Introduction
The FDA approval of the splicing modifier Risdiplam [ 1 ] demonstrated that it is possible to pharmaceutically target functional RNA with small molecules. Since this technology would generate immense new pharmaceutical opportunities [ 2 – 4 ], more than a billion dollars was invested over the last 10 years in biotech and the pharmaceutical industry to achieve this goal [ 4 ]. However, these investments have not produced any successful exit so far, and new patent filings have revealed little new drug-like chemistry [ 5 ]. Most industry projects have been terminated, and earlier enthusiasm among investors and in pharma has been replaced with skepticism.
A critical knowledge gap is represented by the scarcity of high-quality structures that would provide atomic insight into recognition principles required to target RNA with small molecules [ 6 ]. In this study, we reveal the structural basis for single-digit nM binding of the cyclin-dependent kinase inhibitor palbociclib (ibrance) to HIV-1 TAR RNA. We show why this interaction is so sensitive to both RNA sequence and structure and how the small molecule achieves potent inhibition of the assembly of the positive transcription elongation factor b (P-TEFb) on TAR element.
RNA has been targeted pharmaceutically for many decades: natural product antibiotics bind to and inhibit the bacterial ribosomal RNA [ 7 – 9 ], and structural biology has demonstrated that the complex tertiary structures of the ribosome, ribozymes, and riboswitches contain druggable binding pockets [ 9 – 11 ]. However, such structures are neither frequent nor stable in mRNAs or noncoding RNAs, and RNA structure is further destabilized in cells by helicases and active translation [ 12 , 13 ]. Most RNA elements demonstrated to have biological function in regulation of eukaryotic gene expression are instead single-stranded motifs recognized by RNA-binding proteins or simpler and often dynamic secondary structure elements that generally lack druggable binding pockets [ 2 , 14 ]. Therefore, extending the approach to mRNAs coding for disease-causing proteins or noncoding RNAs has remained a major challenge due to the poor selectivity and non-drug-like properties of drug discovery candidates, as well as mechanistically unclear cellular responses.
The transactivation response (TAR) element of HIV-1 represents a case study in these challenges as a long-standing paradigm for RNA recognition by peptides and small molecules [ 15 ]. It also plays a pivotal role in the emergence of the virus from latency by binding to the HIV Tat protein to recruit P-TEFb to the nascent RNA transcript and promote efficient transcription of the viral genome [ 16 , 17 ]. Hundreds of manuscripts have been dedicated to disrupting the Tat–TAR interaction using small molecules or peptide mimics [ 15 ]. However, most reported compounds have been flexible and charged peptide mimetics [ 18 ], natural products, or other molecules that bind RNA nonspecifically through intercalation or electrostatics (e.g. aminoglycosides [ 19 , 20 ]), which might not even penetrate cells where antiviral effects are observed [ 21 ].
Understandably, it has been stated that TAR is unlikely to be drugged with small molecules [ 2 ]. This conclusion is too broad, since work in biotech [ 22 , 23 ], pharma [ 24 , 25 ], and others [ 26 ] has identified drug-like molecules with interesting properties, but it correctly reflects the overwhelming balance of the academic and patent literature.
In the accompanying manuscript, we show that the blockbuster cdk4/cdk6 inhibitor palbociclib/ibrance binds to HIV TAR with single-digit nM affinity (3–30 nM, depending on the assay and buffer conditions) and inhibits assembly of the super elongation complex (SEC) on TAR at low nM concentration. The interaction is exquisitely sensitive to single nucleotide changes or small chemical modifications of the small molecule. We show here that palbociclib induces a completely new structure in TAR, unlike the “Tat free” or “Tat bound” structures described before [ 27 – 32 ], to create a deep binding pocket reminiscent of small-molecule binding sites in enzymes such as its clinical target cdk6 [ 33 ]. Binding requirements are distinct from those observed in previous structures of complexes with cyclic peptide mimics of Tat [ 34 , 35 ]. This new conformation provides a pocket composed of multiple required nonconsecutive nucleotides, which must be presented in a defined structural context. Thus, this work and the accompanying manuscript demonstrate that even relatively simple RNA motifs can be targeted with the high affinity and specificity required to initiate sustained preclinical efforts with drug-like molecules and demonstrate how this occurs [ 36 ]. They also challenge the now prevailing notion that RNA is an inherently difficult drug target by providing direct structural evidence of how single-digit nM ligands can be identified.
Materials and methods
RNA transcription
To establish and validate the 3D structure of the complex, we recorded a large set of nuclear magnetic resonance (NMR) experiments on wild-type and several mutated versions of TAR. Supplementary Table S1 summarizes all RNAs that were studied and the NMR experiments conducted for each of them. Unlabeled, fully deuterated, and selectively deuterated TAR and the corresponding 13 C and/or 15 N uniformly labeled RNA samples were synthesized using in vitro transcription on a 5–10 ml scale using DNA templates (from Integrated DNA Technologies) and T7 RNA polymerase purified in-house [ 37 , 38 ]. Single-stranded DNA corresponding to the phage T7 RNA polymerase promoter was annealed to 80 µl of 100 µM template sequences, heated to 95°C for 4 min, and then annealed at room temperature over 40 min. The annealing mixture was incubated with 5 mM of each of the four NTPs (Sigma), transcription buffer, 10% dimethyl sulfoxide, 20% PEG-8000, and 22 mM magnesium chloride. All RNA samples were purified from crude transcription mixtures by 18% denaturing polyacrylamide gels in 8 M urea and tris/borate/ethylenediaminetetraacetic acid (EDTA) buffer, which were run for ~18 h. The RNA was excised from the gel, purified by electro-elution, and concentrated by ethanol precipitation. The samples were re-dissolved in ∼10 ml of high salt wash (500 mM NaCl, 250 mM KCl in 50 mM potassium phosphate at pH 6.5, with 10 µM EDTA to chelate any divalent ions) and concentrated using Centriprep conical concentrators (3 kDa cutoff, Millipore), then slowly exchanged into storage buffer (10-20 mM potassium phosphate at pH 6.5).
Compound acquisition and synthesis
Palbociclib was purchased from Selleckchem and used without further purification. Purity documentation can be found at https://file.selleckchem.com/downloads/hplc/S111614%20Palbociclib%20HCl%20HPLC-selleck.pdf .
NMR
Prior to NMR experiments, RNA samples were buffer-exchanged into the desired buffer (see figure captions and text for details). For experiments investigating non-exchangeable protons, samples were prepared in 99.99% D 2 O; samples used to study exchangeable protons were prepared instead in 90% H 2 O/10% D 2 O.
Experiments were recorded on Bruker 600 and 800 AVANCE HD spectrometers equipped with cryo-cooled TCI probes at temperatures of 5 and/or 10°C (data recorded in water) and 25 and/or 37°C (data recorded in D 2 O). All NMR data were collected and processed using Topspin 3.6 from Bruker and analyzed using Sparky [ 39 ].
Concentrations of the NMR samples were ~1 mM, except for titrations and heteronuclear correlation experiments, which were recorded at 0.2–0.3 mM concentrations to reduce sample requirements. Experiments used for assignments were acquired with 1 H and 13 C spectral widths of 22 and 100 ppm, respectively, using excitation sculpting for water suppression [ 40 ]. 1-3-3-1 water suppression [ 41 ] was used to observe signals from rapidly exchanging resonances.
1 H- 15 N HSQC spectra were acquired for both the imino and amino regions at 5 and/or 10°C. In order to detect broad peaks from protonated C/A bases, 1 H- 15 N SoFastHMQC [ 42 ] spectra were collected as well, while HNN-COSY [ 43 ] experiments were used for verification of canonical base pairs and putative base triples. NMR experiments were collected at pH’s between 5.0 and 7.2 (most data were collected at pH 6.5, 5.8, and 5.0); data quality improves below pH 6.5, presumably due to the stabilization of two protonated base triples.
The 2D NOESY spectra were acquired at three mixing times (100, 200, and 300 ms), while TOCSY experiments were acquired with a 90 ms mixing time. 1 H- 13 C HMQC spectra were recorded on uniformly 13 C/ 15 N-enriched RNAs for assignment of aromatic and ribose protons. For HMQC and CT-HSQC experiments, the 13 C offset was set at 106.5 ppm using a typical 1 H 90° pulse width of ∼13.5 µs with a 1.5 s relaxation delay. The 13 C-CT-HSQC constant time experiments (Bruker’s standard sequence “hsqcctetgpsp.2”) incorporated a fixed delay of 13.3 ms to refocus evolution under the CC scalar coupling, in addition to the normal incremented evolution delay. Evolution under the 13 C chemical shift during this fixed delay period cancels out, leaving the 13 C chemical shift evolution time in the indirect dimension determined as usual by the number and size of the incremented evolution delays (400 t 1 increments for a total acquisition time of 21.6 ms).
In addition to 13 C-edited 3D NOESY-HMQC experiments, 13 C-edited/filtered 2D NOESY experiments [ 44 ] were acquired with a mixing time of 300 ms in about 24 hours of instrument time in interleaved fashion to generate four separate spectra: 13 C (ω 1 )- 13 C (ω 2 ) doubly selected (correlating 13 C-bound protons in ω 1 and ω 2 ), 13 C (ω 2 )-filtered/ 13 C (ω 1 ) selected ( 13 C-bound in ω 1 and 12 C-bound in ω 2 ), 13 C (ω 1 )-filtered/ 13 C (ω 2 ) selected ( 12 C-bound in ω 1 and 13 C-bound in ω 2 ), and 13 C(ω 1 )- 13 C (ω 2 ) doubly filtered ( 12 C-bound protons in both ω 1 and ω 2 ).
Free and bound palbociclib resonances were unambiguously assigned from TOCSY, NOESY, and (for the bound molecule) doubly filtered NOESY experiments, showing only 12 C-bound 1 H resonances in ω 1 and ω 2 . Complexes with fully deuterated TAR were also used for assignments of bound palbociclib resonances and to conduct ligand-detected titrations in which 1 H spectra of palbociclib were collected following addition of increasing concentrations of fully deuterated wild type and A22GA35G TAR.
Due to the high quality of the NMR data and the large number of data sets collected, RNA assignments were completed to a very high degree ( Supplementary Table S2 ); all protons of palbociclib were also assigned for the free and two independently bound forms ( Supplementary Table S3 ).
Structure calculation and analysis
Structure calculations for three complexes were performed using XPLOR-NIH. Initial folding was carried out using NOE-derived distance restraints, which were obtained from analysis of 2D and 3D NOESY spectra (in both H 2 O and D 2 O) collected at multiple mixing times and classified as strong, medium, or weak based on comparison with peaks corresponding to fixed or well-defined distances (e.g. H5–H6 in pyrimidine). Hydrogen bonding and planarity restraints were applied to base pairs and base triples whose NHs were protected from solvent exchange after identifying acceptors through HNN-COSY experiments and/or proximity inferred from the NOE information. This process was independently repeated for complexes of palbociclib with three TAR constructs: wild-type, A35G mutant, and A22G–A35G double mutant.
NOESY spectra recorded at pH 5.0 and pH 5.8 were used jointly to generate the distance restraints table for structure calculations. On average, >20 experimental NOE interactions were identified for each nucleotide, including about 6 long-range restraints per residue. Most importantly, ~70 intermolecular NOEs were unambiguously identified for each of the three complexes, an unusually large number. Dihedral restraints were only included for base-paired nucleotides that reside in helices and conform to A-form helical structures, as established from the pattern of NOE interactions and chemical shifts. Sugar puckers were constrained to C3′-endo for all ribose in the helices, which also had no H1′−H2′ cross-peaks in TOCSY spectra. Residues with C2′-endo pucker were identified by strong H1′−H2′ and H1′–H3′ correlations in TOCSY spectra. The remaining backbone torsion angles were left unrestrained due to the absence of experimental information.
Refinement of the structure was done by independently calculating 200 structures for each of three complexes starting from conformations obtained from an initial folding step. The models converge well for all three complexes, with root mean square deviations (RMSD) for all backbone atoms in the 10 lowest energy structures as low as 1.1 Å across the complete structure ( Supplementary Table S4 ).
Results
The kinase inhibitor palbociclib binds potently to HIV-1 TAR ( K d of 3–30 nM depending on assay and buffer conditions), is sensitive to single nucleotide substitutions or small chemical modifications of the small molecule, and inhibits assembly of the SEC (see accompanying manuscript). To establish how binding and inhibition arise, we determined the structure of the palbociclib complex with wild-type TAR (Fig. 1 ) and with two mutated RNAs ( Supplementary Fig. S1 ). By collecting a very extensive NMR dataset for four RNA constructs, we obtained a large number of experimental restraints, including ~70 unambiguously assigned intermolecular NOEs, which allowed the determination of three structures of high precision (Fig. 1 and Supplementary Table S4 ). The structure of the small molecule and a schematic illustration of the recognition mechanism are shown in Fig. 2 .
Three mutants (A22G, A35G, and A22GA35G) were introduced to better define the molecular interface and validate the structure. For each of them, the imino and TOCSY spectra remain in the slow exchange regime, indicative of sub- µM binding, and their TOCSY spectra are very similar to each other ( Supplementary Figs 2 and 3 ). These mutants were introduced because some resonances of palbociclib and near RNA residues A35 and U40 are broadened in the complex (perhaps due to incomplete protonation of A22 and C24 nearby; see below). The A22G and A22GA35G double mutants were designed based on initial inspection of the wild-type structure to test the hypothesis that the A22G substitution would avoid the need to protonate A22 to form a base triple with G26-C39 (see below). This hypothesis proved correct and allowed a more precise insight into the palbociclib binding pocket, because resonances of both U40 and palbociclib are sharpened significantly by this mutation. In the end, we established the 3D structures of the complexes of palbociclib with wild-type, A35G, and A22GA35G RNAs ( Supplementary Table S4 ). The three structures are very similar to each other (Fig. 1h ) and report on the same new conformation of TAR induced by palbociclib.
NMR analysis, spectral assignments, and structure calculations
Nearly complete NMR assignments were obtained for each of the three sequences ( Supplementary Table S2 ) by recording spectra at different pH and temperature, by using different isotope labeling patterns, through spectral editing/filtering [ 44 ], and through the comparison of spectra containing the nucleotide substitutions discussed above. All palbociclib resonances were assigned for the free molecule, and two bound forms were observed in the 2:1 complex, corresponding to two distinct complexes with stronger (low nM) and weaker (low µM) affinity ( Supplementary Table S3 and accompanying manuscript). Inter-proton distance restraints derived from the NOE peak intensities were prepared independently for the three complexes of Fig. 1 by combining data acquired at pH 5 and 5.8, as were dihedral angle and sugar pucker restraints ( Supplementary Fig. S4 ). Hydrogen bond identification was facilitated by the acquisition of 1 H- 15 N HSQC and HNN-COSY experiments ( Supplementary Fig. S5 ).
Supplementary Table S4 reports experimental and structural statistics. On average, >20 NOE interactions were assigned for each nucleotide. Most importantly, many intermolecular NOEs were unambiguously identified because the binding pocket is so deep and well defined. Structures calculated using XPLOR NIH converge for all three complexes to 1–1.5 Å RMSD.
Overview of the structure of the complex
A schematic representation of the conformational rearrangement in TAR is shown in Fig. 2 , and close-up views of the palbociclib binding pocket are shown in Fig. 3 . The small molecule opens up the unstable A22-U40 base pair by forming two hydrogen bonds with the Watson–Crick face of U40 (Fig. 3a ), with a third possible CH-O hydrogen bond. While we do not have direct evidence supporting the formation of the CH-O hydrogen bond, the geometry observed in the structure is consistent with it. Changing U40 to C, inversion of the base pair to UA, or methylation of the palbociclib bridging NH all abolish nM binding (accompanying manuscript), while the A22G substitution is well tolerated. The small molecule stacks between the lower (G21-C41 base pair) and upper (G26-C39 base pair) helices, while the UCU bulge is extruded from the partially ordered stacked-in conformation observed in the free RNA [ 27 , 30 ], thereby creating a single extended 11 base pair helix, with A22 replaced by palbociclib (Fig. 1 ). A22 relocates to form a protonated A22+G26C39 base triple and stacks between the aromatic ring of the small molecule and U25 (Fig. 3b ). The extended conformation of the ribose-phosphate-ribose backbone between G21 and A22 is stabilized by the cyclopentane, which forms hydrophobic interactions with both riboses from the minor groove side. With a pKa of around 8, the piperazine secondary amine is mostly protonated at all pHs we investigated (5 to 7.2) and forms electrostatic interactions in the major groove (Fig. 3d and e).
The UCU bulge nucleotides do not interact directly with palbociclib but stabilize the binding pocket by interacting with other RNA nucleotides (Fig. 4 ). U23 links A22 and C24, which forms a protonated reverse Hoogsteen C24 + -G36-C29 base triple (Fig. 4a ) that also stabilizes a new cross-loop C30-G34 base pair that extrudes A35 (Fig. 4c ); when A35 is replaced with G35, the cross-loop base pair becomes C30-G35 (Fig. 4d ), retaining very similar chemical shifts and structure to the wild-type RNA. U25 forms a base triple with the G28-C37 base pair (Fig. 4b ). Thus, all bulge nucleotides interact with other bases, except U23, which acts as a spacer and can be replaced with other nucleotides without disrupting the complex (see accompanying manuscript), while it is essential for binding to TAR by Tat or peptide mimics of the protein.
We hypothesized that partial protonation of A22 and C24 leads to the line broadening for A35 and U40 and for the aromatic protons of palbociclib. The A22G mutant does not require protonation to stabilize the base triple with G26-C39 (Fig. 3c ), while the A35G substitution indirectly stabilizes the C24 + -C29-G36 base triple by facilitating formation of the CG cross loop base pair (Fig. 4 ). Thus, the A22GA35G double mutant provides the highest quality NMR data, but the same structural features are observed in all complexes (Fig. 1h ).
Palbociclib forms a pseudo base pair with U40
Palbociclib forms hydrogen bonds between its pyrimidine ring nitrogen and bridging NH and the U40 H3 and O4, respectively, with an additional potential CH-O hydrogen bond between its aromatic H a and the O2 of U40 (Fig. 3a ). NH signals corresponding to eight stable G-C and two A-U base pairs appear in the HSQC spectrum of A35G TAR ( Supplementary Fig. S5 ). NH signals are also observed for U31, G32, and G33 in the upper loop and for two bulge residues (U23 and U25) in the region where non-hydrogen-bonded and GU base pairs resonate (11 to 12 ppm). We did not observe a peak for U40 H3 in the wild-type complex, but when we examined the A22G or A35GA22G mutants, we observed an additional NH signal. Since all other Us are assigned, it must be U40 NH (Fig. 5 ), a conclusion confirmed by the pattern of NOE interactions. No corresponding peak is observed in the HNN COSY because palbociclib is not isotopically labeled.
Formation of the pseudo base pair is unambiguously established by strong intermolecular NOEs between U40 H3 and the palbociclib bridging NH and aromatic H a resonances (the assignment of palbociclib in the complex is presented next). The acidic bridging NH, invisible in the free small molecule, becomes clearly observable at 9.5 ppm in the complex because it is hydrogen bonded (Fig. 6 and Supplementary Fig. S6 ); its methylation abolishes nM complex formation (accompanying manuscript). The U40 H3 and bridging NH and H a of palbociclib exhibit very similar patterns of NOEs with G21, A22/G22, G26, C39, and C41, unambiguously placing U40 and the aromatic ring of palbociclib between the G21-C41 and G26-C39 base pairs (Fig. 6 and Supplementary Figs S6 and S7 ).
Palbociclib forms two different complexes with TAR
To trace chemical shift changes in palbociclib and assign its spectrum ( Supplementary Fig. S8 and Supplementary Table S5 ), we “reverse titrated” 100 µM of palbociclib with fully 2 H-labelled wild-type and double mutant TAR ( Supplementary Fig. S9 and accompanying manuscript). The spectra of the small molecules display similar behavior with both RNAs but are sharper for the double mutant. At lower concentrations of RNA (0–2 µM), the free palbociclib peaks broaden and shift moderately due to exchange with the lower affinity µM binding site (see accompanying manuscript). Signals for ligands bound to the higher affinity binding site only become visible at RNA concentrations beyond 2 µM (1:50 ligand:RNA), because the bound population remains below the limit of detection when only small amounts of RNA are added. For example, the well-resolved H a resonance around 9 ppm progressively shifts to 8.9 ppm and decreases in amplitude as increasing amounts of RNA are introduced, while peaks corresponding to ligand bound to the high-affinity site progressively increase near 8.0 ppm; near 1:1 stoichiometry, only the peak corresponding to the high-affinity site remains visible ( Supplementary Fig. S9 ), because essentially only the high-affinity site is populated at 1:1 ratio.
Several aromatic protons exhibit large upfield shifts of between 0.5 and 2 ppm ( Supplementary Fig. S10 ), likely due to the ring currents of neighboring bases (G21-C41 and A22+-G26-C39), which sandwich the aromatic rings of palbociclib. In contrast, the acetyl group methyl (H l ) undergoes only a small downfield shift (−0.1 ppm); it projects outside the RNA cavity, has only a few NOE interactions with the G26 ribose, and its removal is tolerated (accompanying manuscript).
NOESY spectra were collected to complete assignments of palbociclib in the complex ( Supplementary Fig. S11 ), while the H c -H d pyridine pair was unambiguously identified in TOCSY spectra ( Supplementary Fig. S12 ). The 13 C (ω 1 )- 13 C (ω 2 ) doubly filtered experiment, which contains only NOE signals for the small molecule, confirmed these findings. Two sets of peaks are observed for palbociclib H a , H e , H m , and H l protons when the small molecule is in 2:1 excess. In this second set of signals, the resonances are very close, but not identical, to free palbociclib chemical shifts; these arise from the weaker complex observed in the reverse titration experiment.
Intermolecular NOEs were unambiguously attributed based on these assignments and, when necessary, unambiguously distinguished from intramolecular NOEs via filtered NOESY experiments ( Supplementary Fig. S13 ). A graphic summary of the intermolecular NOEs observed in the complex is provided in Supplementary Fig. S14 , and their complete tabulation is in Supplementary Table S5 .
Hydrophobic interactions from the cyclopentane and electrostatic interactions from the piperazine stabilize the RNA structure
The phosphodiester backbone linking G21 and A22 is in an extended conformation to allow insertion of palbociclib (Figs 3 and 4 ). The cyclopentane interacts with the ribose of G21, A22/G22 in the minor groove (Fig. 3 ), as demonstrated by numerous intermolecular NOEs with the A22/G22 ribose and bases ( Supplementary Figs S8 , S10 , and S14 ). Large upfield shifts, relative to a freshly prepared sample of free TAR, were observed for C1′ and C4′ of A22, while the H5′/H5′′ experience large downfield shifts ( Supplementary Fig. S15 and Supplementary Table S6 ). The signals for these ribose resonances resemble those observed for nucleotides adopting extended backbone conformations in single-stranded loops [ 45 , 46 ].
The piperazine adopts the more stable chair conformation and emerges into the major groove in a region of strong electronegative potential [ 23 ]. Its secondary amine is near the phosphate linking G21 and A22 (Fig. 3 ), while its aromatic CHs form hydrophobic interactions with the A22 ribose, as demonstrated by NOEs between the piperazine and the G21 and A22 bases and ribose ( Supplementary Figs S8 , S10 , and S14 ).
The A22+G26-C39 base triple
Intra- and intermolecular NOE interactions position A22 near the G26-C39 base pair, in an orientation consistent with the formation of a protonated Watson–Crick/Hoogsteen base triple stabilized by two hydrogen bonds: NH1+(A22)-N7(G26) and NH 2 6(A22)-O6(G26) (Fig. 3b ), as observed in an unrelated crystal structure [ 47 ]. The A22 C2 resonance is shifted ~4 ppm upfield to 145.5 ppm, compared with the free RNA, as observed in adenine protonated at the N1 position [ 48 ], and was assigned in an adenine-only labeled sample of the A35G mutant, which only contains three adenines ( Supplementary Fig. S15e and f ), two of which are unchanged helical residues. Based on this geometry, we hypothesized that the A22G substitution would form a similar Hoogsteen base triple without protonation (Fig. 3c ). The G22-G26-C39 base triple is much closer to ideal geometry and better defined than the A22+ base triple, presumably because of imperfect NOE information for the wild-type complex due to line broadening.
Hoogsteen geometry is established by the proximity of A22 H2 and G26 H8 ( Supplementary Fig. S16 ). Relatively sharp amino nitrogen resonances (89.4 ppm) with unusually downfield shifted protons at 9.65 ppm and 8.9 ppm are consistent with hydrogen bonding and protonation. Their NMR signal is different from Watson–Crick paired adenine NH 2 , which are observed at 8–8.5 and 6–6.5 ppm and broadened by rotation around the CN bond. The downfield-shifted amino resonances were unambiguously attributed to A22 by using the A35G mutant. A NOESY collected with jump and return water suppression reveals weak NOEs between A22 N6H2s and G26, C39 and U38 base protons, confirming the Watson–Crick/Hoogsteen geometry.
The protonated base triple is further supported by the observation of two broad peaks near 15.6 and 16.6 ppm and ~180 ppm ( Supplementary Fig. S17 ), where resonances from protonated base triples have previously been reported [ 49 , 50 ]. The peak at 16.5/180 ppm is A22 N1H1+, while the peak at 15.6/180 ppm corresponds to C24 N3H3+, as described below. The 180 ppm 15 N shift has been unambiguously assigned before in a protonated adenine [ 48 ]. Unfortunately, these two peaks are too broad to display NOE interactions under any experimental conditions we investigated.
The 15 N amino resonance of C39 exhibits a downfield shift of nearly 10 ppm from regular Watson–Crick base pairs, consistent with its participation in the protonated base triple. These peaks are more clearly observed when the A22G or A35GA22G mutations are introduced because of increased spectral quality ( Supplementary Fig. S18 ). Lastly, the 15 N2 amino resonances for five guanines (G26, G28, G32, G33, and G35) are also observed in wild type or A35G TAR ( Supplementary Fig. S19 ). G32 and G33 are single-stranded nucleotides in the apical loop, and we observe a single intense peak since the NH 2 is not involved in hydrogen bonding. Guanine NH 2 ’s involved in regular base pairs are exchange broadened due to rotation about the CN bond, but G28 and G35, near the C24 + -G36-C29 base triple (see below), are sharp. The amino protons of G26 are downfield shifted as well ( 15 N2 at 77 ppm), and their rotation slowed, leading to observation of relatively sharp resonances, presumably due to base triple formation.
The stabilization of the base triple by the A22G substitution was confirmed by examination of the A22G and A22GA35G mutants, for which almost all exchangeable peaks were assigned (Figs 5 and 6 ). A sharp peak corresponding to G22 H1 appears at 10.7/144.3 ppm, consistent with hydrogen bonding to a carbonyl. NOESY spectra for the A22GA35G double mutant (and for A22G) contain many NOEs involving G22 H1 (with G21, G26 U40, U38, G22, and C39; Supplementary Figs S7 , S8 , and S14 ), which position G22 above the palbociclib-U40 pseudo base pair and in the plane of the G26-C39 base pairs, with which it hydrogen bonds in the Hoogsteen orientation.
The C24 and U25 bases form base triples with the C29-G36 and G28-C37 base pairs
The UCU bulge nucleotides are not directly in contact with palbociclib but contribute to recognition by stabilizing the new RNA structure. The UCU bulge is extruded, as observed in the structure of TAR bound to peptide mimics of Tat [ 28 , 29 , 32 ], but its conformation is completely different. U23 links A22 to C24 to position this cytosine in the reverse Hoogsteen orientation [ 51 ] relative to the C29-G36 base pair, with which it forms a protonated base triple (Fig. 4a ), which also stabilizes a cross-loop base pair between C30 and G34, bulging out A35 (Fig. 4c ). U25 is in the plane of the G28-C37 base pair and forms a single hydrogen bond with it as well as a potential CH-O hydrogen bond (Fig. 4b ), for which, however, we have no independent corroboration. Consistent with the structure, when we replaced C24 or U25 with 2-amino purine to conduct a well-established fluorescence assay [ 52 ], we observed only small changes in fluorescence and altered binding behavior (see accompanying manuscript). These base triples explain why the largest chemical shift changes in TOCSY spectra were observed near the apical loop and bulge, which initially prompted us to think this would be the palbociclib binding site.
Formation of the cross-loop C30-G34 base pair is established by the observation of a weak G34 H1N1 peak with NOE interactions to a pair of hydrogen-bonded N4H 2 ’s ( Supplementary Fig. S20 ). The G34 H1 also exhibits NOEs with G36 and C29, supporting the bulging out of A35 (Fig. 4c ), whose base resonances are broadened beyond detection.
The broad signal at 15.6 ( Supplementary Figs S17 and 21 ) is at the chemical shifts of cytidine protons in C+GC base triples [ 49 , 50 ], but the 180 ppm 15 N shift is unexpected. Nevertheless, all resonances for the C24 base appear as one major and one minor peak ( Supplementary Fig. S17 ), including N3H3+, identifying the 15.6/180 ppm peak as belonging to C24. We do not know the origin of the minor conformer, which is not present in freshly prepared samples. The C24 N4H 2 resonances are observed at 12.1 and 9.4 ppm, respectively, 3 ppm downfield compared to cytosines in canonical Watson–Crick base pairs, and the chemical shift of the 15 N4 is also unusual (114.5 ppm; Supplementary Fig. S21 ). The observation of NOE interactions between C24 and C29, G35, G36, and C37 identifies the C29-G36 base pair as the recipient of the protonated C24, which hydrogen bonds to it in the reverse Hoogsteen orientation [ 51 ] generated by a sharp turn in the RNA backbone between U23 and C24. Evidence for a turn in the backbone is provided by unusual C24 chemical shifts, particularly C1′ and H5′/C5′.
Hydrogen bonding between the C24 N4H 2 s and G36 was demonstrated conclusively using an HNN-COSY [ 43 ] ( Supplementary Fig. S22 ). In this spectrum, the C24 N4 is correlated with a Guanine N7 at 229.3 ppm; the C24 N3H3 + proton is positioned near the O6 of G36 and in hydrogen bonding distance (Fig. 4a ). The trans/reverse orientation of the base triple was further demonstrated by NOE interactions observed for the A35G mutant, which provides higher quality data ( Supplementary Fig. 18 ) because G35 more stably base pairs with C30.
In wild type and A35G TAR, U25 is in the plane of G28-C37, forming a single hydrogen bond or electrostatic interaction between U25 O4 and C37 N4H 2 (Fig. 4b ). This conclusion is supported by strong NOEs from the U25 base to C37, C29, U38, and G28 ( Supplementary Fig. S23 ), while NOEs between U25 and A22 H2 locate the U25-G28-C37 base triple directly above A22+G26-C39. The ribose of U25 experiences large upfield shifts for H1′/C1′ and downfield shifts for H5′/H5′′/C5′. We also observe a significant upfield shift for C41 H1′/C1′, probably due to the replacement of the metastable A22U40 base pair with the palbociclib–U40 pseudo base pair, which alters ring current effects ( Supplementary Fig. S24 ).
The structure forms at near-neutral pH
Three protonation events occur as palbociclib binds to TAR. The piperazine secondary amine (pKa of ~8) is mostly protonated under all conditions we investigated (pH 5–7 or lower; the other nitrogens in the small molecule all have predicted pKa’s below 3). This group can be methylated without loss of binding because the pKa does not change significantly (accompanying manuscript). In addition, protonation of A22 and C24 stabilizes the complex through its effect on RNA structure.
At pH values between 5 and 7, we observe the large and characteristic chemical shift changes associated with complex formation in TOCSY spectra ( Supplementary Fig. S25 ) and in 1D titrations ( Supplementary Fig. S26 ). We show in the previous manuscript that low nM affinity (20–30 nM) is retained at pH 7. At higher pH, we expect that partial deprotonation of the piperazine will reduce affinity, as we have observed for RNA binding of other molecules sharing the pyridyl-piperazine motif. The pKa’s of A22 and C24 in the complex are likely to be significantly shifted from 3.8/4.2 through the formation of Hoogsteen base triples. We plan to systematically address the pH dependence of this interaction in follow-up work. We collected data at lower pH to fully protonate the base triples, which leads to sharpening of nearby resonances and better definition of the structure, like any crystallographer would optimize crystal packing to obtain higher resolution.
Stabilization of the RNA structure
A final surprising observation is the protection from exchange with solvent of many 2′-OH resonances in the complex, which are invisible in the free RNA under the same conditions ( Supplementary Fig. S27 ). In the NOESY spectra of all three complexes, the 2′-OH resonances from G18, G21, A27, G28, C29, G35, G36, C37, C41, G43, and C44 are clearly observable. These signals were unambiguously identified based on the absence of splitting in 15 N-labeled samples and strong NOEs with H1′ and H2′ of the corresponding ribose spin systems ( Supplementary Fig. S27 ). These hydroxyl groups are protected from solvent exchange not just where palbociclib binds and near the base triples, but throughout the entire helix. We do not know whether this observation is attributable to thermodynamic stabilization upon ligand binding [ 53 , 54 ] or a specific feature of this new structure.
Discussion
Simple secondary structures such as HIV TAR, a paradigmatic system for RNA-small molecule recognition, regulate RNA biology in healthy and diseased cellular states, yet are widely believed to be difficult or perhaps impossible to drug with small molecules [ 2 ]. Therefore, they are sometimes not pursued in drug discovery projects, in favor of more complex structures, which might not be stable in the cell and/or might lack functional validation. In the accompanying manuscript, we show that the cdk4/cdk6 kinase inhibitor palbociclib (and ribociclib, but not abemaciclib) binds to TAR with 3–30 nM affinity in a manner that is exquisitely sensitive to single nucleotide substitutions or small changes in small molecule chemistry. It also inhibits recruitment of P-TEFb at low nM concentration, which we could not achieve with peptide mimics of Tat with 30 pM binding affinity [ 29 ]. Here, we demonstrate that binding and biological response are achieved through a completely new RNA structure induced by the small molecule (Fig. 2 ).
Experienced industry chemists openly express skepticism as to whether small molecules with drug-like characteristics can bind to RNA specifically and elicit a target-specific response in the cell. To address the first point, we show that 13 nonconsecutive nucleotides form the palbociclib binding pocket and interact directly with it (Fig. 7 ) or stabilize the RNA structure required for binding (Fig. 2 ). Low nanomolar binding is compromised by certain single nucleotide changes, inversion of base pairs, or simple modifications of the small molecule (accompanying manuscript, Fig. 8 ), but it is retained up to pH 7, suggesting that the pKa of A22 and C24 is significantly increased by the formation of the base triples. Deprotonation of the piperazine will significantly reduce affinity above pH 7 (acetylation affects binding to other RNAs of similar compounds sharing the pyridyl-piperazine motif), but this limitation could be addressed by subsequent rounds of structure-driven chemistry. It is highly unlikely that so many nucleotides would be presented elsewhere in the genome in the correct context to replicate this structure and that such a motif, if it existed, would also be functional. Because the TAR RNA sequence and structure are highly conserved, the virus might pay a significant replicative cost to escape treatment. For example, U40 is never mutated to C, and the A22U40 base pair is very seldom inverted (both would abolish binding); when this occurs, nucleotide U31 is nearly invariably changed to C [ 55 ]. However, loss of C24 to generate HIV-2 TAR would provide an escape pathway for the virus because this mutant does not bind to palbociclib (accompanying manuscript), as the structure clearly demonstrates.
With regard to the second point, palbociclib inhibits HIV replication in cells that do not express cdk4/cdk6 [ 56 ], but caution must be exercised in interpreting cellular data without careful validation of target engagement. “Reverse palbociclib” was synthesized precisely to enable such studies because it retains binding to TAR without inhibiting kinases, but it is toxic at single-digit µM concentrations, preventing meaningful cellular investigations (accompanying manuscript).
Importantly, two molecules of palbociclib bind to TAR. The first molecule binds specifically with single-digit nM affinity and is deeply buried within the binding pocket, resulting in large changes in its NMR chemical shifts. The second molecule binds only superficially with 2–10 µM affinity and exchange-averaged chemical shifts that are very close to those of the free small molecule. When mutations disrupt high-affinity binding, and with pre-miR-21 or HIV-2 TAR, we observe only reduced affinity binding manifested through intermediate or fast NMR exchange (accompanying manuscript). The pervasive and facile observation of promiscuous binding events of this kind in screening campaigns contributes significantly to the widespread skepticism that RNA can be targeted specifically.
The structure of the TAR palbociclib complex is completely different from previous structures of TAR and confounded our expectations as well as computational predictions, even when the binding site was pre-defined. In free TAR, the UCU bulge nucleotides are stacked into the helix and partially disordered [ 27 ], while in its complexes with Tat or Tat-derived peptides, U23 forms a base triple with A27-U38 [ 29 , 57 ]. The two other bulge nucleotides are extruded from the helix, and A35 flips to the syn conformation, rearranging the apical loop to form the structure recognized by P-TEFb [ 29 ].
The conformational rearrangement observed in the present structure is facilitated by hydrophobic interactions with the ribose of G21 and A22, mediated by the cyclopentane and aromatic CH3, and stacking of the aromatic ring and pyridine between the upper and lower helix, which becomes a continuous 11 base-pair helix, including palbociclib-U40 as a pseudo base pair. Hydrophobic interactions were also essential in the binding of peptide mimics of Tat, where the most important amino acid side chain was an isoleucine [ 28 , 29 , 34 ]. In addition, the secondary amine at the tip of the piperazine sits in a region of deep electronegative potential in the RNA major groove [ 23 ], but without hydrogen bonding (methylation of this group preserves binding because the pKa is only slightly affected). A22 forms a protonated A22+G26-C39 base triple [ 47 ]; the A22G substitution allows the same base triple to form without protonation. U23 is the only nucleotide that does not interact with other bases, acting instead as a spacer, while C24 protonates to form the C24 + C29G36 base triple, which stabilizes the new cross-loop C30G34 base pair and extrudes A35 from the helix. U25 then interacts with the G28-C37 base pair to form a third base triple. Based on the observation of large chemical shift changes in the upper helix and apical loop of TAR and inhibition of P-TEFb, we expected palbociclib to bind there, although the RNA chemical shifts are very different from those observed for Tat peptide mimics. We strongly caution against making statements on what RNA binding sites for small molecules might look like based on limited data from SHAPE or chemical shift mapping, or from computational modeling.
The free RNA presents the flexible UCU loop and partly stable A22U40 base pair within the framework of a stable secondary structure, and the small molecule exploits this conformational flexibility to create a deep binding pocket (Fig. 8 ). In the case of peptide mimics of Tat, it is likely that the new conformation is already present at low population and binding occurs by conformational capture [ 58 , 59 ]. We believe the present structural rearrangement is instead a true induced fit but will investigate this interesting mechanistic question in the future.
The conformational rearrangement is key not just to specific binding but also to inhibiting SEC assembly on TAR. Tat-derived peptides modulate the RNA structure to generate the conformation recognized by P-TEFb, and therefore peptide mimics of Tat fail to inhibit the SEC despite pM affinity [ 29 ]. A similar mechanism was also observed for inhibition of the Hepatitis C IRES by benzimidazole derivatives, albeit with much weaker affinity [ 60 ].
In summary, the present structure provides a framework for understanding how a well-known enzyme-targeting drug achieves high specificity and affinity for a simple RNA stem-loop. It also confirms that the chemistry of RNA-binding small molecules is not different from protein-binding small molecules [ 61 , 62 ]; therefore, there is little justification in pursuing chemistries that lie far outside the limits of drug-likeness [ 5 ] and are not likely to lead to successful clinical development. We emphasize that the affinity of palbociclib for TAR, 3-30 nM, is comparable to its K i for cdk4/cdk6, 60 nM, which was achieved only after a decade-long effort [ 33 ]. We believe it would be straightforward, with this structure in hand, for motivated medicinal chemists to evolve additional potency and separate kinase inhibition from RNA binding in a non-cytotoxic analogue.
Supplementary Material
Acknowledgements
We wish to thank all members of the Varani group for discussion and support; Thomas Pavelitz for preparation of the T7 RNA polymerase; and Drs Ursula Schulze-Gahmen and James Hurley for donation of P-TEFb samples. We apologize to the many colleagues whose work was not cited; the literature on TAR is immense and it would be impossible to provide references for even a fraction of the many deserving manuscripts.
Author contributions : Ramireddy Ravikanth Reddy: Synthesized all RNAs, prepared samples, collected NMR data, determined the structure, optimized methodology, investigation, wrote and edited the manuscript; generated graphics. Venkata Narayana Vidadala: Procured and synthesized small molecules, collected data for imino titrations, edited the manuscript. Gregory L. Olsen: Assisted with Xplor-NIH calculations and set up of NMR experiments, edited the manuscript. Bhawna Chaubey: Assisted with spectral assignments; edited the manuscript. Gabriele Varani: Designed experiments and assisted in data analysis, wrote, reviewed, and edited the manuscript.
Contributor Information
Ravikanth Reddy Ramireddy, Department of Chemistry, University of Washington, Seattle, WA 98195-1700, United States.
Venkata Vidalala, Department of Chemistry, University of Washington, Seattle, WA 98195-1700, United States.
Bhawna Chaubey, Department of Chemistry, University of Washington, Seattle, WA 98195-1700, United States.
Gregory L Olsen, Department of Chemistry, University of Washington, Seattle, WA 98195-1700, United States.
Gabriele Varani, Department of Chemistry, University of Washington, Seattle, WA 98195-1700, United States.
Supplementary data
Supplementary data is available at NAR online.
Conflict of interest
G.V. is a co-founder of Ithax Pharmaceuticals and retains a professional association with the company.
Funding
This work was supported by the National Institutes of Health, National Institute of General Medical Sciences, grant number 1R35GM126942 as well as by a gift from the Washington Research Foundation to G.V. Funding to pay the Open Access publication charges for this article was provided by National Institutes of Health.
Data availability
Coordinates were deposited at the PDB under accession codes 9PDG (wild-type), 9PDF (A35G mutant), and 9PDE (A22GA35G mutant). NMR chemical shift assignments were deposited at BMRB under accession codes 31259 (wild-type), 31258 (A35G mutant), and 31257 (A22GA35G mutant). Raw data and processing scripts underlying this article will be shared on reasonable request to the corresponding author.
This paper is linked to: https://doi.org/10.1093/nar/gkag186 .
References
1. Ratni H, Ebeling M, Baird J et al. Discovery of risdiplam, a selective survival of motor neuron-2 (SMN2) gene splicing modifier for the treatment of spinal muscular atrophy (SMA). J Med Chem. 2018;61:6501–17. 10.1021/acs.jmedchem.8b00741.
2. Warner KD, Hajdin CE, Weeks KM. Principles for targeting RNA with drug-like small molecules. Nat Rev Drug Discov. 2018;17:547–58. 10.1038/nrd.2018.93.
3. Gallego J, Varani G. Targeting RNA with small molecule drugs: therapeutic promises and chemical challenges. Acc Chem Res. 2001;34:836–43. 10.1021/ar000118k.
4. Cross R. The RNA drug hunters. Chem Eng News. 2017;95:16–8.
5. Ravegnini E, Trabocchi A, Lenci E. Small-molecule RNA ligands: a patent review (2018-2024). Expt Op Therap Patents. 2025;35:675–93. 10.1080/13543776.2025.2492759.
6. Shortridge MD, Varani G. Structure based approaches for targeting non-coding RNAs with small molecules. Curr Opin Struct Biol. 2015;30:79–88. 10.1016/j.sbi.2015.01.008.
7. Hermann T. Drugs targeting the ribosome. Curr Opin Struct Biol. 2005;15:355–66. 10.1016/j.sbi.2005.05.001.
8. Yonath A. Antibiotics targeting ribosomes: resistance, selectivity, synergism, and cellular regulation. Annu Rev Biochem. 2005;74:649–79. 10.1146/annurev.biochem.74.082803.133130.
9. Carter AP, Clemons WMJ, Brodersen DE et al. Functional insights from the structure of the 30S ribosomal subunit and its interaction with antibiotics. Nature. 2000;407:340–8. 10.1038/35030019.
10. Howe JA, Wang H, Fischmann TO et al. Selective small-molecule inhibition of an RNA structural element. Nature. 2015;526:672–7. 10.1038/nature15542.
11. Liu T, Xu L, Chung K et al. Molecular insights into de novo small-molecule recognition by an intron RNA structure. Proc Natl Acad Sci USA. 2025;122:e2502425122. 10.1073/pnas.2502425122.
12. Wan Y, Qu K, Zhang QC et al. Landscape and variation of RNA secondary structure across the human transcriptome. Nature. 2014;505:706–9. 10.1038/nature12946.
13. Flynn RA, Zhang QC, Spitale RC et al. Transcriptome-wide interrogation of RNA secondary structure in living cells with icSHAPE. Nat Protoc. 2016;11:273–90. 10.1038/nprot.2016.011.
14. Varani L, Hasegawa M, Spillantini MG et al. Structure of tau exon 10 splicing regulatory element RNA and destabilization by mutations of frontotemporal dementia and parkinsonism linked to chromosome 17. Proc Natl Acad Sciences USA. 1999;96:8229–34. 10.1073/pnas.96.14.8229.
15. Abulwerdi FA, Le Grice SFJ. Recent advances in targeting the HIV-1 tat/TAR complex. Curr Pharm Design. 2017;23:4112–21. 10.2174/1381612823666170616081736.
16. Karn J. Tackling Tat. J Mol Biol. 1999;293:235–54. 10.1006/jmbi.1999.3060.
17. Mbonye U, Karn J. The molecular basis for human immunodeficiency virus latency. Ann Rev Virol. 2017;4:261–85. 10.1146/annurev-virology-101416-041646.
18. Hamy F, Felder ER, Heizmann G et al. An inhibitor of the tat/TAR RNA interaction that effectively suppresses HIV-1 replication. Proc Natl Acad Sci USA. 1997;94:3548–53. 10.1073/pnas.94.8.3548.
19. Wang S, Huber PW, Cui M et al. Binding of neomycin to TAR element of HIV-1 RNA induces dissociation of Tat protein by an allosteric mechanism. Biochem. 1998;37:5549–57. 10.1021/bi972808a.
20. Stelzer AC, Frank AT, Kratz JD et al. Discovery of selective bioactive small molecules by targeting an RNA dynamic ensemble. Nat Chem Biol. 2011;7:553–9. 10.1038/nchembio.596.
21. Daelemans D, Schols D, Witvrouw M et al. A second target for the peptoid tat/transactivation response element inhibitor CGP64222: inhibition of human immunodeficiency virus replication by blocking CXC-chemokine receptor 4-mediated virus entry. Mol Pharmacol. 2000;57:116–24. 10.1016/S0026-895X(24)26449-X.
22. Murchie AIH, Davis B, Isel C et al. Structure-based drug design targeting an inactive RNA conformation: exploiting the flexibility of HIV-1 TAR RNA. J Mol Biol. 2004;336:625–38. 10.1016/j.jmb.2003.12.028.
23. Davis B, Afshar M, Varani G et al. Rational design of inhibitors of HIV-1 TAR RNA through the stabilisation of electrostatic “hot spots”. J Mol Biol. 2004;336:343–56. 10.1016/j.jmb.2003.12.046.
24. Mei H-Y, Mack DP, Galan AA et al. Discovery of selective, small-molecule inhibitors of RNA complexes - I. The tat protein/TAR RNA complexes required for HIV-1 transcription. Bioorg Med Chem. 1997;5:1173–84. 10.1016/S0968-0896(97)00064-3.
25. Mei H-Y, Cui M, Heldsinger A et al. Inhibitors of protein-RNA complexation that target the RNA: specific recognition of human immunodeficiency virus type I TAR RNA by small organic molecules. Biochem. 1998;37:14204–12. 10.1021/bi981308u.
26. Abulwerdi FA, Shortridge MD, Sztuba-Solinska J et al. Development of small molecules with a noncanonical binding mode to HIV-1 trans activation response (TAR) RNA. J Med Chem. 2016;59:11148–60. 10.1021/acs.jmedchem.6b01450.
27. Aboul-ela F, Karn J, Varani G. Structure of HIV-1 TAR RNA in the absence of ligands reveals a novel conformation of the trinucleotide bulge. Nucleic Acids Res. 1996;24:3974–81. 10.1093/nar/24.20.3974.
28. Davidson A, Leeper TC, Athanassiou Z et al. Simultaneous recognition of HIV-1 TAR RNA bulge and loop sequences by cyclic peptide mimics of Tat protein. Proc Natl Acad Sci USA. 2009;106:11931–6. 10.1073/pnas.0900629106.
29. Shortridge MD, Wille PT, Jones AN et al. An ultra-high affinity ligand of HIV-1 TAR reveals the RNA structure recognized by P-TEFb. Nucleic Acids Res. 2018;47:1523–31. 10.1093/nar/gky1197.
30. Bou-Nader C, Link KA, Suddala KC et al. Structure of a complete HIV-1 TAR RNA portray a dynamic platform poised for protein binding and structural remodeling. Nat Commun. 2025;6:2252. 10.1038/s41467-025-57519-w.
31. Pham VV, Salguero C, Khan SN et al. HIV-1 Tat interactions with cellular 7SK and viral TAR RNAs identifies dual structural mimicry. Nat Commun. 2018;9:4266. 10.1038/s41467-018-06591-6.
32. Aboul-ela F, Karn J, Varani G. The structure of the human immunodeficiency Virus type-1 TAR RNA reveals principles of RNA recognition by Tat protein. J Mol Biol. 1995;253:313–32. 10.1006/jmbi.1995.0555.
33. Toogood PL, Harvey PJ, Repine JT et al. Discovery of a potent and selective inhibitor of cyclin-dependent kinase 4/6. J Med Chem. 2005;48:2388–406. 10.1021/jm049354h.
34. Davidson A, Patora-Komisarska K, Robinson JA et al. Essential structural requirements for specific recognition of HIV TAR RNA by peptide mimetics of Tat protein. Nucleic Acids Res. 2011;39:248–56. 10.1093/nar/gkq713.
35. Bardaro MF, Shajani Z, Patora-Komisarska K et al. How binding of small molecule and peptide ligands to HIV-1 TAR alters the RNA motional landscape. Nucleic Acids Res. 2009;37:1529–40. 10.1093/nar/gkn1074.
36. Ramireddy RR, Shortridge MD, Vidalala V et al. The kinase inhibitor palbociclib binds to HIV TAR RNA with very low nanomolar affinity and exquisite specificity. Nucleic Acids Res. 2026;54. 10.1093/nar/gkag186.
37. Barnwal RP, Yang F, Varani G. Applications of NMR to structure determination of RNAs large and small. Arch Biochem Biophys. 2017;628:42–56. 10.1016/j.abb.2017.06.003.
38. Varani G, Aboul-ela F, Allain FH-T. NMR investigations of RNA structure. Progr NMR Spectr. 1996;29:51–127. 10.1016/0079-6565(96)01028-X.
39. Lee W, Tonelli M, Markley J. NMRFAM-Sparky: enhanced software for biomolecular NMR spectroscopy. Bioinformatics. 2015;31:1325–7. 10.1093/bioinformatics/btu830.
40. Hwang TL, Shaka AJ. Water suppression that works. Excitation sculpting using arbitrary waveforms and pulsed field gradients. J Magn Reson. 1995;112:275–9. 10.1006/jmra.1995.1047.
41. Stonehouse J, Shaw GL, Keeler J. Improving solvent suppression in jump-return NOESY experiments. J Biomol NMR. 1994;4:799–805. 10.1007/BF00398410.
42. Amero C, Schanda P, Dura MA et al. Fast two-dimensional NMR spectroscopy of high molecular weight protein assemblies. J Am Chem Soc. 2009;131:3448–9. 10.1021/ja809880p.
43. Wohnert J, Dingley AJ, Stoldt M et al. Direct identification of NH…N hydrogen bonds in non-canonical base pairs of RNA by NMR spectroscopy. Nucleic Acids Res. 1999;27:3104–10. 10.1093/nar/27.15.3104.
44. Peterson RD, Theimer CA, Wu H et al. New applications of 2D filtered/edited NOESY for assignment and structure elucidation of RNA and RNA-protein complexes. J Biomol NMR. 2004;28:59–67. 10.1023/B:JNMR.0000012861.95939.05.
45. Varani G, Tinoco I Jr. Carbon assignments and heteronuclear coupling constants for an RNA oligonucleotide from natural abundance 13 C- 1 H corelated experiments. J Am Chem Soc. 1991;113:9349–54. 10.1021/ja00024a044.
46. Hines JV, Landry SM, Varani G et al. Carbon-proton scalar couplings in RNA: 3D heteronuclear and 2D isotope-edited NMR of 13 C-labeled extra stable hairpin. J Am Chem Soc. 1994;116:5823–31. 10.1021/ja00092a037.
47. Su L, Chen L, Egli M et al. Minor groove RNA triplex in the crystal structure of a ribosomal frameshifting viral pseudoknot. Nat Struct Biol. 1999;6:285–92.
48. Toews S, Wacker A, Faison EM et al. The 5′-terminal stem–loop RNA element of SARS-CoV-2 features highly dynamic structural elements that are sensitive to differences in cellular pH. Nucleic Acids Res. 2024;22:7971–86. 10.1093/nar/gkae477.
49. Van Dongen MJP, Heus HA, Wymenga SS et al. Unambiguous characterization of a DNA-RNA triple helix by 15 N- and 13 C-filtered spectroscopy. Biochem. 1996;35:1733–9. 10.1021/bi952002f.
50. Macaya R, Wang E, Schultze P et al. Proton nuclear magnetic resonance assignments and structural characterization of an intramolecular DNA triplex. J Mol Biol. 1992;225:755–73. 10.1016/0022-2836(92)90399-5.
51. Almakarem ASA, Petrov AI, Stombaugh J et al. Comprehensive survey and geometric classification of base triples in RNA structures. Nucleic Acids Res. 2012;40:1407–23. 10.1093/nar/gkr810.
52. Bradrick TD, Marino JP. Ligand-induced changes in 2-aminopurine flourescence as a probe for small molecule binding to HIV-1 TAR RNA. RNA. 2004;10:1459–68. 10.1261/rna.7620304.
53. Conte MR, Conn GL, Brown T et al. Hydration of the RNA duplex r(CGCAAAUUUGCG) 2 determined by NMR. Nucleic Acids Res. 1996;24:3693–9. 10.1093/nar/24.19.3693.
54. Fohrer J, Hennig M, Carlomagno T. Influence of the 2′-hydroxyl group conformation on the stability of A-form helices in RNA. J Mol Biol. 2006;356:280–7. 10.1016/j.jmb.2005.11.043.
55. Huthoff H, Girard F, Wijmenga SS et al. Evidence for a base triple in the free HIV-1 TAR RNA. RNA. 2003;10:412–23. 10.1261/rna.5161304.
56. Pauls E, Badia R, Torres-Torronteras J et al. Palbociclib, a selective inhibitor of cyclin-dependent kinase4/6, blocks HIV-1 reverse transcription through the control of sterile alpha motif and HD domain-containing protein-1 (SAMHD1) activity. AIDS. 2014;28:2213–22. 10.1097/QAD.0000000000000399.
57. Puglisi JD, Tan R, Canlan BJ et al. Conformation of the TAR-arginine complex by NMR. Science. 1992;257:76–80. 10.1126/science.1621097.
58. Borkar AN, Bardaro MF, Camilloni C et al. Structure of a low-population binding intermediate in protein-RNA recognition. Proc Natl Acad Sci USA. 2016;113:7171–6. 10.1073/pnas.1521349113.
59. Dethoff EA, Petzold K, Chugh J et al. Visualizing transient low-populated structures of RNA. Nature. 2012;491:724–8. 10.1038/nature11498.
60. Parsons J, Castaldi MP, Dutta S et al. Conformational inhibition of the hepatitis C virus internal ribosome entry site RNA. Nat Chem Biol. 2009;5:823–5. 10.1038/nchembio.217.
61. Rizvi NF, Santa Maria JP, Nahvi A et al. Targeting RNA with small molecules: identification of selective, RNA-binding small molecules occupying drug-like chemical space. SLAS Discov. 2020;25:384–96. 10.1177/2472555219885373.
62. Fang L, Velema WA, Lee Y et al. Pervasive transcriptome interactions of protein-targeted drugs. Nat Chem. 2023;15:1374–83. 10.1038/s41557-023-01309-8.
Associated Data
Supplementary Materials
Data Availability Statement
Coordinates were deposited at the PDB under accession codes 9PDG (wild-type), 9PDF (A35G mutant), and 9PDE (A22GA35G mutant). NMR chemical shift assignments were deposited at BMRB under accession codes 31259 (wild-type), 31258 (A35G mutant), and 31257 (A22GA35G mutant). Raw data and processing scripts underlying this article will be shared on reasonable request to the corresponding author.
This paper is linked to: https://doi.org/10.1093/nar/gkag186 .