Guanine riboswitch variants from Mesoplasma florum selectively recognize 2′-deoxyguanosine

Abstract

Several mRNA aptamers have been identified in Mesoplasma florum that have sequence and structural features resembling those of guanine and adenine riboswitches. Two features distinguish these RNAs from established purine-sensing riboswitches. All possess shortened hairpin-loop sequences expected to alter tertiary contacts known to be critical for aptamer folding. The RNAs also carry nucleotide changes in the core of each aptamer that otherwise is strictly conserved in guanine and adenine riboswitches. Some aptamers retain the ability to selectively bind guanine or adenine despite these mutations. However, one variant type exhibits selective and high-affinity binding of 2′-deoxyguanosine, which is consistent with its occurrence in the 5′ untranslated region of an operon containing ribonucleotide reductase genes. The identification of riboswitch variants that bind nucleosides and reject nucleobases reveals that natural metabolite-sensing RNA motifs can accrue mutations that expand the diversity of ligand detection in bacteria.

Gene-control elements called riboswitches (1) are mRNA motifs typically found in the 5′ untranslated regions of bacterial mRNAs. Riboswitches selectively bind small molecules, and structural changes within the 5′ untranslated regions are usually harnessed to control the expression of the adjoining ORF. The architectures of these RNAs are commonly formed from a metabolite-binding aptamer domain and an expression platform (2–5), although more diverse assemblies of aptamers and expression platforms have been found that yield more complex gene-control characteristics (6–9).

Each aptamer adopts a complex secondary- and tertiary-structured fold to form a conserved receptor for the ligand (10–17). This demand for precise structure formation and specific molecular recognition causes the aptamer domains to be highly conserved even among distantly related species. In contrast, the expression platform can adopt a variety of different structures provided it maintains its responsiveness to the occupation state of the aptamer domain.

Computer-aided searches based on conserved RNA sequences and structures have been used to identify representatives of numerous riboswitch classes (18–24). However, these bioinformatics algorithms can fail to identify variants of known riboswitch classes that differ substantially from the established aptamer consensus. Furthermore, there could be exceedingly rare classes of riboswitch aptamers or exceptionally small aptamers that will be missed by existing bioinformatics algorithms because there are too few representatives for comparison or they have too few conserved features. Given these limitations, many new riboswitch classes might remain undiscovered, and the true number of metabolite-sensing aptamer folds could greatly exceed the many types published to date (25).

Indications that there are many variant, small, or rare riboswitch classes to be discovered come from several recent reports of new riboswitch classes. For example, a single C-to-U mutation within the core of guanine riboswitches can change the specificity of ligand binding to adenine (11, 26–30). Also, two related types of riboswitch aptamers for the modified nucleobase 7-aminomethyl-7-deazaguanine have been identified that require as few as 34 nucleotides to form a selective and high-affinity binding pocket (31). Moreover, there are three classes of aptamers for S-adenosylmethionine (SAM) (23, 24, 32) whose representatives are more rare in bacteria than the SAM-I class of riboswitches commonly found in Gram-positive bacteria (33–35). These findings indicate that a far greater diversity of metabolite-sensing riboswitches exists that might be difficult for existing search strategies to definitively identify and classify.

One possibility is that some organisms will have recently evolved riboswitch classes with aptamers that are unique in architecture or ligand specificity. If such boutique riboswitches can easily emerge through evolution, there could be far more riboswitches than have been discovered to date. In this report, we describe a series of aptamers that are exceedingly rare among sequenced genomes and have been identified only in the bacterial species Mesoplasma florum. One subclass of these RNAs is selective for 2′-deoxyguanosine (2′-dG). Our findings highlight the capacity for metabolite-binding RNAs to evolve specificities toward structurally related derivatives and further demonstrate that exceptionally rare riboswitch classes are likely to be present in some organisms.

Results and Discussion

Consensus Sequences and Structures of Guanine- and Adenine-Sensing Riboswitches.

Guanine-sensing riboswitches usually reside upstream of genes involved in purine biosynthesis, salvage, and transport (26). The guanine riboswitch aptamer from the Bacillus subtilis xpt-pbuX mRNA exhibits a K_d for guanine of ≈5 nM. Ligand binding to this aptamer causes transcription termination, and a similar gene-control mechanism is predicted for most other guanine riboswitches as well. X-ray crystallography has been performed on this aptamer bound to either guanine or the functional analog hypoxanthine (10, 11). In both instances, the ligands are almost completely enveloped by the RNA. Similarly, a related aptamer that binds adenine by using a similar architecture also engulfs the ligand (11).

The tight ligand-binding pocket of this aptamer class is formed by conserved nucleotides at the junction of three stems termed P1, P2, and P3 (Fig. 1A). When the ligand is bound, the aptamer adopts a conformation with the P2 and P3 stems extending parallel to one another. This structure is held in place by Watson–Crick base-pairing interactions and other hydrogen bonds formed between the loops of these stems, called L2 and L3. Most of the highly conserved nucleotides forming the ligand-binding core of the aptamer are present in joining regions J1-2, J2-3, and J3-1, which link the three stems together (Fig. 1B).

Fig. 1.

Sequence and structural features of guanine riboswitch aptamers and several newly found RNAs. (A) Sequence alignment comparing the xpt guanine riboswitch aptamer sequence from the xpt-pbuX mRNA from B. subtilis with related sequences from Mesoplasma florum (types I, II, III, and IV) and from Oenococcus oeni, Vibrio sp., Vibrio splendidus, and Leuconostoc mesenteroides (types V-A, V-B1, V-B2, and V-C, respectively). The known or putative functions of the genes immediately downstream of each sequence are noted as predicted elsewhere (54). Nucleotides corresponding to pairing regions P1, P2, and P3 are shaded blue, green, and orange, respectively. Nucleotides corresponding to loop regions (L) and joining regions (J) also are identified. The asterisk identifies the C nucleotide in xpt that forms a Watson–Crick base pair with the guanine ligand. Nucleotides shaded gray are mutated relative to the highly conserved nucleotides denoted in red that are typically found in the J and L regions of guanine and adenine riboswitches or nucleotides that are inserted or deleted in these regions. (B) Consensus sequence and secondary structure of guanine riboswitch aptamers. Nucleotides in red are present in >90% of the known representatives. Circles identify nucleotides whose base identities are not conserved, and lines indicate Watson–Crick base pairing. Nucleotides that form hydrogen-bonding interactions with the guanine ligand are identified according to the numbering system used previously for the xpt aptamer (10, 11, 26). (C) Structural model of the guanine-binding site formed by the xpt aptamer docked to guanine (10, 11). Dashed lines identify hydrogen-bonding contacts between the aptamer nucleotides (numbered as described in B) and the ligand. The shaded area identifies the space that would be occupied by the sugar moiety of a guanine nucleoside. (D) Sequence and secondary structure of the type I-A aptamer from M. florum. Boxed nucleotides depicted in blue identify variations from the xpt aptamer that occur at otherwise highly conserved positions. Dashed line represents a 3-nucleotide deletion compared with the L3 sequence of xpt. Nucleotide numbers are as described in Fig. 2, with the equivalent positions for the xpt aptamer depicted in parentheses. Other notations are as defined for B.

The joining regions carry four nucleotides that form hydrogen bonds with functional groups of the purine ligand (Fig. 1C). One key interaction is made by nucleotide C74 of the xpt aptamer, which forms a Watson–Crick base pair with guanine. Interestingly, several variants of this RNA motif were found that carry a C-to-U mutation at the equivalent position in the structure, and these RNAs reject guanine and bind adenine with affinities measured in the mid nanomolar range (27, 28). An atomic-resolution model of an adenine riboswitch aptamer from the add gene in B. subtilis confirmed that adenine forms a Watson–Crick base-pairing interaction with the variant U nucleotide, and that other features typical of guanine riboswitches remained essentially identical (11). Indeed, this single-nucleotide change at position 74 is sufficient to change the specificity of guanine aptamers to adenine and vice versa.

Discovering More Distant Homologs of Guanine and Adenine Riboswitches.

We used a bioinformatics search strategy to discover variant purine riboswitch candidates. This process was achieved with an algorithm that identifies sequences that closely correspond to the consensus sequence and secondary structure features of known purine-sensing aptamers (see Materials and Methods). The parameters of this search were set to allow recovery of low-quality matches, and we focused the most attention on sequences that deviate substantially from the consensus, but that nonetheless exist in genomic contexts consistent with riboswitch function.

We noticed one sequence in M. florum that could be threaded to conform reasonably well to the consensus structure, but that deviated in sequence at several key positions in the core and in loops L2 and L3. Despite these significant differences, the location of this sequence in the apparent 5′ untranslated region of the guaAB operon suggested that it might function as a purine-sensing riboswitch. A BLAST search for related sequences uncovered several more examples of this motif in M. florum. This bacterium is a nonparasitic member of the class Mollicutes, and organisms of this class are notable for their simplified cell structures and small genomes (36).

One of the M. florum sequences differed from characterized purine riboswitches only in the L2 and L3 regions, with the joining regions otherwise adhering to the consensus. To determine whether there might be other variant purine aptamers analogous to this RNA, we manually inspected sequences generated from the original search, scanning for sequence or structure irregularities. Three additional riboswitches were identified that contained shortened L3 sequences relative to the consensus, in addition to the eight M. florum examples mentioned previously.

Subsequently, we performed automated searches by using algorithms trained on all known purine riboswitches, as well as algorithms trained more narrowly on the variant sequences, but no additional variants were identified. In total, 12 new putative riboswitch examples were found in bacteria (Fig. 1A), all of which bear close similarity to the consensus sequence and structure established for guanine aptamers. These RNAs were classified into five types (I–V) based on the mutations they carry relative to the guanine aptamer consensus. Interestingly, eight of the RNAs representing types I-IV are present in M. florum.

One of the eight RNAs (IV-A) carries a C-to-U mutation at the position equivalent to nucleotide 74 of the xpt riboswitch in B. subtilis. Therefore, this RNA was predicted to sense adenine. The seven remaining RNAs from M. florum carry mutations at two or more positions that are highly conserved among known guanine and adenine riboswitch representatives (Fig. 1A). All seven RNAs carry mutations in otherwise conserved nucleotides in J1-2, J2-3, and L2 of the aptamer, and they also carry an L3 loop that is four nucleotides, rather than the seven or eight nucleotides normally present in known guanine-binding aptamers. The four remaining RNAs, classified as type V, are found in other bacterial species and carry the distinctive nucleotide changes in L3 and, in some instances, L2. Specific interactions between the L2 and L3 regions are known to be important for folding and function of guanine and adenine riboswitches (29), and therefore the loop mutations likely cause the variant RNAs to adopt a different structure for this tertiary interaction.

We speculated that the aptamer core mutations in the M. florum RNAs might substantially alter the ligand-binding pocket of each riboswitch, allowing it to recognize a metabolite other than guanine. For example, RNA I-A carries 39 nucleotide changes (including insertions and deletions) relative to the xpt RNA (Fig. 1A), and 10 of these changes occur at positions with nucleotide identities that are conserved in >90% of the known guanine riboswitch aptamers (Fig. 1D). In addition, three of the four nucleotides known to contact guanine in the xpt aptamer (nucleotides 22, 47, and 51) (Fig. 1 B and C) are mutated in the I-A (nucleotides 31, 54, and 58) (Fig. 1D) and I-B aptamers. Although the nucleotide corresponding to C74 of xpt in the I-A and I-B aptamers could retain its recognition of the base-pairing face of guanine, the other core mutations likely recognize other portions of a guanine-containing ligand. Moreover, these core mutations typically convert A and U residues to G and C residues, despite the fact that the M. florum genome has only 27% GC content. The acquisition of additional G and C residues in some variant aptamers suggests adaptation to a new function.

A Guanine Riboswitch Variant Binds 2′-dG.

Frequently, the metabolite that is sensed by a riboswitch can be discerned by noting the function of the protein product of the downstream gene. Some of the RNA motifs are located upstream of either unannotated genes or genes with functions that appear to be unrelated to purine metabolism, and thus did not provide clues for possible ligands. However, some reside upstream of genes involved in purine biosynthesis or transport (Fig. 1A), suggesting that the variant RNAs bind guanine or a ligand that includes this nucleobase. Of particular interest was aptamer I-A, which resides upstream of genes encoding ribonucleotide reductase subunits. Ribonucleotide reductase enzymes convert ribonucleotides into their deoxyribonucleotide counterparts (37, 38). Given that three of the nucleotides mutated in the I-A aptamer are in the immediate vicinity of the N9 position of guanine in known guanine riboswitches (Fig. 1C), we speculated that this variant riboswitch might respond to 2′-dG or one of its 5′-phosphorylated derivatives.

In-line probing (39) was performed by using a series of guanine and guanosine derivatives (see Materials and Methods for a complete list) to determine the ligand specificity for all variant riboswitch types. In-line probing assays reveal shape changes in an aptamer that occur upon ligand binding. For example, I-A exhibits substantial structural modulation when 100 μM 2′-dG is present (Fig. 2A). Importantly, the pattern of spontaneous cleavage products is consistent with the formation of a three-stem junction similar to guanine and adenine aptamers (26–28), and the majority of the internucleotide linkages that become more structured upon 2′-dG addition (Fig. 2B) are in the predicted ligand-binding core of I-A RNA. Furthermore, in-line probing data collected at various concentrations of 2′-dG indicate changes in the extent of RNA cleavage at specific sites in the RNA that are consistent with a 1:1 binding of ligand with an apparent K_d of ≈80 nM (Fig. 2C).

Fig. 2.

Structural modulation of the I-A aptamer caused by binding of 2′-dG. (A) In-line probing analysis of the I-A aptamer. Fragmentation patterns of 5′ ³²P-labeled precursor RNAs (Pre) were established by incubating the RNAs in the absence (−) or presence (+) of 100 μM 2′-dG and separating spontaneous RNA cleavage products by using denaturing 6% PAGE. Lanes designated NR, T1, and ⁻OH identify RNA samples loaded after subjecting to no reaction, partial digestion with RNase T1, or partial digestion with alkali, respectively. Left arrowheads identify select bands corresponding to RNase T1 cleavage after G residues, and right arrowheads identify regions within the aptamer that undergo structural stabilization on ligand binding. Band intensity changes at sites 1 and 2 from a separate inline probing analysis (PAGE image not shown) were used to establish the apparent K_d as depicted in C. (B) Sites of structural flexibility and 2′-dG-mediated structural modulation. Data were derived from the image depicted in A. (C) Plot of the normalized fraction of RNA cleaved at sites 1 and 2 versus the logarithm of the concentration (c) of 2′-dG. The concentration of ligand required to cause half-maximal change in fraction cleaved (dashed lines) reflects the apparent K_d.

Affinities and Specificities of Natural Variants of Guanine Riboswitch Aptamers.

A previous study (26) revealed that the apparent K_d value of the xpt riboswitch aptamer for guanine is ≈5 nM. We conducted similar K_d determinations for the RNAs shown in Fig. 1A and expanded this process to include compounds similar to 2′-dG (Fig. 3 and data not shown). As expected, we find that the xpt RNA binds guanine most tightly, whereas the addition of a ribose or deoxyribose moiety on the N9 position of the purine ring causes a loss of binding affinity of nearly two orders of magnitude or more (Fig. 3A Left). Similarly, types III and V prefer binding guanine over various nucleoside derivatives by ≈100-fold or more. This observation is expected because these RNA types retain a U residue corresponding to nucleotide 51 in the xpt aptamer. The retention of this residue and a C at the position analogous to C74 (Fig. 1C) suggests that these RNAs can form at least six of the seven hydrogen bonds formed between known guanine aptamers and their guanine ligand. The type V RNAs retain the base identities for all ligand-binding nucleotides, and these RNAs exhibit K_d values that are most similar to those of the xpt aptamer for all of the ligands tested. Also as expected, the type IV RNA, which carries the C-to-U mutation at position 74, binds adenine more tightly than guanine and exhibits binding affinities for these ligands that are consistent with a previously studied adenine riboswitch aptamer (27, 28).

Fig. 3.

Affinities and specificities of natural guanine aptamer variants. (A) Values for apparent K_d for various ligands were determined for the guanine riboswitch aptamer from the B. subtilis xpt RNA and with each variant aptamer type. An arrow above a circle indicates that the exact K_d value is either lower or higher than the scale of the plot. (B) Chemical structures of 2′-dA and the analog 2′-d-2,6-DAP and the apparent K_d values for 2′-dG and 2′-d-2,6-DAP with I-A RNA or a mutant version of this aptamer (I-A*) that carries a single C-to-U mutation at a position equivalent to nucleotide 74 (Fig. 1B). Compounds tested that failed to bind are not presented.

In contrast, type I RNAs most tightly bind 2′-dG relative to guanine and various nucleoside and nucleotide analogs (Fig. 3A Right). Most strikingly, both I-A and I-B RNAs discriminate by approximately two orders of magnitude against guanosine, which differs from 2′-dG by a single oxygen atom. This level of discrimination might be required by the cell to ensure that the expression of ribonucleotide reductase is controlled only by changing concentrations of a deoxyribonucleoside, and that expression is not inappropriately repressed by normal concentrations of the corresponding ribonucleoside. Similarly, type II RNAs exhibit the highest affinities for 2′-dG, but are far less selective for this compound than type I aptamers. Perhaps type II RNAs are intentionally less selective to permit changing concentrations of several guanine-containing compounds to modulate gene expression.

These findings also are consistent with the genomic arrangement of variant RNAs in M. florum. Specifically, aptamer types I-B, II-B, and III-A control individual genes located immediately adjacent to each other in the genome. If all three genes were controlled by only one compound, then a three-gene operon arrangement controlled by one riboswitch would be optimal. However, the presence of three types of aptamers next to these adjacent genes strongly suggests that their expression is under the control of riboswitches with three distinct specificities or affinities.

A Single Mutation Swaps Ligand Specificity of the 2′-dG Aptamer I-A.

A C-to-U mutation at nucleotide 74 of the xpt aptamer can change its ligand specificity from guanine to adenine (27). This nucleotide forms a Watson–Crick base pair with the purine moiety, whereas other nucleotides make hydrogen-bonding interactions with other positions on the ligand (Fig. 1C) (10, 11, 40). Although most purine-sensing riboswitches respond to guanine, six natural examples of adenine-sensing riboswitches carrying a U at the equivalent position 74 have been identified (27, 41, 42), and a seventh example is represented by the type IV RNA reported in this study (Figs. 1A and 3A).

If the I-A RNA binds 2′-dG by using a similar core structure adopted by guanine and adenine riboswitch aptamers, it is expected that a C-to-U mutation at the equivalent nucleotide 74 position should alter the specificity for the purine moiety of the ligand. We conducted this test by using the ligand candidate 2′-dG and its analogs, 2′-deoxyadenosine (2′-dA) and 2′-deoxy-2,6-diaminopurine nucleoside (2′-d-2,6-DAP) (Fig. 3B). Although 2′-dA should compensate for the aptamer C-to-U mutation, previous studies with guanine- and nucleobase 7-aminomethyl-7-deazaguanine-sensing riboswitches have revealed that 2,6-DAP binds more tightly to the mutant aptamers (27, 31). Furthermore, it has been shown that an adenine riboswitch binds 2,6-DAP with an affinity ≈30-fold better than that for adenine (27). Aptamers carrying the C-to-U mutation likely exhibit preferences for ligands that carry 2,6-DAP because of the formation of hydrogen bonds between other nucleotides in the aptamer core and the exocyclic amine at position 2 of the purine ring like those normally occurring in guanine riboswitches (Fig. 1C).

As expected, the unaltered I-A RNA tightly binds 2′-dG and rejects both 2′-dA and 2′-d-2,6-DAP (Fig. 3B). In contrast, the I-A* aptamer carrying the C-to-U mutation rejects both 2′-dG and 2′-dA (K_d values >1 mM), but binds 2′-d-2,6-DAP with a K_d of ≈8 μM. Thus, the 2,6-DAP analog also is preferred by the I-A* RNA as observed for other aptamers carrying similar C-to-U mutations. With the common form of guanine- and adenine-sensing aptamers, a U residue at the position equivalent to nucleotide 51 forms a hydrogen bond with the exocyclic amine present in guanine and 2,6-DAP (Fig. 1C). However, the I-A and I-A* aptamers carry a different nucleotide at the position equivalent to nucleotide 51, and therefore they likely recognize this extra amine group differently despite similarities elsewhere in the aptamer structure.

Molecular Recognition Determinants of a 2′-dG Aptamer.

In addition to the various nucleoside and nucleotide analogs described earlier, we examined nucleobase analogs of guanine to further define the functional groups recognized by the I-A aptamer. Although guanine is bound by the aptamer with a poorer affinity than 2′-dG (Fig. 3A), guanine induces structural changes in the RNA (Fig. 4A) with characteristics (Fig. 4B) of a typical 1:1 RNA–ligand interaction (8). An important observation from the in-line probing data is that guanine does not induce the full spectrum of changes in spontaneous RNA cleavage in the J2-3 region, which are observed when 2′-dG is bound (Fig. 4B). This finding is consistent with the fact that the J2-3 region likely carries the nucleotides required to recognize the deoxyribose moiety of 2′-dG and, therefore, does not undergo the same level of structural stabilization with guanine that is induced by the cognate ligand. This hypothesis is further supported by the observation that guanosine, which binds with an affinity equal to that of guanine, also does not induce complete structural stabilization of the J2-3 region (Fig. 4A).

Fig. 4.

Molecular recognition by a riboswitch aptamer that senses 2′-dG. (A) In-line probing analysis of the I-A aptamer depicted in Fig. 2B in guanine concentrations ranging from 0 to 100 μM or with 330 μM guanosine or 2′-dG as indicated. Other notations are as described for Fig. 2A. (B) Plot of the normalized fraction of RNA cleaved at nucleotides 59 and 60 (Fig. 2B) versus the logarithm of the concentration (c) of guanine. (C) In-line probing analysis of the I-A aptamer with 2′-dG, guanine, and four guanine analogs at 100 μM. Other annotations are as described for Fig. 2A. (D) Schematic representation of the molecular recognition contacts used by the 2′-dG aptamer to selectively recognize its ligand.

If the guanine base occupies the same site as does the guanine moiety of 2′-dG, then guanine analogs can be surveyed for binding activity to reveal other functional groups that are important for recognition by the aptamer. However, several guanine analogs that carry modifications of functional groups on the purine ring are not bound by the I-A aptamer when present at 100 μM (Fig. 4C). These findings, and those presented earlier, reveal that the I-A aptamer recognizes nearly every available functional group to form a precise binding pocket (Fig. 4D).

Transcription of RNAs Carrying I-A and III-B Aptamers Reveals Metabolite-Mediated Termination.

The majority of guanine and adenine riboswitches are predicted to control gene expression by regulating transcription termination (J. Barrick and R.R.B., unpublished data). In these instances, the aptamer resides a short distance upstream of a predicted intrinsic transcription terminator (43, 44), which forms a strong base-paired stem followed by a run of U residues. All of the newly found aptamer variants in M. florum lie immediately upstream of putative intrinsic terminator stems (data not shown), indicating that they are components of riboswitches that control transcription termination.

To assess how these variant riboswitches might control gene expression, single-round transcription termination assays (45) were performed by using DNA templates containing either I-A or III-B riboswitch sequences. The amounts of terminated and full-length RNA transcripts should change in a ligand-dependent manner if the riboswitch controls this process and retains activity in an in vitro assay. As observed with members of several other riboswitch classes (33, 46–48), in vitro transcription assays measuring riboswitch control do not range the full spectrum between 100% termination and 100% full-length (FL) transcription. This result could be due to numerous differences between the reaction conditions used and the natural conditions in M. florum cells, such as the use of RNA polymerase from a different species, the absence of certain ions and small molecules, and the absence of proteins that might influence RNA folding. However, we do observe changes in the percentages of transcripts that are terminated when the target ligands are present in the in vitro transcription reaction (Fig. 5A). For example, the highest levels of termination for the I-A and III-B constructs occur when 2′-dG and guanine are added to the reactions, respectively. This finding matches the preferred ligands for these RNAs as determined by in-line probing assays (Fig. 3A).

Fig. 5.

In vitro transcription termination assays with types I-A and III-B riboswitches. (A) PAGE analysis of single-round transcription termination assays. Bands representing terminated transcripts (T) or full-length (FL) runoff transcripts are identified. Single-round transcription assays (see Materials and Methods) are conducted in the absence (−) of added ligand candidate or in the presence of 25 μM of the compounds indicated. The asterisk identifies a product of transcription pausing that appears to be unrelated to riboswitch function. (B) Plots of the fraction of RNAs terminated versus the logarithm of the concentration of ligand added to single-round transcription termination assays.

It has been reported that some riboswitches function as kinetically driven, rather than thermodynamically driven, gene-control elements (28, 49, 50). For example, to yield half-maximal modulation of transcription termination, kinetically driven riboswitches require a concentration of ligand much higher than the apparent K_d of the aptamer for the metabolite. The metabolite concentration needed to half-maximally modulate transcription termination, called T₅₀ (49), was determined for both the I-A and III-B riboswitch constructs by examining single-round in vitro transcription termination assays conducted in the presence of various concentrations of 2′-dG or guanine (Fig. 5B). For the I-A construct, T₅₀ values of 2 μM and >100 μM were observed for 2′-dG and guanine, respectively. In contrast, III-B exhibited T₅₀ values for these two compounds that were nearly exactly opposite. Again, the differences in the T₅₀ values for the two constructs and two ligands indicate that the type I riboswitches most likely respond to cellular 2′-dG concentrations, whereas the type III riboswitches most likely respond to guanine.

The specificities exhibited by the I-A and III-B RNAs in both assay types correspond to those predicted based on the functions of the gene products they likely control (Fig. 1A). The rationale for controlling expression of ribonucleotide reductase genes with 2′-dG is straightforward. In contrast, the rationale for controlling the expression of a GMP synthase gene with a guanine-sensing III-B RNA is less obvious because it might seem more logical for the riboswitch to sense the ribonucleotide product of the enzyme. However, 15 previously identified guanine riboswitches are associated with genes for GMP synthase in other organisms (ref. 26 and data not shown). Therefore, M. florum appears to use a variant guanine riboswitch to control this homologous gene much like other bacteria.

The discovery of variant riboswitches in M. florum that have altered ligand specificities and affinities suggests that a far greater diversity of riboswitches is present in bacteria and that some of these RNAs might be exceedingly narrow in phylogenetic distribution. Although highly selective for their cognate ligands, at least some riboswitch aptamers appear to be versatile and can change ligand affinities by accumulating one or only a few mutations. It seems reasonable to speculate that the common fold and consensus sequence for guanine riboswitches (Fig. 1B) could undergo other mutations that would allow it to bind other purine-related compounds while maintaining its secondary structure.

The two riboswitch variants that selectively sense 2′-dG have been found in only one bacterial species. Given the apparent structural versatility of RNA, there might be many unidentified riboswitch classes that reside in a single species. The aptamers for these RNAs could be close variants of known riboswitch classes, which would permit their discovery by comparative sequence analysis. However, we have evidence that structurally distinct classes of riboswitches also can be found to exist in only a few species (24). Searching for new riboswitch classes by bioinformatics methods that recognize conserved sequences or structures requires the existence of multiple copies. Therefore, the identification of exceedingly rare riboswitch classes might be best pursued by using approaches that involve direct experimental testing.

Materials and Methods

Bioinformatics.

The original search for purine riboswitches was performed with the SequenceSniffer program (J. E. Barrick and R.R.B., unpublished data), with an E-value cutoff of 10,000. To determine whether additional purine riboswitch variants could be identified by using automated homology searches, the National Center for Biotechnology Information RefSeq database (51) was searched by using the RaveNnA extension (52) to the software package INFERNAL (www.infernal.janelia.org). A covariance model derived from all known purine riboswitches and one derived exclusively from the variant purine riboswitches were used as inputs for the searches.

Chemicals and Oligonucleotides.

2′-deoxyguanosine, 3′-deoxyguanosine, 2′-deoxyadenosine, 2′-deoxyguanosine-5′-phosphate, 2′-deoxyadenosine-5′-phosphate, 2′-deoxyguanosine-5′-diphosphate, 2′-deoxyadenosine-5′-diphosphate, 2′-deoxyguanosine-5′-triphosphate, 2′-deoxyadenosine-5′-triphosphate, guanosine-5′-phosphate, guanosine-2′-phosphate, guanosine-5′-diphosphate, guanosine-5′-triphosphate, adenosine-5′-triphosphate, 2′-deoxy-2,6-diaminopurine nucleoside (2,6-diaminopurine 2′-deoxyriboside), guanine, adenine, 2,6-diaminopurine, N²-methylguanine, O⁶-methylguanine, and 7-methylguanine were purchased from Sigma–Aldrich (St. Louis, MO). DNA oligonucleotides were synthesized by the Howard Hughes Medical Institute Keck Foundation Biotechnology Resource Center at Yale University; purified by denaturing PAGE; eluted from the gel by crush-soaking in 10 mM Tris·HCl (pH 7.5 at 23°C), 200 mM NaCl, and 1 mM EDTA; and precipitated with ethanol.

In-Line Probing Assays.

RNA constructs were prepared from synthetic double-stranded DNA templates by in vitro transcription by using methods similar to those described previously (53). The resulting RNAs were dephosphorylated by using alkaline phosphatase (Roche Diagnostics, Indianapolis, IN) and subsequently labeled with ³²P by using T4 polynucleotide kinase (New England Biolabs, Ipswich, MA) following the manufacturer's instructions. Radiolabeled RNAs (≈2 nM) were subjected to in-line probing by incubation with or without various ligands for 40 h in 10-μl reactions containing 50 mM Tris·HCl (pH 8.5 at 23°C), 20 mM MgCl₂, and 100 mM KCl. Denaturing 10% PAGE was used to separate spontaneously cleaved products, which were visualized by using a Molecular Dynamics PhosphorImager (Sunnyvale, CA). ImageQuaNT software was used to quantitate spontaneous cleavage amounts.

In Vitro Transcription Termination Assays.

The protocol for single-round transcription assays was adapted from that described previously (45). The lysC promoter of B. subtilis was used to facilitate greater transcription yield with the I-A construct. Transcription reactions contained 100 nM DNA template in 20 mM Tris·HCl (pH 8.0 at 23°C)/20 mM NaCl/14 mM MgCl₂/0.1 mM EDTA/1 mg/ml BSA/50% glycerol 2.2 μM E. coli RNA polymerase holoenzyme (Epicenter Technologies, Madison, WI). Transcription was initiated by adding 2.5 μM GTP and UTP, 1 μM ATP, 4 μCi [α-³²P]ATP, and 1.35 μM ApA dinucleotide. After incubating for 10 min at 37°C, 0.075 mM GTP, ATP, and CTP; 0.025 mM UTP; and 0.1 mg/ml heparin were added, and the resulting mixture was allowed to incubate for 20 min at 37°C. Products were separated by denaturing 6% PAGE and imaged and quantitated by using a PhosphorImager and ImageQuaNT software.

The FL and T transcript amounts were established by correcting for the differences in the number of A residues in the molecules. The percentage of [α-³²P]ATP compared with total ATP concentration in the initiation and elongation reactions (7% and 0.4%, respectively) was established, and the relative amount of radioactivity per T (U_T) and FL (U_FL) transcripts was calculated for each transcript size by using the following equation: [(Number of A residues in initiation region)(7%) + (Number of A residues in elongation region)(0.4%)] = U.

U_T/U_FL is equal to the correction factor (X%) that accounts for the increased number of radiolabeled adenosine residues in the FL transcript. The equation used to establish the percentage of transcription termination was: [T/(T + FL)(X%)] = percentage termination.

Abbreviations

2′-dA

2′-deoxyadenosine

2′-dG

2′-deoxyguanosine

2′-d-2,6-DAP

2′-deoxy-2,6-diathinopurine nucleoside.