The subject of the present analysis is a small RNA aptamer (1,2) identified by in vitro selection (3,4) for affinity to the p50₂ form of mammalian transcription factor NF-κB. Because of its role in activating genes involved in inflammation, inhibition of apoptosis and HIV-1 activation, this dimeric transcription factor is of interest as a potential target for therapeutic inhibition (5,6). Previous work showed that the anti-NF-κB RNA aptamer bound with nanomolar affinity to NF-κB p50₂ in a manner that competed with DNA binding, that the binding interaction could be detected in the yeast three-hybrid system (7), and that aptamer variants with improved properties could be identified by yeast genetic selections (2). The X-ray crystal structure of a 29-nt version of the RNA was determined in complex with NF-κB p50₂ (8). The features of this complex have subsequently been reviewed and compared to the prior structure of NF-κB p50₂ bound to DNA (9,10).

The essential features of the RNA aptamer structure derived from its complex with NF-κB p50₂ are shown in Figure 1A. The RNA aptamer was predicted to fold as a stem-loop structure with an asymmetric internal loop, and a U–G wobble pair adjacent to a terminal GUAA tetraloop. Examination of the RNA aptamer within the NF-κB-aptamer complex (Figure 1B) validated secondary structure predictions, and revealed structural features deviating from A-form geometry. The latter included a large overall kink in the RNA and a complex pattern of base interactions within the asymmetric internal loop. These interactions included non-canonical pairing of A9–G22 and U6–C24, with stacking of unpaired bases U7, G8 and G23. Together with the U13–G18 wobble pair, these interactions provided a series of hydrogen bonding and van der Waals contacts with the DNA binding surface of NF-κB p50. In the crystal complex with NF-κB p50₂, one aptamer binds to each NF-κB subunit, requiring a large opening of the Rel homology domains in the protein dimer relative to the structure bound to DNA (Figure 1C and D). In fact, the aptamer structure mimics the major groove of the normal DNA binding sequence in such a manner that the interacting protein side chain conformations are largely preserved between the two complexes (8,10). Figure 1.

Anti-NF-κB RNA aptamer and crystal structure in complex with NF-κB p50₂. (A) Proposed secondary structure of the 29-nt anti-NF-κB RNA aptamer in solution. The RNA hairpin studied here is identical in sequence to the RNA aptamer co-crystallized with NF-κB p50₂ (with exception of the inversion of terminal nucleotides G1 and C29). Canonical RNA Watson–Crick base pairs are red, the 5′ region of the internal loop is cyan, the wobble pair is green, the GNRA-type tetraloop is gray, and the 3′ region of the internal loop is blue. (B) RNA aptamer structure extracted from the crystal structure and colored as in (A). (C and D) Comparison of the crystal complex of NF-κB p50₂ with the RNA aptamer (PDB ID 1OOA) or with bound DNA (PDB ID 1NFK). RNA binding requires substantial opening of the Rel homology domains (8,9). The two p50 subunits are shown in orange and yellow. Nucleic acids are shown as space filling models, where the RNA aptamer structures (C) are colored as in (A) and DNA structure (D) is colored red and pink.

The anti-NF-κB RNA aptamer was transcribed in vitro using purified His₆-tagged T7 RNA polymerase and synthetic DNA oligonucleotides (Integrated DNA Technologies), as previously described (23,24). RNA was purified by denaturing 15% polyacrylamide gel electrophoresis, identified by UV absorbance, excised from the gel, recovered by diffusion into 0.3 M sodium acetate and precipitated from ethanol. The RNA was further purified on an anion exchange column [MonoQ (5 × 1 cm²)] and desalted on a gel filtration column [BioRad P6 (5 × 1 cm²)] performed on a BioRad LC system. The purified RNA was lyophilized, resuspended in water and brought to pH 6.8 by the addition of 1 M NaOH. The sample was further lyophilized to enable preparation of a 99.99% D₂O solution for the acquisition of non-exchangeable proton data. ¹³C/¹⁵N-labeled RNA was prepared using ¹³C/¹⁵N-labeled rNTPs [Isotec Inc. (Sigma Aldrich) Miamisburg, OH]. All NMR samples were ∼1 mM RNA and 50 mM NaCl.

NMR spectra were collected on 750 MHz Bruker Avance DMX or 900 MHz Varian Inova spectrometers at the National Magnetic Resonance Facility at Madison (NMRFAM). The spectrometers were equipped with a single z-axis gradient HCN cryoprobe and conventional room-temperature probe, respectively. Exchangeable resonances were assigned by reference to 2D NOESY spectra (50, 100 and 150 ms mixing time) in 90% H₂O/10% D₂O at 283 K using a 1-1 spin echo water suppression scheme (25). Non-exchangeable resonances were assigned by reference to 2D NOESY spectra (100, 200, 250 and 300 ms mixing times), 2D (¹H–¹H) TOCSY, 2D ¹H–¹³C HSQC (aromatic and ribose optimized), 3D (¹H–¹³C–¹H) TOCSY, 3D (¹H–¹³C–¹H) COSY and 3D (¹H–¹³C–¹H) NOESY-HMQC spectra of the RNA in 99.99% D₂O at 293 and 303 K, as previously described (26). The completeness of the ¹H chemical shift assignments was 97% for non-exchangeable and 67% for exchangeable resonances. For experiments in 99.99% D₂O, the residual HDO resonance was suppressed with a low power pre-saturation pulse. Hydrogen bonds for the ¹³C, ¹⁵N-labeled sample in 90% H₂O/10% D₂O at 298 K were detected using a ²J_NN 2D HNN COSY experiment (27). Thermal unfolding of the RNA aptamer was monitored by acquiring a series of 1D ¹H spectra at increasing temperatures (4, 8, 12, 18, 22, 26, 30, 35, 40, 45 and 50°C). Partial alignment of RNA for residual dipolar coupling (RDC) measurements was achieved by adding 17 mg/ml Pf1 filamentous bacteriophage (ASLA Ltd, Riga, Latvia) to the ¹³C/¹⁵N-labeled samples (28). The ¹J_C–H couplings were measured under isotropic and anisotropic conditions using a J-modulated [¹³C–¹H] CT-HSQC with a total of 20 and 10 J-modulated planes for aromatic and ribose optimized HSQC experiments, respectively (28,29).

Structural constraints for the solution structure of the NF-κB RNA aptamer were categorized by the qualitative peak volumes of the NOE [strong (1.8–3.0 Å), medium (2.0–4.5 Å) or weak (3.0–6.0 Å)] obtained from 2D NOESY spectra (99.99% D₂O) with mixing times of 100, 200, 250 and 300 ms. Torsion angle restraints for residues in the lower (G1–C5, G25–C29) and upper (A10–G14, A17–U21) stem were constrained to A-form values (± 15°), which were consistent with NOESY, TOCSY, HNN-COSY and RDC data. Hydrogen bonds and weak planarity restraints (1 kcal mol⁻¹ Å⁻²) were enforced for the 8 Watson–Crick and 1 U–G wobble pair, since these base pairs were unambiguously identified by NOESY and HNN-COSY experiments. Two hydrogen bonds in the G–A pair within the GUAA tetraloop were also enforced, but only after structure calculations in the complete absence of non-experimental restraints indicated formation of a G–A pair in the majority of the accepted structures (based on acceptance criteria, vide infra). Aside from the two hydrogen bond restraints in the G–A pair in the tetraloop, no non-experimentally derived restraints (distance, dihedral or planarity) were used in the structure calculations for any of the internal loop or tetraloop nucleotides.

An extended structure generated in CNS 1.1 (30) was used to calculate 100 starting structures, as previously described (31). The 50 structures with lowest energy were refined using Xplor-NIH (32) with conformational database potentials (33). Use of the conformational database potentials did not alter the structure of the RNA loop regions, as determined by comparison of structures calculated in their absence, and were validated by an improved fit to the RDC data. PALES software (http://spin.niddk.nih.gov/bax/software/PALES) was used to estimate the values of the axial (D_a) and rhombic (R) components of the alignment tensor from the converged, low energy structures calculated by Xplor-NIH, as previously described (24). Optimal D_a/R-values were –25/0.14 for the 19 lower stem (G1–C5, G25–C29) RDC values. RDC values for the upper helix, internal loop and GUAA tetraloop could not be fit to the lower helix alignment tensor, and were therefore not used in the final structure calculations. The lowest-energy structures with no NOE violations (>0.5 Å), dihedral violations (> 5°) and RDC violations (>3 Hz) were accepted. Structures were analyzed in MOLMOL (34) and figures were created using PyMOL (www.pymol.org).

The 29-nt anti-NF-κB RNA aptamer sequence (Figure 1A) binds tightly to the p50₂ form of NF-κB, and competitively blocks NF-κB p50₂ binding to duplex DNA in vitro and in vivo (1,2,7). In the RNA–p50₂ complex (8), the RNA has a pronounced kink within its asymmetric internal loop (Figure 1B), which makes extensive interactions with protein residues.

Analysis of the free RNA aptamer under NMR conditions (1 mM RNA, 50 mM NaCl, pH 6.8) by native gel electrophoresis showed a single conformation indicative of a monomeric hairpin stem-loop structure (Supplementary Figure S1). Multi-dimensional TOCSY, NOESY and HSQC data exhibit disperse resonances throughout the RNA, suggesting that the molecule is well structured in a single hairpin conformation. The 1D ¹H NMR and 2D HNN COSY NMR spectra reveal that the RNA contains eight Watson–Crick base pairs, as expected (Figure 2A). It should be noted that the minor RNA conformational difference at the 5′ and 3′ termini is due to the presence of a mixture of n and n + 1-transcription product lengths. Additional resonances noted in Figure 2A likely correspond to alternative G1 and U28 imino protons and were not incorporated into the NMR structure calculations. Figure 2.

NMR evidence for a single, pre-organized RNA aptamer conformation. (A) Base pairing interactions of the RNA aptamer in solution. A 1D 1H NMR spectrum and resonance assignments for the imino protons (top). A 750 MHz 2D HNN COSY experiment showing cross-hydrogen bond scalar couplings (²J_N,N) (bottom). Resonances and chemical shift assignments are colored as in Figure 1. The number and asterisk symbols correspond to a minor n + 1 transcription product, which has different chemical shifts for the U28 (#) and G1 (*) imino protons, but does not otherwise interfere with analysis. (B) A 900 MHz 2D ¹H–¹H NOESY spectrum (250 ms mixing time) of the RNA aptamer. Intranucleotide H1′–H6/8 NOE assignments are indicated by number and a sequential internucleotide ‘walk’ is illustrated. Numbers are colored as in Figure 1. The parenthesis for nucleotide G18 denotes that its H1′ chemical shift is not in the region displayed.

Structures of the anti-NF-κB RNA aptamer were calculated with 541 NOE-derived distance restraints, 160 backbone torsion angle restraints for the helical regions, 25 hydrogen bond restraints based on NOESY and ²J_NN COSY analysis, and 19 residual dipolar couplings incorporated at the refinement stage (Figure 2 and Table 1). The structural statistics and stereo superimposition of the 10 lowest-energy structures are shown (Table 1 and Figure 3A). The lower (nt 1–5, 25–29) and upper (nt 10–13, 18–21) helical regions adopt predominantly A-form geometries, and the GUAA fold has the same stacking and hydrogen bonding patterns as previously observed for GNRA tetraloop structures, despite the presence of an adjacent U13-G18 wobble pair (Figures 3 and 4A) (36,37). In contrast, the internal loop deviates from a typical A-form fold (Figures 3B and 4B). Unpaired pyrimidines 6 and 7 contain aromatic stacking interactions within the helix while C24 is stacked between the two uridines. Cross-strand stacking of G8 and G23 reduces the interphosphate cross-strand distance by 3 Å, down to 15 Å in comparison to a regular A-form strand width of 18.5 Å. The non-A-form helical geometry of the internal loop structure appears to be largely due to this cross-strand stack. Cross-strand stacking interactions involving G-A pairs are relatively common in RNA structures (38–41). Figure 3.

Global structure of the free and bound anti-NF-κB RNA conformations. (A) Stereo view of the superimposition over all non-hydrogen atoms of the 10 lowest energy solution structures (RMSD ∼0.94 Å). The 5′ (nt 1–13) and 3′ (nt 18–29) termini of the RNA are colored cyan and blue, respectively, while the GUAA tetraloop is colored gray. (B) Stereo view of the superimposition over just the lower helix. The NMR structure is colored according to Figure 3A and the crystal structure is shown in red (nt 1–17) and pink (nt 18–29).

This study enables a comparison of the detailed structures of RNA and protein partners before and after complex formation. We show that the anti-NF-κB RNA aptamer is pre-structured, specifically through non-canonical base pairing and a cross-strand stacking arrangement within the asymmetric internal loop. This structure is likely crucial for engagement of the DNA-binding surface of the p50 protein. We find that RNA structural changes induced upon protein binding are (i) a hinge-like bending within the internal loop to change the orientation of the two Watson–Crick stems, and (ii) a protein-induced perturbation of the GUAA tetraloop, which otherwise adopts a canonical fold in the free RNA. Further, we propose that the cross-strand guanine stack within the major groove of the internal loop is a critical structural element for p50₂ recognition and specificity, assisting in the molecular mimicry of the κB DNA target sequence by expanding the major groove. A comparison of the overall fold and helical parameters between the lowest energy NMR structure, B-form DNA, and A-form RNA support this hypothesis, illustrating that the anti- NF-κB RNA aptamer closely imitates the exposed major groove face of the DNA (Table 2, Figure 6). Figure 6.

Comparison of B-form DNA, the NMR structure of the free NF-κB RNA aptamer and A-form RNA. (A) Idealized B-form duplex DNA is colored as in Figure 1D and contains the sequences 5′-GATACTTGAACGT-3′ and 5′-ACGTTCAAGTATC-3′. (B) Anti-NF-κB RNA aptamer NMR structure (blue), capped by the GUAA tetraloop (gray). (C) Idealized A-form duplex RNA is colored brown and tan and contains the sequence 5′-GAUACUUGAACGU-3′ and 5′-ACGUUCAAGUAUC-3′. The 26-nt DNA and RNA coordinates were generated using Insight (Biosym).

Although an increasing number of cases in the literature investigate RNA folding in free and protein-bound states, there is a limited number of examples where both RNA conformations are structurally defined (13–22). How does the anti-p50 RNA aptamer compare with other RNAs with respect to folding rearrangement upon target binding? There is a general trend toward mutually induced fit between RNAs and their binding partners, but important examples exist where binding is not accompanied by major changes in RNA folding. In these examples, much or all of the interaction surface of the RNA molecule is preformed in the absence of the binding partner. The tRNA/tRNA synthetase complexes are examples where structured and pre-folded RNAs are observed (46–54). Upon protein binding, changes in secondary and tertiary tRNA structure are typically minor, and are limited to the flexible anticodon loop and the 3′ acceptor nucleotide. The overall geometry of the folded tRNA appears relatively rigid and autonomous, and this appears to also be true for a tRNA aptamer selected for increased synthetase affinity (55).

Other helical RNAs have also been found to be pre-structured prior to binding to their protein targets. Specifically, two examples (the SAM protein domain–SRE RNA and Rnt1p protein–snR47h RNA complex) reveal protein recognition through structured RNA pentaloops or tetraloops (17,18). In the case of the SAM domain–SRE RNA interaction, both the protein and the RNA are structured and complex formation occurs upon binding to the major groove of a pre-formed pentaloop structure (17). For the Rnt1p–snR47h–RNA complex, the AGNN tetraloop conformation also retains the identical fold in the free and bound states (18). Thus, unlike the tetraloop conformational change that occurs in the anti-NF-κB aptamer, these two examples show rigid, pre-formed loop structures that do not undergo conformational re-organization upon protein binding.

The concept of protein-induced RNA folding has been reviewed by Williamson (14), who highlighted structures of RNA binding partners determined both alone and in a complex. Cases include the ribosomal S15 and L11 proteins inducing large structural changes upon ribosomal RNAs, ribonucleoprotein U1A changing the folding of the 5′ untranslated region of its own mRNA upon binding, and a similar example for ribosomal protein L30 binding to its own mRNA. In each of these examples, both binding partners undergo conformational changes upon interaction, though the effect is most striking for the RNA. In the present case, protein contacts at the RNA/p50₂ interface are nearly identical to those observed in the κB-DNA/p50₂ interface, while only the overall N-terminal protein domain orientation has changed to accommodate aptamer binding (Figure 1C and D).

The increasing number of detailed structural studies for protein and RNA partners alone and in complex provides additional opportunity for comparison. In one example, structure determination of the free protein (56), free RNA (13) and complex (57) enabled a thorough analysis of nucleolin binding to a consensus RNA aptamer. The authors show that certain elements of RNA stem structure were detectable in the absence of protein, but a dynamic equilibrium of RNA loop structures was also observed. Nucleolin binding stabilized one particular RNA form, clearly confirming the existence of a mutually induced protein–RNA complex (13). Thus, like the anti-p50 RNA aptamer, the RNA reorganization upon protein complex formation might be characterized as moderate, with pre-formation of long-range RNA structure that is retained in the complex.

Coordinates for the 29-nt anti-NF-κB-RNA aptamer have been deposited into the RCSB Protein Data Bank (accession code 2jwv) and NMR resonance assignments have been deposited into BioMagResBank (accession code 15538).