Fluoride ion encapsulation by Mg²⁺ and phosphates in a fluoride riboswitch

Abstract

Significant advances in our understanding of RNA architecture, folding and recognition have emerged from structure-function studies on riboswicthes, non-coding RNAs whose sensing domains bind small ligands and whose adjacent expression platforms contain RNA elements involved in the control of gene regulation. We now report on the ligand-bound structure of the Thermotoga petrophila fluoride riboswitch, which adopts a higher-order RNA architecture stabilized by pseudoknot and long-range reversed Watson-Crick and Hoogsteen A•U pair formation. The bound fluoride ion is encapsulated within the junctional architecture, anchored in place through direct coordination to three Mg²⁺ ions, which in turn are octahedrally coordinated to waters and five inwardly-pointing backbone phosphates. Our structure of the fluoride riboswitch in the bound state defines how RNA can form a binding pocket selective for fluoride, while discriminating against larger halide ions. The T. petrophila fluoride riboswitch most likely functions in gene regulation through a transcription termination mechanism.

The field of RNA structure, folding and recognition has been propelled forward by the discovery of metabolite-sensing bacterial non-coding RNA elements, termed riboswitches, that have been shown to impact on RNA-mediated gene regulation^1,2. Most riboswitches are positioned within 5′-untranslated regions of genes associated with transport and metabolism of their cognate metabolites. Riboswitches interconvert between metabolite-free and -bound conformations depending on metabolite concentration, with the sensing domain involved in metabolite recognition, while the adjacent expression platform contains RNA elements that control translational initiation, transcription termination or ribozyme-mediated cleavage. To date, structure-function studies have been undertaken on riboswitches that target purines and their analogs, amino acids, coenzymes and phosphoamino sugars, as well as Mg²⁺ ions^3,4. Structural studies on the compact ligand-bound sensing domains of the above riboswitches have elucidated the structural principles underlying RNA folding topology and recognition, whereby endogenous RNA molecules can fold and form pockets that recognize small metabolites and discriminate against closely related analogs^5,6.

Of particular interest is how RNA as a negatively charged polyphosphate can bind ligands that are also negatively charged. The earliest studies to address this question focused on the cofactor thiamine pyrophosphate (TPP) riboswitch², where structural insights established that a pair of hydrated Mg²⁺ ions mediated interactions between the diphosphates of TPP and guanine base edges (rather than backbone phosphates) of the RNA^7,8. This structural information was critical for guiding subsequent studies on ligand-induced folding of the TPP riboswitch^9,10. Similar structural principles involving a bridging hydrated Mg²⁺ ion were also utilized for recognition of the monophosphate of flavin mononucleotide (FMN) by its riboswitch^11,12.

Recently, a riboswitch associated with crcB motif non-coding RNAs from Pseudomonas syringae has been identified that targets fluoride ion with a K_d of approx. 60 μM and discriminates against other halogen ions¹³. This riboswitch is common to bacterial and archaeal species and found to activate the expression of genes that encode putative fluoride transporters. Given the small size and negative charge of the fluoride ion, it seems remarkable that RNA can form a small enough pocket to target it and discriminate against larger halide ions.

Structure of fluoride-bound riboswitch

The conserved secondary fold of crcB RNA motif fluoride-sensing riboswitches was deduced following sequence conservation and covariational analysis amongst bacterial and archaeal species, as well as in-line probing and mutational studies¹³. The resulting analysis identified two helical stems connected by a large asymmetric internal loop, with overhangs at both ends, with that at the 5′-end capable of adopting pseudoknot-like higher-order interactions.

We used isothermal titration calorimetry (ITC) to establish that the 52-mer RNA sequence corresponding to the sensing domain of the T. petrophila fluoride riboswitch (Fig. 1a) bound fluoride ion (on addition of KF) with a K_d of 134 ± 9 μM under 5 mM Mg²⁺ conditions (Fig. 1b). The stoichiometry of binding approaches 1:1 (N = 0.87), with estimated thermodynamic parameters of ΔH = −2.5 ± 0.1 Kcal/mol and ΔS = 9.3 cal/mol/deg. Similar binding parameters were observed under 1 mM Mg²⁺ conditions, but no binding was observed in the absence of Mg²⁺ ion (Supplementary Fig. 1a, b, respectively).

Figure 1. Sequence, binding affinity and structure of the sensing domain of the T. petrophila fluoride riboswitch in the ligand bound state.

a, Secondary structure schematic of the 52-nt sensing domain of the T. petrophila fluoride riboswitch used in this study. The color-coding highlights regular stem segments (blue and green) and long-range pseudoknot pairing interactions (magenta and pink; connected by lines), and bases not involved in pairing interactions (gold). Five phosphates, whose non-bridging oxygens form a shell around the three hydrated metal ions (in cyan), with the metal ions in turn coordinated to the fluoride ion (in red), are labeled by red stars. b, ITC-binding curve for binding of KF to the T. petrophila fluoride riboswitch in 5 mM Mg²⁺-containing buffer (for experimental conditions, see Methods section). c, 2.3 Å crystal structure of the fluoride riboswitch in the ligand bound state. The color-coding of RNA segments follows that shown in panel a, with the fluoride ion shown by a red ball and directly-coordinated metal ions by cyan balls. d, A close-up stereo view of the ligand-binding pocket in the same perspective as in panel c, with the emphasis on the fluoride ion, three coordinating metal ions and five inwardly-pointing backbone phosphates. e, F_o-F_c omit electron density map contoured a 4σ level calculated following deletion of the fluoride ion. f, Anomalous density at 9σ level at positions of metal ions 1, 2 and 3 for crystals of the complex soaked in 50 mM Mn²⁺-containing solution.

The sensing domain of the T. petrophila fluoride riboswitch (Fig. 1a) yielded crystals in the presence of fluoride ion that diffracted to 2.3 Å resolution. We solved the structure of this fluoride riboswitch by cocrystallizing it with Ir(NH₃)₆³⁺, and capitalizing on the anomalous properties of iridium to solve the phase problem. (Supplementary Fig. 2). The three dimensional structure in the bound state is shown in Fig. 1c with different elements color-coded as shown in Fig. 1a. The most striking feature is that the bound fluoride ion (red ball, Fig. 1c), positioned within the center of the riboswitch fold, is surrounded and directly coordinated by three metal ions (cyan balls, Fig. 1c; metal-fluoride distances of 1.8–2.0 Å). A close-up stereo view of the ligand-binding pocket in the same perspective as in Fig. 1c, with the emphasis on the fluoride ion (in red), three coordinating metal ions (in cyan) and five inwardly-pointing backbone phosphates (non-bridging oxygens in pink and phosphorus atoms in yellow) is shown in stereo in Fig. 1d.

Fluoride coordinated to three Mg²⁺ ions

We have identified the fluoride ion binding site in the F_o-F_c omit map for fluoride ion (Fig. 1e) and assign all three metal ions coordinated to the fluoride ion, to Mg²⁺ ions, based on observation of anomalous signals following soaking of the crystals in 50 mM Mn²⁺ ion-containing solution (Fig. 1f and Supplementary Fig. 3). The additional metal ion (in orange, Fig. 1c) positioned close by but not coordinated to the fluoride ion (metal-fluoride distance of 4.0 Å), is assigned to a K⁺ cation (buffer contained K-acetate), based on the observed anomalous signal at its position following soaking of crystals in Cs⁺ ion-containing (Supplementary Fig. 4a) and Tl⁺ ion-containing (Supplementary Fig. 4b) solutions.

Riboswitch adopts pseudoknot scaffold

The higher order fold of the fluoride-bound riboswitch is stabilized by pseudoknot formation involving residues G2-G3-C4-G5 of the 5′-overhang segment and residues C14-G15-C16-C17 of the internal loop, thereby forming a regular duplex composed of four stacked Watson-Crick G-C base pairs (in magenta, Fig. 2a). This validates the prediction of pseudoknot formation amongst the same residues¹³, though our crystal structure shows stabilization by four rather than the postulated five base pairs anticipated in solution. In this regard, 5′-terminal G1 and 3′-terminal U51 and G52 are involved in crystal packing contacts in our structure of the fluoride riboswitch (Supplementary Fig. 5). None of the residues in the C18 to U23 segment of the internal loop are involved in pairing interactions, but rather C18 stacks on terminal Watson-Crick G2-C17 pair, with a sharp turn at the C18-A19 step, followed by continuous stacking within the A19-A20-A21 and C22-U23 steps (Fig. 2a). The junctional architecture in the vicinity of the fluoride-binding site is additionally stabilized by formation of long-range single-base pseudo-knot like pairing between A6 and U38 (predicted previously¹³) and between A40 and U48 (Fig. 2b). Both form non-canonical pairs, with A6•U38 pairing through a reversed Watson-Crick alignment (Fig. 2c) and A40•U48 pairing through a reversed Hoogsteen alignment (Fig. 2d). Note that unpaired U7 and G39 are mutually interdigitated (Fig. 2b) and contribute to formation of the junctional architecture. The tracing of these RNA segments in the 2F_o-F_c electron density maps of the fluoride riboswitch are shown in Supplementary Fig. 6a, b.

Figure 2. Details of long-range interactions within the structure of the T. petrophila fluoride riboswitch in the ligand bound state.

a, An expanded region of the structure highlighting pseudoknot formation involving residues G2-G3-C4-G5 of the 5′-overhang segment and residues C14-G15-C16-C17 of the large internal loop (in magenta), as well as continuous stacking within the A19-A20-A21 and C22-U23 steps. b, An expanded region of the structure highlighting long-range single-base pseudo-knot like pairing between A6 and U38, and between A40 and U48. Note the mutual interdigitation between unpaired U7 and G39 that contribute to formation of the junctional architecture. c, Long-range reversed Watson-Crick A6•U38 pair formation. d, Long-range reversed Hoogsteen A40•U48 pair formation.

Coordination of Mg²⁺ ions by phosphates

The negatively charged fluoride ion (in red) is directly coordinated to three Mg²⁺ ions (in cyan) labeled M1, M2 and M3 (stereo view in Fig. 3a), with the fluoride ion positioned somewhat out of the plane (0.49 Å) formed by the three Mg²⁺ ions (Fig. 3b). A stereo-view of the corresponding F_o-F_c omit electron density maps for fluoride ion, three Mg²⁺ ions and bound waters are shown in Supplementary Fig. 7a (contoured at 3σ) and 7b (contoured at 7σ). The three Mg²⁺ ions in turn are coordinated by waters and five inwardly-pointing non-bridging phosphate oxygens (in pink; phosphorus in yellow), with three of these phosphates involved in bidentate coordination (pA6, pU7 and pG42), while two others involved in monodentate coordination (pU8 and pU41) (Fig. 3a). A stereo-view of the corresponding F_o-F_c omit electron density maps for the five nucleotides with inwardly-pointing phosphates are shown in Supplementary Fig. 7c (contoured at 3σ) and 7d (contoured at 5σ).

Figure 3. Details of fluoride ion binding site in the T. petrophila fluoride riboswitch in the ligand bound state.

a, An expanded stereo view highlighting the non-bridging phosphate oxygens directly coordinated to the three Mg²⁺ ions labeled M1, M2 and M3, that are in turn directly coordinated to the fluoride ion. In terms of nomenclature, the phosphate labeled pA6 corresponds to the G5pA6 step. Three of the phosphates are involved in bidentate coordination with the metals, while two are involved in mondentate coordination. b, The fluoride ion is positioned 0.49 Å above the plane formed by the three metal ions. c, A stereo view of the coordination geometries of fluoride ion with surrounding Mg²⁺ ions, and in turn of Mg²⁺ with surrounding non-bridging phosphate oxygens and waters. Metal coordination within square-planar positions are shown by dashed blue lines, while coordination in apical positions are shown by dashed red lines in panels a, c and d. d, These panels show the octahedral coordination geometries around Mg²⁺ ions M1, M2 and M3, together with coordination distances in Å.

In essence, our structure identifies a unique solution for how a negatively-charged RNA scaffold can target a negatively-charged fluoride ligand. The fluoride ion is surrounded by and coordinated to an inner shell of three Mg²⁺ ions, which in turn are surrounded and coordinated to an outer shell of five backbone phosphates and waters. Notably, the five participating phosphates are located within two distinct segments of the sequence, with three of them residing within the 5′-overhang G5pA6pU7pG8 segment, while two other reside within the A40pU41pG42 internal loop segment, as labeled in Fig. 1a by red asterisks. Because of the unique orientation of these five inwardly-directing backbone phosphates, 8 of the 14 internal loop residues are stacked but not involved in hydrogen-bond pairing, and differ from large internal loops of other structurally characterized riboswitches, where ligand-binding results in a compaction mediated by maximal hydrogen-bond pairing and stacking of loop residues^14–19.

Metal ions and their coordinated waters were identified based on 2F_o-F_c and F_o-F_c maps guided by the coordination geometries. The positioning of fluoride ion, non-bridging phosphate oxygens and water molecules that constitute the three Mg²⁺ ion cluster that is coordinated to the fluoride ion is shown in stereo in Fig. 3c, with individual Mg²⁺ ions adopting the anticipated octahedral-like alignments as shown in Fig. 3d. Mg²⁺ coordination within the square-planar arrangements are shown by dashed blue lines, while coordination in apical positions are shown by dashed red lines, together with distances listed in Å in Fig. 3d. Notably, the fluoride riboswitch does not bind chloride ion (added KCl) in the presence of 20 mM Mg²⁺ as monitored by ITC (Supplementary Fig. 1c). In addition, it does not bind fluoride ion (added KF) in the presence of 20 mM Li⁺ ions (Supplementary Fig. 1d), in contrast to the observed binding under Mg²⁺ ion conditions (Fig. 1b).

Fluoride-Mg²⁺ coordination in proteins

Of additional note, the fluoride ion adopts an apical position in the octahedral coordination geometries for all three Mg²⁺ ions (Fig. 3d). Mg²⁺ ion M1 is coordinated by four non-bridging phosphate oxygens aligned in a square planar arrangement, while Mg²⁺ ions M2 and M3 are coordinated by two non-bridging phosphate oxygens (Fig. 3d). Additional support for our model of the coordination geometry of three Mg²⁺ ions around fluoride in the fluoride riboswitch comes from the highly similar coordination geometry observed in the 1.9 Å structure of fluoride-inhibited pyrophosphotase (crystals grown from 1 mM MnCl₂ and 5 mM NaF-containing buffer)²⁰. Here, a fluoride ion is coordinated by a similar three-metal ion arrangement in the pyrophosphase system (Supplementary Fig. 8). In the fluoride-inhibited pyrophosphotase structure, two Mn²⁺ ions and one Na⁺ ion are coordinated by oxygens atoms from one POP molecule and from four aspartate carboxylate groups, and water molecules (Supplementary Fig. 8a).

We are aware that fluoride anion is a potent hydrogen bond acceptor^13,21, but currently have no evidence in support of direct F^--H hydrogen bonding in the structure of the fluoride riboswitch, nor was such hydrogen bonding reported for the fluoride-inhibited pyrophosphatase system²⁰.

Previous studies of metals in RNA folding, stability and catalysis have focused on monovalent and divalent cations²², including Mg²⁺ clusters bridging closely-positioned phosphates in 5S RNA²³, P4-P5-P6 fragment of group I introns²⁴ and Mg²⁺-sensing riboswitches^25,26. Our current contribution based on the discovery of a fluoride riboswitch¹³, has defined how a RNA scaffold combines an inner shell of metal ions and an outer shell of phosphates to completely encapsulate a fluoride ion.

Folding and mechanism of action

We have recorded 900 MHz imino proton NMR spectra of the fluoride riboswitch in the free and fluoride bound states, which establish a conformational transition (chemical shift changes) and compaction (additional peaks) through higher order structure generation on complex formation, with slow exchange between the free and bound forms (Supplementary Fig. 9). We have not been able to crystallize the ligand-free form of the fluoride riboswitch in the presence of Mg²⁺ ions. In the absence of this structure, we can only speculate on the folding energy landscape of the fluoride riboswitch as to how it dynamically assembles to recognize its ligand. This could occur through either hierarchical or simultaneous positioning of Mg²⁺ ions, RNA functional groups and ligand for recognition. This issue of folding energy landscape, as well as energetic costs associated with transfer of the fluoride ion from free to the riboswitch-bound state, could perhaps be addressed in the future by comprehensive single molecule studies.

Previous functional studies implied that the B. cereus and P. syringae fluoride-responsive riboswitches control gene expression by regulating transcription termination and translational initiation, respectively¹³. To deduce mechanistic insights into gene regulation by the T. petrophila fluoride riboswitch from a structural perspective would require crystallographic characterization of the complete riboswitch containing both the sensing domain and expression platform in the ligand-free and bound states, which is beyond our current capabilities. Nevertheless, we note that that the sensing domain and adjacent expression platform of the T. petrophila fluoride riboswitch could interconvert between two conformations, one in which stem 1 forms in the presence of bound fluoride (ON state) and an alternate conformation at low fluoride concentrations where stem 1 is disrupted as a result of forming a rho-independent transcription terminator (OFF state) involving a stable stem loop ending in uracils at the 3′-end (Supplementary Fig. 10). Thus, it appears that both the B. cereus (supported by functional data)¹³ and T. petrophila fluoride riboswitches are likely to adopt a transcription termination mechanism for gene regulation.

METHODS SUMMARY

Details of RNA preparation, purification and complex formation, as well as crystallization, x-ray data collection and refinement are described in detail in Full Methods.

Full Methods and any associated reference are available in the on line version of the paper at www.nature.com/nature

METHODS

RNA preparation, purification and complex formation

The crcB motif of the T. petrophila fluoride riboswitch followed by the HDV ribozyme was transcribed in vitro using T7 RNA polymerase²⁷. The transcribed RNA was purified by denaturing polyacrylamide gel electrophoresis (PAGE), followed by anion-exchange chromatography and ethanol precipitation. To generate the complex of the sensing domain of the T. petrophila fluoride riboswitch with fluoride ion, 10 mM KF was added to the buffer consisting of 100 mM potassium-acetate, pH 6.8 and 5 mM MgCl₂. After annealing at 60 °C for 10 min, the complex was purified by gel-filtration chromatography, prior to setting up crystallization trials.

Crystallization

Crystals of the fluoride anion-bound T. petrophila fluoride riboswitch grew at 20 °C over a period of 1 week using sitting-drop vapor diffusion approach after mixing the complex at an equimolar ratio with the reservoir solution containing 0.1 M MES-sodium, pH 6.5, 35–40% (w/v) MPD. To generate heavy-atom-bound crystals, 5 mM Ir(NH₃)₆³⁺ was mixed with 0.5 mM complex before setting up crystallization trials. The Ir(NH₃)₆³⁺ bound crystals grew from a solution containing 0.1 M sodium-cacodylate, pH 7.0, 20 mM spermine, 0.2 M strontium chloride and 20% MPD over a period of 4 days.

For heavy atom and cation soaking, crystals grown from 0.1 M sodium-cacodylate, pH 7.0, 20 mM spermine, 160 mM KCl, 50 mM MgCl₂ and 20% MPD were transferred into the mother solution of the crystals with MgCl₂ replaced with 50 mM MnCl₂, or with KCl replaced with 50 mM Tl-acetate or 100 mM CsCl for 24 hr.

For cryoprotection, crystals were passed through several 5 μL drops of the stabilizing solution, which was the reservoir solution with MPD replaced by 30% MPD and 5% glycerol.

X-ray data collection and refinement

X-ray diffraction data were collected on flash frozen crystals of the fluoride anion-bound T. petrophila fluoride riboswitch at NE-CAT beamlines at the Advanced Photon Source, Argonne National Laboratory and processed using the HKL2000 program (HKL Research). The structure (space group: P2₁2₁2) was determined using SAD technique employing anomalous signal from four iridium atoms using HKL2MAP program²⁸ and PHENIX²⁹ suite. The RNA model was built in COOT³⁰ and refined in PHENIX²⁹ and REFMAC³¹ using 2.3 Å native data set of the ligand-bound fluoride riboswitch. Metal ions and their coordinated waters were identified based on 2F_o-F_c and F_o-F_c maps guided by the coordination geometries. Fluoride ion was added to the model at the last stage based on the experimental and refined maps, coupled with electrostatic analysis. The x-ray statistics of the native and iridium-containing crystals are listed in Supplementary Table 1. Anomalous data sets of Mn²⁺, Cs⁺, and Tl⁺ were collected at the wavelength of 1.7712 Å. The x-ray statistics of the complex crystals soaked in Mn²⁺-, Cs⁺-, and Tl⁺-containing solutions are listed in Supplementary Table 1. Mn²⁺, Cs⁺, and Tl⁺ cations were positioned based on the anomalous electron density maps (Supplementary Figs. 3, 4a and 4b). Cations were interpreted as Mg²⁺ or K⁺ on the basis of the anomalous maps of their mimics, coordination geometry and distances³².

NMR Spectra

Imino proton NMR spectra (10 to 15 ppm) of the fluoride riboswitch in the free and bound state were recorded on a 900 MHz Bruker NMR spectrometer with cryoprobe using a jump-and-return pulse for H₂O solvent suppression. NMR spectra were recorded in buffer containing 50 mM K-acetate-d₃, 5 mM Mg-sulfate, 90%H₂O/10%D₂O, pH 6.8 at 25 °C.

Isothermal Titration Calorimetry

All experiments were performed on a Microcal ITC200 calorimeter at 20 °C. Prior to titration, 0.2–0.3 mM RNA samples of the fluoride-bound riboswitch were dialyzed overnight at 4 °C against experimental buffer containing 50 mM potassium acetate, pH 6.8, and 0 to 20 mM MgSO₄ or other cations to remove bound fluoride. RNAs were refolded by heating at 60 °C for 10 min and followed by cooling on ice. For measurements, KF dissolved in the dialysis buffer at 5 mM concentration, was typically titrated into the RNA in the sample cell (V = 207 μL) by 20 serial injections of 2 μl each, with a 0.5 μl/s rate, 180 s intervals between injections, and a reference power of 6 μcal/s. The thermograms were integrated and analyzed by using Origin 7.0 software (Microcal, Inc.).