Transcriptional and translational S-box riboswitches differ in ligand-binding properties

There are a number of riboswitches that utilize the same ligand-binding domain to regulate transcription or translation. S-box (SAM-I) riboswitches, including the riboswitch present in the Bacillus subtilis metI gene, which encodes cystathionine γ-synthase, regulate the expression of genes involved in methionine metabolism in response to SAM, primarily at the level of transcriptional attenuation. A rarer class of S-box riboswitches is predicted to regulate translation initiation. Here we identified and characterized a translational S-box riboswitch in the metI gene from Desulfurispirillum indicum. The regulatory mechanisms of riboswitches are influenced by the kinetics of ligand interaction. The half-life of the translational D. indicum metI RNA–SAM complex is significantly shorter than that of the transcriptional B. subtilis metI RNA. This finding suggests that, unlike the transcriptional RNA, the translational metI riboswitch can make multiple reversible regulatory decisions. Comparison of both RNAs revealed that the second internal loop of helix P3 in the transcriptional RNA usually contains an A residue, whereas the translational RNA contains a C residue that is conserved in other S-box RNAs that are predicted to regulate translation. Mutational analysis indicated that the presence of an A or C residue correlates with RNA–SAM complex stability. Biochemical analyses indicate that the internal loop sequence critically determines the stability of the RNA–SAM complex by influencing the flexibility of residues involved in SAM binding and thereby affects the molecular mechanism of riboswitch function.

Riboswitches are cis-acting RNAs that regulate gene expression in response to specific effector molecules (1,2). Each riboswitch class contains an aptamer domain that recognizes its cognate ligand, which includes vitamins, amino acids, cofactors, nucleotides, metal ions, and tRNAs (2)(3)(4). In most riboswitches, binding of the ligand to the aptamer affects the structure of an expression platform, which regulates expression of the downstream genes.
As a general rule, in an ON switch, binding of the ligand promotes expression of the regulated genes, whereas in an OFF switch, binding of the ligand leads to reduced expression of the regulated genes. In most ON switches, the effector molecule is utilized either as a substrate or exported out of the cell, resulting in a "feedforward" mechanism. In contrast, in most OFF switches, the associated genes are involved in biosynthesis or uptake of the effector molecule, resulting in a negative feedback control. The diversity of riboswitch-regulated genes and their recognized ligands leads to a number of exceptions to these general principles.
Riboswitches are found primarily in bacteria, where they regulate gene expression at the level of transcription attenuation or translation initiation. Transcriptional riboswitches generally contain an intrinsic transcriptional terminator that leads to premature transcription termination. Folding of the RNA into an alternate antiterminator structure allows transcription to continue, and the gene is expressed. In an OFF switch, the antiterminator is more thermodynamically stable, and binding of the ligand causes stabilization of the terminator helix to repress gene expression. Translational riboswitches operate similarly, except that the terminator helix is replaced by a sequestrator helix that occludes the Shine-Dalgarno (SD) 2 region and prevents translation of the downstream gene, and the antiterminator is replaced by an antisequestrator helix that allows the ribosome to bind to the RNA and initiate translation (2).
The simplest model of riboswitch-mediated regulation suggests that the RNA senses physiological levels of the ligand and that the resulting change in RNA structure triggers the regulatory activity. However, for some transcriptional riboswitches, a higher ligand concentration is required to direct gene regulation than expected based on the apparent equilibrium dissociation constant (K D ) (5-7). It has been proposed that, for transcriptional riboswitches, the kinetics of ligand binding and the rate of transcription play an important role in the regulatory decision (6,8). During transcription, the aptamer domain is transcribed and folded first, with only a short window of opportunity for the ligand to bind before RNA polymerase (RNAP) reaches the decision point between termination and antitermination. For an OFF switch, if the aptamer domain folds and binds the ligand before RNAP completes transcription of the antiterminator sequence, the competing terminator helix forms, and termination occurs. However, if the ligand does not bind to the RNA in this timeframe, the more stable antitermi-nator forms, and transcription continues. Because of the speed of transcription, most transcriptional riboswitch RNAs do not reach equilibrium with the ligand before RNAP reaches the decision point (8).
Occupancy is also affected by the rate of dissociation of the ligand from the RNA (9,10). Translational riboswitches differ from their transcriptional counterparts in that they can bind the ligand cotranscriptionally and posttranscriptionally. They therefore have the potential to establish thermodynamic equilibrium with their ligand (9,11). If the half-life of the RNAligand complex is shorter than the half-life of the fully transcribed mRNA, then a translational riboswitch has the potential to make multiple reversible regulatory decisions in its lifetime.
SAM-responsive S-box (SAM-I) RNAs are one of the most prevalent classes of riboswitches and are predominantly found in Firmicutes (3). S-box riboswitches regulate genes that are involved in biosynthesis and uptake of methionine and SAM and operate primarily at the level of transcription attenuation (12). The well-characterized Bacillus subtilis yitJ transcriptional S-box riboswitch forms a highly stable complex with SAM in vitro, with a half-life of 7.2 min (7). The high stability of the complex indicates that, when bound, SAM is not likely to dissociate in the timescale of transcription, making the binding and regulatory decision irreversible (13). The S MK -box (SAM-III) RNAs, another class of SAM-sensing riboswitches that regulate SAM synthetase (metK) genes at the level of translation initiation, exhibit a much less stable RNA-SAM complex with a half-life of ϳ8 s (9). A long S MK -box transcript half-life (ϳ3 min) provides the opportunity for multiple rounds of SAM association and dissociation, suggesting that the translational S MK -box riboswitch can act as a true reversible riboswitch and make multiple regulatory decisions in its lifetime (9,14,15).
SAM-responsive transcriptional S-box and translational S MK -box riboswitches employ different regulatory strategies (irreversible versus reversible) to regulate their respective genes. However, the two riboswitches are structurally different and recognize SAM using different mechanisms. Hence, it is unclear whether it is the level of regulation or the riboswitch structure that dictates the differences in the regulatory mechanisms. We therefore chose to compare the properties of transcriptional and translational SAM-responsive riboswitches from the same class in different organisms. We previously showed that different B. subtilis S-box genes show different properties (7); we therefore compared properties of the same gene.
Our data show that the stability of the translational S-box RNA-SAM complex is significantly (8.6-fold) lower than the stability of the transcriptional S-box RNA-SAM complex, which may enable the RNA to make more than one reversible regulatory decision, unlike its transcriptional counterpart. We investigated elements within the riboswitch that might contribute to the differences in SAM binding properties and identified a nucleotide in the second internal loop of helix P3 whose identity correlates with predicted transcriptional versus translational regulation. Mutational analysis indicated that the identity of this residue affects the stability of the RNA-SAM complex, and selective 2Ј-hydroxyl acylation analyzed by primer extension (SHAPE) analysis provided insight into the structural basis of the differences in ligand binding properties. These results indicate that this internal loop sequence plays an important role in determining the stability of the RNA-SAM complex in transcriptional and translational RNAs and, therefore, affects the molecular mechanism of S-box riboswitch function.

Identification of a translational S-box riboswitch
Bioinformatic analyses have suggested that, in some non-Firmicutes, S-box riboswitches regulate gene expression by mechanisms other than transcription attenuation, based on the absence of a distinct terminator helix in the expression platform (16 -18). We used the Rfam database (19) to identify S-box sequences that were likely to regulate gene expression at the level of translation initiation and for which the riboswitch of the equivalent gene has been characterized in B. subtilis.
The metI S-box riboswitch from Desulfurispirillum indicum, a member of the phylum Chrysiogenetes (20), has all of the conserved features (18,21) of the S-box RNA family (Fig. 1A). The expression platform of this RNA is predicted to fold into two competing structures, a sequestrator helix that occludes the putative SD sequence and an antisequestrator helix that exposes the SD region. The metI gene encodes a putative cystathionine ␥-synthase that catalyzes conversion of cysteine to cystathionine. This gene is followed by the gene encoding cystathionine ␤-lyase, which catalyzes conversion of cystathionine to homocysteine. In B. subtilis, these enzymes are encoded by the metIC operon (7), which is regulated by a transcriptional S-box riboswitch (Fig. 1B).

Transcriptional and translational metI RNAs bind to SAM with similar affinity and specificity
The aptamer domain of the D. indicum metI RNA contains elements that have been shown previously to be important for SAM binding and for formation of structural elements such as the kink-turn and the pseudoknot (13,22). We therefore compared the SAM binding properties of the D. indicum metI aptamer RNA with those of B. subtilis metI and other previously characterized S-box riboswitches (7).
Size-exclusion filtration assays were performed to determine the SAM binding affinities of the two metI riboswitches. The B. subtilis glyQS RNA, which is a T-box riboswitch that binds tRNA Gly instead of SAM, was used as a negative control. The D. indicum metI aptamer RNA bound SAM with an affinity comparable with that of the B. subtilis metI RNA ( Fig. 2A). A G residue in the junction between helices P1 and P2 (J1/2) has been shown previously to interact directly with SAM, and mutation of that G to C causes loss of SAM binding in the yitJ S-box aptamer RNA (13). Analogous mutations (G19C for D. indicum metI RNA and G39C for B. subtilis metI RNA) resulted in undetectable levels of SAM binding (Fig. 2B), consistent with previous results for yitJ. S-box riboswitches have also been shown to strongly discriminate against SAH, a natural analog of SAM (12). To test whether D. indicum S-box RNA exhibits similar selectivity against SAH, competition binding assays were performed with [ 3 H]SAM, unlabeled SAM, and SAH. We found that unlabeled SAM but not SAH competed The sequence is shown in the sequestrator conformation. Green and red residues represent the alternate pairing required for formation of the antisequestrator, shown above the sequestrator. The SD sequence is shown in blue. B, metI sequence from B. subtilis. The sequence is shown in the terminator conformation. Green and red residues represent the alternate pairing required for formation of the antiterminator, shown above the terminator.

Transcriptional versus translational S-box riboswitches
efficiently with [ 3 H]SAM for binding to the aptamer RNAs, which indicates that the binding is specific for SAM (Fig. 2B). These data show that the aptamer domains of transcriptional and translational metI S-box riboswitches exhibit affinity and specificity for SAM as well as sensitivity to a mutation in J1/2, similar to that of the previously characterized S-box riboswitches (7,12,13).

The D. indicum metI riboswitch undergoes a SAM-dependent structural transition and binds 30S ribosomal subunits in vitro
The translational regulation model suggests that the sequestrator helix in D. indicum metI RNA undergoes a SAM-dependent structural transition that results in stabilization of the sequestrator helix that contains the SD region. We performed an RNase H cleavage assay to determine whether accessibility of the SD region decreases in response to SAM, as predicted by the model. The RNA was incubated with a DNA oligonucleotide complementary to the SD region in the presence or absence of SAM, and the availability of the SD region was monitored by addition of RNase H, which cleaves RNA-DNA hybrids. A cleavage product was observed only in reactions that contained the oligonucleotide but lacked SAM (Fig. 3A), consistent with SAM-dependent stabilization of the sequestrator helix. These results indicate that the SD sequence is accessible to oligonucleotide binding (and therefore RNase H cleavage) in the absence but not in the presence of SAM, supporting the translational regulation model.
SAM-dependent stabilization of the sequestrator helix is predicted to inhibit binding of the ribosome to the SD region, whereas in the absence of SAM, the formation of the antisequestrator helix should make the SD sequence available for ribosome binding. To test this hypothesis in vitro, we performed a primer extension inhibition ("toeprint") assay. In this assay, when a transcribing RT encounters an obstacle on the  A ␥-32 P-labeled primer was annealed to the 3Ј end of the transcript and incubated with 30S ribosomal subunits and tRNA fmet in the presence or absence of SAM. RT was added to initiate cDNA synthesis. The reaction products were resolved on a 6% denaturing polyacrylamide gel. The bands labeled ϩ16 and ϩ17 are RT stops that correspond to their positions with respect to the A of the AUG start codon in the transcript.

Transcriptional versus translational S-box riboswitches
RNA (e.g. the 30S ribosomal subunit or a strong secondary structure), it stalls, and primer extension is inhibited. When the initiator tRNA fMet is positioned in the P site of a bound 30S subunit, an RT stop (toeprint) is usually observed 16 -17 nt downstream of the A of the AUG start codon (23,24). D. indicum metI RNA was incubated with Escherichia coli 30S ribosomal subunits and initiator tRNA fMet in the presence or absence of SAM. Strong RT stops were observed at positions ϩ16 and ϩ17 nt downstream of the A in the AUG start codon in the absence of SAM. The intensity of the toeprint signal decreased ϳ2-fold (2.1 Ϯ 0.2; S.D., n ϭ 3) in the presence of SAM (Fig. 3B). This suggests that addition of SAM shifts the RNA population into a conformation that inhibits ribosome binding, supporting the hypothesis that the SAM binding represses translation initiation. In the presence of SAM (Fig. S1, lane 2), RT stops were also observed at the 3Ј ends of the aptamer domain (G125, C127) and the sequestrator helix (G188, A189). This suggests that both structures are stabilized in the presence of SAM, which supports our model of the OFF conformation. Interestingly, the intensity of these stops decreased upon addition of the 30S subunit and tRNA fMet (Fig.  S1, lane 5), consistent with the model that binding of the 30S subunit can stabilize the ON conformation.

Stability of the transcriptional and translational RNA-SAM complexes in vitro
Although the SAM-binding affinities of the transcriptional and translational RNAs were similar, the kinetics of SAM interaction may play an important role in dictating the regulatory strategies (25)(26)(27). The kinetics of ligand dissociation determine the stability of the RNA-ligand complex. A less stable RNA-ligand complex may provide an opportunity for multiple rounds of regulation by the riboswitch, whereas a more stable RNA-ligand complex could make the regulatory decision irreversible.
We determined the stability of the translational metI S-box RNA-SAM complex using a 2-aminopurine (2AP)-based fluorescence assay. 2AP fluorescence is sensitive to local conformational changes in RNA (28); the fluorescence intensity decreases in response to base-stacking and ligand-binding interactions. A14 of helix P1, which is part of the SAM-binding core (Fig. 4A), was replaced with 2AP. This residue stacks with U15 and base-pairs with U121, both of which specifically recognize the sulfonium ion in SAM (13,29). SHAPE-based structural analyses have shown that the A at this position undergoes a SAM-dependent decrease in reactivity (13,30). 2AP at position A14 is predicted to fluoresce in the absence of SAM and exhibit quenching in the presence of SAM. Previous work from our laboratory showed that a 2AP substitution in the same position in the yitJ aptamer domain resulted in SAM-dependent quenching of fluorescence (31). A bipartite RNA consisting of two pieces of RNA (a chemically synthesized 5Ј half with 2AP at the A14 position and a 3Ј half synthesized by T7 RNAP transcription) that were cofolded to reconstitute the aptamer domain was used to assess SAM binding to the translational metI S-box RNA. The bipartite RNA bound SAM with an affinity (K D ϭ 99 nM) that was within 2-fold of that of the intact RNA (K D ϭ 53 nM). A similar bipartite system was used to study the kinetics of SAM dissociation from the S MK -box riboswitch (9).
To determine the rate of SAM dissociation, the equilibrated bipartite metI RNA-SAM complex was mixed rapidly with a 10-fold molar excess of competitor B. subtilis yitJ S-box RNA. The yitJ RNA binds SAM with a high affinity (K D ϭ 19 nM) and forms a very stable complex with a half-life of 7.2 min (7). As the SAM dissociates from the metI RNA-SAM complex, it is sequestered by the yitJ RNA, and the fraction of SAM-bound RNA decreases. This leads to an increase in 2AP fluorescence emission over time, and the data are plotted as the fraction of SAM-bound RNA as a function of time.
The data were initially fit with a single-exponential equation, but improvement in the fit was observed by using a doubleexponential equation. The fast and slow dissociation rates (k off ) were determined to be 0.23 s Ϫ1 and 0.014 s Ϫ1 , respectively (Fig.  4C). The observed fast rate is likely to be due to a dynamic intermediate state that might exist between the bound and unbound states of the S-box aptamer. The t1 ⁄ 2 value, which indicates the stability of the RNA-ligand complex, was determined from the slow rate as 50 Ϯ 2 s. The association rate (k on ) derived from the K D and the k off values is 2.6 ϫ 10 5 M Ϫ1 s Ϫ1 . The time required for the ligand to bind the RNA depends on the physiological concentration of SAM in D. indicum. For comparison, the half-life of the translational Enterococcus faecalis S MK -box RNA in vivo was determined to be ϳ3-4 min (9). The average half-life of mRNAs in a Gram-negative non-Firmicute like E. coli is ϳ5-7 min (32,33). Assuming a half-life value in a similar range for the metI transcript, the t1 ⁄ 2 value suggests that SAM can bind and dissociate more than once before the metI mRNA is degraded. The ability for multiple rounds of SAM association and dissociation implies that the riboswitch can make multiple reversible regulatory decisions in its lifetime.
Next we determined the stability of the transcriptional metI RNA-SAM complex (Fig. 4D). We could not derive an off-rate from a similar 2AP fluorescence-based stopped-flow assay, as SAM failed to dissociate in the timescale of the experiment. We therefore used the size-exclusion filtration-based assay with radioactive SAM as the ligand and nonradioactive SAM as the competitor ligand. In this assay, the preformed RNAradioactive SAM complex was mixed with excess cold SAM, and aliquots were removed at various time points and passed through the filter to separate RNA-bound SAM and unbound SAM. The amount of radioactive SAM bound to RNA was determined for each aliquot, and data were plotted as the fraction of SAM bound to RNA as a function of time. The t1 ⁄ 2 value calculated from the plot was 7.2 Ϯ 0.2 min (Fig. 4D), which is similar to the previously reported value of 5.1 min (7). The k off value is 0.0016 s Ϫ1 , and the k on value is 3.5 ϫ 10 4 M Ϫ1 s Ϫ1 . Because the rate of prokaryotic transcription is estimated to be ϳ25-30 nt s Ϫ1 (34 -36), a complex half-life in minutes suggests that binding and regulatory decision would be irreversible in the timeframe of transcription.
The translational metI riboswitch-SAM complex is 8.6-fold less stable than the transcriptional metI riboswitch-SAM complex despite similarity in the overall secondary structure, core binding sequences, and binding affinities. The low stability suggests that it can reach thermodynamic equilibrium with its

Transcriptional versus translational S-box riboswitches
ligand. Hence, the transcribed mRNA is predicted to monitor and respond to changes in the physiological levels of SAM that are closer to the K D . In contrast, the high stability of the transcriptional metI riboswitch-SAM complex suggests that it will likely never reach equilibrium with SAM. Consequently, a concentration of SAM higher than the K D will be required to trigger the termination response. These data suggest that the translational and transcriptional S-box riboswitches can adopt different regulatory strategies to regulate the same gene in different organisms.

Role of helix P3 in RNA-SAM complex stability
The overall secondary structure and identity of the nucleotides in the SAM-binding core are conserved among S-box sequences, whereas elements outside of the binding core show variability (7,21,37,38). For example, although the residues at the proximal end of helix P3 that form the first internal loop (IL3a) are highly conserved and directly contact SAM (13,29), the distal end of helix P3 varies in length and sequence. Helix P3 contains a second internal loop (IL3b), whose size and nucleotide composition are variable (39). It has been shown previously that deletion of IL3b is detrimental to the function of the B. subtilis metE S-box riboswitch and that the presence of an A in the 3Ј side of IL3b promotes SAM binding and riboswitch activity (39). The adenine corresponding to A104 in the 3Ј side of IL3b in B. subtilis metI riboswitch (Fig. 4B, red) is highly conserved and is present in more than 75% of S-box sequences (19).
We compared the sequence of IL3b between the transcriptional and translational RNAs. Although the 3Ј side of IL3b of the transcriptional metI RNA contains the more common 2-nt bulge, including the conserved A (A104, Fig. 4B), the translational metI RNA lacks this conserved A and contained only a single cytosine residue (C84, Fig. 4A) at the same position. Analysis of the Rfam seed sequence alignment showed that a single C in this position is found in Ͻ1% of S-box sequences, all of which are predicted to regulate translationally based on the lack of a predicted transcriptional terminator and the presence of putative sequestrator structures. We hypothesized that the difference in the composition between the second internal loops of the two RNAs could contribute to the difference in their RNA-SAM complex stabilities.
To test this hypothesis, we first replaced C84 in the translational metI RNA with two adenine residues (C84A/⍀A85), to mimic IL3b of the transcriptional metI RNA. This mutant has an RNA-SAM complex half-life of 280 Ϯ 10 s (Fig. 4C), a 5.6fold increase. Replacing A104 and A105 in the transcriptional RNA with a single C (A104C/⌬A105) resulted in an RNA-SAM complex half-life of 2.0 Ϯ 0.4 min (Fig. 4D), a 3.6-fold decrease in complex stability. These results support our hypothesis that the difference in the kinetics of ligand dissociation between the transcriptional and translational RNAs is, at least in part, influenced by the difference in the size and nucleotide composition of the IL3b region.

SHAPE analysis of transcriptional and translational RNAs
To investigate the structural basis of the difference in the stabilities of the RNA-SAM complexes between the two RNAs, we examined the structure of the WT and mutant translational and transcriptional metI RNAs using SHAPE (40). SHAPE measures the backbone flexibility of each nucleotide in an RNA; unconstrained nucleotides are preferentially acylated and show higher reactivity relative to constrained nucleotides. Comparison of the SHAPE reactivity profiles of the WT transcriptional and translational RNAs showed significant differences in the flexibility of the conserved A residues in the 5Ј side of IL3a. The nucleotide identity of the position corresponding to the first A (translational, A54; transcriptional, A74) is 97% conserved (19,38), whereas that corresponding to the second A (translational, A55; transcriptional, A75) is 96% conserved (39). In the B. subtilis yitJ S-box RNA crystal structure with SAM, the first A in IL3a is a part of the conserved A⅐U sheared base pair (translational, A54⅐U87; transcriptional, A74⅐U108) that directly contacts SAM by forming a base triple with its adenosyl moiety (13,29). The second A in IL3a (A75 in Fig. 4B) interacts with the conserved A in IL3b (A104 in Fig. 4B, red).
In the translational metI RNA, the A residues in IL3a (A54 and A55; green nucleotides in Fig. 4A) were significantly more reactive (Fig. 5A) than the corresponding A residues (A74 and A75; green nucleotides in Fig. 4B) in the transcriptional metI RNA (Fig. 5C). Introduction of mutations (C84A/⍀A85) in IL3b of the translational metI RNA led to a 2-fold and 4.5-fold decrease in the reactivities of A54 and A55, respectively (Fig.  5B). In contrast, in the transcriptional metI RNA, A74 in IL3a was reactive, whereas A75 was minimally reactive (Fig. 5C). The mutations (A104C/⌬A105) in IL3b of the transcriptional metI RNA led to a 3.4-fold increase in the reactivity of A75 (Fig. 5D). In both cases, the second A in IL3a exhibited a larger change in reactivity in response to the mutations.
These results indicate that the presence of A residues in the 3Ј side of IL3b decreases flexibility of the A residues (Fig. 4, A  and B, green) in the 5Ј side of IL3a. In contrast, the presence of C at the same position increases flexibility of the A residues in the 5Ј side of IL3a. The change is most prominent in the second A residue of IL3a, which contacts the 3Ј side of IL3b. This suggests that the nucleotide identity and composition of IL3b may influence the dynamics of residues in IL3a. Because the residues in IL3a directly contact SAM, the change in local flexibility of these residues could affect the kinetics of ligand interaction. These data suggest that the residues in IL3b, through their interaction with residues in IL3a, could indirectly affect the stability of the RNA-SAM complex. The composition of IL3b may therefore play a role in determining the regulatory mechanisms employed by transcriptional versus translational S-box riboswitches.

Discussion
Few studies have directly compared transcriptional and translational riboswitches for a specific riboswitch class. For example, transcriptional and translational preQ1 riboswitches fold via two distinct pathways of conformational selection and induced fit mechanisms, respectively, despite their structural similarity (41). Comparison of the adenine-binding translational add and transcriptional pbuE riboswitches suggests that the level of gene regulation might be important in determining riboswitch regulatory strategies (42). It has been shown that the add riboswitch can switch reversibly between the ON and OFF

Transcriptional versus translational S-box riboswitches
states but pbuE RNA cannot (42-44). These riboswitches regulate genes that are functionally distinct (adenosine deaminase versus an adenine efflux pump), so it is unclear whether their regulatory mechanisms correlate with gene function. More importantly, the role of RNA-ligand complex stability in the regulatory strategies of transcriptional versus translational riboswitches was not investigated.
The rate of association is ϳ10 times slower for the transcriptional RNA compared with the translational RNA. The k on value of 3.5 ϫ 10 4 M Ϫ1 s Ϫ1 for the B. subtilis transcriptional riboswitch suggests that, in the presence of 1 M SAM, it would take ϳ20 s for half of the metI transcripts to bind SAM. Increasing the SAM concentration would proportionately decrease the association time constant. Under SAM-limiting conditions (Ͻ25 M) (7), binding will take around ϳ1 s, whereas under saturating levels of SAM (300 M), it could take ϳ0.1 s. For B. subtilis metI, there are 24 nt between the end of the aptamer domain and the end of the antiterminator helix. This means that after the aptamer is transcribed, it has ϳ1 s (assuming a transcription elongation rate of 25 nt/s) to fold and bind SAM. Riboswitch aptamers have been shown to fold in less than 100 ms (45,46). The probability of SAM binding to the aptamer

Transcriptional versus translational S-box riboswitches
before antiterminator helix formation is therefore higher when SAM is at saturating levels. After the aptamer is transcribed, it will take ϳ1.5 s for RNAP to transcribe the stretch until terminator and the run of Us (ϳ40 nt). The half-life of 7.2 min suggests that, if SAM is bound to the aptamer RNA, then it is unlikely to dissociate in this interval. Hence, the binding and regulatory decision for the metI transcriptional riboswitch is irreversible in the timeframe of transcription. The SAM pools of D. indicum are not known. However, the ϳ10 times faster on and off rates support the possibility of multiple rounds of SAM association and dissociation.
Our data indicate that the presence of a C in the 3Ј side of IL3b imparts low stability to RNA-SAM complexes. A metE S-box riboswitch mutant that contains a C in IL3b exhibited loss of SAM binding and riboswitch function (39), suggesting that a C at this position can be detrimental to SAM binding in other S-box riboswitches as well. It remains to be seen how the presence of other residues in IL3b would affect the dissociation kinetics and, hence, the regulatory mechanism of S-box riboswitches. The presence of a single C residue in the 3Ј side of IL3b appears to be associated only with the translational S-box riboswitches. Based on the low stability of the RNA-SAM complex, we predict that the translational riboswitches with this sequence will regulate their genes reversibly. However, there are some S-box sequences that, we predict, regulate translationally based on the lack of a terminator helix and that do not have the single C residue in the 3Ј side of IL3b. Characterizing RNAs of this type may provide additional insight into the role of this domain in S-box RNA function.
Like the metI transcriptional riboswitch, most of the other B. subtilis transcriptional S-box riboswitches have an A in the IL3b and form stable complexes with SAM, with half-lives ranging from 4.3-11 min (7). One exception is yxjG, which is predicted to form a weaker pseudoknot base-pairing interaction (2 nt versus 3-4 nt), which could contribute to the lower complex stability (1.6 min). The B. subtilis yoaD RNA that lacks IL3b has an RNA-SAM complex half-life of 1.8 min. These observations provide additional support for the hypothesis that the presence and composition of IL3b play a role in the ligand dissociation kinetics. D. indicum contains a second S-box riboswitch in the metA gene that regulates at the level of transcription attenuation in vitro (Fig. S3). In contrast to the translational metI riboswitch, the 3Ј side of IL3b of the metA transcriptional RNA contains the highly conserved and more predominant A residue, which is associated with highly stable RNA-SAM complexes and irreversible regulation.
Interestingly, the differences in the composition of an internal loop in the P2 helix of thiamine pyrophosphate (TPP) riboswitch have been suggested to contribute to differences in the regulatory mechanisms between the E. coli thiM and Arabidopsis thaliana thiC TPP riboswitches (47). The nucleotides in this internal loop are either part of or adjacent to the binding pocket and important for TPP binding and folding of the riboswitch (48). Hence, the contribution of an internal loop to regulatory differences between riboswitches in the same class could be a feature found in other riboswitch families.
Residues adjacent to the ligand binding pocket have been suggested to modulate ligand binding kinetics in the purine and flavin mononucleotide (FMN) riboswitches (10,49). However, no correlation was observed between the variation in ligand binding kinetics and the degree of local nucleotide flexibility (10). Our SHAPE data link the kinetics of SAM dissociation to the dynamics of nucleotides adjacent to the SAM binding pocket in the S-box riboswitch. In the SAM-bound yitJ crystal structure (13), the residues corresponding to A74 and U108 (Fig. 4B) form a base triple with the adenosyl moiety of SAM.
The residue corresponding to A75 in IL3a is part of a dinucleotide platform (A75⅐C76-G107) that has been suggested to further stabilize the position of the adenine ring of SAM (29,50). The residues corresponding to A75 in IL3a and A104 in IL3b interact via hydrogen bonds and stacking interactions. The amino groups of A75 and A104 form hydrogen bonds with the 2Ј-hydroxyl groups of the other. In the SAM-free metF-H2 S-box crystal structure (51), which lacks IL3b, this interaction is missing. Instead, the A corresponding to A75 in IL3a blocks the SAM-binding pocket by forming a base triple with residues corresponding to A74 and U108.
Unlike the A75-A104 interaction in the transcriptional metI RNA, the interaction between C84 in IL3b and A55 in IL3a in the translational metI RNA could be weak or absent. This is because A55 is 4.5-fold more flexible in the presence of C in IL3b (Fig. 5, A and B). Additionally, a ⌬C84 deletion mutant exhibited an RNA-SAM complex half-life (Fig. S2B, 54 s) similar to that of the WT translational RNA. In the absence of this interaction, the translational aptamer could behave similarly as the SAM-free metF crystal structure that lacks IL3b. The residue corresponding to A55 has been proposed to transition rapidly between two major equilibrium conformations, one that occupies the binding pocket and another that is flipped out of the pocket (51). This means that A55 in the metI translational RNA could compete with SAM for the binding pocket, which could explain the faster dissociation kinetics. In contrast, the A104-A75 interaction in the B. subtilis metI transcriptional RNA would prevent A75 from occupying the binding pocket, making the binding pocket less dynamic, and could encourage a SAM-bound state by stabilizing a binding-competent aptamer RNA.
The regulatory mechanism of the translational riboswitch is likely to be influenced by a combination of additional factors, such as the strength of the SD sequence, the dynamics of the expression platform, the cellular concentration of the ligand, and competition between the kinetics of ligand binding to the aptamer domain and ribosome recruitment to the expression platform. The variation in ligand binding properties of the 11 B. subtilis transcriptional S-box riboswitches have been linked to the function of the genes they regulate (7). Similarly, the reversibility of translational S MK -box and add riboswitches has been speculated to be linked to their physiological roles (9,42). It is possible that the regulatory strategy adopted by the translational metI S-box riboswitch provides an advantage to its host organism in its niche. D. indicum has been isolated from arsenate-rich environments and could depend on detoxification by SAM-dependent methylation of arsenite (20,52,53). Therefore, rapid sensing of physiological levels of SAM and maintaining SAM homeostasis may be critical for its survival. Investigation into these aspects, in vitro and in the context of the native Transcriptional versus translational S-box riboswitches host, will enable a better understanding of the basis of the regulatory differences observed between transcriptional and translational S-box riboswitches.

DNA constructs and RNA synthesis
The sequences for translational metI constructs were derived from the genome of D. indicum strain S5 (NCBI CP002432). Transcriptional S-box sequences were obtained from the genome of B. subtilis strain 168 (NCBI AL009126). DNA constructs were generated by using genomic DNA or ligated pairs of complementary oligonucleotides (Integrated DNA Technologies) as templates for PCR amplification. Constructs containing mutations were made by replacing a pair of oligonucleotides with an alternate pair that contained the sequence changes. All constructs were confirmed by DNA sequencing. RNA was synthesized using laboratory-prepared T7 RNAP or the Tran-scriptAid T7 High Yield Transcription Kit (Thermo Scientific). All transcripts synthesized using T7 RNAP were designed to contain a G residue as the first nucleotide to facilitate transcription initiation. The RNA products were gel-purified using denaturing PAGE. RNA concentration was determined using an ND-1000 spectrophotometer (NanoDrop Technologies, Inc.).

SAM binding assays
Equilibrium dissociation constants were determined using a size-exclusion filtration assay as described previously (7).

RNase H cleavage assay
RNase H cleavage experiments were carried out with D. indicum metI RNA uniformly labeled by transcription in the presence of [␣-32 P]UTP (800 Ci/mmol, PerkinElmer Life Sciences). An antisense oligonucleotide (5Ј-TCCCCCT-3Ј) complementary to the SD region (nucleotides 180 -186) was added to SAM-bound and SAM-free RNA (40 nM) at 100-fold molar excess, and the reactions were incubated at 37°C for 5 min. RNase H (10 units/l, Ambion) was added, and the reactions were incubated at 37°C for 10 min. The reactions were terminated by phenol-chloroform extraction, and products were resolved by denaturing 6% (w/v) PAGE and visualized by PhosphorImager analysis (Molecular Dynamics).

Primer extension inhibition assay
A DNA primer complementary to the positions 277-302 of the 3Ј end of the D. indicum metI transcript was 5Ј end-labeled (54) with [␥-32 P]ATP (6000 Ci/mmol, PerkinElmer Life Sciences). The labeled oligonucleotide (10 nM) was annealed to the full-length T7 RNAP-transcribed RNA (10 nM) corresponding to positions 7-302 of the metI RNA. SAM (160 M) or water was added, and the samples were incubated for 10 min at 37°C. Purified E. coli 30S ribosomal subunits (1.5 M, kindly provided by the Kurt Fredrick laboratory) and E. coli tRNA fMet (0.9 M, tRNA Probes) were added, and the complexes were incubated at 37°C for 30 min. Primer extension reactions were carried out as described previously (55). Reactions were quenched by addition of gel-loading buffer (Ambion) and resolved using 6% denaturing PAGE. A DNA sequencing ladder was generated by using a DNA Sequenase 2.0 Kit (USB), a DNA template corresponding to positions 7-302 of the D. indicum metI RNA, and the same downstream primer as used in the primer extension assay. Products were visualized using PhosphorImager analysis (Molecular Dynamics) and quantified using ImageQuant 5.2 software.

Measurement of RNA-SAM complex stability in vitro
The size-exclusion filtration binding assay was modified to determine the stabilities of the WT and mutant transcriptional metI RNA-SAM complexes. Aptamer RNA (1 M) was denatured, slowly cooled, and incubated with [ 3 H]SAM (2 M) at 37°C for 30 min. Next, unlabeled SAM (200 M) was added as a competitor ligand. Samples were removed at intervals and were passed through BSA-treated Nanosep 10K filters. The amount of [ 3 H]SAM retained by the filter was counted in a liquid scintillation counter. Complex half-life was determined by nonlinear regression analysis using GraphPad Prism 8 software.
The stabilities of the WT and mutant translational metI RNA-SAM complexes were measured using stopped-flow spectroscopy with a bipartite RNA as described previously (9). The upstream RNA, corresponding to positions 7-60 of the D. indicum metI leader RNA, contained a 2AP substitution at position A14 (purchased from Dharmacon). The downstream RNA corresponded to positions 61-128 of the D. indicum metI leader RNA and was generated by T7 RNAP transcription. The first base pair of this half of RNA was changed from C-G to G-C to introduce G61 as the initiating nucleotide to facilitate transcription. The RNA segments (2 M each) were cofolded in 1ϫ buffer (20 mM Tris HCl (pH 8.0), 20 mM NaCl, 10 mM MgCl 2 , and 0.1 mM EDTA) by heating the mixture at 65°C for 5 min and slowly cooling to 40°C. SAM (2 M) was added after slowly cooling, and reactions were incubated at 37°C for 30 min to allow the complex to reach equilibrium. A competitor RNA (10 or 20 M) corresponding to positions 14 -151 relative to the transcription start site of the B. subtilis yitJ leader RNA was folded separately using similar folding conditions. Kinetics experiments were performed in an SX20-MV spectrometer Transcriptional versus translational S-box riboswitches (Applied Photophysics) using an excitation wavelength of 310 nm. Equal volumes of preformed metI S-box RNA-SAM complex and competitor yitJ leader RNA were mixed rapidly, and fluorescence emission was monitored after passage through a 360-nm cutoff filter. The raw trace (Fig. S2A) represents an increase in fluorescence over time as the 2AP fluorescence unquenches because of SAM dissociation. The fraction of RNA bound to SAM was set to 1 at time 0, and the data were plotted as fraction bound versus time. The reported rate is the mean of three independent experiments. For each experiment, at least five individual traces were measured. The data from these experiments were averaged and fit to a single-or double-exponential equation using GraphPad Prism 8 software to determine k off . The complex half-lives were determined from the off rate as t1 ⁄ 2 ϭ ln (2)/k off. The k on value was determined from the equation K D ϭ k off /k on . For the mutant translational RNA, the data were collected for 1000 s, and the reported half-life value was derived from this plot.

SHAPE
D. indicum metI aptamer RNA (nucleotides 7-128) and B. subtilis metI RNA (nucleotides 28 -150) were introduced into the SHAPE cassette (40). The RNAs (0.4 M) were folded by heating and snap-cooling in a modified SHAPE buffer (100 mM HEPES (pH 8.0), 10 mM MgCl 2 , and 20 mM NaCl). The RNAs were modified with N-methylisatoic anhydride (NMIA; 3.25 mM, Invitrogen) and precipitated as described previously (40). Primer extension inhibition and RNA-Seq reactions were carried out in parallel using a fluorescently labeled primer (5Ј-GAACCGGACCGAAGCCCG-3Ј; Applied Biosystems, Ther-moFisher Scientific) complementary to the SHAPE cassette. RNAs were denatured before annealing to the primer, and extension reactions were carried out for 20 min at 52°C, followed by 5 min at 65°C. Capillary electrophoresis was carried out at the Ohio State University Comprehensive Cancer Center Genomic Shared Resource facility. The data were processed using QuShape software (56). SHAPE reactivity was derived from the difference between the frequencies of RT stops at each nucleotide in the presence and absence of NMIA. The data for each nucleotide were averaged after removing statistical outliers and normalized using the 2%-8% rule (57).

Data availability
All data described are contained within the article.