|Home | About | Journals | Submit | Contact Us | Français|
Regulatory mRNAs elements termed riboswitches respond to elevated concentrations of cellular metabolites by modulating expression of associated genes. Riboswitches attain their high metabolite selectivity by capitalizing on the intrinsic tertiary structures of their sensor domains. Over the years, riboswitch structure and folding have been amongst the most researched topics in the RNA field. Most recently, novel structures of single- and cooperative double-ligand sensors have broadened our knowledge of architectural and molecular recognition principles exploited by riboswitches. The structural information has been complemented by extensive folding studies, which have provided several important clues on the formation of ligand-competent conformations and mechanisms of ligand discrimination. These studies have greatly improved our understanding of molecular events in riboswitch-mediated gene expression control and provided the molecular basis for intervention into riboswitch-controlled genetic circuits.
Regulatory mRNA elements termed riboswitches have evolved for specific sensing of cellular metabolites and modulation of gene expression in species from all three kingdoms of life. Following their landmark discovery in 2002 as missing links in the feedback control of vitamin-related genes [1,2], riboswitches in the subsequent decade have materialized into one of the most important mechanisms of gene expression control in bacteria. Most bacterial riboswitches are located within the 5′-untranslated regions of genes associated with metabolism and transport of their cognate metabolites. Typically, riboswitches are composed of two domains, an evolutionary conserved metabolite-sensing module and a variable expression platform that contains RNA elements required for gene repression or activation. Depending on metabolite availability, riboswitches can adopt mutually exclusive metabolite-bound and metabolite-free conformations that oppositely affect gene expression (Figure 1a,b). This conformational interplay is often based on the ability of a segment that forms the closing helix P1 of the sensing domain to alternately pair with an expression platform region.
Riboswitches respond to a variety of metabolites ranging from large coenzymes to small cations and anions [3,4]. Given the low complexity of RNA molecules, the intriguingly high selectivity of riboswitches has been the subject of intensive structural research . This review focuses on the three-dimensional structures and mode of action of seven additional riboswitch classes investigated recently. Aside from the direct structural approach, several laboratories have pursued biochemical and biophysical studies, also discussed in this review, which have uncovered folding trajectories of several small and medium sized riboswitches, thereby demonstrating the variety of approaches implemented by RNA for adopting binding-competent conformations. These recent structural and structure-guided studies have provided a more comprehensive overview of factors underlying the molecular mechanisms of riboswitches action, thereby establishing a solid foundation for the development of metabolite-like antibacterials and re-engineered RNA-based molecular sensors.
Purine and related riboswitches constitute a diverse group of RNA elements responding to purine bases and their derivatives. The first reported structures of riboswitches, adenine and guanine sensors [6,7], have been recently supplemented with structures of natural  and hybrid  riboswitches specific to the nucleoside 2′-deoxyguanosine (dG), structures [10–13] of the class I riboswitch responding to the modified purine base pre-queuosine-1 (preQ1–I), and structures of two classes of sensors selective for the cyclic purine dinucleotide c-di-GMP [14–16]. The above riboswitches adopt either three-way junctional or pseudoknot folds reinforced by long-distance tertiary interactions and, in addition, stabilized by ligand binding.
Despite shortened hairpins P2 and P3 and numerous nucleotide changes , the dG riboswitch adopts a three-way junctional fold reminiscent of the tuning fork-like conformation of the canonical guanine riboswitch (Figure 1c, left). Not surprisingly, several mutations in the conserved guanine riboswitch core can switch the guanine selectivity to a modest specificity for dG  by providing room for the deoxyribose moiety of the ligand. The most critical difference between guanine and dG riboswitches is the U58C substitution, which causes sliding of the pyrimidine along the ligand and alternative base pairing with dG (Figure 1c, bottom) [8,9]. Additional dissimilarities, that likely account for the specificity and affinity of the dG riboswitch, have been observed in regions outside of the ligand-binding pocket. They include (i) involvement of a purine (G33) in formation of the (G34•U52)•G33 triple within stem P2, thereby replacing an unpaired stacked purine (A24) within stem P3 of the guanine riboswitch [8,18] (Figure 1c, middle) and (ii) novel ‘key-and-lock’ inter-loop interactions formed by adenine A71 insertion into the L3 loop , thereby replacing base quartets observed in the guanine riboswitch (Figure 1c, top).
The smallest riboswitch to date, which binds to preQ1, forms a tight H-type pseudoknot conformation (Figure 1a) where loops L1 and L3 are positioned in the major and minor grooves of co-axially stacked stems S2 and S1, respectively (Figure 1d, top) [10–13]. PreQ1 is intercalated into the interhelical interface and maintains continuous stacking of the two stems. Similar to purine riboswitches [6,7], preQ1 is almost completely encapsulated in the ligand-binding pocket. As in guanine [6,7] and dG  riboswitches, preQ1 forms a Watson-Crick pair with a cytosine (C17), whereas other edges are specifically recognized by base and backbone residues of the riboswitch (Figure 1d, bottom), facilitated by bound Ca2+ cations [12,19].
Like S-adenosyl-(L)-methionine (SAM) riboswitches , the two riboswitch classes responding to the secondary messenger c-di-GMP exemplify alternative structural solutions for recognition of the same ligand. The c-di-GMP-I riboswitch adopts a Y-shaped three-way junctional structure, with helices P2 and P3 joined together by tertiary tetraloop-tetraloop receptor interactions (Figure 1e, top) [14,15]. In the c-di-GMP-II riboswitch , the P2–P3 stem is bent towards J2/4 where it forms tertiary pseudoknot base pairing (P4), which could be considered as a third helix of the pseudo three-way junction (Figure 1f, top). In both riboswitches, ligands bind the core regions through intensive stacking interactions that involve invariant adenines A47 and A70 sandwiched between ligand bases. Nevertheless, both ligand bases and the phosphodiester backbone provide more extensive hydrogen bonding to the RNA in the c-di-GMP-I riboswitch, thereby contributing to the higher affinity of this class for the ligand [20,21] (Figures 1e,f, bottom).
The S-adenosyl-(L)-homocysteine (SAH)-specific riboswitch appears to exclusively up-regulate expression of genes involved in utilization of the by-product of SAM-dependent chemical reactions. The SAH riboswitch adopts an infrequent “LL-type” pseudoknot fold where the P1 helix lies perpendicular to coaxially stacked P2, P2b and P4 helices (Figure 1g, top) . The fold is stabilized by docking J1/4 into the minor groove of P1 and by ligand binding in the cleft created by the minor grooves of P2b and P1 (Figure 1g, bottom). The adenine ring participates in stacking and specific base pairing with guanine G15, whereas the methionine moiety interacts primarily with the RNA backbone.
RNA molecules usually employ cooperativity for folding/unfolding and binding of proteins or Mg2+ cations. Surprisingly, glycine riboswitches that are composed of two adjacent glycine-sensing domains showed cooperative binding of two glycine molecules  (Figure 1b). The structure of the Vibrio cholerae sensing domain II  and the moderate resolution structure of the Fusobacterium nucleatum tandem domain glycine riboswitch  shed light on the cooperative nature of glycine recognition.
The glycine riboswitch features a three-way junctional architecture with glycine positioned above the junction in the widened Mg2+-stabilized helical region of P3 (Figure 2a, top). Glycine is specifically recognized by conserved U69 and purine bases with the help of a Mg2+ cation, which similarly to other riboswitches , neutralizes the negative charge of the ligand (Figure 2a, bottom). The junction and regulatory helix P1 are stabilized by tertiary intercalation of an adenine (A33) extruded from the glycine-binding pocket.
Intermolecular interactions in the asymmetric unit of the domain II structure  and the interdomain interactions in the tandem riboswitch structure  (Figure 2b, top and middle), coupled with biochemical data [26,27], strongly suggest the formation of three interdomain contact regions (quartenary interactions) likely involved in the cooperative response. Two pseudo-symmetrical contacts, designated α–α′ and β–β′, are formed by the insertion of non-paired adenine-rich segments into the minor groove of helices P1 (Figure 2b, top and bottom). Additional interactions, designated γ–γ′, involve a non-canonical U•A base pair. Since the contact regions are located in proximity to the glycine binding pockets, it is conceivable that glycine binding to one domain could induce structural rearrangements that facilitate ligand binding to the second domain.
The structure of the Sterptococcus mutants tetrahydrofolate (THF) riboswitch uncovered another cooperative system, which, in contrast to the tandem glycine riboswitch, employs binding of two ligands to one RNA sensor . The THF riboswitch adopts an ‘inverted’ junctional architecture where the three-way junction and long-distance tertiary contacts have switched places  (Figure 2c, top). One of the THF lignads stabilizes the three-way junction by intercalating into the adjacent widened helix, whereas the other ligand is sandwiched between two helical stems in the vicinity of tertiary base pairing interactions, stabilizing the pseudoknot and helix P1. Ligand binding is similar in both sites and, in contrast to many other riboswitches, mostly involves the recognition of a small part of the ligand moiety by conserved pyrimidines (Figure 2c, middle and bottom), whereas the glutamate moiety appears not to interact with the RNA. Cooperative ligand binding was observed at physiological Mg2+ concentration of 0.5 mM, whereas higher concentrations (6 mM) prompted independent binding .
The Eubacterium siraeum THF riboswitch structure has captured a conformation with a disrupted helix P1, which could be considered an intermediate state of riboswitch folding . This structure lacks the ligand positioned adjacent to the pseudoknot, and additional studies are required to clarify whether this riboswitch binds only one ligand or alternately whether the second binding site is disrupted by crystal packing interactions.
Riboswitches demonstrate high selectivity towards their cognate ligands mostly through shape complementarity and specific interactions. Recent riboswitch structures have revealed several variations on these themes. The dG riboswitch discriminates against guanosine because its ribose adopts the favorable C3′-endo conformation instead of the C2′-endo conformation, eliminating a hydrogen bond with RNA (Figure 3a) . The potential for strong interactions between the dG riboswitch and dGTP is possibly prevented by the flexibility of the dGTP sugar-phosphate moiety and the lack of a dedicated binding pocket on the riboswitch . The SAH riboswitch forms a tight pocket so that SAM, which has an additional methyl group, would clash with the RNA (Figure 1g, bottom) . In the preQ1 riboswitch, the preQ1 ligand forms hydrogen bonds that would be eliminated when its methylamine moiety is replaced by the cyano moiety of the related preQ0 compound [11–13].
Contrary to the conventional view that a riboswitch responds to just a single cognate metabolite, the glmS riboswitch can respond to an array of chemical compounds related to glucosamine N6-phosphate (GlcN6P), thereby assessing information of the overall metabolic state of the cell . Additionally, more than one effector could possibly participate in preQ1 and THF riboswitch control [11,28,29].
Along with these observations, several riboswitches were reported to interact with metabolite-like compounds including antibiotics. Efforts have been undertaken for the structure-guided design of novel riboswitch ligands [31,32], which may exert antimicrobial properties through gene expression suppression. Such studies have been successful in the case of the guanine riboswitch, where artificial compounds were found to inhibit growth of pathogenic bacteria in rich media [33,34] and in a mammalian model of staphylococcal infection .
An additional advance was contributed by Dixon et al., who re-engineered an adenine riboswitch for binding to the non-natural compounds ammeline and azacytosine, using a limited number of core mutations  (Figure 3b). The resulting riboswitches discriminated against purines and responded to synthetic ligands in vivo. Structural studies of the novel riboswitch-ligand pair demonstrated small adjustments within the ligand-binding pocket, including a lateral movement of C51, reminiscent of the dG riboswitch.
Since many ligands are encapsulated in the ‘closed’ binding pockets, riboswitches should adopt a ligand-competent ‘open’ conformation prior to metabolite binding. The surprising resemblance of ligand-free and ligand-bound states described for lysine and glmS riboswitches [36–38] was also observed in the novel structures of THF , glycine  and FMN  riboswitches. In the SAM-I  (Figure 3c), preQ1  and mutant adenine [41,42] riboswitches, internal residues occupy sites of bound ligands in the free state. Nevertheless, ‘closed’ and ‘blocked’ conformations can open up for metabolites soaked into the unliganded riboswitch crystals  and, therefore, are in equilibrium with open riboswitch states. However, local conformational dynamics of RNA cannot adequately explain small-angle X-ray scattering (SAXS) , fluorescent  and other data, which suggest that the apo-form of riboswitches may adopt an ensemble of substantially dissimilar interconverting states in the absence of ligands . These results imply that riboswitches likely employ ‘conformational selection’ to recognize and bind their cognate ligands. The conformational selection concept postulates that the ligand selects ligand-binding competent conformation(s) from the dynamic ensemble of multiple pre-existing conformations and shifts the equilibrium towards the bound state via favorable binding energetics. The closing of the ligand-binding pockets also suggests post-binding conformational adjustments, consistent with the ‘induced fit model’.
Some riboswitches, for example TPP  and c-di-GMP-I  riboswitches, appear to populate near ligand-bound conformations less frequently than other riboswitch classes, therefore their free and bound states are considerably distinct. The free states also differ amongst representatives of the same riboswitch class [11,40,45,46] and it would be interesting to understand whether the similarity of the free and bound states correlate better with environmental factors (for instance, growth temperature) or mechanisms of gene expression control (transcription or translation).
In solution, ligand-bound riboswitches are not frozen in a single conformation and demonstrate motions on different time scales . RNA dynamics is not restricted to flexible loop regions and can be observed within the ligand-binding pocket. According to time-resolved fluorescence spectroscopy of purine-sensing riboswitches , crystal structures represent the global minimum that accounts for only 50–60% of the population, thereby suggesting the elastic nature of the ligand recognition mode that provides high affinity and specificity for the sensor.
Most riboswitches have to commit to the metabolite-bound or metabolite-free folding pathway in a short temporal window of riboswitch transcription. Therefore, much attention has been devoted to understanding the global folding of metabolite-sensing domains and entire riboswitches. Recent studies performed on purine [43,48–50], preQ1 [46,51], SAM-I [40,45], SAM-II , and SAM-III  riboswitches have revealed various folding trajectories for different riboswitch classes, as well as species within the same class. The vast majority of these studies emphasize the importance of Mg2+ cations for the pre-formation of ligand-binding competent conformations, by for instance, aligning helical segments and promoting base pairing, as in the SAM-I riboswitch (Figure 4) [40,45]. Ligand binding further contributes to the stabilization of the bound riboswitch fold through either local adjustments or global changes, for instance, stabilization of stacking and tertiary pairing contacts [40,45,54,55]. A full understanding of the mechanics of riboswitch folding will require studies performed under cotranscriptional conditions.
Many critical questions in the riboswitch field, such as how riboswitches sense cognate ligands and how they choose folding pathways, have been recently addressed by numerous studies on many riboswitch systems. These studies have provided the structural basis for cooperative ligand binding by glycine and THF riboswitches, discrimination of closely related compounds by dG, preQ1 and SAH riboswitches, and differential ligand binding by c-di-GMP-I and c-di-GMP-II riboswitches. Despite various strategies adopted by riboswitches for binding of cognate ligands, many experimental results converge on the ‘conformational selection’ principle of metabolite recognition, possibly with some impact from ‘induced fit’ events after initial sensing. Future studies should explore and address the sensing and folding processes in a cotranscriptional context.
This work was supported by National Institutes of Health grant GM073618 to D.J.P. and New York University Medical Center start-up funds to A.S.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Papers of particular interest, published within the annual period of review, have been highlighted as:
• of special interest
•• of outstanding interest