Deciphering structure-activity relationships in a series of Tat/TAR inhibitors

A series of pentameric “Polyamide Amino Acids” (PAAs) compounds derived from the same trimeric precursor have been synthesized and investigated as HIV TAR RNA ligands, in the absence and in the presence of a Tat fragment. All PAAs bind TAR with similar sub-micromolar affinities but their ability to compete efficiently with the Tat fragment strongly differs, IC50 ranging from 35 nM to >2 μM. While NMR and CD studies reveal that all PAA interact with TAR at the same site and induce globally the same RNA conformational change upon binding, a comparative thermodynamic study of PAA/TAR equilibria highlights distinct TAR binding modes for Tat competitor and non-competitor PAAs. This led us to suggest two distinct interaction modes that have been further validated by molecular modeling studies. While the binding of Tat competitor PAAs induces a contraction at the TAR bulge region, the binding of non-competitor ones widens it. This could account for the distinct PAA ability to compete with Tat fragment. Our work illustrates how comparative thermodynamic studies of a series of RNA ligands of same chemical family are of value for understanding their binding modes and for rationalizing structure-activity relationships.


Introduction
Today, it is becoming increasingly clear that non-coding RNAs play essential roles in a variety of fundamental cellular and pathological functions. For a multitude of diseases, targeting such RNAs constitute an alternative strategy that complements traditional protein-based targeting. These RNAs adopt folded hairpin structures leading to binding pockets suitable for the formation of specific interactions with their natural counterparts (RNAs, proteins, metabolites …). Small molecules able to selectively inhibit these interactions are of considerable interest both as therapeutic agents and as chemical probes. The ability of some antibiotics to exert their activity by binding to defined regions of bacterial ribosomal RNAs provides the proof of concept that it is possible to target specifically some RNA structures with small molecules (Poehlsgaard & Douthwaite, 2005). However, even if numerous RNA-directed small ligands, mainly identified by standard or virtual screening approaches, have been reported in the literature so far (see reviews (Blond, Ennifar, Tisné, & Micouin, 2014;Disney, Yildirim, & Childs-Disney, 2014;Guan & Disney, 2012)), none of them has yet led to commercially approved drugs, except antibiotics. Efforts for the discovery of small synthetic molecules that combine high affinity and selectivity for a particular RNA target should be pursued. A better knowledge of the factors that govern small molecule-RNA recognition is therefore of great importance to achieve this goal. For several years, an extensive research has been dedicated to the highly conserved HIV-1 TAR RNA hairpin fragment (Blond et al., 2014;Massari, Sabatini, & Tabarrini, 2013;Yang, 2005), which plays a crucial role in the viral transcription step of HIV via its complexation with the transactivating transcription (Tat) protein and cellular factors (Stevens, De Clercq, & Balzarini, 2006). Even if up to now these efforts have been unsuccessful in leading to an antiviral drug onto the market, the TAR RNA fragment remains nevertheless an excellent model to better understand the main molecular forces that drive RNA-ligand associations (Kumar & Maiti, 2013;Suryawanshi, Sabharwal, & Maiti, 2010).
Previously, we reported a comparative TAR interaction study with three series of original ligands called PAAs "Polyamide Amino Acids" (α-PAA, β-PAA and C-α-PAA). These compounds were trimeric structures, each unit comprising an amide backbone (2-aminoethylglycyl or 2-aminoethyl β-alanyl) onto which an amino acid residue (L-α or L-β Phenylalanine, Arginine and Lysine) is condensed (Figure 1(A)). In addition to the identification of some lead structures, this study also revealed that the nature of the PAA (α-, βor C-α-PAA) and to a lesser extent, its sequence, had a large impact both on the TAR binding mode and on the ability to displace a preformed Tat/TAR complex (Pascale et al., 2013). In an attempt to increase both TAR affinity and Tat competitor behavior, we have designed a series of pentameric C-α-PAA deriving from one tri-C-α-PAA lead, identified in our previous work. In this paper, we describe the penta-PAAs preparation as well as the TAR interaction studies in the absence and in the presence of a Tat fragment, by fluorescence and fluorescence resonance energy transfer (FRET) spectroscopies. TAR-PAA interactions have been further characterized by UV-melting and circular dichroism experiments and the RNA recognition site confirmed by NMR. A detailed thermodynamic analysis has allowed us to rationalize structure-activity relationships by discriminating two distinct groups of PAAs ligands that interact with TAR by two different interaction modes, which have been further supported by molecular modeling studies.

Fluorescence binding assays
Ligand solutions were prepared as serial dilutions by an epMotion automated pipetting system (eppendorf) in buffer A (20 mM HEPES (pH 7.4 at 25°C), 20 mM NaCl, 140 mM KCl, and 3 mM MgCl 2 ) at a concentration twice higher than the desired final concentration to allow for the subsequent dilution during the addition of the RNA solution. The appropriate ligand solution (30 μL) was then added to a well of a non-treated black 384-well plate (Nunc 237105), in triplicate. Refolding of the RNA was performed using a thermocycler (ThermoS-tatPlus Eppendorf ) as follows: the RNA, diluted in 1 mL of buffer A, was first denatured by heating to 90°C for 2 min then cooled to 4°C for 10 min followed by incubation at 20°C for 15 min. After refolding, the RNA was diluted to a working concentration of 10 nM through addition of the appropriate amount of buffer A. The tube was mixed and 30 μL of the RNA solution was added to each well containing ligand. This subsequent dilution lowered the final RNA concentration to 5 nM. The fluorescence was measured on a GeniosPro (Tecan) with an excitation filter of 485 ± 10 nm and an emission filter of 535 ± 15 nm. Each point was measured 5 times with a 500 μs integration time and averaged. Binding was allowed to proceed at least 30 min at room temperature to achieve equilibrium.
To study the temperature dependence, the plates were incubated after 30 min equilibrium at different temperature ranging from 5°C to 35°C.
The salt dependence was studied in 20 mM HEPES (pH 7.4 at 25°C), 20 mM NaCl, 3 mM MgCl2 with the KCl concentration varied between 70 and 250 nM.
For competitive experiments in the presence of a dsDNA, a 15-mer sequence (5′-CGTTTTTATTTTTGC-3′) and its complement, annealed beforehand, were added to buffer A to obtain a 100-fold nucleotide excess over TAR RNA (900 nM duplex; 5 nM RNA). For competitive experiments in the presence of tRNA, a mixture of preand mature yeast tRNAs (containing >30 different species from baker's yeast (Saccharomyces cerevisiae, Sigma, type X-SA) was added to buffer A to obtain a 100-fold nucleotide excess over TAR RNA.

FRET displacement assays
Ligand solutions and RNA (40 nM working solutions) were prepared as described above in buffer B [50 mM tris buffer (pH 7.4 at 25°C), 20 mM KCl, and .005% Tween 20]. Labeled Tat peptide (40 nM in buffer B) was mixed to an equal volume of TAR RNA for 20 min at room temperature to form the Tat/TAR complex before adding the ligand. The appropriate ligand solution (30 μL) was added to a well of a 384-well plate, in triplicate, followed by 30 μL of the Tat/TAR solution. Fluorescence was measured as described above after 30 min of incubation at room temperature.
2.4. Temperature-dependent UV spectroscopy (UV melting) Thermal denaturation scans were obtained using a Cary 300 (Varian) spectrophotometer equipped with an electrothermal multicell holder. Absorbance vs. temperature profiles were recorded at 260 nm. After structuration of TAR RNA and incubation (1 h) with the corresponding ligand, the temperature was raised from 20 to 90°C, with a heating rate of .5°C/min. Thermal denaturation studies were carried out at 2 μM TAR RNA with 2 μM PAA or without PAA (TAR alone). The experiments were performed in buffer C [10 mM sodium cacodylate, 10 mM NaCl (pH 7.5), and .1 mM EDTA]. The melting temperature (Tm) value was taken as the midpoint of the melting transition as determined by the maximum of the first-derivative plot with Prism software.

CD study
CD measurements were performed with a Jasco J-810 spectropolarimeter equipped with a Jasco PTC 423S Peltier temperature controller. Samples were prepared in buffer D [20 mM potassium phosphate buffer (pH 7.4 at 25°C), 10 mM NaCl, and 1 mM MgCl 2 ]. Spectra were obtained at 3 μM RNA or PAA (for individual spectra) or at a molar ratio of 1:1, 1:2, and 1:5 RNA:PAA for the complexes. Spectra were recorded at 20°C from 360 to 200 nm at 1-nm intervals, with an integration time of 4s and a 50-nm/min speed. CD scans were repeated five times and then averaged and corrected by the subtraction of the buffer background.
Salt dependence of K D was analyzed by the following equation: where Knel is the dissociation constant at the standard state in 1 M KCl, Z is the number of ions displaced from the nucleic acid, and ψ is the fractional probability of a counterion being thermodynamically associated with each phosphate of the RNA number of cations. Knel and Zψ were treated as fitting parameters. For thermodynamic analysis, ΔG°values were plotted vs. T. Nonlinear regression using the three-parameter fit in Prism 5 was used to fit the following equation to the data: where Tr is a constant reference temperature (in our study Tr = 293.15 K), and the three fit parameters are DH Tr the change in enthalpy upon binding at Tr; DS Tr , the change in entropy upon binding at Tr; and ΔCp, the change in heat capacity. ΔCp was assumed to be independent of temperature; inclusion of a ΔCp/ΔT term in the analysis did not improve the quality of the fits and gave larger standard errors for the returned parameters.
DH T and DS T were calculated from the result derived from the fitting of the curve ΔG°values vs. T by the Equations (3) using: where DH T is the change in enthalpy upon binding at T (25°C) and DS T is the change in entropy upon binding at T.
For compensation analyses, graphs of ΔH°values vs. TΔS°ones, ΔG°vs. ΔH°, and ΔH°vs. ΔCp were plotted. Linear regression in Prism 5 was used to fit the following Equations (4) to the data:

Molecular dynamics simulations
AutoDock Vina (Trott & Olson, 2010) was used for generating 40 initial complex conformations for each PAA using PDB ID 2kx5 as the receptor structure.
Starting from the best scored docking poses a molecular dynamics (MD) system was prepared using explicit solvent for its equilibration and free energy evaluation. AM1-bcc partial charges were assigned to each PAA using antechamber from the Amber suite. Then, the complexes were prepared with tleap using the standard ff10 Amber force-field (AMBER12, University of California, San Francisco). Each structure was minimized in two steps: first, all residues except solvent were held fixed with a restraint force of 100 kcal/mol-Å 2 . Steepest descent minimization followed by conjugate gradient was performed using 2500 steps in both cases. The same minimization protocol was applied in 10,000 steps without positional restraints. After minimization, the temperature was raised from 0 to 300 K in 100 ps using constant volume dynamics. A restraint force of 10 kcal/mol-Å 2 was applied to the complex and SHAKE was turned on for bonds involving hydrogen atoms. Then, 200 ps of constant pressure dynamics were applied for density equilibration. Finally, each production run was performed using Langevin dynamics with a collision frequency of 1 ps −1 . An atom-based long range cut-off of 9 Å was applied during all the simulations.

Results and discussion
The promising results that we previously obtained with trimeric-C-α-PAA structures as Tat/TAR inhibitors prompted us to design longer structures in order to increase their efficiency and their activity (Pascale et al., 2013). Starting from one of the most promising compound, namely "C-α-FRF", we arbitrarily decided to elongate it from its N-terminal extremity to form a pentameric PAA, using three C-α-PAA monomers (R, K, and F). As already noticed (Bonnard et al., 2010) the two basic residues (K and R) were selected for their ability to form hydrogen bonds and/or electrostatic interactions. Moreover, it is well known that arginine-rich peptides and peptidomimetics incorporating guanidinium motives recognize TAR at the major groove, between the bulge and the loop, as does the basic domain of the Tat protein ( 49 RKKRRQRRR 57 ) (Blond et al., 2014;Massari et al., 2013;Richter & Palu, 2006). Furthermore, the aromatic F residue was selected for increasing π-stacking and van der Waals interactions. Thus, 9 penta-C-α-PAAs "YX-FRF" (X and Y = "R" or "F" or "K" PAA monomer) were synthesized (Ia-c, IIa-c, and IIIa-c, Figure 1).

Chemistry
Penta-C-α-PAAs were prepared following a standard solid-phase strategy, using a β-alanine functionalized MBHA resin and starting from fully and orthogonally N-protected C-α-PAA monomers (Pascale et al., 2013) (Scheme 1(A)). This procedure includes: (i) successive elongation/deprotection steps involving selected monomers, (ii) acetylation of the last residue, and (iii) acidic cleavage of the protecting groups and simultaneous release from the resin (Scheme 1(B)). Crude penta-C-α-PAAs were obtained in high HPLC yields (from 70 to 98%) and then purified by semi-preparative HPLC. Structures were confirmed by HRMS experiments (Supplementary Table S1).

TAR affinity and FRET displacement assays
With the penta-C-α-PAAs in hand, we first evaluated TAR affinities (K D ) by monitoring the fluorescence change of a fluorescently labeled (Alexa 488) TAR fragment (18-44 nt), as previously described (Pascale et al., 2013). We also assessed their ability to displace a Tat fragment from a preformed TAR/Tat complex (IC 50 ) via a FRET assay, using a fluorescein-tagged Tat peptide fragment (amino acids 48-57) and a Dabcyl-labeled TAR fragment (18-44 nt) (Murchie et al., 2004) Dissociation constants (K D ) and inhibitory concentrations (IC 50 ) associated with each PAA are given in Table 1 and compared to the trimeric C-α-FRF precursor, used as a reference.
All penta-C-α-PAAs strongly bind to TAR with similar affinities (48.6 nM < K D < 116.8 nM) but they do not have the same ability to displace the Tat/TAR complex. Clearly, they can be divided into two distinct groups: the first one (series I: YF-FRF), "F-rich", has no ability to displace the Tat/TAR complex up to 2 μM, while the second one (series II: YR-FRF and series III: YK-FRF), "R/K-rich", displays low IC 50 values, except IIa which reveals an intermediate behavior. Thus, the nature of the penultimate residue X in YX-FRF pentamers has a major influence on IC 50 values. Similarly, the nature of the last PAA Y has also an impact on IC 50 values (see Table 1, series II and III). Therefore, it seems that the N-terminal extremity of penta-PAAs mainly governs their mode of interaction with TAR which drives or not the displacement of Tat protein. In addition, as previously reported with tri-PAAs, there is no correlation between K D and IC 50 values. It is also worth noting that even though the increase in affinity is moderate (2-to 5-fold) compared to the original trimeric structure C-α-PAA (see Table 1), the impact on IC 50 values is far more spectacular, from 67 to >2000 times (in the case of the most active compound IIIb). This underscores the fact that the determination of a K D value is not sufficient to predict the ability of a ligand to inhibit the interaction between an RNA fragment and its cognate partner.
To assess the specificity of penta-C-α-PAAs, K D values of some representative compounds of series I and II were measured in the presence of a large excess of tRNA (K D ′) and dsDNA (K D ′′) ( Table 1). All penta-PAA retain a specificity for TAR in the presence of both tRNA and DNA, this specificity being higher in the latter case. Moreover, it is noteworthy that penta-C-α-PAA are more specific than tri-C-α-PAA (1.7 < K D ′/K D < 6.5; 1.8 < K D ′′/K D < 2.4) (Pascale et al., 2013), whatever their sequence. But as for trimers, penta-PAA containing F-rich sequences are less specific than PAA containing more cationic K/R-rich sequences (see PAA YFFRF vs. PAA YRFRF II in Table 1), likely indicating that cationic residues are preferentially involved in specific interactions rather than in ionic ones.
For better understanding the factors that account for the distinct behavior of penta-PAAs as Tat competitors, structural and thermodynamic studies were undertaken.

NMR experiments
Since tri-C-α-PAAs interact with TAR at the bulge level, it is likely that longer PAAs also bind TAR at the same site. However, we conducted NMR studies not only for verifying this hypothesis but also for comparing the mode of interaction for two PAAs, i.e. Ic and IIIb, representative of the two groups of PAAs ("F-rich" and "R/K-rich") as defined in Tat competition assays.
The titration of TAR RNA with PAA Ic or IIIb by 2D TOCSY experiments corroborated the specific interaction around the bulge for both compounds. Indeed, the overlaid TOCSY spectra with increasing amounts of Ic (Figure 2(A1)) or IIIb (Figure 2(A2)) are very similar as they both show large chemical shift changes (Δd ≥ .1 ppm) for U23, C24, U25, C29, C39, U40, and C41, smaller changes (Δd ≤ .1 ppm) for residues C30, C37, U38, and U42, and no significant changes for residues C18, C19, and U31 more remote from the bulge. The slight differences between the two ligands concern residues C30, U38, C41, and U42 which undergo larger chemical shift changes in IIIb than in Ic probably due to the higher affinity of IIIb for TAR. Comparison of the TOCSY experiments of penta-PAAs with those previously reported for a tri-PAAs sharing the same FRF sequence (4b in Pascale et al. (2013)) did not show any significant differences except that the large chemical shift changes (Δd ≥ .1 ppm) occur earlier (from Equation (1)) for penta-PAAs, probably reflecting once again the higher affinity of penta-PAAs vs. tri-PAAs for TAR RNA.
Stacked 1D NMR spectra of the imino resonance region of TAR in the presence of increasing amount of Ic or IIIb (Figure 2(B1) and (B2/)) are very similar and confirm the interaction of both compounds around the bulge as the imino protons of residues G21, G26, G28, and U38 undergo changes of chemical shifts (Δd ≥ .1 ppm) upon addition of ligands. At 25 μM, imino protons of these residues are in a slow rate of exchange at the NMR time scale. As the concentration of penta-PAA (IIIb and Ic) increases (75 μM) the rate of exchange remains slow for G28 and U38 while imino signals of G26 and G21 become coalescent. If we compare the RNA imino resonances spectra obtained by the penta-PAA (Ic or IIIb) titration with the ones previously reported with the tri-PAA 4b titration (Pascale et al., 2013) we can notice that penta-PAA binding seems to slow down the rate of exchange of the interactive imino protons. This could possibly reflect the higher affinity of the penta-PAAs for TAR (≈10-fold) as compared to the tri-PAAs.
Overall, these NMR data unambiguously reveal that the binding site of both penta-PAA compounds (Ic and IIIb) is centered on the bulge region of TAR. Although TOCSY experiments seem to indicate a stronger interaction for IIIb than Ic (some residues of IIIb display a larger chemical shift), these differences are not enough marked to explain the distinct ability of Ic and IIIb to compete with Tat.

Circular dichroism and UV melting studies
Even if penta-PAAs interact with TAR in the same region, their binding could induce distinct TAR conformational changes that could be related to their different ability to compete with Tat. Circular dichroism studies were undertaken to compare the TAR structural changes occurring upon binding of the two representative penta-PAA Ic and IIIb (Supplementary Figure S1). For PAA alone, only negligible CD intensity in the 200-350 nm region was observed, suggesting the lack of structuration or secondary shape. By contrast, the CD spectrum of TAR alone is characteristic of an A-form double helix, with a strong positive band at 260 nm, very sensitive to base-stacking, and a strong negative band at 210 nm, related to the A-form of the TAR RNA helical structure (Loret, Georgel, Johnson, & Ho, 1992). Addition of increasing concentration of PAA Ic or IIIb induces a strong reduction in ellipticity at 260 nm. This reflects that upon binding of Ic or IIIb, C24 and U25 residues of the bulge are exteriorized and concomitantly destacked as in the case of argininamide, Tat basic domain, and Tat derived peptides (Kumar & Maiti, 2013;Murchie et al., 2004;Tan & Frankel, 1992). Nevertheless, the stronger decrease in intensity associated with a red shift displacement for the 210 nm band observed with IIIb compared to Ic could suggest a more pronounced effect of the latter on the TAR helicity (Davidson, Patora-Komisarska, Robinson, & Varani, 2011). On the other hand, melting temperature studies conducted in the presence of Ic or IIIb at a 1:1 PAA/TAR molar ratio, shows that the binding of IIIb induces a slightly higher stabilization of the TAR structure than compound Ic (temperature increased by 4°C and 2°C upon binding, respectively (Supplementary Figure S2). However, it is unlikely that this small variation could account for the distinct ability of the two PAA to displace the Tat/TAR complex.

Thermodynamic studies
Thermodynamic characterization is essential for understanding molecular interactions and their impact in biological processes (Chodera & Mobley, 2013;Lane & Jenkins, 2000;Pilch, Kaul, Barbieri, & Kerrigan, 2003). To gain further insights on the TAR binding modes of penta-C-α-PAAs, thermodynamic binding profiles associated with each PAA/TAR equilibrium were determined. Enthalpy (ΔH°) and entropy changes (ΔS°) were calculated from Equations (1) and (2) (see experimental section) after determination of DG T at several temperatures (278-308 K). We also determined the electrostatic (ΔG°el) and non-electrostatic (ΔG°nel) components of the Gibbs energy according to the polyelectrolyte theory (Record, Zhang, & Anderson, 1998). In the case of the two representative compounds Ic and IIIb, the K D dependency on the ionic strength of the solution was studied over a range of KCl concentration from 70 to 250 mM. The results of thermodynamic analyses are summarized in Table 2.
At first glance, all penta-PAA/TAR equilibria display very similar thermodynamic signatures, with ΔG°varying from −39.6 to 41.7 kJ/mol. As for tri-C-α-PAA, non-electrostatic interactions dominate the binding by contributing to about 95% of the total binding free energy, whatever the PAA sequence (cf ΔG°nel/ΔG°r atio, Table 2). This strongly suggests that the majority of ammonium and guanidinium groups of penta-PAA interact with TAR RNA via specific hydrogen bonding and/or π-cation interactions rather than via ionic interactions with the phosphate backbone.
A closer analysis reveals that each series I-III constitutes in fact a three ligands-set distinct from the others. Indeed, examination of the following relationships: (ΔH°vs. TΔS°), (ΔG°vs. ΔH°), and (ΔH°vs. ΔCp) within each series (Figures 4(A)-(C)) points out strong correlations for two of them (I and III). Undoubtedly, the thermodynamic signatures of series I and III differ,   proving that their interaction mode is distinct. Thus for example, in series I, the ligand of the highest affinity (Ib) corresponds to the tighter one (the lower ΔH°and TΔS°of the series) whereas in series III, the looser ligand (IIIc, the higher ΔH°and TΔS°of the series) displays the best affinity. In these two series, the correlation between ΔH°a nd ΔCp demonstrates that a part of the heat capacity change reflects the degree of tightness of the complex. Thus, a loosening in the structure complex (ΔH°increase) is associated with a ΔCp increase.
Concerning series II, ΔH°and TΔS°values range in small intervals, as compared with the two other series. No linear correlation could be deduced, maybe because in all cases, compound IIc caused deviations. However, from a thermodynamic point of view, series II is closer to series III than to I.
As all penta-C-α-PAAs derive from the same C-terminal FRF sequence, these observations clearly demonstrate that the nature of the N-terminal residues has a strong impact on their TAR recognition mode and, consequently, on their ability to displace the Tat/TAR complex. Thus, compounds of series I, unable to displace the Tat/TAR complex (IC 50 > 2000 nM), have a thermodynamic behavior which clearly differs from R/Krich compounds of series II and III that are effective Tat competitors (IC 50 < 400 nM). On the other hand, in the case of I and III, the nature of the last PAA residue Y has an impact on ΔH°and TΔS°values, whereas it has only little effect in II.
For a better comprehension on the comparative binding modes of the two PAA models, i.e. "RFFRF" Ic and "KRFRF" IIIb, we synthesized two truncated derivatives devoid of the F-β-alaninamide moiety, namely tetra-C-α-PAA "RFFR" and "KRFR", and compared their thermodynamic profiles with Ic and IIIb, respectively (Table 3).
Even if the TAR affinity of RFFR (F-rich) is higher than that of KRFR (K/R rich), its RNA complex is considerably looser than the one formed with KRFR, as demonstrated by ΔH°, TΔS°, and ΔCp values. This is in line with our previous results concerning F-rich and K/R-rich tri-PAAs and likely means that for F-rich PAAs, interactions are fewer and/or less optimized than for K/-rich PAA. Lengthening the loose tetra-RFFR binder by adding the F-β-alaninamide moiety (leading to RFFRF Ic) fails to maximize interactions and slightly destabilizes the complex, as demonstrated by the enthalpy loss and the entropy gain of 6.2 and 6.7 kJ/mol, respectively. By contrast, adding the F-β-alaninamide moiety to the tight tetra-KRFR ligand (leading to KRFRF IIIb) induced a considerable enthalpy loss of 43.4 kJ/mol, overbalanced by an entropy gain of 47.6 kJ/mol. We rationalized this thermodynamic behavior by hypothesizing that only the N-terminal tetrameric sequence KRFR tightly interacts with TAR in the interaction site, whereas the C-terminal moiety (i.e. F-β-alaninamide) remains free and exposed to the solvent. In such case, the mobility of this unbound fragment would result in the destabilization of the whole interaction network, leading to a decrease in tightness at  the ligand/RNA interface. This phenomena would result in less unfavorable TΔS°(due to greater mobility) and less favorable ΔH°(due to weaker interactions) than for the corresponding tetra-PAA, as observed here. To support these hypotheses, we performed a molecular modeling study concerning the TAR interaction of these two penta-PAAs.
3.6. TAR molecular recognition of penta-C-α-PAA Ic and IIIb MD is a powerful tool to explore molecular complexes formation at an atomistic level and analyze the conformational energy landscape accessible to these molecules. Such protocols have been applied in this study in order to predict the interaction modes between C-α-PAA and TAR that may complement experimental data. Visual inspection of MD results is in good agreement with NMR data, which suggested a preferred binding site in the bulge region of TAR for both Ic and IIIb compounds. However, the orientation differs from both complexes: IIIb points towards the inter-helical junction of the bulged TAR, whereas Ic provides a more packed conformation and buries into the TAR major groove similarly to previously reported structures (Davidson et al., 2009(Davidson et al., , 2011. The dynamic behavior of Ic and IIIb . F-β-alaninamide and acetyl terminus (in yellow) are framed. In the two cases, F-β-alaninamide is free of interaction. For IIIb, it remains exposed to the solvent while the rest of the penta-PAA buries into the TAR major groove. The bulge closure distance has been calculated as N2(G34)-N4(C24) distance for Ic and N2(G32)-N4(C24) distance for IIIb. (B) Main interactions provided by MD simulations for Ic (a-d) and IIIb (e-i). Henceforth, PAA are numbered as Ac-R1-F2-F3-R4-F5 and Ac-K1-R2-F3-R4-F5. HIV-1 TAR is represented as a white surface with red nucleotides and PAA are represented in green. On the one side, Ic interacts (a) with C24 into the bulge region through a F2 π-stacking interaction. R4 interacts at the same time with (b) phosphate U23 group and (c) F3 via intra-residue cation-π stacking interaction. (d) Arginine R1 stacks with C30 providing another cation-π interaction which stabilizes the complex. On the other side, IIIb (e) stacks with C24 via a R2 cation-π interaction. (f) Ac group buries into the major groove and interacts with the 2′OH group of U23. (g) G33 also provides a hydrogen bond interaction with R4 through the 2′OH group. (h) K1 provides charge-based interactions with the RNA backbone. (i) F3-NH 3 + group (yellow arrow) anchors to the TAR opening, which contains a high negative charge density.
complexes along the MD is noteworthy. Analyses of the full trajectory suggest that Ic widens the bulge region up to 15.2 Å while IIIb produces a contraction of this region up to 9.0 Å with subsequent RNA helicity loss ( Figure 5(A)). According to these results, both compounds induce non-negligible TAR conformational changes, as pointed out previously by CD spectra. In addition to major structural changes of TAR upon ligand binding, MD confirms the positive influence of the N-terminal sequence since F-β-alaninamide terminal moiety remains exposed to solvent during almost all the trajectory. Particularly, molecular modeling results suggest a remarkable difference in the exposition to solvent of the F-β-alaninamide moiety in PAA Ic and IIIb, due to the conformational changes in the RNA previously described. Moreover, no specific interactions have been characterized for this unbound fragment that might contribute to ΔHº. A closer inspection of the complex trajectories suggests that specificity is clearly achieved by side-chain interactions based on π stacking, cation-π stacking, and charge-based interactions as well as the orientation of the compound along the major groove ( Figure 5(B)). Compound Ic buries into the major groove of TAR by means of intimate charge-based contacts. Interestingly, U25 is flipped out from the RNA complex and drowns down the whole UCU loop. This is counterbalanced by C24 stabilization into the bulge region through a F2 π-stacking interaction and R4 charge-based interaction with the phosphate group of U23. R4 also yields intramolecular contacts with F3 via a cation-π stacking interaction. This interaction is reproduced by R1 with the stack of the guanidinium group on top of C30, which significantly contributes to complex stabilization. Despite the sequence dissimilarity between compounds Ic and IIIb, both provide similar interactions with TAR. For instance, C24 stacks with R2 via a cation-π interaction, while the acetyl moiety buries into the major groove of the apical loop and interacts with the 2′OH group of U23 via hydrogen bonding. The same interaction is hypothesized for R4 through the sugar moiety of G33. Moreover, carbonyl and ammonium groups of penta-PAA backbone significantly contribute to binding through electrostatic interactions. For instance, K1 provides a charge-based interaction with the RNA backbone and F3 anchors to the C30/U31/G34/G36 cluster.
To conclude, these molecular modeling studies validate our assumptions based on experimental data. Thus, contrary to neomycine B, which binds in the minor groove at the junction between the bulge and the lower Scheme 1. (A) Protected C-α-PAA monomers used for (B) the solid-phase synthesis of penta-C-α-PAAs. stem of TAR and allosterically inhibits Tat binding (Lu et al., 2011), both penta-PAA Ic and IIIb recognize TAR at the major groove, between the bulge and the loop, as does the arginine-rich domain of the Tat protein ( 49 RKKRRQRRR 57 ). This is not surprising since Ic and IIIb contain at least one arginine residue and it is well known that even a single argininamide residue has a specificity for this binding pocket (Frankel, 1992;Long & Crothers, 1995;Olsen et al., 2005). Upon binding of both PAA Ic and IIIb, the bulge nucleotides become unstacked, as observed in the case of arginamide, Tat basic domain and Tat derived peptides (Bardaro, Shajani, Patora-Komisarska, Robinson, & Varani, 2009;Davidson et al., 2011). However, while the binding of the tight ligand IIIb induces a strong contraction at the bulge region, the loose one, Ic, widens it. This may account for their distinct ability to inhibit the Tat/TAR complex. Indeed, IIIb is a strong antagonist of Tat (IC 50 = 35 nM) whereas Ic has no ability to displace Tat from a preformed Tat/TAR complex until 2 μM. Our results also demonstrate that in the case of penta-PAA IIIb, only the tetrameric N-terminal moiety interacts tightly with RNA, while the C-terminal part, i.e. F-β-alaninamide, is excluded from the interaction site and stays exposed to the solvent. Therefore, it appears that R-rich PAA tetramers would be of optimum length to allow a tight interaction with TAR and to compete efficiently with the Tat fragment for TAR binding.

Conclusion
Many RNA ligands have been developed in view of inhibiting the interaction of a target RNA with its cognate partner. Studies related to RNA/ligand interactions are often limited to the determination of ligand affinity and/or of ligand ability to inhibit the natural complex. However, these studies do not provide insights into structure-activity relationships, which are important for a rational design of specific RNA ligands.
In the present work, increasing the length of previously studied trimeric C-α-PAAs to pentameric structures led in all cases to an increase in TAR affinity as one would expect but surprisingly the ability to compete with Tat was strongly improved only in the case of R/K rich-PAA. While NMR, CD, and stability studies could not reveal any significant difference in the TAR binding of penta-PAAs, only a comparative analysis of thermodynamic profiles allowed us to highlight different binding features in the PAA series. This led us to propose two distinct interaction modes for F-rich and K/R-rich penta-PAA that were validated by molecular modeling studies. According to these computational models, distinct conformational changes would occur upon binding of the two kinds of ligands that could result in different ability to displace Tat fragment. Our PAA study illustrates how comparative thermodynamic and structural studies of a series of RNA ligands of same chemical family are of value for understanding their binding modes and for rationalizing structure-activity relationships. Such studies should be expended in this challenging research area to improve both the design of new ligands and the knowledge of their interactions.
On the other hand, although the conformational flexibility of PAA is well suited for the induced-fit mechanism by which RNA recognition occurs, it makes structure prediction difficult and may induce a lake of specificity for the target RNA. So, based on all these new results, it would be interesting to reduce the flexibility of the PAA polyamide backbone and to study the influence on the TAR binding of this conformational restriction. Works are in progress and results will be reported in due course.

Supplementary material
The supplementary material for this paper is available online at http://dx.doi.org/10.1080/07391102.2015. 1114971.