Structural basis for control of bacterial RNA polymerase pausing by a riboswitch and its ligand

Folding of nascent transcripts can be modulated by the RNA polymerase (RNAP) that carries out their transcription, and vice versa. A pause of RNAP during transcription of a preQ1 riboswitch (termed que-PEC) is stabilized by a previously characterized template consensus sequence and the ligand-free conformation of the nascent RNA. Ligand binding to the riboswitch induces RNAP pause release and downstream transcription termination; however, the mechanism by which riboswitch folding modulates pausing is unclear. Here, we report single-particle cryo-electron microscopy reconstructions of que-PEC in ligand-free and ligand-bound states. In the absence of preQ1, the RNA transcript is in an unexpected hyper-translocated state, preventing downstream nucleotide incorporation. Strikingly, on ligand binding, the riboswitch rotates around its helical axis, expanding the surrounding RNAP exit channel and repositioning the transcript for elongation. Our study reveals the tight coupling by which nascent RNA structures and their ligands can functionally regulate the macromolecular transcription machinery. Using a combination of biochemical, structural and computational analyses, the authors unveil the mechanism of bacterial RNA polymerase pausing and restart mediated by a small-molecule-dependent RNA structural element termed a riboswitch.

Folding of nascent transcripts can be modulated by the RNA polymerase (RNAP) that carries out their transcription, and vice versa. A pause of RNAP during transcription of a preQ 1 riboswitch (termed que-PEC) is stabilized by a previously characterized template consensus sequence and the ligand-free conformation of the nascent RNA. Ligand binding to the riboswitch induces RNAP pause release and downstream transcription termination; however, the mechanism by which riboswitch folding modulates pausing is unclear. Here, we report single-particle cryo-electron microscopy reconstructions of que-PEC in ligand-free and ligand-bound states. In the absence of preQ 1 , the RNA transcript is in an unexpected hyper-translocated state, preventing downstream nucleotide incorporation. Strikingly, on ligand binding, the riboswitch rotates around its helical axis, expanding the surrounding RNAP exit channel and repositioning the transcript for elongation. Our study reveals the tight coupling by which nascent RNA structures and their ligands can functionally regulate the macromolecular transcription machinery.
In bacteria, genetic information is transcribed by the five-subunit (α 1 α 2 ββ′ω) protein RNA polymerase (RNAP) that forms a universal core containing all transcription functions 1 .
During transcription elongation, the incorporation of nucleotides into the RNA chain is coupled to the translocation of RNAP 2 . However, elongation is often temporarily interrupted due to an off-path state of the enzyme, which competes with the addition of the subsequent nucleotide. This so-called transcriptional pausing is universally involved in numerous biological processes, including RNA folding 3 , transcriptiontranslation coupling in bacteria 4 , transcription factor recruitment 5 , messenger RNA processing 6 and transcription termination 7 . Transcriptional pausing is triggered when RNAP encounters a consensus DNA sequence (G −11 G −10 Y −1 G +1 in Escherichia coli and G −11 G −10 Y −1 A +1 in Bacillus subtilis) 8 , causing the enzyme to enter into an elemental paused state 9 .
The elemental paused elongation complex (ePEC) has been found to be further stabilized through at least three distinct mechanisms. Class I pauses are stabilized by the presence of an RNA hairpin within the RNAP exit channel, whereas class II pauses involve RNAP reverse translocation (or backtracking) along the DNA template 10 . In addition, these two classes are sensitive to transcription factors that will further modulate pausing efficiency, such as N-utilization substance A (NusA) protein for class I (ref. 11) and GreB for class II pauses 12 . In contrast, while the consensus DNA sequence is essential at RNA structure-stabilized pauses, the latter were shown to be insensitive to NusA and GreB, instead responding to a conformational change of a nascent RNA motif termed a riboswitch (Fig. 1a) (ref. 13).
Riboswitches are structural RNA elements embedded in the 5′ untranslated regions (5′ UTR) of mRNAs that are predicted to regulate Article https://doi.org/10.1038/s41594-023-01002-x expression level through the modulation of transcription termination or translation initiation 16 . In Bacillus subtilis (Bsu), the riboswitch localized upstream to the queCDEF operon binds the transfer RNA nucleotide precursor 7-methylamino-7-deazaguanine (preQ 1 ) as a ligand and represents one of the smallest riboswitches identified so the expression of up to 4% of genes in certain bacteria 14 , making them an attractive target for the design of new antibiotics 15 . A typical riboswitch is composed of two interconnected domains: an aptamer that binds a specific ligand, followed by an expression platform that undergoes conformational changes upon ligand binding, altering the genetic Upon encountering a consensus pause sequence, RNAP enters an offline ePEC that can be either stabilized (undocked state) or released through a riboswitch conformational change (docked state) upon preQ 1 ligand binding. Green star represents potential contact between the RNA and RNAP β-flap domain. b, Nucleic acid scaffold used for cryo-EM data acquisition. Sequence and secondary structure of the Bsu preQ 1 riboswitch used in this study are shown. c, Lifetimes of the que pause determined in the absence and presence of preQ 1 . que-PEC was assembled under conditions similar to those used to assemble the sample for cryo-EM data collection. The rate of pause escape was determined after addition of the next templated rNTP (rGTP). Samples were taken 15, 30, 45, 60, 90, 120, 240 and 480 s after the addition of 5 µM rGTP (band labeled as +2). Ch lanes are chased reactions collected after addition of 100 µM rGTP for an additional 5 min. Error bars are s.d. (standard deviation) of the mean from independent replicates (n = 2). d, Overall fold and cryo-EM density of the que-PEC obtained in the absence of preQ 1 . The 3.3-Å resolution cryo-EM map is rendered as a transparent surface and the refined model of the que-PEC is colored as labeled. The RNAP backbone is represented as a ribbon diagram. The RNA transcript is colored gold, template DNA is colored black and nontemplate DNA is colored dark red. e, Overall fold and cryo-EM density of the que-PEC obtained in the presence of preQ 1 . The 3.8-Å resolution cryo-EM map is rendered as a transparent surface and the refined model of the que-PEC is colored as in d.

Article
https://doi.org/10.1038/s41594-023-01002-x far 17 . This Bsu preQ 1 riboswitch operates at the transcriptional level, in which ligand binding stabilizes an H-type pseudoknot structure (the 'docked' conformation), which favors the formation of a terminator hairpin in the expression platform for premature termination (Supplementary Fig. 1a), thus decreasing the expression level of genes involved in queuosine biosynthesis 17 .
In previous work, we identified a specific cross-talk between riboswitch folding and RNAP processivity that is modulated by preQ 1 binding as a characteristic feature of the que pause 13 . RNAP pausing in the expression platform (the que-PEC) is stabilized in the absence of ligand through interactions between the partially folded pseudoknot and the RNAP exit channel (Fig. 1a). Ligand binding then disrupts these interactions to promote RNAP release from the paused state 13 . Although single-molecule assays probing riboswitch dynamics revealed that the presence of RNAP at the que pause substantially alters the ligand-dependent RNA conformational change, we lack a high-resolution mechanistic understanding of this pause regulation mediated by coupling between nascent RNA and RNAP in cis. We used single-particle cryo-electron microscopy (cryo-EM) to determine the structures of multiple conformations of ligand-free que-PEC with a consensus resolution of 3.3 Å, together with the structures of the corresponding preQ 1 -bound que-PEC with a consensus resolution of 3.8 Å. A comparison of the structures, in combination with biochemical assays and mutational analysis, demonstrates that the ligand-free riboswitch in the exit channel induces retraction of the RNA strand from the active site into a hyper-translocated state, explaining the inhibition of catalysis. Upon preQ 1 binding, the riboswitch twists, expanding the exit channel, inducing a counter-rotation of the RNAP swivel module and jutting the RNA 3′ end back into the active site to release pausing. The que-PEC structures thus invoke a mechanism for the RNA-based control of transcriptional pausing and subsequent pause release that may apply to numerous other bacterial transcription elongation complexes (ECs).

The in vitro-assembled que-PEC is functional
Previously, we found that the E. coli and B. subtilis RNAPs respond to the que pause similarly 13 , motivating us to use the better characterized E. coli RNAP to structurally study the que-PEC. The que-PEC was formed using nucleic scaffolds comprising DNA and RNA oligonucleotides mimicking a transcription bubble (Fig. 1b). Pause escape was measured upon rGTP addition to extend the RNA by two nucleotides (Fig. 1b,c). In the presence of preQ 1 , pause escape increased by twofold, consistent with our previous observation of pause release upon ligand binding to the riboswitch during promoter-initiated transcription 13 .
For structural studies, we similarly assembled the que-PEC and purified it using size-exclusion chromatography. Addition of 8 mM CHAPSO into the cryo-EM buffer allowed the particles to adopt random orientations in the vitrified ice, and, consistent with previous observations 18 , the detergent did not interfere with preQ 1 -mediated pause release ( Supplementary Fig. 1b,c). Furthermore, ligand-mediated pause release is also detected during promoter-initiated transcription of the same RNA construct ( Supplementary Fig. 1d,e). We conclude that the reconstituted que-PEC affects reversible RNAP pausing, making it suitable for structural mechanistic studies via single-particle cryo-EM.

Structure determination shows similar protein conformations
For each que-PEC (minus and plus ligand), a single consensus three-dimensional (3D) reconstruction (Table 1) was determined and showed density associated with the upstream and downstream DNA. Each reconstruction shows details of the active site, and the RNA-DNA hybrid and the riboswitch transcript within the RNAP exit channel are also clearly visible in the density maps (Fig. 1d,e and Supplementary Figs. 2 and 3). For the ligand-free que-PEC, 3D classification of the unmasked complexes led to two 3D classes; however, the conformations of these 3D classes were indistinguishable when overlayed and were therefore combined back into a single consensus structure that was refined to a nominal resolution of 3.3 Å (Fig. 1d, Supplementary Fig. 2 and Table 1). For the preQ 1 -bound data, traditional 3D classification techniques yielded a single RNAP class, along with a second class that is associated with grid particulates and ice ( Supplementary Fig. 3e). Particles in the preQ 1 -bound dataset were submitted to nonuniform refinement as a single consensus volume and reached an overall resolution of 3.8 Å (Fig. 1e, Supplementary Fig. 3 and Table 1).
Local resolution calculations indicated that the RNAP structures have resolutions spanning 3.1-12.0 Å and 3.3-10.0 Å in the absence and presence of preQ 1 , respectively (Supplementary Figs. 2f and 3f). The core RNAP structure, which includes the RNA-DNA hybrid, and the active site are the most stable regions with the highest resolutions, while the RNA exit channel and the peripheral regions nearing the solvent interface are not as well resolved.
To build the model for the initial que-PEC consensus structure without preQ 1 , we initially placed E. coli RNAP in its elongation complex (PDB 6ALF) 19 into the EM density map. Some regions, including the hinge regions between helices and the RNA emerging from the RNAP exit channel, needed to be rebuilt by hand. The que-PEC consensus is similar to the post-elongation complex structure, with a root mean squared deviation (r.m.s.d.) of 1.6 Å; however, this value represents a conformational average of the que-PEC that we will analyze further in the following sections. Using the post-elongation complex also allowed us to mark the position of the active site (Mg 2+ ) that sits between the 3′ RNA nucleotide and the catalytic triad of aspartic acid residues in the β′ subunit. To build the initial model for the preQ 1 -bound consensus structure, coordinates for the hisPEC RNAP (PDB 6ASX) 18 were placed into the refined consensus map. The major differences between the two structures are at the primary channel, mainly the β-SI3 domain, and at the RNA exit channel, where the β′-zinc-binding domain (ZBD) of the preQ 1 -bound structure is shifted ~2 Å from the exit channel. Otherwise, the two structures are similar. with an overall r.m.s.d. of 2.1 Å.

3D variability analysis reveals continuous heterogeneity within the que-PECs
Even reaching sub-4-Å resolutions for the que-PEC, both in the absence and presence of preQ 1 , there were no obvious distinct conformations found within the datasets using conventional 3D classification approaches, which often work well for separating discrete conformations of a structure, but fail when conformational heterogeneity leads to continuous motion. We wondered whether the apparent similarities between the que-PEC structures could be due to a continuity motion of RNAP conformers rather than adopting discrete conformations, as observed previously during transcription-translation coupling 20 . To test this hypothesis, we submitted particles refined by both two-dimensional (2D) classification and ab initio 3D reconstruction to 3D variability analysis (3DVA) (ref. 21) (Supplementary Fig. 4). For this analysis, 20 structural intermediates were calculated.
To resolve the conformational heterogeneity for both datasets, 142,410 particles from the ligand-free que-PEC consensus refinement (Supplementary Fig. 4a and Supplementary Video 1) and 51,824 particles from the ligand-bound consensus refinement (Supplementary Fig. 4b and Supplementary Video 2) were submitted to 3DVA with a filter resolution of 5 Å and 7 Å, respectively. For the −preQ 1 dataset, three eigenvector modes were indicated, each of which consisted of a series of 20 volumes that were used for making movies in ChimeraX. From the volume series, the first and last frames (frames 0 and 19) represent the negative and positive values along the reaction coordinate for each variability component.
The first variability component was seen in the β′-SI3, β-lobe, β-flap helices, exit channel domains and the upstream DNA helix. For this component, a scissoring movement between the β′-SI3 and the β-lobe can also be observed (Supplementary Video 1). Additionally, the first component shows movement of the exit channel domains, primarily  Similarly, to resolve the continuous heterogeneity in the ligand-bound que-PEC, the consensus refined particles were submitted to the 3DVA clustering routine. Particles from each component from 3DVA and their associated volumes were submitted to nonuniform refinement and refined separately. Due to the lower number of particles for this dataset, a filter resolution of 7 Å was necessary while defining two principal components to resolve the conformational heterogeneity. In the presence of preQ 1 , two 3DVA components both refined to 3.8 Å overall (Supplementary Fig. 4b and Supplementary Table 2). For component 0, which shows the most variability, the major region of conformational heterogeneity is seen at the β-SI3 region at the primary channel, whereas for component 1 the variability lies at the helices of the β-flap domain and the upstream DNA helix.

Distal ligand binding modulates the RNAP active site
We compared our que-PEC structures to the rigid core module of previously obtained EC structures 18,19 . In the absence of preQ 1 , the RNA-DNA hybrid is in an unconventional translocation state that most closely resembles the posttranslocated E. coli RNAP elongation complex 19 . Notably, the RNA 3′ end is shifted ~1.6 Å upstream from the i (or product) site into the i − 1 site relative to this posttranslocated EC (Fig. 2a). The template DNA (tDNA) still resides in a fully posttranslocated state so that tDNA residue T16 remains in the i site while still pairing with A47 of the RNA 3′ end. The next tDNA nucleotide, C15, is no longer paired with the nontemplate DNA (ntDNA), while occupying the rNTP binding i + 1 site, but is too far from the RNA 3′ end to allow it to template the next incoming rNTP (Fig. 2b) (ref. 22). To test this model, we used exonuclease III (Exo III) footprinting 23 and found that the tDNA in the que-PEC indeed primarily resides in a posttranslocated register ( Supplementary Fig. 5). This active site conformation is consistent with the functional disruption of the nucleotide addition cycle and provides further structural rationalization for the inhibition of catalysis in the absence of preQ 1 . In further support, a 'slow RNAP mutant' that impairs nucleotide substrate stabilization (S1105A mutant) 22 shows a significant increase in pause half-life only in the absence of ligand ( Supplementary Fig. 6a,b).
Strikingly, in the presence of ligand, the RNA 3′ end moves closer to the RNAP catalytic center and now occupies the i site in a posttranslocated register while remaining base paired with the tDNA residue T16 (Fig. 2a,b).
In the absence of ligand, the geometry of the translocation state suggests a weak RNA-DNA hybrid that could lead to transcription termination 24 . To further probe the observed translocation switch upon ligand binding to the riboswitch, we performed in vitro transcription of DNA templates in which the RNA-DNA hybrid in the context of the que pause was altered by replacing key nucleotides with rU:dA base pairs to decrease its stability and favor termination of transcription (Fig. 2c). As predicted, efficient termination was observed when the RNA-DNA hybrid is weak, with seven rUs out of the nine total residues (variant 7U). As the stability of the RNA-DNA hybrid is strengthened by serially changing back these excess rU:dA base pairs, a significant proportion of readthrough product was detected in the presence of preQ 1 , with the most efficient ligand-mediated anti-termination observed with the 4U variant (Fig. 2c,d and Supplementary Fig. 6c-e). These observations further support that the ligand induces a reverse RNA translocation, which stabilizes the RNA-DNA contacts with RNAP and prevents termination when the RNA-DNA hybrid is additionally weakened.
Together, our structural and functional observations support a mechanism where binding of preQ 1 triggers a series of riboswitch conformational changes relative to paused RNAP that modulate the translocation register and realign the RNA 3′ end with the active site to release the paused RNAP.

Global conformational changes within the que-PECs
At the elemental pause sequence, a group of RNAP structural modules, including the clamp, dock, shelf, SI3 and C-terminal segment of the β′ subunit, has been found to exhibit a rotation (or swiveling) roughly about an axis perpendicular to the plane defined by the helical axes of the RNA-DNA hybrid and the downstream DNA 18 . The state of this 'swivel module' constitutes a structural feature defining a paused state of the RNAP 12,18,25 .
Overlaying the ligand-free que-PEC structures from the intermediate spaces between the 3DVA components, during the continuous motion outlined by the 3DVA, the final frames of the videos show that RNAP is in the most swiveled state while in the absence of ligand. The total swiveling motion covers a range of ~3˚ relative to the initial frames about an axis centered near a hinge previously described in the his-PEC 11,18 (Fig. 3a, Supplementary Table 2 and Supplementary Videos 1 and 2). As the 3DVA components are ranked in order of the most variability, the first 3DVA component in the absence of preQ 1 (component 0) undergoes a notable amount of swiveling and clamp opening movement. Also, accompanying the clamp movement is the shifting of the exit channel domains as the RNA exit cleft is going back from an open to a semiclosed state that bars the nascent riboswitch from emerging (Supplementary Video 2). The second 3DVA component (component 1) is unswiveled and shows motion mostly in the β-SI3 region and that of the upstream DNA helix ( Supplementary Fig. 4). Finally, the third 3DVA component (component 2) is in a similar conformation to that of component 1, but shows considerable variability at the RNA exit channel, where the riboswitch is found growing from the channel (Supplementary Fig. 4a and Supplementary Video 2). Simultaneously, the exit channel domains, β′-ZBD and the β′-dock are shifted away from the exit channel and toward the upstream DNA helix. In summary, the que-PEC has large conformational freedom in the absence of ligand, which is transduced to the RNAP swivel module motion stabilizing transcriptional pausing.
In the presence of preQ 1 , the RNAP is in the unswiveled state, and adopts a similar conformation as in the active elongation complexes with a closed clamp (Fig. 3a) (refs. 18,19,25). As a result, the exit channel is in a closed conformation allowing more contacts with the nascent transcript. Strikingly, riboswitch folding within the RNAP exit channel seems to profoundly impact the global RNAP conformation, a feature previously observed within other PECs in the presence of transcription factors such as NusA and NusG 11,25,26 . To further test this observation, we performed in vitro time-pausing assays of the que pause in the presence of NusA or NusG transcription factors. Interestingly, we observed that NusA increases the que pause efficiency only in the presence of preQ 1 (Fig. 3b,c), while NusG decreases the pause half-life only in the absence of ligand (Fig. 3d,e). These results suggest that the RNAP is in a different conformation depending on the riboswitch docking state, and support an active role of the nascent RNA structure in modulating RNAP pausing.

Ligand binding induces riboswitch rotation
From the consensus 3D refinements, densities associated with the P2 helix of the riboswitch were identifiable in the vicinity of the RNA exit channel as a double-helical stem structure in both the absence and presence of preQ 1 (Fig. 1d). However, since it is in a solvent-accessible region and conformationally flexible, it was not possible to build a detailed de novo model of the entire transcript. To this end, we took advantage of a previously obtained NMR structure of the same preQ 1 riboswitch 27 and performed molecular dynamic flexible fitting (MDFF) simulations on the electron density associated with the riboswitch emerging from the RNAP exit channel. From 100 MDFF trajectories generated from each structure, the 20 best-fit models were selected on the basis of their global cross-correlation coefficients (Supplementary Fig. 7a and Supplementary Videos 3 and 4). Analysis of these models revealed that, upon preQ 1 binding, the riboswitch aptamer undergoes a notable twisting motion around the P1 helical axis within the RNAP exit channel (Fig. 4). Specifically, ligand binding induces an ~42° rotation of the aptamer toward the β′-ZBD (Fig. 4a and Supplementary Videos 5 and 6), leading to an ~1.4-Å shift of the RNA 3′ end in the downstream Article https://doi.org/10.1038/s41594-023-01002-x direction of the active site (Fig. 4b). Since we could not fully resolve the β-flap tip because of its well-documented flexibility 28 , we asked whether its inclusion in the MDFF simulations would affect the observed RNA twisting motion. Additional models generated with the β-flap tip included behave similarly (Supplementary Figs. 8 and 9), suggesting that the preQ 1 -induced aptamer rotation is a robust feature.
Overall, MDFF unveils an unexpected, ligand-induced twisting and insertion motion of the preQ 1 riboswitch relative to RNAP, which has not been visualized in previous structural analyses performed with the isolated RNA 27,29 .

The RNA exit channel responds to riboswitch rearrangement
As the newly synthesized transcript emerges from the RNAP main cleft, the first five nucleotides past the RNA-DNA hybrid (RNA residues C33, U34, A35, A36 and G37) reside in the RNAP exit channel and can potentially form duplex structures and modulate transcriptional pausing efficiency through RNA-protein interactions 30 . Additionally, the presence of a preformed positively charged surface in the RNAP exit channel complementary to an A form helix has been suggested to constitute a path for the nascent RNA that guides the formation of RNA duplexes (Fig. 5a)    we hypothesized that folding of the que-PEC may directly affect the conformation of this subdomain.
In the absence of preQ 1 , the transcript is sterically hindered to form a stable RNA structure due to clashes between the β-flap tip and the riboswitch P2 loop (RNA residues C14, U15, A16, C17, A18 and C19) and the 3′ end of the A stretch (RNA residues A27 to A32). Inside the RNAP exit channel the nucleotides C10, U11, A12 and G13 align with a path of positively charged residues involving arginine and lysine residues in the β′-ZBD, specifically residue β′-R77, which could establish multiple hydrogen bonds with the nucleotide A16 ( Fig. 5b and Supplementary Table 3). This structural observation provides molecular context for why mutation of positive residues in the β′-ZBD to alanine (R77A, K79A and R81A) significantly reduces the efficiency of the que pause in the absence of ligand and supports the functional relevance of this interaction 13 .
Upon ligand binding, due to the twist of the aptamer in the exit channel, the riboswitch shifts closer to the β′-ZBD domain and is in a position where it probably forms a hydrogen bond between β′-K79 and Article https://doi.org/10.1038/s41594-023-01002-x G4 (Fig. 5b,d and Supplementary Table 3). Moreover, K395 within the β′-dock domain could now also form hydrogen bonds with C19 and C33 nucleotides. An additional contact could be formed between β′-K398, which is also conserved in eukaryotic RNAP II (ref. 1), and nucleotide C10 ( Fig. 5b and Supplementary Table 3), suggesting that ligand binding not only stabilizes the pseudoknot, but also increases the extent of RNA-protein interactions, accompanying the RNA twisting motion. Conversely, analysis of the β-flap tip, β′-dock and ZBD reveals that the RNAP undergoes substantial movements as a function of the riboswitch docking state. In particular, the preQ 1 -induced twisting pushes the β-flap, β′-dock and ZBD away from the riboswitch, effectively opening the RNAP exit channel to accommodate the nascent transcript (Figs. 4c and 5c,d and Supplementary Video 2). Thus, ligand binding and the resulting folding of the riboswitch have direct effects on the adjacent RNAP subdomains, providing a structural explanation for the inhibition of transcriptional pausing in the presence of preQ 1 .

Discussion
Understanding both the fundamental mechanism of transcriptional pausing and the dynamic interplay between nascent RNA folding and gene expression is essential due to their profound regulatory roles in bacteria 31 . In particular, the directional 5′ to 3′ transcription of RNA is often paused to temporally and spatially program the cotranscriptional, hierarchical folding of RNA structures within or near the RNA exit channel 32 . We report high-resolution cryo-EM structures of elongation complexes at the riboswitch-mediated que pause site (que-PEC) in the absence and presence of cognate ligand at 3.3 Å and 3.8 Å global resolutions, respectively, revealing how RNA and RNAP affect each other's conformation in a ligand-dependent mechanism (Fig. 6). Binding of the 252-Da preQ 1 ligand to the aptamer domain triggers a cascade reaction, initiated within the RNAP exit channel as it swings open (Fig. 4c), which propagates downstream to the active site of the ~400-kDa RNAP (Fig. 2a) to ultimately release the enzyme from the paused state (Fig. 6). With the riboswitch embedded within the RNAP exit channel, MDFF analysis reveals an intriguing RNA folding pathway initiated by ligand binding that allosterically alters the global RNAP conformation (Figs. 3 and 4). Our structural and biochemical studies rationalize how RNA halts catalysis in the absence of preQ 1 , and how ligand-induced riboswitch folding reactivates RNAP from its paused state, thereby providing direct mechanistic insights into RNA structure-stabilized transcriptional pausing regulation.
The RNAP exit channel is known to act as the gateway for regulatory processes that occur cotranscriptionally as a result of interactions with the translational machinery during transcription-translation coupling 4,33 , transcription factors 11,34,35 and folding of RNA structures within it 30 . In addition, leveraging its positive charges that align with the phosphates of A form RNA, the RNAP exit channel forms a route that guides RNA duplex formation, acting as a basis for a regulatory connection with the downstream active site 18 .
In the absence of ligand, the 5′ segment of the P2 helix (specifically RNA residue C17), is found in close proximity to the β-flap tip and threads toward the β′-ZBD (Fig. 5), as predicted by molecular dynamic simulation 13 . Accordingly, both deletion of the β-flap tip (residues 890-914) and point mutations in the RNA-binding region of the β′-ZBD (residues R77, K79 and R81) decrease the que pause efficiency substantially, supporting the functional relevance of this interaction 13 . The 3′ segment of P2 (in particular RNA residue U34) resides on the proximal face of the β′-dock (contacting residue K395; Fig. 5), in agreement with previous studies showing the equivalent interactions between the 5′ segment of the his-paused RNA hairpin and RNAP 18 . Therefore, following translocation at the elemental pause site, the ligand-free aptamer pseudoknot appears to constitute a physical barrier preventing RNA extrusion from the exit channel, thereby holding RNAP in the swiveled conformation (Fig. 3a). Conversely, this particular feature may also contribute to the fast response of the que-PEC to ligand binding, since Article https://doi.org/10.1038/s41594-023-01002-x the pseudoknot structure is held in place by RNAP for dynamic sensing of preQ 1 (refs.13,36). Following ligand binding, the riboswitch rotates ~42° along its P1 helical axis (Fig. 4) and alters the que-PEC at multiple architectural levels, leading to the release of transcriptional pausing. First, preQ 1 -induced stabilization of the P2 helix remodels RNAP, causing its RNA exit channel subdomains to shift away, effectively 'opening' the path and providing clearance for the nascent transcript (Figs. 4 and 5). To this end, the β-flap moves away, expanding the channel's inner diameter, as has also been seen to allow for the accommodation of a more simple regulatory RNA hairpin 28 . In addition, the β′-dock and ZBD shift, counteracting with the swivel module rotation, as proposed for an active elongation complex 9,19 . In the que-PEC, this remodeling is expected to expose the RNAP binding surface for NusA domain S1 ( Supplementary Fig. 10) (ref.11). In support of these observations, we find that the que-PEC is sensitive to NusA only in the presence of preQ 1 (Fig. 3b,c).
Next, extensive contacts and physical proximity of the riboswitch L2 loop to the β′-dock and ZBD rationalize the impact that riboswitch folding has on the global RNAP conformation. In the absence of preQ 1 , as the 3′ segment of P2 is abutted by the β′-dock, stabilization of the preQ 1 -folded pseudoknot, coupled with riboswitch twisting, promotes pause release by pushing the swivel module toward the elongation-active nonswiveled conformation to operate through an induce-fit mechanism (Fig. 6) (ref.29). Interestingly, the que pause is decreased in the presence of NusG factor only in the absence of preQ 1 , and to the same extent as in the ligand-bound condition (Fig. 3d,e), supporting the notion of an anti-swiveling motion triggered by ligand binding and docking of the riboswitch 25,26 . While this renders the que-PEC structure in the docked riboswitch state most similar to the canonical PEC at a class I pause 18 , this anti-swiveling motion can potentially occur concomitantly with nucleotide addition and therefore constitutes a structural intermediate during transcription reactivation. In support of these observations, the β′-ZBD domain, to which most of the RNA contacts are detected, has previously been implicated in stabilizing the EC 37 and in modulating transcription termination 38 .
Several mechanisms have been proposed for how RNAPs translocate along the DNA template during transcription elongation 2 . An obstacle to this movement is presented by structural intermediates that are transient and difficult to capture because of their inherent dynamics. Such intermediates have been suggested to also occur in yeast RNAP II bound to α-amanitin and in viral RNAP 39,40 and were recently observed in paused bacterial RNAP complexes. For example, an asymmetric movement of the RNA-DNA hybrid (half-translocation) explains the stabilization of RNAP pausing at hairpin-stabilized pauses 11,18 . Strikingly, even if the template DNA is posttranslocated in the absence of ligand, the que-PEC RNA is retracted from the active site, a geometry preventing subsequent nucleotide incorporation (Fig. 6). Importantly, a weaker RNA-DNA hybrid converts the que-PEC from a paused complex to a complex prone to termination (Fig. 2c,d and Supplementary  Fig. 6) suggesting that this retracted intermediate is also present at the early steps of intrinsic or protein-mediated transcription termination 41 . During intrinsic termination, the asymmetric movement of the RNA would directly follow terminator hairpin nucleation in the hybrid-shearing mechanism 41 . However, in the ligand-bound que-PEC, the downstream melting of the hybrid would be disfavored due to  Supplementary Table 3). d, preQ 1 binding and subsequent steric hindrance shift the riboswitch closer to the β′-ZBD. Shaded oval indicates the region of proximity between the riboswitch and RNAP domains and key nucleotides are indicated in red (see also Supplementary Table 3).
Article https://doi.org/10.1038/s41594-023-01002-x the lack of rU:dA base pairs in the hybrid, leading to pause escape in the wild-type construct and transcription readthrough in the Mut4U variant (Fig. 2c,d).
To date, only RNA hairpin folding within the RNAP exit channel has been found to allosterically alter RNAP structure in the context of transcriptional pausing and termination mediated by the RNA transcript 32,41 . For example, folding of an RNA hairpin within the RNAP exit channel could favor forward translocation of the enzyme by pulling out the nascent transcript in the RNA-DNA hybrid, which has been proposed to help rescue backtracked RNAPs 32 . Other transcription reactivation mechanisms involve the action of additional proteins, such as the bacteriophage Q 42 or λN protein 38 . The ligand-mediated regulation of the que pause operates through a distinct mechanism in which riboswitch folding in the docked state induces a reverse RNA translocation, pushing the RNA 3′ end closer to the RNAP active site (Fig. 2a,b). A similar transcription reactivation mechanism has been reported for the DNA translocase RapA 43 suggesting that the que-PEC pause release mechanism is applicable to other regulatory mechanisms controlling transcriptional pausing. In addition, an analogous rotation of the P1 helix in the S-adenosyl methionine-sensing riboswitch has been found previously 44 , suggesting that 'RNA twisting' mechanisms could more generally be employed by structured RNAs to execute their regulatory functions.
The que-PEC is still competent for ligand binding to the aptamer, even though the que pause is situated in the expression platform (gene regulatory domain), therefore, it may constitute a transcriptional checkpoint for late ligand-binding events 36 . Because the riboswitch is very small and will be transcribed in vivo in a short amount of time, pause stabilization in the absence of preQ 1 will allow more time for the aptamer to recognize and bind low concentrations of the ligand, preventing transcription readthrough more efficiently 45 . This regulatory mechanism would also allow for the recruitment of regulatory proteins, such as NusA and NusG, as a function of the riboswitch state (Fig. 3) to fine-tune the downstream gene regulatory response. For example, NusA is known to regulate transcription termination at suboptimal terminators in Bsu 46 , thus rationalizing the effect we observe for NusA on the que pause only in the presence of preQ 1 to further enhance downstream transcription termination 47 . Overall, the high-resolution structures and biochemical evidence obtained here reveal the adaptability of bacterial RNAP to funnel a variety of distinct molecular RNAP can convert to a PEC once encountering a consensus pause sequence that is stabilized by riboswitch folding. Binding of preQ 1 ligand induces pseudoknot stabilization to release RNAP from the paused state. The docked state (induced by ligand binding) leads to riboswitch rotation within the RNAP exit channel, ultimately leading to RNA exit channel expansion to accommodate the nascent transcript. Active site schematics are shown as oval insets.
Article https://doi.org/10.1038/s41594-023-01002-x inputs into achieving a desired gene regulatory output, and promise to guide the development of novel antibacterial therapies against the transcription machinery.

Online content
Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41594-023-01002-x.

Oligonucleotides used in this study
DNA oligonucleotides were obtained from IDT and the RNA construct used for paused elongation complex formation was obtained from GE Dharmacon (Horizon Discovery). A complete list of oligonucleotides used in this study is provided in Supplementary Table 1.

RNAP expression and purification
E. coli core RNAP bearing an AviTag biotinylation tag on the C terminus of the β′ subunit and S1105A RNAP mutant were prepared as described previously 30 . Briefly, wild-type and mutant RNAPs were purified from E. coli strain BLR λDE3 transformed with the appropriate plasmids. Cells were grown in 6 l Luria medium supplemented with kanamycin (25 µg ml −1 ) or ampicillin (100 µg ml −1 ) on a platform shaker (250 r.p.m. at 37 °C) to an apparent optical density (OD 600 ) of 0.5-0.6. Protein expression was induced by addition of IPTG to a final concentration of 1 mM. The induced cells were then incubated at 37 °C for 3-4 h at 200 r.p.m. and collected by centrifugation (3,440g, 15 min, 4 °C). The cell pellet was resuspended in 25 ml lysis buffer (50 mM Tris-Cl, pH 8.0, 2 mM EDTA, 5% v/v glycerol, 1 mM β-mercaptoethanol, 1 mM DTT, 300 mM NaCl, 0.25 ml of 10 mg PMSF per ml in ethanol and 0.5 ml of a protease inhibitor cocktail containing 31.2 mg benzamide, 0.5 mg chymostatin, 0.5 mg leupeptin, 0.1 mg pepstatin, 1 mg aprotonin and 1 mg antipain per ml in ethanol). The resuspended cells were lysed by sonication. Subsequent purification steps were carried out at 4 °C unless otherwise indicated. Crude RNAPs were enriched by polyethylenimine precipitation. Polyethylenimine (average molecular weight 60 K; Acros Organics, catalog no. 17857) was added to 0.6% final with gentle stirring and the precipitate was recovered by centrifugation (20,000g, 15 min, 4 °C). The polyethylenimine pellets were washed by gentle resuspension in 25 ml TGEDZ buffer (10 mM Tris-Cl pH 8, 0.1 mM EDTA, 5 µM ZnCl 2 , 1 mM DTT, 5% glycerol) plus 500 mM NaCl followed by centrifugation (20,000g, 15 min, 4 °C). RNAPs were eluted by gentle resuspension of the pellets in 25 ml TGEDZ buffer plus 1 M NaCl followed by centrifugation (20,000g, 15 min, 4 °C). RNAP was precipitated from the supernatant at 4 °C by slow addition of solid ammonium sulfate with gentle stirring to 37% w/v final, allowed to stand at 4 °C overnight and then recovered by centrifugation (20,000g, 15 min, 4 °C). The precipitated RNAP was redissolved in 20 ml of buffer A (50 mM NaH 2 PO 4 , 300 mM NaCl, 1 mM DTT, 5% glycerol, pH 8.0) containing 10 mM imidazole, loaded slowly to a column prepared with 10 ml of Ni 2+ -NTA agarose resin (Qiagen), washed with 50 ml of buffer A containing 20 mM imidazole and then eluted with buffer A containing 250 mM imidazole. Fractions containing RNAP were located by Bradford assay, pooled and dialyzed against 2 l TGEDZ buffer plus 200 mM NaCl for 3-4 h at 4 °C. The dialyzed RNAP was then loaded at 3 ml min −1 onto a heparin-sepharose column (5 ml HiTrap) using an AKTA purifier (GE Healthcare), washed with 25 ml of TGEDZ buffer plus 200 mM NaCl and eluted with TGEDZ buffer plus 500 mM NaCl. Purified RNAPs were dialyzed into storage buffer (20 mM Tris-Cl, pH 8, 250 mM NaCl, 20 µM ZnCl 2 , 1 mM MgCl 2 , 0.1 mM EDTA, 1 mM DTT, 25% glycerol) and kept in small aliquots at −80 °C until use.

Preparation of paused elongation complex for cryo-EM
The que-PEC RNA was transcribed in vitro using T7 RNAP transcription and gel purified before use. Synthetic DNA oligonucleotides were obtained from IDT. The nucleic scaffold (tDNA, ntDNA and RNA) was annealed at a 1:1:1 ratio in a buffer containing 100 mM KCl and 50 mM Tris-HCl pH 7.5 (90 °C for 2 min, 37 °C for 10 min and room temperature for 10 min). RNAP core was mixed to the nucleic scaffold at a molar ratio of 1:3 in assembly buffer (100 mM KCl, 50 mM Tris-HCl pH 7.5 and 1 mM MgCl 2 ) and incubated for 15 min at 37 °C. CHAPSO (Sigma Aldrich, catalog no. 82473-24-3) was added at 8 mM, and the complex was concentrated by centrifugal filtration (Amicon, 100 kDa cutoff column) to 5-6 mg ml −1 RNAP concentration before grid preparation. PreQ 1 ligand, when present, was added to 10 µM final concentration and incubated with the complex for 5 min at room temperature.

Cryo-EM data collection and processing
For both the (−) and (+) preQ 1 samples, 3.5 µl of the complex was applied to a glow-discharged C-flat 400-mesh Au grid (Ted Pella). The sample was vitrified by plunge freezing in a liquid ethane slurry using a Ther-moFisher Vitrobot at 4 °C and 100% humidity. All images for each dataset were collected on a Titan Krios electron microscope (Ther-moFisher) equipped with a K2 Summit direct electron detector (Gatan) operated at 300 keV and a nominal pixel size of 1.01 Å per pixel. Images were acquired using Leginon software 49 . The total exposure time was 8 s and frames were recorded every 0.2 s, giving an accumulated dose of 62 e − /Å 2 using a defocus range of −0.5 µm to −3.5 µm.
Raw movie frames pertaining to each dataset were dose weighted and corrected for beam-induced drift using MotionCor2 (ref. 50). The contrast transfer function (CTF) parameters were determined using CTFFIND4 (ref. 51). All image processing was done in RELION v.3.0 and cryoSPARC 21,52 . Following the determination of the CTF parameters, nontemplate-based particle picking was done in crYOLO 53 . Particles were extracted into 300 px 2 (1.01 Å per pixel) boxes and imported into cryoSPARC for 2D classification. These particles were used to generate an initial 3D reconstruction ab initio, while defining two classes resulted in one RNAP class and another containing nonsample-related particles. Before 3D refinement, per-particle drift correction was carried out. The ab initio 3D volume was refined against the particle data while correcting for the CTF higher order aberrations using the nonuniform (NU) refinement procedure in cryoSPARC 21 . Following NU refinement, the data were further refined using the local refinement procedure while applying a tight mask at the protein-solvent interface. For all datasets, the same procedure was carried out to generate a consensus volume. For the consensus refinements, gold standard global Fourier shell correlation (FSC) calculations gave overall resolutions of 3.65 Å and 3.9 Å in the absence and presence of preQ 1 , respectively. After this, a round of 3D classification produced indistinguishable volumes for both (−) and (+) preQ 1 datasets. Consequently, the particles from the two 3D classes from the (−) preQ 1 dataset were combined and subjected to a final round of non-uniform refinement in cryoSPARC ( Supplementary Fig. 2). The one 3D class from the (+) preQ1 dataset was also used in a final round of non-uniform refinement in cryoSPARC ( Supplementary Fig. 3). The best 3D consensus volumes for each dataset were determined to have global FSC resolutions of 3.3 Å and 3.8 Å in the absence and presence of preQ 1 , respectively.
To better resolve the conformational heterogeneity of the samples, the refined particles from each dataset were submitted to the 3DVA routine in cryoSPARC. For the (−) preQ 1 data, a filter resolution of 5 Å was employed while solving for three principal components. The output particles were submitted to the 3DVA clustering routine. The result was three particle sets representing the three variability clusters (that is, components). Each component was then submitted to nonuniform refinement, followed by local refinement. For the (+) preQ 1 data, the same 3DVA pipeline was followed. However, due to a smaller number of particles, a filter resolution of 10 Å was used and only two 3DVA components were solved for and clustered.

Model building, refinement and validation
The protein coordinates from the post-translocated elongation complex were used as a starting point for coordinate fitting and refinement for unliganded structure (PDB 6ALF) and the hisPEC complex (PDB 6ASX) for the ligand bound complex. The nucleotide scaffold was built into each map using a combination of Coot 54 and Phenix real-space refinement 55 . The coordinates were first placed into the refined maps using the fit to map procedure in UCSF ChimeraX (ref. 56). Once placed, the coordinates were submitted to a round of unrestrained, all-atom refinement using the Phenix real-space refinement procedure. Then, another https://doi.org/10.1038/s41594-023-01002-x round of all-atom refinement was done while applying secondary structure and Ramachandran restraints. Finally, the fitted coordinates were validated in MolProbity 57 .

Molecular dynamics flexible fitting simulations
The MDFF is a simulation approach to flexibly fit atomic coordinates into EM maps [58][59][60] .
The MDFF integrates the EM density map as a potential so that high density areas in the map are minima on the potential energy surface. To achieve this, guiding forces are applied to the atoms in a molecular system that are proportional to the gradient of the EM map potential. Specifically, the EM map potential (U EM ), which is defined on a 3D grid, is given by: where j runs over the atoms in the system and the MDFF potential map (V EM ) is given by: Here w j are atomwise weights, ξ is a force scaling, Φ(r) is the EM density at position r, Φ max is the maximum value of the EM density map and Φ thr is a density threshold. The density threshold serves to eliminate EM data corresponding to the solvent contribution to the map. The actual MDFF guiding forces (f i ) that bring the structure into correspondence with the EM density map are given by: The initial structures of the que-PEC contained the template DNA and a ten-residue-long downstream riboswitch RNA forming the RNA-DNA hybrid, but lacked the complete coordinates for the residues corresponding to the preQ 1 riboswitch. Complete models of the que-PEC in the absence and presence of preQ 1 were generated by attaching a structural model of the 37-nucleotide preQ 1 riboswitch to the 5′ end of the RNAs present in the initial structure. The structural model of the riboswitch corresponded to conformer 1 in the NMR structure of a class I preQ 1 riboswitch aptamer bound to its cognate ligand (PDB 2L1V) (ref. 27). In the case of the model in the absence of preQ1, the ligand present in the NMR structure was removed. Missing hydrogens and terminal patches were added to the initial models using the CHARMM-GUI web server. The resulting structures were energy minimized in vacuum with 100 steps of steepest descent and 500 steps of adopted basis Newton-Raphson method with a gradient tolerance of 0.01 to remove the initial clashes. A cutoff of 16 Å was used to generate the nonbonded list. The nonbonded list was updated heuristically. Switching function was used to treat the Lennard-Jones and electrostatic interactions within the complexes. During minimization, the heavy atoms in the protein backbone (C, O, N and Cα), nucleic acid backbone (P, O1P, O2P, O5′, C5′, C4′, C3′ and O3′) and the ligand were harmonically restrained with a force constant of 1.0 kcal mol −1 Å −2 and the heavy atoms in the protein side chain and the sugar and base of the nucleic acid residues were harmonically restrained using a force constant 0.1 kcal mol −1 Å −2 . During CHARMM energy minimization and subsequent MDFF simulations (see below), we used the CHARMM36 force field 61,62 for the protein and nucleic acid components of the PEC, and the CHARMM general force field (CGenFF) 63 for the preQ 1 ligand.
The energy-minimized coordinates of the initial models, along with the cryo-EM density maps into which the initial structures were already docked, were used to set up the MDFF simulations. The input files for the MDFF simulations were generated using the MDFF plugin within the VMD software 64 . These files included restraint parameters that we used to maintain the secondary structure of the individual components in the complex and prevent overfitting during the MDFF simulations. The MDFF simulations were run in vacuum for 2 ns with a time step of 1 fs using NAMD 65 . Before the production run, 1,000 steps of energy minimization were performed. The temperature was maintained using the Langevin thermostat at 300 K. During the initial MDFF simulations, the coordinates of the protein components, template DNA and ten-residue-long downstream riboswitch RNA that forms the RNA-DNA hybrid in the initial structures were held fixed. Only the upstream 37 residues of the RNA were free to move during the simulations. For all the MDFF simulations, atomwise weights, w j , were set to atomic masses, the scaling factor (ξ) was set to 0.3 and Φ thr is a density threshold. For both conditions (−/+ preQ 1 ), 100 independent traditional MDFF trajectories were generated. Although, in principle, advanced flexible fitting methods, such as cascade MDFF simulations, can be useful in improving the quality of the models by overcoming the problem of entrapment of structures in the local minima of cryo-EM density-dependent potentials, our approach of traditional MDFF simulations ensured resolving close-to-global minimum conformations (Supplementary Note 1).

Pausing assays
A 148-nucleotide DNA template including the preQ 1 riboswitch from B. subtilis under the control of the T7A1 promoter was generated using an overlapping PCR strategy. In addition, 25 nucleotides not found in the wild-type sequence were inserted after the promoter sequence to generate a 25-nucleotide stretch in which the RNA transcript lacks any uracil residues (EC-25) except for the +2 position dependent on the ApU dinucleotide used to initiate the transcription. Transcription templates for transcription in vitro were generated by PCR using the 'T7A1-PreQ1-RNA0p (1)' forward oligonucleotide and the complementing reverse oligonucleotides (Supplementary Table 1).
Halted complexes were prepared in transcription buffer (20 mM Tris-HCl, pH 8.0, 20 mM NaCl, 20 mM MgCl 2 , 14 mM 2-mercaptoethanol, 0.1 mM EDTA) containing 25 µM ATP/CTP mix, 50 nM α-[ 32 P]GTP (3,000 Ci mmol −1 ), 10 µM ApU dinucleotide primer (Trilink, catalog no. O-31004) and 50 nM DNA template. A portion of 100 nM E. coli RNAP holoenzyme (New England Biolabs, catalog no. M0551S) was added to the reaction mixture and incubated for 10 min at 37 °C. The reaction mixture was then passed through a G50 (GE Healthcare, catalog no. 27533001) gel filtration column to remove any free nucleotides. To complete the transcription reaction, a mixture containing all four rNTPs (25 µM for time-pausing experiments and 100 µM for termination assays) was added concomitantly with heparin (450 µg ml −1 ) to prevent the re-initiation of transcription. In the case of the construct DNA template used in the cryo-EM studies, 10 µM rNTPs were used to perform the time-pausing assay. preQ 1 (when present) was added to 10 µM or ranged from 100 nM to 250 µM for the T50 determination in Supplementary Fig. 8e. The reaction mixture was incubated at 37 °C, and aliquots were quenched at the desired times into an equal volume of loading buffer (95% formamide, 1 mM EDTA, 0.1% SDS, 0.2% bromophenol blue, 0.2% xylene cyanol). Reaction aliquots were denatured before loading 5 µl each onto a denaturing 8 M urea, 6% polyacrylamide sequencing gel. The gel was dried and exposed to a phosphor screen (typically overnight), which was then scanned on an Amersham Typhoon PhosphorImager (GE Lifesciences). Gel images were analyzed with ImageLab (Bio-Rad) software.

Time-pausing analysis
The half-life of transcriptional pausing was determined by calculating the fraction of the RNA pause species compared with the total amount of RNA for each time point, which was analyzed with pseudo-first-order kinetics to extract the half-life 66 . For each