochemistry and biophysics of this enzyme have been ongoing for several decades (7)(8)(9)(10)(11). There is now increasing focus on the impact of genomic differences among subtypes and circular recombinant forms (CRFs) in viral spreading (12)(13)(14), where CRFs are the recombinant mosaic of HIV-1 genomes that have been spread to at least three or more persons who are not epidemiologically related (www.hiv.lanl.gov). Because of globalization, different subtypes and CRFs circulate between different geographical regions. The circulation is of great importance because natural polymorphisms carried inside subtypes and CRFs determine the characteristic responses to drug treatment and may accelerate drug resistance development by reducing the time associated with the fitness-restoring processes (15)(16)(17)(18).
To date, much knowledge has been gained regarding the effects of drug pressure-selected mutations on protease inhibitor (PI) susceptibility, i.e. gaining an understanding of the effects that primary mutations have on ligand/protease interactions and how secondary mutations both inside and outside the active site region lead to restoring catalytic efficiency with concomitant multidrug resistance (11, 18 -26). Natural polymorphisms, which are characteristic of the epidemic map of a given subtype/CRF in a specific geographical region (12)(13)(14), are oftentimes similar to these secondary drug pressure-selected mutations. We and others are particularly interested in understanding the roles that natural polymorphisms in non-active site locations have on mechanisms that may lead to drug resistance and resistance-emergent pathways, which include altering HIV-1 PR conformation sampling, modulating PR dynamics, changes in PR stability, and the substrate groove (15,16,21,26,(27)(28)(29)(30)(31)(32)(33)(34)(35)(36)(37)(38)(39)(40)(41).
Access of substrates and inhibitors to the catalytic site of HIV-1 PR is regulated by the ␤-hairpin turns, referred to as the "flaps," which cover the active site pocket. In the absence and presence of substrate/inhibitors, the flaps and the enzyme adopt different configurations that have been characterized by molecular dynamics (MD) simulation (42,43) and x-ray crystallography (44,45). Fig. 1 shows conformations described as "closed," "semi-open," and "wide open" ensembles ( Fig. 1A) (44,45). Our laboratory has pioneered the application of site-directed spin labeling with electron paramagnetic resonance (EPR) spectroscopy, particularly double electron-electron resonance (DEER), to characterize the protein's conformational sampling profile of PR in solution (30,34,46). Analysis of DEER distance profiles from spin labels incorporated into the flaps leads to a description of the fractional occupancy of the conformational sampling scheme, as shown in Fig. 1A. For subtype B unbound enzyme (without substrate/inhibitor), three characteristic conformations have been observed, which are com-monly described as the closed, semi-open, and wide-open states (31,33,47). However, our work has shown that drug pressureselected mutations (30,31,33,34) and natural polymorphisms (30,33,47) not only alter the fractional occupancy of the conformational sampling of unbound-subtype B but can also engender an additional distance that we postulate is a "curled" protein conformation (32,33), where the flaps are proposed to adopt a curling motion that opens up access to the active site (31)(32)(33)47).
The fractional occupancy of this curled conformation was shown to be particularly apparent in CRF01_AE (47) and protease construct PR5 (32), which has five amino acid substitutions relative to subtype B (sequences in Fig. 1B). Both of these constructs include the natural polymorphism at residue 35, glutamate to aspartate (E35D), and both show an increase in the population of the curled state, which is not destabilized upon addition of inhibitors. NMR studies of both of these constructs reveal increased backbone dynamics compared with wild-type subtype B (32). Hence, we suggested that disrupted or altered salt bridge interactions in the hinge regions of these constructs underlie the molecular basis of these altered conformations and dynamics. Hydrogen-deuterium exchange/mass spectrometry (48) also implicated the Glu-35/Arg-57 interaction as important in modulating dynamics in subtype C. This study investigates the impact of the substitution E35D, which is a major "wide-open (MD coordinates)," and "curled (MD coordinates)," with 1-oxyl-2,2,5,5-tetramethyl-⌬3-pyrroline-3-methyl, MTSL, attached to residue 55 using Multiscale Modeling of Macromolecular systems to generate the anticipated distances. Colored segments with residue span in parentheses in the "closed" conformation highlight distinct regions that are referred to throughout the text: flaps (green, residues 43-59), elbow/hinge (red, residues 34 -42), fulcrum (yellow, residues 9 -24), and cantilever (blue, residues 60 -74). B, amino acid sequences of HIV-1 PR variants B, E35D, PR5, CRF01_AE, and subtype A constructs where the catalytic residue (Asp-25) is highlight in red and natural polymorphisms are underlined in blue. All sequences except for A contain the three stabilizing mutations and CYS substitutions (Q7K, L33I, L63I, C67A, and C95A), where alternative sequences for subtype A are shown in a green background. Note for x-ray studies, EPR, and NMR measurements, the substitution D25N was included to render the enzyme inactive. C, locations and identities of natural polymorphisms of each construct are shown as spheres on HIV-1 PR (PDB code 2PK5 (86)).
natural polymorphism that occurs roughly 30% of the time in subtype B drug-naive patients, increases in patients treated with the inhibitor tipranavir (TPV), and occurs with Ͼ90% frequency in subtype A and F drug-naive patients (Stanford University HIV Drug Resistance Database). Additionally, the alterations in protein dynamics and conformational sampling may be exploited in optimizing new inhibitors that more effectively stabilize constructs that harbor these substitutions.

Results and Discussion
E35D Stabilizes Open-like States-The effect of the E35D natural polymorphism on flap conformational sampling was explored by DEER spectroscopy. DEER distance profiles and resultant population analyses for subtype B and the subtype B construct that harbors the single E35D substitution in the absence and presence of the select inhibitors saquinavir (SQV), darunavir (DRV), and a non-hydrolyzable substrate analog Ca-P2 (with sequence H-Arg-Val-Leu-r-Phe-Glu-Ala-Nle-NH 2 ; r ϭ reduced) are given in Fig. 2. These inhibitors were chosen for spectroscopic studies because they were those for which successful crystals were obtained for structural studies. Full data analyses are given in supplemental Figs. S1-S8. Although the conformational sampling of unbound E35D is dominated by the semi-open conformation (52% fractional occupancy, distance near 36 Å), this relative percentage is decreased by 34% relative to subtype B (47), with associated increases in the populations of the other states (Fig. 2C). Specifically, relative to subtype B, re-distribution in the conforma-tional ensemble leads to an increase in the fractional occupancy of the wide-open and closed-like states by 18 and 15%, respectively (Fig. 2C). Compared with the conformational sampling of PR5 and CRF01-AE, E35D possesses less of the curled state with larger fractional occupancy of the wide-open state. Taken together, these comparisons imply that the single amino acid substitution of E35D destabilizes the semi-open conformation with stabilization of the open-like and closed-like states. These results suggest that additional mutations, such as those in PR5 and CRF01_AE, act in concert with E35D to increase fractional occupancy and hence stabilize the curled conformation.
DEER data also show that for E35D, the addition of inhibitors/substrate analogs shifts the predominant conformation to the closed state (distance centered ϳ33 Å), an observation also seen for other HIV-1 PR constructs, including subtype B (50). The binding of inhibitors to stabilize the closed state is consistent with results from kinetic and inhibition assays and is comparable with conformational shifts observed for other nondrug-resistant forms of PR (50). For unbound-subtype B, the DEER distance profiles do not contain a detectable population at a distance Ͻ30 Å in both the presence and absence of inhibitor, i.e. no evidence for flap curling. In contrast, for E35D, a detectable population at 25 Å appears upon addition of SQV, DRV, and Ca-P2 (Fig. 2, C and D). It is noteworthy to mention that the fractional occupancy of the curled distance in unbound E35D (Fig. 2C) is outside the error of the DEER distance mea- surements, indicating the result is statistically valid (51). Interestingly, the breadth of the population for this distance is narrowed upon addition of inhibitor or substrate analog (Fig. 2B). This effect was also observed in DEER studies of CRF01_AE and PR5 (32). Additionally, in the presence of both DRV and SQV, E35D contains a lower fractional occupancy of the closed conformation when compared with subtype B (Fig. 2D), potentially indicating that inhibitors may bind to a non-closed or curled conformation as seen for darunavir-resistant PR (52). We propose the persistence of this distance in the presence of inhibitor may suggest that this peak is a signature of inhibitor binding in an altered conformation that contrasts the typical inhibitorbound closed HIV-1 PR state. Evidence for an asymmetric flap orientation in unbound and inhibitor-bound DRV-resistant PR has been observed crystallographically (52,53). In these structures, one flap was seen to tuck into the active site cleft where the other sits over the inhibitor, and in one case, DRV was also found in a non-standard orientation sitting perpendicular to the active site cleft (53).
E35D Increases Protein Backbone Dynamics-We have recently shown that PR backbone dynamics vary for constructs that have alterations in conformational sampling profiles, where more rigid dynamics are observed when the protein occupies a predominantly closed-like state, and conversely where increased dynamics are found for constructs that have open-like states stabilized (32,33). Analysis of NMR-derived heteronuclear order parameter values (S 2 ), reflective of amide bond fluctuations, determined from multifrequency NMR investigations reveal regions of the protein that undergo enhanced or diminished motion relative to subtype B (supplemental Figs. S9 and 10 and supplemental Table S1). Fig. 3A shows color-coded ribbon diagram for subtype B, E35D, and PR5 where the color is indicative of the value of S 2 . Fig. 3B compares differences in S 2 values for these constructs. In gen-eral, most sites in E35D exhibit increased backbone dynamics relative to subtype B (indicated by red and orange color along the diagonal in Fig. 3B). E35D has quite similar dynamics to PR5, with some sites (indicated by blue along the diagonal in Fig. 3B) that are more rigid. Strikingly, the tips of the flaps in E35D exhibit slowed dynamics compared with both subtype B and PR5, whereas nearly all other regions of E35D and PR5 are similar to each other and different from subtype B, indicating that the single E35D substitution can account for the majority of the increased backbone dynamics observed.
E35D Minimally Modifies the Overall Backbone Structures-Suitable crystals of HIV-1 PR E35D and PR5 with inhibitors DRV, CaP2, SQV, and amprenavir (APV) were obtained, and x-ray structures were determined and are discussed in reference to four previously reported structures as follows: the DRVbound subtype B construct (PDB code 3BVB) (54), the subtype A construct (which bears R57K natural polymorphism and 14 other mutations (PDB code 3IXO)) (45), and two DRV-resistant subtype B constructs that each contain Glu-35 (PDB codes 3U7S and 4NPT) (52,55). The x-ray crystallographic statistics for current and previously reported structures are summarized in Table 1. All crystal structures were refined to reasonable residual factors for both R free and R cryst and Debye-Waller factors (B-factors) of main chains, side chains, ligand, and waters (Table 1).
Structural imposition of ligand-bound subtype B, E35D, PR5, and unbound subtype A demonstrates a little structural variation, with only minor backbone deviations observed for E35D compared with subtype B and slightly larger deviations occurring upon addition of mutations present in PR5 (Fig. 4A) and subtype A (supplemental Fig. S11). The larger backbone deviation of SQV-bound PR5 and unbound subtype A may partly come from the crystal packing effect in terms of the distinct crystal symmetry (56). For a quantitative assessment of the suggests that the E35D natural polymorphism slightly decreases intra-monomer distances between residues 35 and 50, with a minimal decrease of the inter-monomer distances among residues 35 to 16 and 81, and where a similar pattern is observed for APV-bound E35D (supplemental Fig. S11C). However, the impact of E35D when bound with CaP2/SQV leads to slightly increased structural differences where residues from 45 to 55 in the flap region and residues from 80 to 83 near the active pocket of the SQV-bound E35D construct exhibit increased inter-monomer distances. Two common features can be distinguished from DRV-bound and SQV-bound PR5 constructs in supplemental Fig. S11, D and E. The majority of the negative structural alterations (ϽϪ1.5 Å) are located inside each monomer around residues from site 15 to site 19 and residues from site 38 to site 42. The majority of the positive structural alterations (Ͼ1.5 Å) are around residue 46 between different monomers.
Additional Mutations in PR5 Slightly Alter the Structure of the Hinge and Fulcrum-The additional mutations in the hinge and flap regions of PR5 and subtype A are seen to lead to moderately larger structural deviations in the flap, hinge, and fulcrum regions when compared with subtype B (Fig. 4B) S11, C-F), it is apparent that most of the prominent changes of the PR5 construct occur around two natural polymorphisms of I15V and R41K, with only minor changes around sites 35 and 57. More structural alterations (ϽϪ1.5 or Ͼ1.5 Å) are generated in the subtype A construct (supplemental Fig. S11F) around amino acid residues of 7, 17, 36, 40, 45, 67, and 80 close to these mutation sites in the protein. Nevertheless, the backbone distance between residues 35 and 57 was not significantly changed for E35D, PR5, and subtype A referring to subtype B construct, with a distance variation less than 1 Å.
Salt Bridge Interactions Are Modulated by E35D/R57K Substitutions-The salt bridge interaction in PR between residues Glu-35 and Arg-57 likely serves as an important interaction to modulate the flap and hinge-region interactions that can consequently impact the rigidity and conformational flexibility of the protein backbone. Fig. 5 and supplemental Fig. S11 show the local structure of residues 35 and 57 of the various PR constructs characterized. The results clearly show a systematic disordering of the interactions between residues 35 and 57 as additional mutations are accumulated in the order of subtype B Ͻ E35D Ͻ PR5 Ϸ subtype A. We semi-quantitatively describe the strength of this salt bridge interaction by the N-O distances between the terminal nitrogen and oxygen atoms of side chains. A schematic model of this interaction is depicted in Fig. 6. Results from this analysis are given in Table 2. "Strong" interactions describe N-O distances of Ͻ3 Å; "weak" refers to cases when the smallest N-O distance is 3-4 Å; and "absent" reflects situations where all N-O distances are Ͼ4 Å. These definitions are consistent with methods reported by others (57).
Both "staggered" and "paired" orientations of residues 35 and 57 in the DRV-bound subtype B PR structure (Fig. 5A) can be seen. The staggered configuration contains only one strong N-O interaction with a distance of 2.7 Å, whereas in the paired configuration, two N-O pairs occur with distances of 2.8 Å. The strong salt bridge interactions in the subtype B construct may stabilize the flap with respect to the cantilever region resulting in a more rigid structure with relatively lower backbone dynamics that may be consistent with the  Red indicates that backbone distances of the HIV-1 PR constructs between the corresponding residues are shorter than that of the subtype B construct. Sequence numbers run from 1 to 198 to indicate the residues in monomer A (1-99) and monomer B (100 -198) such that asymmetry between the two monomer interactions can be readily distinguished. Structures of E35D with DRV, CaP2, SQV, and APV determined show how the single E35D natural polymorphism weakens the salt bridge interaction between Asp-35 and Arg-57 (Fig.  5, B-E, respectively) These structures, to varying degrees, reveal multiple conformations in the orientations of both Asp-35 and Arg-57, with higher frequency of alternative conformations seen for Asp-35 in multiple crystals of PR with varied inhibitors. Some evidence exists for the formation of staggered salt bridge interactions, but paired orientations are nearly absent given perpendicular-like orientations of the nitrogencarbon planes. The weakened interactions give rise to N-O distances of Ͼ3 Å. Alternative conformers are generated at both Asp-35 and Arg-57 residues with a population of about 50% in DRV-and SQV-bound E35D. In addition, one alternative conformer is found at the Asp-35 residue in CaP2-bound E35D.
These phenomena suggest that weakened salt bridge interactions are generated by a single E35D natural polymorphism, and the impact can be perturbed by different ligands inside the HIV-1 PR.
In inhibitor-bound structures of PR5 (Fig. 5, F-H), which contains the double mutations of E35D and R57K, the salt bridge interactions between residues 35 and 57 are eliminated. Replacing the direct salt bridge interaction between residues 35 and 57 is a series of intermolecular interactions mediated by Pro-79, Gly-78, and Val-77. Specifically, N-O distances between Asp-35, Pro-79, and Gly-78 range within 3 and 4 Å, and the N-O distances between Lys-57 and Val-77 near the active region are Ͻ3 Å. These interactions are depicted in Fig. 7 (DRV-bound PR5 and SQV-bound PR5). This network may provide a compensating interaction that retains the stability of the semi-open state but may also contribute to an increase tendency for the flaps to curl (as seen in DEER data) when in solution or interacting with inhibitors.
Subtype A carries the natural polymorphism R57K but not E35D. The effect of R57K on weakening the salt bridge interaction is seen Fig. 5I, where a disrupted salt bridge interaction is observed only in one monomer. Given that the average backbone C␣ positions are similar to subtype B, it is likely that R57K strongly contributes to the disrupted salt bridge. Hence, the effect of the R57K mutation in subtype A supports our argument that the complete dissociation of the salt bridge interaction seen in crystal structures of PR5 is promoted by double mutation of E35D and R57K.
Results from NMR dynamics indicate that overall the backbone dynamics of HIV-1 PR are increased when E35D is incorporated. The structures reveal that changes in the salt bridge and or molecular interactions within the flap-hinge-cantilever region may provide the molecular basis for this change in dynamics. The alternative interaction network seen in PR5 structures also provides a rationale as to why the dynamics of PR5 and E35D are similar given that the salt bridge interaction is completely abolished in the crystal structure of PR5 and only partially diminished in E35D. We acknowledge that NMR dynamics were determined in unbound enzyme, and structural insights are being taken from crystals of inhibitor-bound states. Nevertheless, we believe the two sets of data are congruent and are useful to paint a structurally based model for understanding how natural polymorphisms are impacting dynamics and conformational sampling.
No Impact on Enzymatic Activity of Resistance Was Generated by E35D/R57K-The impact of natural polymorphisms in E35D and PR5 on catalytic parameters K m , k cat , and k cat /K m was determined to be minimal to modest. Values of dissociation constants, K i , for inhibitors SQV, APV, ATV, and DRV were also determined and found not to be modulated by these natural polymorphisms (Tables 3 and 4). Overall, the Michaelis-Menten parameters for E35D and PR5 are very similar to subtype B, which are similar to others reported previously (58,59). Values for k cat and k cat /K m for E35D show a modest decrease of ϳ2and 3-fold, respectively, when compared with subtype B, indicating that disruption of the salt bridge without compensating interactions slightly decreases catalytic turnover. For

TABLE 3 Values for Michaelis-Menten kinetic parameters and inhibition constants (K i ) for HIV-1 PR constructs
Values in parentheses indicate the relative ratio referenced to data for subtype B.

TABLE 4 Statistics of the prevalence of E35D natural polymorphisms, where numbers in parentheses indicate the total number of isolates in analysis (one mutation was extracted from one person)
The data analysis is up to March 2015. -indicates no data available in the database. Overall, these are minor changes, but they reveal that changes in dynamics and conformational sampling that originate from alterations in salt bridge interactions in the hinge and cantilever have slightly more of an impact on catalytic turnover than substrate binding. The K i values for subtype B with all inhibitors tested were subnanomolar and in agreement with other studies (40,59). Overall, the E35D and R57K natural polymorphisms, alone or in combination, are found to have no impact on PI resistance, as evidenced by minor to no significantly different changes in K i values. Kinetic analyses and inhibition assays indicate that E35D and R57K do not modulate interactions of the catalytic pocket. Additionally, drug binding affinity can be assessed by evaluation of the atomic arrangement of inhibitors inside the catalytic pocket of HIV-1 PR (19,20). As discussed above and shown in Fig. 4, polymorphisms E35D and R57K induce little to no backbone rearrangements around the catalytic pocket (Fig.  8). Nearly identical atomic arrangements of SQV are found in E35D and PR5 constructs. One trivial difference of DRV on the aniline group could be distinguished among conformations when bound to subtype B, E35D, and PR5 constructs. Consequently, the structures are consistent with kinetic analyses indicating little to no impact on inhibitor dissociation. Taking into consideration that patterns of secondary mutations are seen to diverge for various subtypes and drug resistance emerges at different rates than for non-B PR (60 -65), our kinetic and inhibition assay results imply that E35D and R57K must act in concert with additional primary mutations and alternative secondary mutations to impact drug resistance.

E35D mutation in HIV-1 PR subtypes with/without treatment
E35D/R57K Are Selected for in Subtype A, F, and G CRF01_AE-The salt bridge interaction between residues 35 and 57 serves as a major interaction between the flap and hinge regions of PR. As reported in the Stanford University HIV Drug Resistance Database, only one natural polymorphism at each of these sites (35 and 57) is found in drug-naive subtypes and CRFs compared with the consensus subtype B sequence, namely E35D and R57K, which on average are found at 41 and 14%, respectively (54). Other amino acids reported for these two residues in drug-naive constructs occur with probabilities of Ͻ1%. Analysis of sequences deposited within the HIV Drug Resistance Database shows that the occurrence probability of E35D and R57K differs among subtypes and CRFs. For drug-naive patients, E35D is predominant (Ͼ80%) in subtype A (prevalent in Uganda), subtype F (prevalent in Brazil), and CRF01_AE (prevalent in Thailand and Vietnam) (Fig. 9, A and B, and Tables 4 and 5). This occurrence is more than twice the frequency of this mutation seen in subtype B (prevalent in the Americas, west Europe, Australia, and Japan), subtype C (prevalent in India and South Africa), subtype D (prevalent in Uganda), and CRF_AG (prevalent in Cameroon and the Ivory Coast). A similar trend is also observed for R57K, which is nearly absent in subtypes B, C, D and CRF_AG, but occurs at 10 -25% prevalence in subtype G and CRF01_AE and with Ͼ40% prevalence in subtypes A and F. Based on the structural analysis of the salt bridge interaction between residues 35 and 57, we would suggest that the local protein structure, conformational sampling, and backbone dynamics of subtypes that harbor the E35D/R57K substitutions will be similar to our findings for PR5, CRF01_AE, and E35D.
E35D/R57K Are Selected for in TPV Resistance and Suppressed in DRV Resistance-Further analysis of the data provided by the Stanford database reveals that PI treatment alters the prevalence of E35D and R57K (Tables 4 and 5). Specifically, for subtype B, it appears that E35D/R57K substitutions are selected in response to TPV-induced drug pressure, although these substitutions are suppressed in DRV drug pressure evolution. Fig. 9C shows the percentage change in E35D and R57K prevalence in subtypes B and F as a function of PI exposure. For subtype B, inhibitors SQV, NFV, RTV, APV, and TPV select for these substitutions in increasing prevalence, whereas inhibitors IDV, LPV, and DRV are seen to deselect for the altered salt bridge interaction, again listed in increasing order. In contrast, for subtype F, which contains the E35D and R57K substitutions as natural polymorphisms, drug-pressure is not yet seen to significantly alter the mutational prevalence at these sites. It is noted that insufficient data exist in the Stanford database for a complete analysis of subtype F (Tables 4 and 5). The meaning of these observations is unclear at the moment but may indicate compensating conformational landscape/dynamics/entropic interactions that occur as mutations accumulate in response to different inhibitors. Future work is focused on obtaining structural, biochemical, and biophysical information of TPV bound to E35D, PR5 among other subtypes, and CRFs to gain structural insights into why TPV may select for the altered salt bridge interaction. The E35D/R57K double mutation may serve as an interesting molecular interaction in non-B subtypes to target in future inhibitor design schemes, and it may provide further insights into the balance between enthalpic/entropic protein/ inhibitor interactions in drug resistance. It was suggested previously that the current antiretroviral PI drugs, which were designed against HIV-1 subtype B virus, were also actively inhibiting the non-subtype B constructs; however, distinct non- drug-resistant mutation patterns developed (12). Based on the above analysis, it may be beneficial to carry out further studies on possible collaborating effects of E35D and R57K natural polymorphisms in combination with common drug pressureselected mutations of subtypes and CRFs.
Analysis of Subtype B DRV-resistant Structures-Based on previous studies (11,54,55,66,67), there are five x-ray crystal structures of DRV-resistant mutants DRV1, DRV2, DRV5, PR20, and P51. Out of these five structures, DRV1 and P51 are the two constructs that not only show DRV resistance but also have similar distances between residues 35 and 57 as that of the subtype B construct (PDB code 3BVB). The local structures around Glu-35 and Arg-57 in DRV1 and P51 are shown in Fig.  5, J and K. A significant salt bridge interaction can be readily observed by the N-O distances (Table 2) and the residue sidechain configurations. Based on the crystal structures, we hypothesize that the DRV-resistant constructs such as P51 and DRV1 should have backbone dynamics more similar to subtype B than PR5 or E35D (future NMR investigations). It is interesting that DRV selects for restabilization of this hinge/flap interaction when eliciting drug resistance, thus potentially implying that the additional mutations may cause destabilization else- FIGURE 9. Bar graphs of the prevalence of E35D (A) and R57K (B) as natural polymorphisms in HIV-1 PR subtypes and CRFs are shown, and data are not shown when the total number of isolates is smaller than 5. C, population percentage change (␦P%) of the E35D and R57K natural polymorphisms under PI treatments, where ␦P% ϭ (P(in PI-treated patients) Ϫ P(in drug-naive patients))/(P(in drug-naive patients)) ϫ 100. The insets show the enlarged bar graphs of subtype F. Data were not included for TPV-, APV-, lopinavir-, and DRV-treated subtype F patients due to the limited number of DNA sequences.

TABLE 5 Statistics of the prevalence of R57K natural polymorphisms, where numbers in parentheses indicate the total number of isolates in analysis (one mutation was extracted from one person)
The data analysis is up to March 2015. -indicates no data available in the database. where in the dimer that needs to be compensated for by regained structural stability between the flap/hinge regions.

R57K mutation in HIV-1 PR subtypes with/without treatment
Conclusions-The presence of E35D alone is shown by DEER results to destabilize the semi-open conformation. When E35D is combined with other natural polymorphisms, the salt bridge between the flaps and cantilever region is destabilized, likely leading to increased percentages of the curled-open population; however, the possibility does exist that other mutations in PR5 combine to enhance the stability of the curled-open state. NMR analysis suggests that E35D alone and in conjunction with R57K natural polymorphisms are able to enhance the overall protease backbone dynamics especially in the flap and hinge regions, which may confer some selectivity under protease inhibitor pressure. Six HIV-1 protease crystal structures, carrying E35D and/or R57K natural polymorphisms, were analyzed and compared with four other crystals. The increased backbone dynamics in the E35D construct is suggested to be caused by the weakened salt bridge interaction between Asp-35 and Arg-57. The combined effect of the dissociated salt bridge interaction with restabilization via compensating interactions between Asp-35 and Lys-57 via intermediate residues from Val-77 to Pro-79 can explain the similarity in backbone dynamics of PR5 and E35D, although DEER-based conformational ensembles differ. Based on the current statistics from the HIV Drug Resistance Database, occurrence probability of E35D and R57K reveals a distinct characteristic pattern across different HIV-1 subtypes and CRFs indicating possible protein dynamics diversity in HIV-1 PR. Drug-selected mutation evolution patterns for the two latest PIs, including TPV and DRV, demonstrate opposite selection patterns in subtype B of these two natural polymorphisms, possibly suggesting a compensating mechanism for protein/inhibitor flexibility. At this time, there is insufficient data for non-B subtypes to fully analyze the impact of E35D/R57K natural polymorphisms on inhibitor drug resistance development.

Experimental Procedures
Cloning and Site-directed Mutagenesis-Escherichia coli codon-optimized genes of HIV-1 PR subtype B and PR5 constructs were purchased from DNA 2.0 (Menlo Park, CA), which were cloned into pET-23a plasmid between two restriction enzyme digestion sites of NdeI and BamHI. Plasmid carrying the gene for the E35D construct was generated by utilizing the QuikChange site-directed mutagenesis kit (Stratagene, La Jolla, CA). The D25N inactive constructs were utilized for all DEER, crystallization, and NMR measurements. Active Asp-25 enzymes were utilized for the kinetics measurements. Additionally, three stabling mutations (Q7K, L33I, and L63I) and two other mutations (C67A and C95A) for eliminating native cysteines were designed in all constructs to avoid auto-proteolysis of the protease and to ensure site-specific spin labeling for DEER measurements, respectively. These five substitutions are present in all constructs studied here. Fig. 1A shows amino acid sequences.
Protein Expression, Purification, Spin Labeling, and Sample Preparation for DEER-For DEER measurements the spin label, 1-oxyl-2,2,5,5-tetramethyl-⌬3-pyrroline-3-methyl, methanethiosulfonate (MTSL), was chemically incorporated into PR by mutation at site K55C. MTSL was purchased from Santa Cruz Biotechnology. Protein expression, purification, and spin labeling were carried out by following our previously developed procedures where buffer pH for one step of purification was modified for E35D given its isoelectric point of 9.39 (30,34). To ensure high spin labeling efficiency, 4-fold molar excess spin label was added into the HIV-1 PR sample solution. The reaction was carried out at 4°C in the dark for 12 h. Protein precipitates were removed by a high speed centrifuge at 12,000 rpm, and the excess free spin label was removed by buffer exchange (2 mM NaOAc with pH at 5.0) by using a HiPrep 26/10 desalting column (GE Healthcare).
DEER and Data Analysis-HIV-1 PR samples were concentrated and buffer-exchanged to 20 mM NaOAc buffer in D 2 O at pH 5.0 by using centrifugal membrane concentrators (Millipore, Billerica, MA) to 140 M. 4-Fold molar excess of inhibitors, proportional to enzyme, were added into HIV-1 PR solutions. The ligand reactions were performed at room temperature for Ͼ1 h. Sequentially, 30% v/v deuterated glycerol was added into the protein solution. The final protease concentration was ϳ100 M. The protein sample was then transferred to an EPR tube (3-mm inner diameter and 4-mm outer diameter quartz) (Norell, Marion, NC), flash-frozen in liquid nitrogen, and then inserted into the dielectric ring resonator (ER 4118X-MD-5) in a frozen state. Four-pulse DEER scheme was adopted in the measurements on Bruker EleXsys E580 spectrometer at a temperature of 65 K, as described previously (34,68). The DEER modulation curve was processed by using Deer-Analysis (69,70), where the inter-molecular background was subtracted; high frequency noise was filtered by low frequencypass digital filter, and consequently, the DEER data were converted into distance distribution profiles via Tikhonov regularization. The distance profiles are regenerated by Gaussianshaped populations that represent conformational states of HIV-1 PR from which DEER modulation curves are reconstructed for error analysis and peak suppression (30,33,34,50). Details regarding DEER data analyses and population validation are provided in supplemental Figs. S1-S8.
Sample Preparation, NMR Spectroscopy, and Data Analysis-Uniformly 15 N-labeled E35D was expressed and purified from E. coli grown in modified minimal media as described previously (32,33). NMR samples contained ϳ150 M 15 N-labeled HIV-1 PR in 2 mM deuterated NaOAc buffer at pH 5.0 with 10% D 2 O and 100 M 4,4-dimethyl-4-silapentane-1-sulfonic acid as an internal reference. NMR relaxation data were collected on Bruker Avance spectrometers at two frequencies of 600 MHz (AMRIS Facility, University of Florida) and 800 MHz (National High Magnetic Field Laboratory, Florida State University) at 293 K, respectively. Spin-lattice relaxation (R 1 ) and spin-spin (R 2 ) relaxation rates of 15 N were measured via HSQC pulse train of hsqct1etf3gpsi and CPMG pulse train of hsqct2etf3gpsi, respectively (33). 1 H-15 N NOEs were measured in an interleaved manner using the pulse sequence of hsqcnoef3gpsi with a recycle delay of 5 s (33). NMRPipe and Sparky were used to process and analyze the data (71,72). R 1 and R 2 relaxation rates were calculated by fitting the peak intensity to a single exponential decay related to delay time by using GUIrelax (73,74). NOE values were calculated by taking the ratio of resonance intensities with and without 1 H presaturation. Model-free analysis was performed using GUIrelax (73,74). Difference plot of NMR order parameters (S 2 ) was calculated by using S 2 ij ϭ S 2 ij_HIV-1 PRcompare Ϫ S 2 i_HIV-1 PRreference . Protein Expression, Purification, and Crystallization-For crystallization trials, the following additional purification steps were added to the NMR/DEER protocols described above. Samples were buffer-exchanged by centrifugal membrane concentrators (Millipore, Billerica, MA) into buffer A (30 mM K 2 HPO 4 , 100 mM NaCl, 4 mM EDTA, and 5% glycerol at pH 7.3) for further purification by size-exclusion chromatography with a HiLoad 16/60 Superdex 75 Prep Grade size column (GE Healthcare). Fractions corresponding to dimeric protein were pooled and concentrated by centrifugal membrane concentrators (Millipore, Billerica, MA) to 3-5 mg/ml in buffer B (50 mM sodium acetate and 5% glycerol at pH 5) (19,40,75,76). Inhibitors or non-reducible substrate mimics were added to the protein sample at 3:1 molar excess and allowed to interact at 4°C for 1 h, after which any precipitant was removed by centrifugation at 14,000 rpm. Hanging drop vapor diffusion was used for crystallization of all samples, using Hampton Crystal Screen Cryo and Crystal Screen (Aliso Viejo, CA) for initial crystallization screening. The gradient reservoir buffer of GRB-1# (ammonium sulfate from 1.0 to 3 M, 20 mM sodium acetate at pH 5) (19,75) and GRB-2# (sodium chloride from 0.8 to 2.7 M, 30 mM citric acid at pH 5) were chosen as "promising precipitation solution" for further optimization in all crystal trials of E35D and PR5 constructs. SQV-bound E35D crystals were obtained from both GRB-1# and GRB-2#, whereas DRV-, CaP2-, and APV-bound E35D crystals could only be obtained in GRB-2#. SQV-and DRV-bound PR5 crystals were obtained from GRB-1# buffer. Regarding the crystal morphology, rectangular/square sheet-like crystals dominated in E35D crystals, although either non-regularly shaped or rod-like crystals were formed in PR5 crystals.
X-ray Diffraction Data Collection and Analysis-All crystals were pre-soaked in a 30% glycerol cryo-protectant solution before flash cooling to 100 K prior to data collection. Data were collected "in-house" using an RU-H3R rotating copper anode ( ϭ 1.5418 Å) operating at 50 kV and 22 mA utilizing an R-Axis IV 2ϩ image plate detector (Rigaku Corp.). IMosflm and Scala from CCP4 suite 6.4.0 were used for indexing the diffraction data and merging the diffraction peaks (77). All data were processed and scaled to a maximum resolution of 1.6 Å, with an overall completeness of Ͼ92% and R merge of Ͻ16%, for each data set. Initial phases were obtained using two previously determined HIV-1 PR structures (PDB code 4NJT (78) and PDB code 3K4V (79)) as search models. Phases were generated using PHENIX version 1.9 for E35D and PR5 structures, respectively (80,81). Structural refinements were also completed using PHENIX.refinement with R free calculated with 10% of the unique reflections selected at random and excluded (81)(82)(83). Manual refitting of all residues and ligands, including DRV, SQV, CaP2, and APV, into the electron density was complete using Coot 0.7.2.1 (84). PyMOL version 1.7 was used to visualize the crystal structures and to generate figures (85). Double difference analysis of the backbone C␣s was calculated by "Crystal-Analysis" using D ij ϭ D ij_HIV-1 PRcompare Ϫ D ij_HIV-1 PRreference , where i and j represent the residue number, and D represents the distance between these two residues. CrystalAnalysis is a Matlabbased program that our laboratory developed and is available upon request or at the Matlab Exchange File on-line website. For the analysis contained within Table 2 regarding HIV-1 P51, we refit the local structure around Glu-35 and Arg-57 using the electron density map in the RCSB Data Bank.
Protein Activity and Inhibition Constants-Protein expression and purification adopted the same protocol as described above except that plasmids encoding active HIV-1 PR were utilized. The Michaelis-Menten constants (K m , k cat , and k cat /K m ) for all three constructs as well as the inhibition constant (K i ) for four inhibitors of SQV, APV, ATV, and DRV were determined as described previously (40,58,76). The reaction assays were carried on Cary 50 Bio UV-visible spectrophotometer with sodium acetate buffer (50 mM NaOAc, 150 mM NaCl, 2 mM EDTA, and 1 mM DTT at pH 4.7) at a temperature of 37°C where chromogenic substrate (Lys-Ala-Arg-Val-Leu*Nph-Glu-Ala-NLe-Gly) was used. All experiments were repeated three times, and the averaged value and standard deviation are reported in Table 3.