2.2. Structural Comparison of HIV-1 B and CRF02_AG Integrases
In order to determine the potential impact of the natural variations on the protein activity and susceptibility to INSTIs, we built models of the IN structures corresponding to the consensus B sequence and the CRF02_AG variant differing from B subtype by twelve residues. The 18-aas C-terminal end containing the S283G was omitted since the structure of this domain was not resolved by X-ray analysis and the folding of this part of protein is extremely difficult to predict in the apo state, due to its essential length and its highly solvent-exposed position.
Comparative structural analysis were performed considering 6 IN models generated by homology modeling (). Models 1(B) and 2 (CRF02_AG) () represent the unbound homodimer of integrase (IN1–270), which depicts the conformational state of the enzyme just before the 3′-processing of vDNA (apo state); models 3′(B) and 4 (CRF02_AG) () represent the IN dimer in complex with vDNA (holo state), which depicts the active unit of the IN·vDNA strand transfer intasome; models 5 (B) and 6 (CRF02_AG) (not shown) were derived from models 3 and 4 by removing vDNA.
Figure 1 Structural models of the HIV-1 INs from B and CRF02_AG strains. (a) Superimposition of models 1 and 2, representing the enzyme before the 3′ processing from B (in blue) and CRF02_AG (in yellow) strains; (b) Superimposition of models 3 and 4, (more ...)
Models 1 and 2 were constructed from the crystallographic structures of HIV-1 IN-isolated domains or pairs of domains. Overall, the analysis of the models representing the HIV-1 IN conformational state before 3′-processing (apo state) did not show any significant structural change between the two subtypes (Figures and ).
Models 3 and 4 were constructed from the crystallographic structure of the IN·vDNA complex of the PFV intasome [19
]. Although the sequence identity between HIV-1 and PFV INs is low (22%), the structure-based alignment of the two proteins demonstrates high conservation of key secondary structural elements and the three PFV IN domains shared with HIV-1 IN have essentially the same structure as the isolated HIV-1 domains. Moreover, the structure of the PFV intasome displays a distance between the reactive 3′ ends of vDNA that corresponds to the expected distance between the integration sites of HIV-1 IN target DNA (4 base pairs). Consequently, we are confident that the PFV IN X-ray structure represents a good template for the HIV-1 IN model generation [21
]. To obtain a robust alignment, we adjusted the targets (HIV-1 INs from B and CRF02_AG subtypes) and template (PFV IN) sequences manually, considering each structural domain separately, in order to take into account the conservation of the secondary structure (see Section 4
Again, models 3 and 4, representing the IN·vDNA intasomes of both strains, superimposed perfectly and no structural dissimilarity was observed (Figures and ). Most of the variations are located far from the active sites, and the nearest two mutated residues to the active site, at positions 134 and 136, are exposed to the solvent and apparently did not affect significantly the structure. Similarly for 3′-processing, strand transfer activities of B and CRF02_AG recombinant proteins were assayed and compared. In agreement with the modeling results, activities of both INs were comparable ().
Figure 2 Purification of recombinant HIV-1 INs from B and CRF02_AG subtypes and comparison of their activities. (a) Purification products N1, N2, N3 and N4 of recombinant HIV-1 INs from B and CRF02_AG subtypes. (b)–(d) Comparison of DNA binding, 3′-processing (more ...)
It is worth noting that large structural and conformational changes are observed between the apo (models 1 and 2) and holo (3 and 4) states regarding the relative positions of the IN domains (RMSD, root mean square deviation, of 31
Å, based on Cα
) (). These structural modifications result in different contacts between IN domains, N-terminal domain (NTD), catalytic core domain (CCD), and C-terminal domain (CDD). As such, in models 1 and 2 (apo state) no interaction was detected between CTD and CCD, whereas the two domains interact tightly in models 3 and 4 (holo state). The NTD-CCD interface also exhibits substantial changes: in the apo form the NTD-CCD interface belongs to the same monomer subunit whereas in the holo form the interface is from two different subunits. Moreover, IN undergoes important structural transformation leading to structural reorganization of the catalytic site loop upon vDNA binding; the coiled portion of the loop reduces from 10 residues (140–149 aas) in the apo form to 5 residues (140–144 aas) in the holo form (). This partial folding of the catalytic loop is probably stabilized through intra-IN domain-domain interactions and interactions with vDNA which contribute in the helix α
2.4. Docking of INSTIs
Although B and CRF02_AG INs are structurally similar, residue variations may impact the interaction and subsequent activity of the inhibitors. To address this hypothesis, the three inhibitors RAL, ELV, and L731,988 () were docked onto INs by using two different docking algorithms, Glide and AutoDock. RAL and ELV coordinates were taken from the crystallographic structures of PFV intasome cocomplexes [19
], L731,988 was built from scratch (see Section 4
). The three compounds were considered in their deprotonated form, as it has been clearly established that diketo acids (DKAs) mainly exist in this form in solution [24
]. The binding energies obtained by Glide and Autodock scoring functions are reported in .
Table 2 Docking binding energies of RAL, ELV and L731,988 on the HIV-1 IN from B and CRF02_AG strains predicted by Autodock and Glide. The targets are the IN model with one Mg2+ cation in the active site (apo state, models 1 and 2) and IN·DNA model with (more ...)
The inhibitors were first docked onto the unbound IN, models 1 and 2 (apo state), with a single Mg2+
ion within the catalytic site. All three inhibitors are positioned at the catalytic site far from the catalytic site flexible loop. For subtype B, values of binding energies obtained with Glide range in a relatively narrow interval from −8.49 to −7.42
kcal/mol while those obtained with AutoDock range from −8.72 to −6.65
kcal/mol. Scores obtained for a given inhibitor display some variations from one strain to another and between the two docking programs. ELV best pose in model 1 (B subtype) predicted by Glide is very close to that in model 2 (CRF02_AG subtype). Small differences relate to an improved affinity of ELV to model 2 evidenced by a better score (−8.20
kcal/mol) and by the formation of an additional H-bond between the hydroxy group of ELV and E152 side chain (Figures and ). RAL poses in models 1 and 2 differ strongly. In both cases RAL coordinates similarly the Mg2+
cations by its ketoenolate functionality, but the inhibitor adopts opposite positions, more specifically in model 1 its fluorobenzyl ring is oriented towards Y143, while in 2 towards Q148. L731,988 poses are also different in models 1 and 2, characterized by distinct pyrrole ring positions, close to E152 in 1 and to Y143 in 2. Such presence of alternative poses is likely due to a large pocket formed by the accessible active site and the open conformation of the folded loop which allow a large number of conformations and orientations with equivalent binding affinity for the flexible RAL and L731,988 molecules. Consequently no significant difference can be assessed between the binding of the three studied inhibitors to the unbound IN from strains B and CRF02_AG.
Figure 4 RAL (green), ELV (magenta), and L731,988 (cyan) best poses predicted by Glide. The inhibitors were docked into the active site of unbound IN (top) and IN·DNA complex (middle) and IN in holo conformation without DNA (bottom) from of the B (in blue) (more ...)
Further the inhibitors were docked onto models 3 and 4 representing preintegration complexes, IN·2Mg2+
·DNA, from B and CRF02_AG subtypes, respectively. Docking resulted in a binding for the three inhibitors with significantly higher scores than those found for the apo IN. This finding agrees well with the previously published experimental data that showed a high affinity of L-731,988 only to the IN conformations adopted after assembly with the viral DNA [25
]. Glide scores ranked in a range from −10.22 to −8.73 kcal/mol, while AutoDock scores range from −13.45 to −11.11
kcal/mol. Comparisons of the poses produced by the two docking software were found similar, and consequently we focus here on the analysis of Glide results.
The three compounds are positioned in the catalytic site and chelate the Mg2+
cations in agreement with the mechanism of action of these molecules, which are strand transfer inhibitors [26
]. RAL binding mode is characterized by higher scores in both models 3 (B subtype) and 4 (CRF02_AG subtype), respectively, to the other two inhibitors. RAL predicted poses are identical in models 3 and 4 (Figures , , and ). It binds bidentaetly both metal cofactors of the active site acting as a 1–5, and 1–4-type ligand, with the enolic oxygen atom as an oxo-bridge between two Mg2+
cations. Additional stabilization of inhibitor RAL is achieved by π
-staking of fluorobenzyl ring upon Cyt16 of DNA substrate. Similar to RAL, ELV coordinates the Mg2+
cofactors bidentantly through the 1–5 type β
-ketoenolate moiety and 1–3 geminal carboxylic oxygen atoms, with a carboxylic oxygen atom as an oxo-bridge at the bicationic cluster. A few differences of ELV binding in models 3 and 4 refer to slightly different conformation of the chlorofluorobenzyl moiety. L731,988 molecule shows different binding poses in models 3 and 4. In model 3 (B subtype) L731,988 coordinates bidentately one Mg2+
cation by the oxygen atoms from keto functionality of ketoenolate and carboxylate groups, acting as a ligand of 1-6 type. The second Mg2+
cation is coordinated only by the carboxylate oxygen atom. In model 4 (CRF02_AG) L731,988 inhibitor shows exclusively one coordination to the one Mg2+
cation (Figures and ).
The predicted binding poses of RAL correlate well with those observed in the X-ray structure of the PFV intasome complex [19
]. Undoubtedly, the presence of the second catalytic Mg2+
cation, the partial loop folding, and the DNA substrate bearing are presumably the driving determinants for the tight binding of ST inhibitors in the catalytic site. It was perfectly evidenced by Cherepanov that a series of INSTIs fixed similarly to the PFV intasome [19
]. Apparently the crystallographic data or static models derived from these data are not suitable means to explain the specificity of inhibitor recognition by a target. Consequently, considering the similar scoring values for a given inhibitor and closed poses, no significant dissimilarity can be assessed between the binding of studied inhibitors to the IN·2Mg2+
·DNA complex from strains B and CRF02_AG.
To validate the in silico
predictions regarding the susceptibility of subtypes B and CRF02_AG INs, the efficiency of INSTIs (RAL, ELV, and L731,988) on recombinant INs proteins was determined by in vitro
strand transfer assay in the presence of increasing concentration of INSTI (see Section 4
). As to all of the three studied INSTIs, no significant difference in IC50
values against recombinant HIV-1 INs from B and CRF02_AG strains was observed (). IC50
of RAL, ELV, and L731,988 against HIV-1 INs from B and CRF02_AG strains are 41.8, 93.4, 855 nM and 13.7–25.9, 48.9–66.8, 193–291
nM, respectively. The experimental ranking of the three compounds was predicted correctly by Glide scoring function.
IC50 of 3 INSTIs against recombinant HIV-1 B IN and CRF02_AG IN.
The docking calculations evidenced that (i) the IN·DNA complex represents the best target for the studied inhibitors and (ii) the co-complexed vDNA partially shapes the inhibitors binding site. To further explore the role of vDNA, substrate was removed from the IN·vDNA complex and inhibitors were docked again on unbound IN with a fold corresponding to the holo state, models 5 and 6. The binding energies of RAL are depreciated upon vDNA removal in B and CR02_AG subtypes while ELV and L731,988 binding scores are less affected.
Docking scores are nearly similar between the two strains while poses display some variations, as already observed on the apo form. Surprisingly, the AutoDock results show the lower score for RAL binding to both models 5 and 6, while the binding of the two other inhibitors are characterized by better scores, closer to those obtained with models 3 and 4. In contrast the scores produced by Glide are identical between the inhibitors and the subtypes. Chelation of the Mg2+ ions by the inhibitors is still maintained but the interaction patterns differ from those predicted in models 3 and 4. Indeed, in model 5 (B subtype) RAL chelates the first Mg2+ cation through the nitrogen atom of the oxadiazole ring, and the oxygen atom of the carboxamide moiety; the second Mg2+ is coordinated by 1–4 oxygen atoms of pyrimidinone fragment. In model 6 RAL mode of coordination resembles that observed in model 4; however, stabilizing π-stacking interactions were vanished. Again, the large volume of the binding pocket and the lack of stabilizing protein-ligand and DNA–ligand interactions can explain such variety. Consequently, unbound IN in the holo conformation, as unbound IN in the apo conformation, does not appear as a suitable target for the inhibitors RAL and ELV. L731,988 appears as a weaker binder, as confirmed by the experimental IC50 values.
Molecular modeling approaches were used to investigate the effect of the natural variations showed by CRF02_AG strain on the in vitro
activities of the enzyme and its susceptibility to INSTIs as compared to the ones of the consensus B integrase. We found that the structural models of unbound (apo state) and viral DNA-bound (holo state) integrase showed very similar folding and tertiary structure for the two studied strains. The structural models of the IN·vDNA complex superimposed perfectly. This similarity was confirmed by comparable strand transfer activity for IN variants in 14, 112, 125, 134, 136, 206, and 283 positions. Consequently, the naturally occurring variations in the HIV-1 IN subtype CRF02_AG – K14R, V31I, L101I, T112V, T124A, T125A, G134N, I135V, K136T, V201I, T206S, V234I, and S283G, which were suggested to modify IN structure, do not affect significantly in vitro
DNA binding activity, either 3′-processing or strand transfer reaction. Furthermore, docking results revealed that the modes of binding and docking conformations of three studied inhibitors are comparable for B and CRF02_AG strains and these INSTIs possessed similar IN inhibitory activity against B and CRF02_AG HIV-1 strains. Altogether these results demonstrate the absence of difference in susceptibility and confirm previously reported observations for subtype B and C HIV-1 INs [12
]. Thus, in contrast to the lower baseline susceptibilities of recombinant A/G subtype virus to protease inhibitors (PIs) and reduced susceptibility of some A/G isolates to abacavir, INSTIs potentially provide an excellent therapeutic options for the treatment of HIV-1 subtype CRF02_AG-infected patients [10
In the targets all three molecules are positioned similarly with keto-enol moiety in an orientation encouraging coordination of the two metal cofactors in the active site. Furthermore, independently of the method, the three INSTIs displayed a more favorable binding onto the IN·vDNA complex (holo state) than on the unbound enzyme (apo state), in good agreement with their mechanism of action [26
]. Same difference in theoretically predicted modes of RAL binding was reported early by Loizidou [27
]. The observed conformational and structural transformation of IN upon DNA binding led to an important change in the folding and conformation of the catalytic site loop which in turn favors a formation of the binding pocket accommodating the INSTIs. The binding modes of ELV and L731,988 were practically not altered by the removal of the viral DNA. Conversely removing vDNA had a significant effect on the docking results RAL, thereby highlighting the role of vDNA for RAL recognition most likely due to the halogenated benzyl moiety that displaces the unpaired 5′-adenine and stacking with the Cyt16 through π
interactions. Although such interaction is thought to be involved in all the IN strand transfer inhibitors examined [19
], our results suggest that ELV and L731,988 binding determinants differed in part from the ones of RAL.
It should be noted that slight differences were observed between the results obtained with Glide and AutoDock scores, which can be ascribed to the impact of electrostatic interactions in the studied molecular systems. Indeed Glide uses higher negative charge localized on the two oxygen atoms of the hydroxypyrimidinone of RAL than AutoDock (−1.22 and −0.5e
versus −0.183 and −0.265e
). Also, within the AutoDock scoring function, the carboxylate charges used for ELV (2 × −0.64e
) and L731,988 (2 × −0.62e
) are more than two oxygen atoms attached to the pyrimidine cycle of RAL. To verify this hypothesis, we repeated the docking calculations of ELV and L731,988 using the charges of two oxygen atoms attached to the pyrimidine ring of RAL instead of those assigned by Gasteiger charges. The new binding energies of both inhibitors increased from −12.45 and −11.50 to −7.95 and −7.80
kcal/mol for ELV and L731,988, respectively. Since these atomic charges contribute highly in the binding energy as the atoms coordinate Mg2+
ions, they are likely responsible for the discrepancies found between the theoretical binding energies and the experimental IC50
values. The experimental ranking of the three inhibitors based on IC50
is RAL > ELV > L731,988, as predicted by Glide while the ranking predicted by the AutoDock is ELV > L731,988 ≥ RAL. The high negative charges of the carboxylate oxygen atoms of ELV and L731,988 may be the obstacle to have inhibitory actions on integrase, as efficient as RAL, since these charges increase the desolvation free energy and so increase the binding penalty for these inhibitors.
Studies investigating the presence and frequency of polymorphisms in the HIV-1 gene of treatment-native patients are extremely important for tracing the virus evolution and the epidemiology of HIV infections worldwide. Associated crucial questions concern the effect of polymorphisms on viral enzymatic activities, susceptibility towards inhibitors, and inhibitor resistance pathways. The absence of accurate experimental data characterising the IN and/or IN·vDNA complex structures essentially perplexes an exploration of these essential topics. Since the beginning of clinical AIDS treatment with RAL in 2007, only a few attempts to probe RAL binding to integrase from different retroviral strains have been reported. Particularly, molecular docking of RAL into the IN catalytic core domain structure with the inhibitor 5CITEP as a viral DNA mimic has depicted different binding modes and affinities of RAL to IN from B and C subtypes [27
]. Differences between the binding modes of several compounds to IN from B and C subtypes were also communicated [28
In this context, our combined theoretical (structural modeling) and experimental (biochemical) evaluation of subtype CRF02_AG variation impact/effect on IN interaction with DNA or IN susceptibility to INSTIs contribute to the understanding of polymorphism effects at the molecular and structural level. Our experiments have revealed that IN from subtype CRF02_AG has similar enzymatic activity to IN from subtype B, and the susceptibility of the two INs to strand transfer inhibitors is comparable. Results from molecular modeling and inhibitor docking were found in agreement with in vitro observations.
Biochemical studies have revealed the impact of HIV-1 natural polymorphism on the susceptibility of protease (PR)—the other retroviral enzyme—to inhibitors [29
]. Recent structural and biophysical studies have also shown that sequence polymorphisms of B and CRF01_AE strains can alter protease activity and PR inhibitors binding [30
]. In this protein, the variations between the two strains directly impact the conformation of the flap hinge region and the protease core region that play crucial roles for the enzyme functions.
By contrast, the residues showing natural variations in the HIV-1 integrases from B and CRF02_AG strains are located outside the catalytic region and outer to the binding site of the strand transfer inhibitors. Such type of polymorphism may allow the virus to preserve the integrase structural and functional properties as observed in this study.
The methods we applied could be used for the study of other retroviral substrains emerging at the moment or to appear in the future in order to evaluate and optimize the efficiency of novel specific antiretrovirals. Consequently, our study contributes particularly to this topic and closely relates to a clinically and therapeutically—significant question—does the HIV-1 integrase polymorphisms influence the susceptibility towards integrase inhibitors?