|Home | About | Journals | Submit | Contact Us | Français|
Regulatory mRNA elements or riboswitches specifically control the expression of a large number of genes in response to various cellular metabolites. The basis for selectivity of regulation is programmed in the evolutionarily conserved metabolite-sensing regions of riboswitches, which display a plethora of sequence and structural variants. Recent X-ray structures of two distinct SAM riboswitches and the sensing domains of the Mg2+, lysine, and FMN riboswitches have uncovered novel recognition principles and provided molecular details underlying the exquisite specificity of metabolite binding by RNA. These and earlier structures constitute the majority of wide-spread riboswitch classes and, together with riboswitch folding studies, improve our understanding of the mechanistic principles involved in riboswitch-mediated gene expression control.
Since the discovery of riboswitches and related regulatory mRNA elements, it has become evident that the sensing and gene controlling functions traditionally attributed to proteins can be performed by RNAs [1,2]. Recent intensive biochemical and biophysical research has increased the appreciation of riboswitches as universal genetic factors that adjust gene expression in response to various chemical cues in evolutionarily diverse species [3,4]. The adaptation of protein biosynthesis to environmental conditions is achieved by riboswitches through the direct sensing of small-molecule metabolites present in cells at elevated concentrations. Most known riboswitches are located in the 5′-untranslated regions of bacterial genes associated with cognate metabolites and are involved in regulating gene expression at the levels of transcription attenuation and translation initiation. The regulation is typically based on the interplay between alternative conformations within an evolutionarily conserved metabolite-sensing domain and a variable expression platform bearing gene control elements (Figure 1). In some bacteria, the riboswitch-based feedback regulatory circuits exert control over a significant portion of the genome, competing in number with metabolite-sensing regulatory proteins . Several thiamine-pyrophosphate-(TPP)-sensing riboswitches have been also found in fungal and plant genes, where they modulate mRNA splicing and stability .
To date, riboswitches encompass more than a dozen classes specific to various types of metabolites. Not surprisingly, the metabolite-sensing domains of riboswitches demonstrate high diversity in composition, size, and complexity of their secondary and tertiary structures . Principles underlying the exquisite selectivity of metabolite recognition became better understood after determination of the three-dimensional structures of guanine- [7,8], adenine- , TPP- [9,10], and S-adenosylmethionine (SAM)- sensing domains and of the glmS riboswitch/ribozyme [12,13]. During the last two years, this list has been extended by five new riboswitches whose X-ray structures range from simple conformations, such as a small pseudoknot, to large multi-helical folds. These recent structural advances along with studies of molecular mechanisms of riboswitch function will be primarily discussed herein, thus complementing earlier reviews on mRNA recognition by metabolites and proteins [14,15].
After 2006, several new classes have been added to the existing collection of riboswitches responsive to adenine, guanine, TPP, SAM, flavin mononucleotide (FMN), adenosylcobalamin, lysine, glycine, and glucosamine-6-phosphate (GlcN6P) . The most important discovery was the identification of the riboswitch specific for the second messenger cyclic di-guanosine monophosphate (di-GMP) in bacteria and bacteriophages [16••]. Cyclic di-GMP-sensing allows for the RNA-based control of genes that are responsible for wide-ranging physiological transformations within bacterial cells and are not directly involved in the metabolism of the compound. Furthermore, the riboswitch family receptive to purines and purine derivatives was expanded by two variants of the queuosine precursor preQ1 sensor [17,18•], the shortest known riboswitch that controls biosynthesis of the hypermodified nucleoside present in certain tRNAs, and by the 2′-deoxyguanosine sensor  that deviates from the classical adenine/guanine riboswitch sequences. In addition, SAM metabolism, dependent on five riboswitch variants in different bacteria , has been found to rely on yet another type of mRNA segment targeted by S-adenosylhomocysteine (SAH) . Lastly, the first metallosensor in Salmonella enterica mgtA mRNA  has been complemented by the findings of a different magnesium sensor [23••] and RNA motifs triggered by molybdenum and related tungsten cofactors Moco and Tuco .
Despite the continuous success in identification of new riboswitch classes, the determination of riboswitch structures remains a tedious and time-consuming process. However, recent structural studies have succeeded in the determination of challenging riboswitch structures, varying in size and complexity.
Intriguingly, one of the large ribosensors, called M-box, is specific for Mg2+ cations, one of the smallest and most abundant cellular components . The sensor controls Mg2+ homeostasis in Bacillus subtilis by attenuating the transcription of the Mg2+ transporter gene mgtE (Figure 1a). The crystal structure (Figure 2a) revealed that the M-box architecture comprises a short stem-loop P6 and two parallel composite stems P1-P2 and P3-P4 connected by a three-way junction. The P3-P4 helix, from which stem P5 branches off, forms a closely-packed three-helical bundle P4/P5/P1-P2 stapled together by extensive tertiary contacts involving J1/2, L4 and L5 regions. Unlike in other riboswitches, Mg2+ cations play a central role in folding the bridging region and mediating long-distance interactions that stabilize base-pairing in the J1/2 region and in the adjacent P1 helix, thereby inducing the formation of the transcription terminator.
Since the structures of the M-box and other riboswitches are primarily built around three- or four-way helical junctions, they are difficult to use for characterization of more complex riboswitches, such as lysine and FMN riboswitches (Figure 1b, c), thereby demanding their structure determination. The structure of the ~170 nt Thermotoga maritima lysine-bound ‘L-box’ domain, independently determined by two laboratories [25••,26••], consists of three-helical and two-helical bundles radiating from a compact five-way helical junction (Figure 2b). The P2/P2a/L2 stem-loop reverses its orientation through two ~90° turns near conserved motifs and is anchored by ‘kissing’ loop-loop interactions between L2 and L3. Parallel stems P2 and P4 are joined by a novel loop (L4)-helix (P2) contact. The junction is organized around the centrally positioned lysine that stabilizes the top region of the regulatory helix P1 in a way most similar to purine riboswitches.
The metabolite-sensing domain of the FMN riboswitch, an RFN element, presents further structural complexity. The structure of the ~110 nt Fusobacterium nucleatum RFN element [27••] (Figure 2c), one of the shortest among FMN riboswitches, has confirmed the six-way junction topology predicted by sequence analysis . Unlike other large RNAs, the structure does not fold by collinear stacking of adjacent helices. Instead, it adopts a unique butterfly-like scaffold, stapled together by two nearly identical domains P2-P6 and P3-P5 that are connected by the FMN-bound junction. The domains are related by the 2-fold symmetry typical for proteins and seldom observed in large RNAs. Each domain contains two stem-loop structures joined by T-loop and A-minor-like motifs, and shows a striking resemblance to a larger architectural module termed “T-loop PK domain,” found in ribosomal RNA [28•]. The pseudo-symmetrical FMN does not take advantage of the riboswitch symmetry and orients its extremities towards different domains of the riboswitch. Such an arrangement is most similar to the TPP riboswitch [9,10,29], although the FMN-binding regions are not as well separated. Contrary to other riboswitches, the alternative pairing in the FMN-sensing domain apparently involves its 5′ region, located much further from the complementary anti-terminator sequence than the 3′ segment transcribed later (Figure 1c). This atypical mechanism might be explained by a conserved substitution that weakens a tertiary contact and provides greater flexibility to the regulatory J1/2 sequence in the 5′ region.
The structures of the SAM-II/SAM-α [30•• ] and SAM-III/SMK [31••] riboswitches, together with the earlier structure of the SAM-I/S-box riboswitch , uncover molecular details of evolutionarily independent and entirely distinct solutions for SAM sensing by RNA. Unlike the complex four-way-junction architecture of SAM-I riboswitch, the crystal structures of the ~50 nt SAM-II and SAM-III riboswitches revealed simpler compact folds based on a H-type pseudoknot and a three-way junction (Figure 2d, e).
The SAM-II structure from Sargasso Sea metagenome comprises a continuous helix P1/P2b/P2a composed of three stacked helical elements, P1, P2a and P2b, and two loops, L1 and L3, which interact with the major and minor grooves of P2a/P2b and P1, respectively (Figure 2d). SAM positions itself inside the riboswitch and stabilizes the core of the RNA molecule. Since the Shine-Dalgarno (SD) sequence is intimately involved in SAM recognition, the SAM-II riboswitch likely represses translation via an occlusion mechanism rather than by more common secondary structure switching.
The Enterococcus faecalis SAM-III riboswitch folds into an inverted Y-shaped molecule, centered on the SAM-bound three-way junction (Figure 2e). A double-strand reversal in the J3/2 segment stabilizes the junction and forms a lid-like structure that constitutes part of the tight SAM-binding pocket. Complex formation shifts the conformational equilibrium towards the ligand-bound state with the SD sequence sequestered by contacts with the ligand and pairing within P1 and P4 helices (Figure 1d).
Mg2+ cations are essential for the folding and function of cellular RNAs, mostly due to their ability to shield negative charges and collapse RNA into a compact conformation. The TPP [9,10,29,32] and glmS riboswitches [12,13] exploit this property to overcome the negative electrostatic character of the ligand phosphates. The M-box takes the neutralizing capability of Mg2+ cations a step further and uses them as sole ligands that affect gene expression through changes in riboswitch folding [23••].
The large Mg2+ sensor binds four (Mg1–Mg4) out of six cations in the region involved in riboregulation [23••] (Figure 3a). These key cations apparently play non-equivalent roles in metallosensing. While Mg1, Mg3 and Mg4 stabilize the local conformations of L5, P2 and L4 regions, Mg2 bridges L5 and P2. Mg1 forms 4 inner-sphere (one nucleobase and three sugar-phosphate backbone) contacts with RNA, a number rarely observed in RNA structures, and, therefore, may crucially contribute to the tertiary interactions that facilitate pairing of the regulatory segment. Though other Mg2+cations form less inner-sphere contacts with RNA, they may also be important for the specificity of metallosensing, given that almost all Mg3 and Mg4 outer-sphere interactions are made with nucleobases.
The high affinity interactions in the FMN-riboswitch complex also depend on magnesium and, to a lesser extent, on monovalent cations [27••]. As in TPP riboswitches [9,10,29], the extremities of FMN, the isoalloxazine ring system and the phosphate, provide the most important contacts with RNA while the middle ribityl moiety seems less critical for binding (Figure 3b). The ring system is sandwiched between purines, similar to the ligands in purine and TPP riboswitches [7–10,29], and is involved in specific hydrogen bonding with the nucleobase of conserved A99, reminiscent of purine, TPP and SAM-I riboswitches [7–11,29]. The phosphate moiety forms the majority of hydrogen bonds with RNA, both direct and Mg2+-mediated. The Mg2+ cation can be substituted by Ba2+, Mn2+, Ca2+and even by a [Co(NH3)6]3+ group which mimics a fully-hydrated Mg2+ cation. Therefore, in contrast to the TPP riboswitch , neither the identity of the divalent cation nor the inner-sphere metal coordination with FMN and RNA are essential for FMN recognition by the riboswitch. Interactions with both the ring system and phosphate of FMN likely restrict the mobility of the J6/1 and J1/2 segments and contribute to the formation of the regulatory P1 helix.
Though the metal-dependent recognition of a phosphate-bearing FMN has been anticipated, the requirement of potassium for ligand recognition by a lysine riboswitch was unforeseen. Lysine is positioned in the middle layer of a tight pocket and is surrounded by evolutionarily conserved nucleotides (Figure 3c) [25••]. The ‘main chain’ and ‘side chain’ functionalities of the amino acid predominantly recognize nucleobases and the sugar-phosphate backbone, respectively. A K+ cation binds a carboxyl oxygen of lysine and zippers up the binding pocket through multiple coordination bonds, thereby explaining the severe impairment of ligand binding when K+ is omitted or replaced by Na+. As in purine riboswitches [7,8], the regulatory P1 helix is stabilized by contacts with the ligand and indirectly via tertiary interactions restricting mobility of the adjacent RNA segments, while discrimination against similar ligands is achieved by a combination of shape complementarity and specific hydrogen bonding.
It was not entirely expected that SAM riboswitches would form very different binding pockets that interact with SAMs adopting distinct conformations. In the SAM-I structure (Figure 3d) , SAM assumes a compact U-shape conformation, wherein the methionine moiety stacks upon the adenine ring and is partially stabilized by an intramolecular π-cation interaction. SAM binds between two helical regions using predominately hydrogen bond interactions on one side and van der Waals surface complementarity on the other. In the SAM-III riboswitch (Figure 3f) [31••], SAM is also bent, but to a lesser extent, so that the methionine moiety is placed aside of the adenine ring and is partially disordered. Therefore, unlike in SAM-I and SAM-II rboswitches, the ligand length is not critical for the SAM-III riboswitch. The adenosine moiety and the adjacent region of the methionine moiety contact RNA within a semi-open cavity formed by a junctional region. A different conformation of SAM was observed in the SAM-II structure (Figure 3e) [30••]. The ligand extends along the major grove face of the P2b-L1 triplex and uses all available functional groups for interactions with RNA. The hydrogen bonding patterns of SAM recognition also vary in all three riboswitches. Similarities are limited to Hoogsteen pairing with uracil and recognition of carboxylate by purines in the SAM-I and SAM-II structures. Other common features include stacking of the adenine moiety with RNA bases and, most importantly, electrostatic interactions of the positively charged sulfur moiety with O4 carbonyls of uracils. The readout of the positive charge is essential for discrimination between SAM and SAH, the by-product of SAM-dependent methylation reactions (Figure 2d).
Since some riboswitches interact with metabolite-like antibiotics and contain mutations providing antibiotic resistance, riboswitch-controlled gene expression, along with other cellular processes , may be targeted by antimicrobial compounds [34•]. The structures of the eukaryotic TPP riboswitch with TPP and its analogues oxythiamine and pyrithiamine pyrophopshates were found to be nearly identical, with a slight conformational adjustment next to the central ring of pyrithiamine pyrophosphate . No changes were identified in the structures of the lysine riboswitch with antibacterial compounds S-(2-aminoethyl)-L-cysteine and L-4-oxalysine (Figure 2c), while extensions in the lysine analogues L-homoarginine and N6-1-iminoethyl-L-lysine replaced water molecules and caused the rotation of RNA ribose [25••]. As in the TPP riboswitch, shortened ligands of the FMN riboswitch, riboflavin and antibiotic roseoflavin (Figure 2b), re-positioned several nucleotides in the binding pocket . Additional spatial adjustment was detected in the roseoflavin-riboswitch complex for accommodation of a methyl substituent in the ligand ring system. These and earlier data indicate that the binding pockets of the TPP and FMN riboswitches [27••,29] demonstrate intrinsic plasticity, while those of purine [7,8], glmS [12,13] and lysine [25••,26••] riboswitches are more rigid.
The folding pathway and ligand-induced conformational re-arrangements are essential aspects for understanding the molecular mechanism of the riboswitch function. Earlier experiments on purine riboswitches (Figure 4a) suggested partial organization of the sensing domain prior to metabolite binding [36–39]. Follow-up chemical probing [40,41], NMR spectroscopy [42,43], fluorescent spectroscopy [44–47], and calorimetric studies  further refined the model that implies the parallel preorientation of P2 and P3 hairpins, preorganization of interacting loops L2 and L3, and local disorder of the junctional region in the absence of the ligand. To bind the junction, a purine likely samples junctional conformations until it finds a binding-competent form maintained by universally conserved nucleotides . Ligand binding then reinforces the tertiary loop interactions  and stabilizes the regulatory helix P1. For effective gene control, the switching decision should be precisely timed. Since the rate of the ligand-riboswitch complex formation depends on the ligand binding affinity and the kinetics of transcription , the regulation mechanism of related purine riboswitches with distinct expression platforms may involve different contributions from thermodynamic and kinetic components .
To gain further insight on the early steps of purine-riboswitch complex formation, a time-resolved NMR strategy was applied to monitor the ligand-induced folding of the guanine-sensing domain [49•]. The approach is based on the laser-triggered release of hypoxanthine from the inactive photolabile ‘caged’ derivative in a solution of isotopically-labeled riboswitch (Figure 4b). The fast hypoxanthine binding events are then tracked by NMR spectroscopy and analyzed using molecular dynamic simulations. The experimental data suggested a three-step binding model, in which an initial fast low-affinity encounter is followed by the specific binding of the ligand to the riboswitch core, leading up to a final slower process that involves tightening of the tertiary structure.
In cells, riboswitch folding is tightly coupled with transcription. To obtain an integrated picture of structure formation and ligand binding after transcription, single nascent metabolite-sensing domains of adenine riboswitch were unfolded and refolded using single-molecule force spectroscopy [50••]. The assay was performed using dual-trap optical tweezers: the first trap held a transcriptionally stalled RNA polymerase, while the initial RNA transcript emerging from the polymerase was hybridized to a DNA handle in the second trap. After RNA transcription, the traps were moved apart to apply force to the RNA transcript, and the molecular extension of the transcript - related to the number of nucleotides folded - was measured as the transcript unfolded and refolded repeatedly, both in the presence and absence of the ligand. The measurements revealed sequential RNA folding (Figure 4c), which begins with formation of P2, the first fully transcribed element, followed by P3 folding. In the next adenine-competent state, the adenine binding pocket is pre-organized by tertiary contacts, sequestering the nucleotides between P2 and P3. The tertiary interactions then facilitate formation of the helix P1, the least-stable element governing conformational switching. Adenine binding stabilizes the folded state and raises the barrier for leaving the folded state; in vivo, without adenine, P1 helix formation would easily be disrupted by terminator hairpin invasion.
Folding studies on other riboswitches have proposed mechanisms of TPP riboswitch action [51–53], demonstrated the ligand-associated structural transitions in the glycine riboswitch [54•], and uncovered the importance of tertiary interactions for control by SAM-I  and lysine  riboswitches. The first ligand-free structure of the glmS riboswitch/ribozyme  has been complemented by the structure of the unliganded lysine riboswitch [25••,26••], found to be virtually identical to the lysine-bound structure. Though the formation of the lysine-free confirmation has been facilitated by the absence of the competing anti-terminator structure and by stable crystal contacts, this result strongly supports the idea that major parts of a riboswitch can be folded prior to interaction with ligand, while ligand binding locks the conformation of key riboswitch regions and regulatory elements. In contrast to the lysine riboswitch, the similarity between ligand-bound and ligand-free forms of the glmS riboswitch/ribozyme is not likely caused by crystallization, since ligand binding stimulates the cleavage of the glmS RNA and does not make extensive conformational re-arrangements which, in other riboswitches, stabilize the regulatory helix P1.
Recent structures of the riboswitches, discussed here, have added novel junction architectures to the repertoire of RNA folds and broadened the collection of RNA structural motifs by novel variants of turns, loop-loop and loop-receptor interactions. Among the main conclusions of recent structural studies is that a variety of principles are utilized by riboswitches of different complexities to recognize their cognate ligands. Interestingly, the sizes of the riboswitches and their ligands do not correlate, suggesting that other factors, such as the number of related compounds to be discriminated against and the dynamic range of the riboswitch response, may contribute to the complexity of the riboswitch folds. A special place belongs to the SAM riboswitches that represent a striking example of convergent evolution, displaying different folds and ligand binding modes to recognize the same ligand and carry out similar regulatory functions. The lysine and FMN riboswitch structures suggest that metal ions are as important for metabolite binding as for RNA folding. However, it is still not clear whether metal ions simply expand upon a poor repertoire of functional groups in RNA to aid specific metabolite binding. It has become evident that the majority of riboswitch-bound ligands are directly involved in riboswitch function through the specific recognition of regulatory sequences or through participation in the tertiary interactions that facilitate formation of regulatory elements. As demonstrated by biochemical and biophysical studies, ligand binding also stabilizes partially pre-formed riboswitch architectures and contributes to the overall stability of the RNA folds. Though many challenging problems in riboswitch regulation still remain, the structural information will undoubtedly lead to a comprehensive picture of riboswitch-based gene expression control and could help in the design of tools that affect this regulation.
I thank O. Pikovskaya (Memorial Sloan-Kettering Cancer Center) for critical reading of the manuscript, and M. Woodside (University of Alberta) and J. Buck (Johann Wolfgang Goethe-University) for providing figures. This work was supported by National Institutes of Health grant GM073618.
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