A targeted DNA substrate mechanism for the inhibition of HIV‐1 integrase by inhibitors with antiretroviral activity

We recently reported that viral DNA could be the primary target of raltegravir (RAL), an efficient anti‐HIV‐1 drug, which acts by inhibiting integrase. To elucidate this mechanism, we conducted a comparative analysis of RAL and TB11, a diketoacid abandoned as an anti‐HIV‐1 drug for its weak efficiency and marked toxicity, and tested the effects of the catalytic cofactor Mg2+ (5 mm) on drug‐binding properties. We used circular dichroism and fluorescence to determine drug affinities for viral DNA long terminal repeats (LTRs) and peptides derived from the integrase active site and DNA retardation assays to assess drug intercalation into DNA base pairs. We found that RAL bound more tightly to LTR ends than did TB11 (a diketo acid bearing an azido group) and that Mg2+ significantly increased the affinity of both RAL and TB11. We also observed a good relationship between drug binding with processed LTR and strand transfer inhibition. This unusual type of inhibition was caused by Mg2+‐assisted binding of drugs to DNA substrate, rather than to enzyme. Notably, while RAL bound exclusively to the cleavable/cleaved site, TB11 further intercalated into DNA base pairs and interacted with the integrase‐derived peptides. These unwanted binding sites explain the weaker bioavailability and higher toxicity of TB11 compared with the more effective RAL.

We recently reported that viral DNA could be the primary target of raltegravir (RAL), an efficient anti-HIV-1 drug, which acts by inhibiting integrase. To elucidate this mechanism, we conducted a comparative analysis of RAL and TB11, a diketoacid abandoned as an anti-HIV-1 drug for its weak efficiency and marked toxicity, and tested the effects of the catalytic cofactor Mg 2+ (5 mM) on drug-binding properties. We used circular dichroism and fluorescence to determine drug affinities for viral DNA long terminal repeats (LTRs) and peptides derived from the integrase active site and DNA retardation assays to assess drug intercalation into DNA base pairs. We found that RAL bound more tightly to LTR ends than did TB11 (a diketo acid bearing an azido group) and that Mg 2+ significantly increased the affinity of both RAL and TB11. We also observed a good relationship between drug binding with processed LTR and strand transfer inhibition. This unusual type of inhibition was caused by Mg 2+ -assisted binding of drugs to DNA substrate, rather than to enzyme. Notably, while RAL bound exclusively to the cleavable/cleaved site, TB11 further intercalated into DNA base pairs and interacted with the integrase-derived peptides. These unwanted binding sites explain the weaker bioavailability and higher toxicity of TB11 compared with the more effective RAL.
The human imunodeficiency virus type I (HIV-1) is known as the causative agent of AIDS [1]. Integrase (IN) [2], encoded by the virus pol gene, catalyses integration sequentially through the deletion of two nucleotides at the conserved 3 0 extremity of the viral long terminal repeats (LTRs) [3 0 processing (3 0 Pr)] and the covalent insertion of the recessed 3 0 viral DNA into the cell genome [strand transfer (ST)] [3][4][5]. The catalytic domain of HIV-1 IN bears the active site acidic residues, Asp64, Asp116 and Glu152 (the socalled DDE catalytic triad). Each of these residues is critical for 3 0 Pr, ST and disintegration reactions [6].
HIV-1 IN has attracted considerable attention as a potential drug target, as it has a key role in the virus life cycle and has no cellular counterpart [7]. As expected, the introduction of IN inhibitors into antiretroviral therapy has contributed significantly to limiting the spread of AIDS. The three approved IN inhibitors, raltegravir (RAL, MK-0518, Isenstress Ò from Merck and Co, Inc., Whitehouse Station, NJ, Research Triangle Park, NC, USA) [10] have only minor resistance mutations and belong to a class of IN inhibitors that act on ST (INSTIs), although they can also affect 3 0 Pr to a certain extent [11]. Chemically, they derive from or are isosteres of diketoacids (DKA), a class of compounds originally designed to chelate divalent cations and mediate the association of the negatively charged carboxylate and phosphate groups of the substrate DNA and serve as cofactors in the catalysis [12].
For many years, the cocrystal of 1-(5-ChloroIndol-3yl)-3-hydroxy-3-(2H-TEtrazol-5-yl)-Propenone (5CITEP) ( Table 1) binding to the catalytic core domain (CCD) of HIV-1 IN was the only available structure of an inhibitor bound to a key domain of the enzyme [13]. 5CITEP shares six interactions with each CCD monomer, five of which have the a4 helix. Of note, in the cocrystal 5CITEP-CCD complex, we do not observe chelated divalent cations. The carboxylic group of the key Glu152 residue located at the beginning of the a4 helix establishes direct interactions with the diketo motif [13]. This structure has raised numerous questions on the functional location of INSTIs and the role of divalent metal ions during inhibition. Actually, the calculated models of viral DNA-IN complexes have invariably placed INSTIs within the cavity formed by the active site and the processed LTR end, with the diketo (or equivalent) motif engaged in interactions with a pair of divalent metal cations [14,15]. It was only recently that x-ray structures of RAL (Table 1) and several other INSTIs in complex with the intasome that is, the complex of IN with both processed and unprocessed DNAs, have been solved [16][17][18], thanks to the use of the highly soluble and robust PFV (prototype foamy virus) enzyme [19]. Regardless of the inhibitor, the cocrystal structure showed the coplanar oxygen (or nitrogen) motif of INSTIs at a chelating distance from a divalent ion pair fixed to the DNA end, and the halobenzyl group within van der Waals interactions with the highly conserved 5 0 CA 3 0 /5 0 TG 3 0 dinucleotide step [16][17][18]20,21].
The x-ray studies devoted to the binding of INSTIs to the intasome elucidated the mechanism of action of potent drugs and the origin of resistance mutations in the HIV-1 enzyme [16,17,[20][21][22][23][24]. However, studies performed in solution have demonstrated the inability of INSTIs to bind the enzyme or CCD alone [25,26], in contrast to 5CITEP [13]. This prompted us to examine the possible role of DNA alone in drug binding in our recent work [27]. In the latter, we showed that RAL was able to tightly bind to the free unprocessed and processed LTR ends. Furthermore, the affinity of RAL for the processed LTR end alone appeared greater compared with that reported for the intasome [27,28], thereby suggesting that the substrate DNA could be the primary target of INSTIs. However, in previous work, we did not examine the crucial role played by Mg 2+ ions in the stabilization of the drug-DNA complex. Mg 2+ is involved in bridging electrostatic interactions between the co-planar oxygens of the drug [29,30] and selected DNA phosphate oxygens [31][32][33].
We initiated this study from the idea that TB11 and RAL, which largely differ in their chemical structure and efficacy (Table 1), could have different binding sites. The TB11 structure somewhat resembles that of 5CITEP, but its efficiency towards HIV-1 infection is considerably greater [12]. Much is already known about the binding and structural properties of RAL, but there was a lack of physicochemical data on TB11. In prior studies, hope was placed on TB11 as a possible antiviral drug [12,34], but it was abandoned because of its low oral bioavailability and toxicity. In contrast to TB11, RAL did not have the same adverse effects. It has dose limiting side effects (50% cytotoxic concentration~60-600 lM) [35] and displays a marked bioavailability. Thus, it was questioned whether TB11 in the range of the high concentrations used had additional targets besides the substrate virus that were responsible for the side effects [8,29].
In this study, we used circular dichroism (CD) and fluorescence (anisotropy) to evaluate the RAL and TB11 binding affinities for: (a) peptides derived from the IN active site (K156 and K156-loop140) and (b) oligonucleotides mimicking processed and unprocessed substrates (LTR32 and LTR34) and other structural analogues. We also performed plasmid gel retardation and unwinding assays to test the ability of drugs to intercalate into the DNA base pairs.

Oligonucleotides
These were purchased from Eurogentec (Li ege, Belgium). All are single stranded oligonucleotides ( Fig. 1) designed to adopt a folded double-stranded hairpin structure, especially at the low concentrations required in fluorescence spectroscopy [36,37]. Fluorescence studies were possible thanks to the fluoresceine fluorophore grafted on the thymine (in red) at the centre of the added 3T forming a loop. The so-called LTR34 folds into a 17-base pair stem, which mimics the unprocessed U5 LTR end of viral DNA. The folded LTR32 mimics the processed U5 LTR end. It has a stem with 15 base pairs and bears a dinucleotide AC overhang (in green) at the 5 0 terminus of the nontransferred strand. LTR32-GT with an added GT overhang (in green) at the 3 0 end and the blunt-ended LTR30 lacking the two cleavable base pairs 5 0 GT3 0 :5 0 AC3 0 were designed to assess the impact of the terminal dinucleotides on binding. In the four oligonucleotides, the highly conserved base pairs 5 0 CA3 0 /5 0 GT3 0 are coloured in blue.

Peptides K156 and K156-loop 140
These two peptides ( Fig. 1), were synthetized by Christophe Piesse at the University Pierre et Marie Curie to a purity grade > 98%. The K156 peptide [36,37] that contains the residues 147-169 includes a stabilized version of the a4 helix (residues 150-166) found in the HIV IN CCD crystal by Maignan et al. [38]. This a4 helix possesses an additional turn at its N terminus compared with that of Dyda et al. [39], and is superimposable with the a4 helix of the avian sarcoma virus IN CCD [40]. The K156-loop140 peptide (residue 136 to residue 168) comprises the modified a4 helix portion and the native unmodified flexible loop140 at its N terminus. The peptide concentration was estimated for K156-loop140 from the UV signal intensity of the intrinsic tyrosine occupying position 143 in IN and for K156 from the signal of tryptophan added at the C-terminal position for this purpose, using a molar absorption coefficient at 280 nm of 1280 and 5600 M À1 Ácm À1 respectively.

Inhibitors
RAL [8] is produced by Merck & Co. and was purchased from CacheSyn Inc. (Mississauga, ON, Canada). TB11 (MA-DKA) was synthetized by the Burke's group (NCI) [30,34]. The powders are soluble in water. The structures of RAL and TB11 are shown in Table 1.

CD measurements
CD spectra between 190 and 260 nm (peptides and peptidedrug complexes) and between 190 and 330 nm (DNA and peptide-DNA complexes) were recorded with a Jobin-Yvon CD6 dichrograph using a quartz cell of 1 mm path length. Measurements were calibrated with (+)-10-camphosulfonic acid. Spectra of samples dissolved in phosphate buffer (10 mM, pH 6.0) with 5 mM Mg 2+ were recorded in 1-nm steps, corrected for the base line, and averaged over 10 scans. Before spectral recording, samples were incubated for 10 min at the chosen temperature to allow the solutions to reach their equilibrium state. For titration experiments desired ratios were obtained by addition of the ligand (generally 1-20 lM) to the compound kept at a constant amount (generally 20 lM) in the thermally jacketed cell. In the particular case of the DNA titration by peptides, peptide aliquots were added to 10 lM samples and the spectrum of LTR free of ligand was subtracted from that of the complex. This subtraction is not necessary for titrations by drugs as both RAL and TB11 are achiral molecules. Most experiments were recorded at 10°C. Spectra were expressed

Fluorescence experiments
The fluorescence anisotropy and intensity studies were recorded on a Jobin-Yvon Fluoromax II instrument. Fluorescence anisotropy was expressed as: where I//and I⊥ referred to parallel and perpendicular emission components. These two components were measured in L-format. The denominator of A is simply the total light that would be observed if no polarizers are used. For the fluorescein reporter, maximum wavelengths from the xenon lamp (150 watts ozone-free) were for excitation k = 488 nm and for emission k = 516 nm in the case of LTR34, k = 488 nm and k = 515 nm in the case of LTR32 and LTR32-GT and k = 488 nm and k = 514 nm in the case LTR30. The slit for excitation width was fixed at 4 nm and emission width was 5 nm for LTR34 and 4 nm for LTR32, LTR32-GT and LTR30. Samples were placed in 1 cm 9 0.5 cm thermally jacketed quartz cells, controlled by a circulating bath. For each titration point, at least 10 data points were recorded at 5°C with an integration time of 1s. In fluorescence anisotropy measurements, I//and I⊥ intensities of background solution (i.e. buffer and solute contributions) were subtracted from the sample value. Oligonucleotides were diluted to the desired concentration in 800 lL assay buffers (10 mM Na/Na 2 PO 4 , pH 6.0) with or without 5 mM Mg 2+ . Samples of, generally 10 nM oligonucleotide were titrated by the addition of aliquots of drugs or peptides. After each addition the solution was allowed to equilibrate for least 2 min. The validity of fluorescence anisotropy experiments was controlled by measuring the total fluorescence intensity in parallel to fluorescence anisotropy. As variations of fluorescence were weak, we considered that the anisotropy signal contained the desired information on complex formation. Finally, K d (equilibrium dissociation) constants were calculated by fitting the sigmoidal curves, using GRAPHPAD PRISM 5 (Graphpad, Software Inc., San Diego, CA, USA) applying the nonlinear regression (curve fit) 'Least Squares' procedure.
In fluorescence intensity studies, fluorescence of TB11 or RAL was measured at the 20 lM concentration in 800 lL of reaction buffer. Excitation at 313 nm for TB11 and for RAL provided an emission between 325 and 500 nm, using 2 and 5 nm excitation and emission slit widths respectively. Maximal emission was measured at 410 nm for TB11 and 413 nM for RAL.

Intercalative binding
Many intercalating drugs such as doxorubicin (or adriamycin) belonging to the anthracycline family [42] as well as the related mitoxantrone [43] continue to be extensively used in cancer treatments [44]. In the intercalation mode, the chromophore is inserted in between adjacent base pairs, where it is stabilized by several types of interactions including polarization and van der Waals forces [45]. Intercalation produces both an extension and left-handed unwinding of the helix [46] and generates electrophoretic retardation of DNA, the extent of which depends on the geometry and chemical structure of the intercalator.

Gel retardation assay
The agarose electrophoretic retardation assay was carried out as previously described [47]. Briefly, TB11 and RAL were incubated at different concentrations with 250 ng of pBR322 plasmid DNA in binding buffer (20 mM HEPES pH 7.5, 5 mM DTT, 10% PEG-4000, 10 mM MgCl 2 , 20 mM NaCl and 20 lM ZnCl 2 ), in 10 lL total volume. Mixtures were incubated 30 min at 4°C before addition of 2 lL of loading buffer (0.25% xylene cyanol, 0.25% bromophenol blue, 50% glycerol and 200 mM Tris-HCl pH 7.5). Samples were then loaded onto 0.8% agarose gels and electrophoresed in 19 TBE for 5 h at 80 V. After electrophoresis, gels were stained in BET and visualized under UV light.

Design and synthesis of peptides K156 and K156-loop140
K156 is a helix stabilized peptide designed to mimic the a4 helix (residues 147-169) observed at the surface of the enzyme at the junction of the active site [36,37]. Several amino acid residues of the a4 peptide deemed unimportant for DNA recognition have been replaced by more helicogenic residues. The obtained K156 peptide now displays a backbone conformation more similar to the protein a4 helix. K156 can therefore be considered more functional and less prone to aggregation than the native a4 peptide, and its binding properties may now better reflect those of the a4 helix included in the protein. K156-loop140 is a hybrid peptide that extends over 33 amino acids (residue 136-168) and results from the fusion of the native flexible loop140 [38,40,48] with the helix stabilized K156. The two peptides were synthesized by solid phase chemistry at a purity grade > 98%.

CD spectroscopy
The CD method is widely used to assess the secondary structures of proteins and nucleic acids and to follow their conformational changes induced by interactions and any modification in the medium [49,50]. This method is therefore well suited to the study of complexes formed between proteins and nucleic acids or between these molecules and organic ligands, such as inhibitors. In this study, we analysed the complexes of LTRs and peptides, LTRs and inhibitors, peptides and inhibitors and combinations thereof, in phosphate buffer with or without 5 mM Mg 2+ . In an earlier CD study, we have shown that the helix content of the a4 helix analogue, K156, was enhanced through its interaction with LTRs [36]. The spectrum of K156-loop140 also displayed two negative components at % 225 nm and % 205 nm, and one positive at % 190 nm, which was rather similar to the spectrum of K156. However, the signal intensity at 225 nm agreed with a helix content of nearly 20%, which is lower than that found for the shorter K156 [37]. This decreased helix average in the K156-loop140 was due to the incorporation of the unordered loop140 moiety beside the ordered K156 helix. We found via binding experiments that the processed LTR32 increased the helix content more than the unprocessed LTR34 (35% versus only 15%) (Fig. 2), showing that the cavity created by 5 0 GT3 0 depletion facilitated contacts favourable to helix conversion of the peptide loop. The cleavage of 5 0 GT3 0 allowed the peptide to enter the cavity and undergo the conformational changes desired for IN activity. According to several crystal structures, these are loop residues at the a4 helix-loop junction which are converted into helices [20,30,42,51]. What are the nucleotide binding partners of the DNA cavity involved in helix induction? To answer this question, we performed CD experiments using the LTR analogues modified on their extremities. These studies indicated that in contrast to LTR34 and LTR32, neither LTR32-GT nor the blunt-ended LTR30 had increased helix content (not shown), highlighting the key contribution of the 5 0 AC3 0 extremity of the unprocessed strand in binding the loop and setting up a functional helix. This binding did not exclude the binding of the processed strand with the peptide loop, through the liberated 3 0 adenine-OH group bound to Mg 2+ .

Fluorescence anisotropy
Fluorescence anisotropy is a highly appropriate method to assess the properties of DNA and proteins to partner together and fix ligands [37,52]. This technique provides thermodynamic parameters such as K d values related to complex stability and stoichiometry [27]. We performed the titrations by fluorescence anisotropy of unprocessed LTR34, processed LTR32 and LTR analogues with the peptide K156-loop140 in phosphate buffer with 5 mM Mg 2+ . With LTR34 and LTR32, we obtained biphasic curves which are typical for two distinct DNA sites (Fig. 3A). We found a K d1 of 18 nM and a K d2 of 10 lM for LTR32 and a K d1 of 140 nM and a K d2 of 45 lM for LTR34. In contrast, the titration curves of LTR32-GT and blunt-ended LTR30 were monophasic (Fig. 3B); in both cases, the K d was in the 10 lM range and was thus similar to the K d2 values obtained with LTR32 and LTR34 standing for nonspecific binding sites.
Remarkably, the suppression of high affinity binding upon elimination of the 5 0 AC3 0 highlights once again the importance of this step in the recognition of the 3 0 Pr site by IN [53,54]. The obtained K d1 values were in the range of a specific recognition of the LTR34 and LTR32 ends by the K156-loop140. The greater affinity of K156-loop140 for LTR32 compared to LTR34 correlated with the larger amount of helix induction in the peptide by LTR32 in CD experiments. The binding of the peptide to the LTR ends and its helix stabilization through interactions with the 5 0 AC3 0 step are linked events that agree with the picture given by the PFV intasome crystal structure [16]. However, there are also metal mediated interactions, including catalytic E152, with the liberated adenine 3 0 OH group being located at the facing active strand extremity. The peptide loop is also able to interact with the 5 0 AC3 0 site of unprocessed LTR34.
It is worth noting the narrow correlation between the binding values of the K156-loop140 peptide to LTR ends as determined by fluorescence anisotropy and the degree of helix stabilization as shown by CD. The helix stabilization is still visible in the case of the unprocessed LTR oligonucleotide, signifying that the K156-loop140 peptide may bind to the 5 0 AC3 0 dinucleotide before the 3 0 Pr step. This could be allowed by the known large fraying affecting the end of unprocessed oligonucleotide, enabling a good fit of the terminal nucleotides on the peptide loop [16,27]. Despite a significant consumption of energy due to the reduction in entropy, the strong interactions taking place between the partners permit the formation of a stable complex.
Binding properties of TB11 and RAL to the K156-loop140 and K156 peptides CD studies CD experiments provided the first evidence of important differences between the abilities of TB11 and RAL to interact with the peptides. We have shown above that the CD spectra of K156 and K156-loop140 were characteristic of peptides containing a rather fair amount of a helices. Upon addition of TB11 (20 lM), the helix content increased at about the same extent in K156 and K156-loop140 (Fig. 4A,B), suggesting that K156, that is, the a4 helix, is a target of TB11. This binding occurred only with TB11. The CD spectra of K156 and K156-loop140 remained unchanged upon RAL addition (not shown). Remarkably, the inability of RAL to interact with the peptides is reminiscent of that observed with the enzyme and CCD alone.

Fluorescence studies
Fluorescence experiments confirmed the CD results. According to the K d values (7.5 and 7.75 lM), TB11 had about the same affinity for K156-loop140 (Fig. 5) and K156 [55]. This argues in favour of TB11 binding centred on K156 rather than the loop. At the same time, no binding sign was detected with RAL, which is also in agreement with the CD results [27]. As mentioned above, RAL does not interact with either CCD or the enzyme alone [25,26].
Thus, both CD and fluorescence anisotropy detected important differences in the binding properties of RAL and TB11 towards the IN peptides. The binding of TB11 to K156 was reminiscent of the binding of 5CITEP to CCD within the cocrystal structure [13]. In this study, we observed four interactions between the drug and the a4 helix (Gln148, Asn155, Lys156 and Lys159). A fifth, but not unimportant, interaction involved the carboxylate group of the catalytic Glu152 residue located at the a4 helix-loop junction in CCD. This shares a direct bifurcate hydrogen bond with the metal binding motif of 5CITEP, while in the cocrystal structure of the PFV intasome-RAL complex, the Glu152 carboxylate group interacted with the RAL metal binding motif and the scissile phosphate via a Mg 2+ ion. We note the involvement of several interactions of the a4 helix residues in the binding of IN to LTRs in the PFV intasome [16]. One part of the 3 0 Pr inhibitory activity of TB11 could be assigned to IN binding, which mimics that of 5CITEP.

Fluorescence studies
Evidence for selective binding of IN inhibitors to LTR ends was provided by fluorescence anisotropy studies conducted with the unprocessed LTR34 and processed LTR32 oligonucleotides and two control oligonucleotides, LTR32-GT and LTR30. As shown in the Material and Methods, in LTR32-GT, the overhang dinucleotide 3 0 CA5 0 was replaced by 5 0 GT3 0 on the inactive strand. LTR30 is a blunt-ended LTR that has been deprived of the terminal 5 0 GT3 0 /3 0 CA5 0 base pairs. This material has already been used for the study of RAL by fluorescence anisotropy [27]. Here, it served in fluorescence studies based on changes in anisotropy, shift and intensity of the emission signal [56].

RAL binding in the presence of Mg 2+
Our previous fluorescence anisotropy experiments recorded in the presence of 5 mM Mg 2+ showed that a single molecule of RAL bound to the processed LTR32 end (K d % 6 nM) [27]. Regardless of the concentration of LTR32, RAL produced the same increase in fluorescence anisotropy, (from 0.13 to almost 0.145), and formed a monophasic curve in the nanomolar range. With unprocessed LTR34, the variation in fluorescence anisotropy was weaker, indicating that in this instance, a smaller fraction of RAL was fixed to the terminal base pairs [51]. In addition, the affinity of RAL for LTR34 was weaker compared with that for LTR32 (K d % 20 nM versus % 6 nM). At the same time, addition of RAL to the structural analogues, LTR32-GT and blunt-ended LTR 30 [53,54], had no effect on the fluorescence anisotropy spectra, even at a high drug concentration.

TB11 binding in the presence of Mg 2+
The titration curves for LTR34 and LTR32 by TB11 were biphasic, while we obtained monophasic curves for RAL. During the first binding of TB11, the fluorescence anisotropy increased from 0.13 up to almost 0.15 for LTR32 (Fig. 6A), but only up to 0.14 for LTR34 (not shown). Curve fitting by a nonlinear least squares procedure provided average K d1 values of 0.1 lM for the binding of TB11 to LTR32 and 1.5 lM for LTR34. Thus, similar to RAL, TB11 binds more tightly to processed LTR32 than unprocessed LTR34, indicating that the depletion of the 5 0 GT3 0 dinucleotide on the active strand facilitated the accommodation of the drug to the LTR end. Note that this was already the case for the binding of K156-loop140; binding was improved by the processing. However, the double depletion of 5 0 GT3 0 and 5 0 AC3 0 (blunt-ended LTR30) completely suppressed the first binding of drugs to DNA (Fig. 6B), a result that highlighted the need for the terminal 5 0 AC3 0 dinucleotide to produce high affinity binding, which is consistent with our CD data. At the same time, these findings highlight the same requirements of the loop and the drugs for high affinity binding to viral DNA.
The binding of TB11 to the low-affinity site was observed with the four LTR oligonucleotides independent of the presence of a high affinity site. These K d values were in the mM range: 0.4, 0.6 and 0.6 mM for LTR34 (not shown), LTR32 (Fig. 6A) and LTR30 (Fig. 6B) respectively. The important increase in anisotropy accompanying this binding indicated that many TB11 molecules fulfil this site, thereby presenting all the features of a nonspecific binding site. We will demonstrate later that this site is exclusive to TB11 with regard to the molecular intercalation into DNA base pairs.

Influence of Mg 2+ on inhibitor binding
The above experiments were performed in the presence of the Mg 2+ cation [27]. Divalent metal cations are known for their role not only as cofactors of catalysis but also for their impact on the local shaping and ligand binding to particular nucleotides in both the major and minor grooves of the DNA double helix [57,58]. A comparison of the K d values obtained with and without Mg 2+ (Fig. 7, Table 2) revealed that the divalent cation (5 mM) significantly reinforced the affinity of inhibitors for LTRs. In the absence of Mg 2+ , the K d1 values for drug binding to unprocessed LTR34 increased from 18 to 70 nM for RAL and from 1.5 to 5 lM for TB11 (not shown); for processed LTR32, it increased from 6 to 43 nM for RAL (Fig. 7) and from 0.1 to 0.6 lM for TB11 (Fig. 6A). Thus, regardless of the inhibitor, Mg 2+ established bridging interactions between the inhibitor and the DNA substrate, and the effect was greater when LTR was processed.

Role of the terminal 5 0 AC3 0
The comparative analysis of the processed LTR32 and unprocessed LTR34 together with the LTR30 and LTR32GT structural analogues helped to establish the important implication of the LTR 5 0 AC3 0 dinucleotide, alongside Mg 2+ , in the protein-DNA complex formation. The binding of the drugs to LTR ends mimics that of the peptide loop, showing that these can compete with the peptide loop in the recognition process of 5 0 AC3 0 . Actually, numerous data reported and    [54,59]. In the recent crystal structures of the PFV intasomes [16] the role of the dinucleotide 5 0 AC3 0 in the shaping of the loop into a helix is rather clear. The change of the backbone conformation could help the carboxylate group of the Glu152 catalytic residue to adopt a spatial orientation permitting the metal assisted interaction with the DNA site. Our studies are also consistent with previous studies performed in solution, indicating that inhibitors bind to the 5 0 AC3 0 overhang in the intasome [54]. Note, that the binding of drugs to the 5 0 AC3 0 dinucleotide is not displayed in the crystal structure of the PFV intasome-drug complex. In this, the 5 0 AC3 0 dinucleotide continues to share interactions with the peptide loop, as those observed with the intasome alone [16,22].
Binding of RAL and TB11 to the K156-loop140-LTR32 complex

Fluorescence anisotropy
As shown above, in the presence of Mg 2+ , TB11 interacted with good affinity to the K156-loop140 (K d : 7.5 lM) (Fig. 5), but with a higher affinity to LTR32 (K d : 0.1 lM). RAL did not interact with the K156-loop140 but bound with very high affinity to LTR32 (K d : 6 nM). At the same time, 5 0 AC3 0 was required for the binding of the K156-loop140 to LTRs. Thus, in all these events, the role of 5 0 AC3 0 appears determinant, but that of the peptide loop is not as clear. We attempted to answer this question using new fluorescence anisotropy experiments where the peptide was incorporated into the complex. The titration profiles of the preformed LTR32-K156-loop140 complex by TB11 and RAL showed that inhibitors did not dissociate but rather bound the complex (Fig. 8). However, the K d value for binding of RAL to the binary complex was three times greater compared with that of LTR32 alone (16 nM versus 6 nM) (Figs 7 and 8) and rather resembles that reported for RAL binding to intasome (K d 19 nM) [28]. Comparatively, the K d1 value for TB11 binding to the binary complex was now 14 times greater than that of LTR32 alone (1.4 lM versus 0.1 lM) (Figs 6 and 8). Concerning the binding of TB11 to the low-affinity site, this was only weakly decreased (K d2 : 0.4 mM versus 0.6 mM). We still did not find a second binding site to RAL.
Overall, the addition of the peptide loop to the DNA-inhibitor complex increased the stability of the specific complex for both RAL and TB11. The affinity of the inhibitor for the LTR-peptide complex was weaker than that for the LTR alone. The loss of TB11 affinity was greater compared to RAL.

Gel electrophoretic retardation assay
The above fluorescence anisotropy experiments showed that the weak inhibitor TB11 had strong and weak affinity binding sites on both the unprocessed and processed LTRs, while the strong inhibitor RAL had only a very strong site. What then, is the origin of this second site? We suspect intercalation into DNA base pairs, so we performed gel electrophoretic retardation and topoisomerase I relaxation assays. These two techniques used together are well suited for detecting drug intercalation into DNA base pairs [60,61]. DNA intercalators such as anthracyclines inhibit DNA replication and are used in the chemotherapeutic treatment of cancers due to their toxic effects on tumour cells.  Fig. 8. Fluorescence anisotropy titration of the LTR32 + K156-loop140 complex by TB11 (red line) and RAL (black line). Samples were dissolved in phosphate buffer pH 6, 10 mM, and spectra were recorded at 5°C.
As expected, only TB11, which displays a second binding site in the present work, retarded the migration of the pBR322 plasmid in the gel electrophoretic retardation experiments (Fig. 9), a property that is attributable to drugs that intercalate into DNA base pairs [62,63]. The mobility of DNA decreases proportionally to the number of drug molecules inserted into DNA base pairs (Fig. 9A). The binding of TB11 coincided with a dramatic reduction in fluorescence intensity, suggesting that TB11 moves BET from its intercalating site and is therefore a good intercalator. The maximum intercalation was achieved at a drug/DNA molar ratio of 300, with the higher TB11 concentration being 30 lM. The intercalation was weakly cooperative and presented some distributive characteristics. In the case of nonretarded and fully retarded interconverting species, we did not observe a gradual decrease in DNA bulk mobility [47]. Unlike TB11, RAL did not exert a gel retardation effect, which highlights its inability to intercalate into DNA base pairs as already suggested by the absence of a second binding site for this drug in fluorescence anisotropy experiments (Fig. 9B).
The incubation of circular DNA together with topoisomerase I and unwinding ligands (intercalating drugs or some proteins) yields DNA of different superhelix densities. In fact, the torsional strain introduced by drug intercalation results in progressive DNA unwinding. The event can be demonstrated, first, by completely relaxing the BET treated DNA with an excess of topoisomerase I. Topoisomerase I removes all of the supercoils in each DNA molecule, regardless of the present degree of unwinding. Extraction of BET reintroduces supercoiling into DNA in direct proportion to the unwinding initially produced, while the control DNA, which has not been exposed to the unwinding agent, remains completely relaxed. Thus, the DNA unwinding induced by intercalating drugs is demonstrated by an increased supercoiling of final products. The results indicated that topoisomerase I completely relaxed the plasmid in the absence of BET or drug. BET at 10 lgÁmL À1 produced a complete supercoiling of DNA, while TB11 produced a similar effect at only 3 lM (drug: DNA molar ratio 300 : 1). At the same time, RAL was completely unable to unwind the plasmid, even at a high concentration (data not shown). The above results demonstrate that the intercalation of the weak inhibitor TB11 into DNA base pairs occurs, while this effect is absent in the strong inhibitor RAL.

Discussion
RAL and TB11 are DKA derivatives that act on HIV-1 replication mainly by inhibiting the ST activity of IN. RAL is an INSTI that has been incorporated into the group of drugs extensively used in highly active antiretroviral treatment (HAART) [8,64]. In contrast, TB11 has been abandoned for its low bio-availability and high toxicity ( Table 1). The two drugs both bind to the 3 0 Pr at the LTR ends, but, whether the LTR is processed or unprocessed, RAL is always the stronger binder. Yet, the two drugs have a neat preference for the processed site. RAL is a potent inhibitor that binds with high specificity to the 3 0 Pr site, while TB11 is a rather poor inhibitor that has significant side effects. For the synthesis of new compounds with improved performance, we thought it interesting to learn about the physicochemical properties of the two different compounds. The comparison was mainly focused on: (a) their binding to viral LTR ends, which emerged as possible primary targets of the INSTIs [27]; (b) the impact of the Mg 2+ ions in the recognition of the 3 0 Pr site by drugs and the stabilization of the substrate-drug complex; (c) the influence of the single stranded 5 0 AC3 0 dinucleotide at the end of the processed LTR on the binding of inhibitors as this influence was not so visible in the available crystallographic studies; (d) the origin of the lower specificity and greater toxicity of TB11 compared with RAL; and (e) the good correlation between the affinity of the drug for the processed LTR bound to Mg 2+ and their IN inhibition efficacy. Such a dependence of IN inhibition versus affinity for processed LTR constitutes a strong argument in favour of viral DNA ends as a primary site of drug interaction. Obviously, the 3 0 processed LTR-drug complex does not preclude the formation of a ternary complex with the enzyme and the induction of resistance mutations by the drug.

Binding of the catalytic loop and a4 helix to the substrate DNAs
The binding affinity of the K156-loop140 peptide (peptide mimicking the stabilized a4 helix connected to the catalytic loop) to the LTR ends is similar to that of inhibitors. In both cases, the binding was tighter with processed LTR than unprocessed LTR, and it was suppressed when the AC5 0 /GT3 0 was removed (bluntended LTR30). Upon binding to the LTR ends, the K156-loop140 peptide undergoes a helix conversion that mainly affects the flexible loop. The phenomenon is more pronounced when the binding partner is the processed LTR32 rather than the unprocessed LTR34. Thus, there is a clear correlation between binding and helix stabilization; the stronger the peptide affinity for the DNA substrate, the greater the peptide helix stabilization. This result is consistent with the crystallographic data on the CCD and intasome structures found in the literature [16,38], which indicate that there are residues converting into helices located in between the N terminus of the a4 helix and the end of the catalytic loop. Actually, the stabilization into a helix by a peptide through specific interactions with DNA is not an uncommon event and may result in a conformation allowing the stimulation of a biological response [65]. Here, the helix stabilization within the catalytic loop at the junction with the a helix is functionally important as it properly orients the carboxylic group of the Glu152 side chain to ligate a Mg 2+ cation ('B' metal ion in [22,66]). Obviously, this implicates other structural and dynamic factors that are difficult to apprehend, including the formation of a hydrogen bonding network with water molecules and monovalent cations [67].

Role of the 5 0 AC3 0 dinucleotide
The results provided by the four LTR oligonucleotides (LTR34, LTR32, LTR32-GT and LTR30) indicated that the 5 0 AC3 0 dinucleotide at the LTR34 and LTR32 extremity is involved in the binding of both the K156-loop140 peptide and inhibitors. This is especially true in the case of the processed LTR32, where access to the binding site was easier than the unprocessed LTR34, although the latter has favourable base fraying at its end. Indeed, our results confirm the prominent role of 5 0 AC3 0 in the association of IN to DNA, as reported in several biochemical and spectroscopic papers [27,54,68] as well as in the recently published X-ray crystallographic data provided by the analysis of PFV intasome complexes [16].
On the origin of Mg 2+ ions in the complexes and their role in the specific binding and cleavage of substrates Our results emphasize an important participation of Mg 2+ ions in stabilizing the binding of inhibitors to LTR ends (Fig. 10). In fact, the X-ray crystallographic data demonstrated the presence of a pair of divalent metal ions in the binary DNA-IN complexes [16] and the ternary complex DNA-IN-inhibitors, but the latter tell us nothing about the energy of binding. Conversely, in most of the crystal structures of unbound INs reported to date, the active site contained a single metal ion. However, for both HIV-1 and PFV, no crystal structures of the entire enzyme alone are available. The catalytic site of HIV-1 IN CCD crystal structures [38,40,48,69] shows a single divalent metal ('A' metal ion) chelated to the Asp 64 and 116 side chains, while the carboxylic group of the Glu 152 is always free of interaction. The crystal structures of the bi-domain proteins, Nter-CCD [70] and CCD-Cter [71], provided a similar result and do not clarify the source of the second cation in the DNA-IN and DNA-IN-drug complexes.
According to a molecular dynamics calculation, the second Mg 2+ ion ('B') could be brought into the complex with the enzyme by the processing site of the viral DNA [72]. The crystal structures of naked oligonucleotides [31,73] and the molecular dynamics calculations based on recent crystal structures [33] agree with the above results. They provide evidence for a marked binding preference of Mg 2+ to purine-purine dinucleotide sequences such 5 0 AG3 0 , the latter offering the best electrostatic environment to recruit divalent metal cations [31,74]. In the viral LTR substrate, 5 0 AG3 0 coincidentally corresponds to the cleavage site and could capture a Mg 2+ ion. This, so-called 'A' would be pre-positioned on the cleavage site. From this position, it could contribute to the recognition and cutting of the scissile phosphodiester linkage by the enzyme active site bound to the Mg 2+ ion 'B' (Fig. 10). Notably, the processing site of the cleavable 5 0 AG3 0 dinucleotide is included in a 5 0 CAG3 0 trinucleotide. The cytosine phosphate of such a triad has been shown to participate in bridging interactions with divalent metal cations [33]. It is certainly no accident that the cleavage site 5 0 AG3 0 and the highly conserved dinucleotide 5 0 CA3 0 are combined within the processing site to make it highly specific. Moreover, the 5 0 CA3 0 step is extremely malleable and may help the 5 0 AG3 0 step to accommodate its phosphate bond in a cleavable conformation within the IN active site [75][76][77][78]. In fact, the 5 0 CA3 0 step is often found in DNA substrates, where it could pilot the cleavage of the connected steps by enzymes belonging to retroviruses, transposable phage Mu and also transposable elements [55,68,78,79].
Moreover, the fraying that is common in the last two base pairs of DNA can be increased at the HIV-1 LTR ends by the motions affecting the flexible 5 0 CA3 0 dinucleotide [68], thereby explaining why drugs also bind to unprocessed LTRs.
To summarize, our results suggest that the binding of one Mg 2+ ion to the 5 0 CAG3 0 triad could make a particular scaffold to ensure a specific recognition and electrostatic fit by the enzyme or the drug. Thus, in the intasome complexes, one Mg 2+ ion could be supplied by the enzyme active site and the other by the DNA cleavage site. The ability of the DNA duplex end to separate into two strands [80,81] permits the cleavage site to adopt the in-line geometry required for bond cutting [82]. Thus, the base pair fraying is certainly an important factor permitting the best fit of inhibitors on the active strand of the cleavage site. Obviously, the deletion of the 5 0 GT3 0 by the 3 0 Pr reaction permits a better Mg 2+ assisted capture of the drug by the liberated 5 0 AC3 0 dinucleotide strand and the adenine-3 0 OH group [16,83].
A relationship between drug binding to processed LTR and ST inhibition X-ray studies [16,22] have shown that apart from metal assisted interactions with the DNA substrate, inhibitors also share direct interactions with DNA, thus enforcing the stability of the DNA-drug complex. RAL and its antiviral congeners use their halogenated aromatic ring to make contacts with the highly conserved 5 0 CA/TG nucleotides, that is, the adenine phosphate and the cytosine base of the conserved 5 0 CA3 0 , as well as the guanine of the facing 5 0 TG3 0 [16,27,84]. These interactions contribute to the good correlation observed between the affinity of inhibitors for the processed LTR32 and ST inhibition, giving credence to the idea that the DNA substrate is the drug target. Such a mechanism of inhibitors first binding the substrate rather than the enzyme is scarce in the literature.
The extra binding sites and the side effects of TB11 In addition to its interaction on the 3 0 Pr sites at LTR ends, the TB11 molecule has other binding sites including: (a) the intercalation into DNA base pairs as revealed by both fluorescence spectroscopy, gel electrophoretic retardation assay and plasmid unwinding assay, and (b) interactions with peptides derived from the enzyme active site. The fluorescence variation induced by the TB11 intercalation into DNA base pairs (low-affinity site) indicated a nearest neighbour exclusion mechanism [88]. The obtained K d value (10 À4 to 10 À5 M) agrees with the TB11 concentration required to produce the unwinding of supercoiled DNA in gel migration assays. Of note, the TB11 K d was similar to those generally obtained with the well-known antitumour drugs anthracyclines and acridines. These drugs are poisons targeting the topoisomerase II enzyme [89] and provoke the death of tumour cells but also healthy cells. Thus, intercalation into DNA base pairs is an event that can contribute to the toxicity of TB11.
In addition, the binding of a fraction of TB11 to the a4 helix and to the loop140 belonging to the enzyme catalytic site in the enzyme tetramer could be the foun- dation of the 3 0 Pr inhibitory activity of this compound. However, TB11 could also bind to the a4 helix and loop 140 at the surface of the two nonfunctional monomers of the tetramer. This binding is ineffective for IN inhibition, but it diminishes the number of TB11 molecules available for binding to the inhibiting site.

Conclusion
The results presented here, comparing the efficient RAL with the weakly efficient TB11 confirm the idea that the viral DNA substrate is the primary binding site of INSTIs (Fig. 10) [27]. This is strongly supported by the finding of a good correlation between the binding affinity of these inhibitors for the processed viral DNA and their efficacy towards the ST inhibition. In fact, successful antiviral drugs such as RAL optimally fit the cavity opened by 3 0 Pr reaction at the viral DNA end, where they share strong electrostatic interactions with the Mg 2+ ions and van der Waals forces with the DNA nucleotide atoms, essential for the success of the ST reaction. Results prove that the 3 0 Pr-Mg 2+ -site complex is in itself a highly specific binding site for RAL and efficient INSTIs, especially as a cavity similar to that left open by the 5 0 GT3 0 deletion at the viral DNA end, certainly does not exist away from the ends of DNA and cannot be a competitive site. In comparison, the smaller TB11 is unable to fully occupy this space and make all the interactions of the 3 0 Pr site. The issue of the origin of the metal cofactors ('A' and 'B'), found at the IN-LTR interface of the intasome crystal [16] is also interesting. There is no doubt that the 'A' ion which interacts with the carboxylate group of the crucial Asp 64 and Asp 116 residues in some crystal structures of the DNA-free CCD, is brought in the IN-DNA and IN-DNA-drug complexes by the enzyme active site itself. Concerning the origin of the 'B' ion, we assume, in agreement with the hypothesis of Lins et al. [72], that it is carried in the intasome by the DNA substrate itself. When the DNA substrate is the unprocessed LTR, the 'B' ion could be pre-positioned on the 5 0 CAG3 0 triad of the active strand, ready to capture INSTI drugs, through mainly electrostatic interactions.