E. coli TraR allosterically regulates transcription initiation by altering RNA polymerase conformation

TraR and its homolog DksA are bacterial proteins that regulate transcription initiation by binding directly to RNA polymerase (RNAP) rather than to promoter DNA. Effects of TraR mimic the combined effects of DksA and its cofactor ppGpp, but the structural basis for regulation by these factors remains unclear. Here, we use cryo-electron microscopy to determine structures of Escherichia coli RNAP, with or without TraR, and of an RNAP-promoter complex. TraR binding induced RNAP conformational changes not seen in previous crystallographic analyses, and a quantitative analysis revealed TraR-induced changes in RNAP conformational heterogeneity. These changes involve mobile regions of RNAP affecting promoter DNA interactions, including the βlobe, the clamp, the bridge helix, and several lineage-specific insertions. Using mutational approaches, we show that these structural changes, as well as effects on σ70 region 1.1, are critical for transcription activation or inhibition, depending on the kinetic features of regulated promoters.


Introduction
Transcription initiation is a major control point for gene expression. In bacteria, a single catalytically active RNA polymerase (RNAP) performs all transcription, but a σ factor is required for promoter utilization (Burgess et al., 1969;Feklistov et al., 2014). In Escherichia coli (Eco), the essential primary σ factor, σ 70 , binds to RNAP to form the σ 70 -holoenzyme (Eσ 70 ) that is capable of recognizing and initiating at promoters for most genes. Upon locating the promoter, Eσ 70 melts a ~13 bp segment of DNA to form the open promoter complex (RPo) in which the DNA template strand (t-strand) is loaded into the RNAP active site, exposing the transcription start site (Bae et al., 2015b;Zuo and Steitz, 2015). A key feature of the RPo formation pathway is that it is a multi-step process, with the RNAP-promoter complex passing through multiple intermediates before the final, transcription competent RPo is formed (Hubin et al., 2017a;Ruff et al., 2015;Saecker et al., 2002).
A variety of transcription factors bind to the promoter DNA and/or to RNAP directly to regulate initiation (Browning and Busby, 2016;Haugen et al., 2008). Bacterial RNAP-binding factors, encoded by the chromosome or by bacteriophage or extrachromosomal elements, interact with several different regions of the enzyme to regulate its functions (Haugen et al., 2008). One such factor is ppGpp, a modified nucleotide that functions together with the RNAP-binding protein DksA in Eco to reprogram bacterial metabolism in response to nutritional stresses during the so-called stringent response. Following amino acid starvation, ppGpp is synthesized by the RelA factor in response to uncharged tRNAs in the ribosomal A site (Brown et al., 2016;Cashel and Gallant, 1969;Ryals et al., 1982). Together, ppGpp and DksA alter the expression of as many as 750 genes within 5 minutes of ppGpp induction (Paul et al., 2004a;2005;Sanchez-Vazquez et al., 2019), inhibiting, for example, promoters responsible for ribosome biogenesis and activating promoters responsible for amino acid synthesis.
The overall RNAP structure is reminiscent of a crab claw, with one pincer comprising primarily the b' subunit, and the other primarily the b subunit (Zhang et al., 1999). Between the two pincers is a large cleft that contains the active site. In Es 70 without nucleic acids, this channel is occupied by the s 70 1.1 domain which is ejected upon entry of the downstream duplex DNA (Bae et al., 2013;Mekler et al., 2002). The Bridge Helix (BH) bridges the two pincers across the cleft, separating the cleft into the main channel, where s 70 1.1 or nucleic acids reside, and the secondary channel, where NTPs enter the RNAP active site.
DksA binds in the RNAP secondary channel (Lennon et al., 2012;Molodtsov et al., 2018;Perederina et al., 2004). ppGpp binds directly to RNAP at two binding sites: site 1, located at the interface of the β′ and ω subunits (Ross et al., 2013;Zuo et al., 2013), and site 2, located at the interface of β′ and DksA (Molodtsov et al., 2018;Ross et al., 2016). The ppGpp bound at site 1 inhibits transcription ~2-fold under conditions where the effects of ppGpp bound at both sites together with DksA are as much as 20fold (Paul et al., 2004b;Ross et al., 2016). By contrast, DksA and ppGpp bound at site 2 are necessary and sufficient for activation (Ross et al., 2016).
TraR is a distant homolog of DksA. Although only half the length of DksA, TraR regulates Eco transcription by binding to the RNAP secondary channel and mimicking the combined effects of DksA and ppGpp (Blankschien et al., 2009;Gopalkrishnan et al., 2017). TraR is encoded by the conjugative F plasmid and is expressed from the pY promoter as part of the major tra operon transcript (Frost et al., 1994;Maneewannakul and Ippen-Ihler, 1993). Like DksA, TraR inhibits Eσ 70 -dependent transcription from ribosomal RNA promoters (e.g. rrnB P1) and ribosomal protein promoters (e.g. rpsT P2, expressing S20), and activates amino acid biosynthesis and transport promoters (e.g. pthrABC, phisG, pargI, plivJ) in vivo and in vitro (Blankschien et al., 2009;Gopalkrishnan et al., 2017). The affinity of TraR for RNAP is only slightly higher than that of DksA, yet its effects on promoters negatively regulated by DksA/ppGpp in vitro are as large or larger than those of DksA/ppGpp (Gopalkrishnan et al., 2017). The effects of TraR on promoters positively regulated by ppGpp/DksA are also independent of ppGpp (Gopalkrishnan et al., 2017).
Models for DksA/ppGpp and TraR binding to RNAP have been proposed based on biochemical and genetic approaches (Gopalkrishnan et al., 2017;Parshin et al., 2015;Ross et al., 2013;. Crystal structures of DksA/ppGpp/RNAP and TraR/RNAP confirmed the general features of these models and provided additional detail about their interactions with RNAP, but did not reveal the mechanism of inhibition or activation, in large part because of crystal packing constraints on the movement of mobile regions of the complex (Molodtsov et al., 2018). Thus, the structural basis for the effects of DksA/ppGpp or TraR on transcription have remained elusive.
To help understand TraR regulation and principles of the regulation of transcription initiation in general, we used single particle cryo-electron microscopy (cryo-EM) to examine structures of Es 70 alone, Eσ 70 bound to TraR (TraR-Es 70 ), and Es 70 bound to a promoter inhibited by TraR [rpsT P2; (Gopalkrishnan et al., 2017)]. Cryo-EM allows the visualization of multiple conformational states populated in solution and in the absence of crystal packing constraints. Furthermore, new software tools allow for the analysis of molecular motions in the cryo-EM data .
The TraR-Es 70 structures show TraR binding in the secondary channel of the RNAP, consistent with the TraR-Es 70 model (Gopalkrishnan et al., 2017) and crystal structure (Molodtsov et al., 2018). However, the cryo-EM structures reveal major TraRinduced changes to the RNAP conformation that were not evident in the crystal structure due to crystal packing constraints. Structural analyses generated mechanistic hypotheses for TraR function in both activation and inhibition of transcription that were then tested biochemically. On the basis of the combined structural and functional analyses, we propose a model in which TraR accelerates multiple steps along the RPo formation pathway while at the same time modulates the relative stability of intermediates in the pathway. Whether a promoter is activated or inhibited by TraR is determined by the intrinsic kinetic properties of the promoter (Galburt, 2018;Haugen et al., 2008;Paul et al., 2005).

Cryo-EM structures of TraR-Es 70
We used single-particle cryo-EM to examine Eco TraR-Es 70 in the absence of crystal packing interactions that could influence the conformational states. TraR function in cryo-EM solution conditions  was indistinguishable from standard  (Gopalkrishnan et al., 2017) and is broadly consistent with the X-ray structure (Molodtsov et al., 2018 (Zhang et al., 1999), the F-loop (Miropolskaya et al., 2009), and the bridge-helix ( Figure 1D). The N-terminal tip of TraRN (TraR residue S2) is only 4.3 Å from the active site Mg 2+ ( Figure 1E). TraRG interacts primarily with the b'rim-helices at the entrance of the secondary channel ( Figure 1D).
The interactions of TraRC with RNAP differ between the cryo-EM and X-ray structures due to conformational changes induced by TraR binding detected by the cryo-EM structure that were not observed in the X-ray structure (see below). Indeed, the cryo-EM and X-ray structures superimpose with an rmsd of 4.26 Å over 3,471 acarbons, indicating significant conformational differences.

Cryo-EM analysis of Es 70 and rpsT P2 RPo
To understand how TraR-induced conformational changes regulate RPo formation, we analysed single-particle cryo-EM data for Es 70 alone and Es 70 bound to the rpsT P2 promoter. Our aim was to explore conformational space and dynamics unencumbered by crystal packing constraints to compare with the TraR-Es 70 data above. Cryo-EM data for Es 70 resolved to a nominal resolution of 4.1 Å (Figure 1 -figure supplements 3, 4; Supplementary file 1).
Analysis of the rpsT P2-RPo cryo-EM data gave rise to two conformational classes that differed only in the disposition of the upstream promoter DNA and aCTDs The closed-clamp RNAP in the rpsT P2-RPo interacts with the promoter DNA in the same way as seen in other RPo structures determined by X-ray crystallography (Bae et al., 2015b;Hubin et al., 2017b) or cryo-EM (Boyaci et al., 2019) and is consistent with the DNase I footprint of the rpsT P2 RPo (Gopalkrishnan et al., 2017). In the rpsT P2-RPo structure we observed an a-subunit C-terminal domain [aCTD; (Ross et al., 1993)] bound to the promoter DNA minor groove (Benoff et al., 2002;Ross et al., 2001) just upstream of the promoter -35 element [-38 to -43, corresponding to the proximal UP element subsite (Estrem et al., 1999)]. This aCTD interacts with s 70 4 through an interface previously characterized through genetic analyses (Ross et al., 2003) (Figures 2B, C). The aCTDs are linked to the a-N-terminal domains (aNTDs) by ~15-residue flexible linkers (Blatter et al., 1994;Jeon et al., 1995). Density for the residues connecting the aCTD and aNTD was not observed in the cryo-EM map.
Comparing the RNAP conformations of the TraR-Es 70 , Es 70 , and rpsT P2-RPo cryo-EM structures revealed key differences that suggest how TraR activates and inhibits transcription. Below we outline these differences and present experients that test their implications for function.

b'Si3 is in two conformations, one of which is important for TraR activation function
The three TraR-Es 70 structures differ from each other only in the disposition of Si3. Si3 comprises two tandem repeats of the sandwich-barrel hybrid motif (SBHM) fold (Chlenov et al., 2005;Iyer et al., 2003), SBHMa and SBHMb ( Figure 3A). Si3 is linked to the TL-helices by extended, flexible linkers. In TraR-Es 70 (I) and TraR-Es 70 (II), Si3 is in two distinct positions with respect to the RNAP (Figures 1A, 1B, 3A), while in TraR-Es 70 (III) Si3 is disordered (Figure 1 -figure supplement 1D). Si3 in the TraR-Es 70 (I) structure [Si3(I)] interacts primarily with the b'shelf (SBHMa) and the b'jaw (SBHMb) in a manner seen in many previous Eco RNAP X-ray (Bae et al., 2013) and cryo-EM structures Kang et al., 2017;Liu et al., 2017  was similar to WT-TraR for TraR-K50A, and mildly impaired for TraR-E46A or R49A ( Figure 3F; legend for IC50 values). Maximal inhibition was achieved at higher E46A or R49A TraR concentrations. However, these same variants exhibited at least ~2-fold reduced activation at the thrABC promoter ( Figure 3G) even at saturating TraR concentrations, indicating a role for the TraR-Si3 interaction in the mechanism of activation. Consistent with these results, these TraR variants were proficient in RNAP binding in a competition assay ( Figure 1D) (Bae et al., 2013;Mekler et al., 2002). TraR binding induces a ~18° rotation of the RNAP blobe-Si1 domains (the two domains move together as a rigid body), shifting the blobe-Si1 towards TraR, allowing the blobe-Si1 to establish an interface with TraRG and TraRC (615 Å 2 interface area; Figure 4A).
The rotation of the blobe-Si1 widens the gap between the bprotrusion and the blobe ( Figure 4A) and changes the shape of the RNAP channel, altering RNAP contacts Es 70 exhibited a low basal level of transcription from the TraR-activated thrABC promoter in the absence of TraR, and transcription was stimulated about ~4-fold by TraR ( Figure 4D). ED1.1s 70 exhibited a striking increase in basal transcription activity (~32-fold) compared to WT-Es 70 in the absence of TraR ( Figure 4D). A small further increase in transcription was observed upon the addition of TraR ( Figure 4D). These results suggest that s 70 1.1 is an obstacle to promoter DNA entering the RNAP channel and that TraR partially overcomes this barrier. In contrast to deletion of region s 70 1.1, which bypasses the requirement for TraR, rotation of the blobe-Si1 does not weaken s 70 1.1-RNAP contacts sufficiently to release s 70 1.1 completely ( Figure 1D). Rather, blobe-Si1 rotation facilitates the competition between promoter DNA and s 70 1.1 during RPo formation. Our results suggest that TraR-activated promoters are defined, in part, by being limited at the s 70 1.1 ejection step.
bSi1 was also required for inhibition of rpsT P2 ( Figure 4B) and rrnB P1 transcription by TraR (Gopalkrishnan et al., 2017). However, in contrast to the effect of

TraR induces b'shelf rotation and a bridge-helix kink, contributing to inhibition
TraR binding induces a ~4.5° rotation of the b'shelf module ( Figure 5A, B). The BH leads directly into the shelf module, and a kink is introduced in the BH, a long a-helix that traverses the RNAP active site cleft from one pincer to the other, directly across from the active site Mg 2+ ( Figure 5B, C). The BH plays critical roles in the RNAP nucleotide addition cycle (Lane and Darst, 2010b), including interacting with the t-strand DNA at the active site ( Figure 5D). TraR causes the BH to kink towards the t-strand DNA ( Figure 5C), similar to BH kinks observed previously (Tagami et al., 2011;Weixlbaumer et al., 2013;Zhang et al., 1999), resulting in a steric clash with the normal position of the t-strand nucleotide at +2 ( Figure 5E). Thus, the TraR-induced BH kink would sterically prevent the proper positioning of the t-strand DNA in RPo, likely contributing to inhibition of transcription.

TraR binding alters clamp dynamics, restricting clamp motions compared to Es 70 , likely stimulating transcription bubble nucleation
TraR induces conformational changes in the RNAP blobe-Si1 ( Figure  rpsT P2-RPo structures. We therefore analysed and compared the heterogeneity of RNAP clamp positions between the Es 70 , TraR-Es 70 , and rpsT P2-RPo datasets using multibody refinement as implemented in RELION 3 . The maps used for multi-body refinement were carefully chosen to be equivalently processed. After initial classification to remove junk particles, particles were 3D auto-refined, then the refinement metadata and post-processing were used as inputs for RELION CTF refinement and Bayesian Polishing (Zivanov et al., 2018). After a final round of 3D autorefinement (but no further classification), the rpsT P2-RPo dataset had the smallest number of particles (370,965), so a random subset of particles from the other datasets (TraR-Es 70 and Es 70 ) were processed so that each map for multi-body refinement was generated from the same number of particles ( Component 2 gave rise to clamp positions that differed by a 4.6° rotation about a rotation axis roughly perpendicular to the open/close rotation axis, a motion we call twisting ( Figure 6F). Finally, component 3 gave rise to clamp positions that differed by a 2.0° rotation about a third rotation axis parallel with the long axis of the clamp, a motion we call rolling ( Figure 6G).
Using the parameters of the Gaussian fits to the Eigenvalue histograms ( Figures 6B-D), we could estimate the full range of clamp rotations for each component, which we defined as the rotation range that accounted for 98% of the particles (excluding 1% of the particles at each tail; Figure 7).
These same motions (opening/closing, twisting, rolling) were represented in major components of clamp motion for the TraR-Es 70 and rpsT P2-RPo particles as well. The same analyses revealed that TraR binding significantly reduced the range of clamp movement for each of the three clamp motions (Figure 7). We propose that the dampening of clamp motions by TraR facilitates the nucleation of strand opening (Feklistov et al., 2017). As expected, the clamp motions for RPo, with nucleic acids stably bound in the downstream duplex channel, were restricted even further for all three of the major clamp motions (Figure 7).

Discussion
Our cryo-EM structural analyses show that TraR modulates The consequences of these TraR-induced conformational changes for promoter function (activation or inhibition) depend on the distinctly different properties of the two types of promoters. The kinetics of RPo formation and the thermodynamic properties of RPo vary by many orders of magnitude among different bacterial promoters (McClure, 1985). Es 70 can complete RPo formation on some promoters in a fraction of a second, while other promoters require ten minutes or more. The RPo half-life can vary from a few minutes to many hours. This tremendous range of promoter properties gives rise to a dynamic range for bacterial transcription initiation of ~4 orders of magnitude and provides rich targets for regulation (Galburt, 2018).
Mechanistic studies of ppGpp/DksA-and TraR-dependent regulation of initiation have revealed general characteristics of promoters that are either activated or inhibited by these factors, and has led to a conceptual model for how TraR activates some promoters while inhibiting others (Gopalkrishnan et al., 2017;Gourse et al., 2018). In the absence of factors, activated promoters generate RPo very slowly Paul et al., 2005). Given sufficient time, however, RPo that is ultimately formed is very stable; the activated promoters pargI, phisG, and pthrABC have half-lives measured in many hours [15 hrs, > 13 hrs, and 6.7 hrs, respectively ].
On the other hand, inhibited promoters generate RPo very rapidly (Rao et al., 1994). The final transcription-competent RPo is, however, relatively unstable; the RPo half-life of the inhibited promoter rrnB P1 is measured in minutes or less . In the absence of either factors or high initiating NTP concentrations, RPo exists in equilibrium with earlier intermediates along the pathway to RPo formation (Gopalkrishnan et al., 2017;Rutherford et al., 2009).
In order for a transcription factor, such as TraR, to achieve differential regulation (that is, activate some promoters but inhibit others through the same effects on RNAP), the factor must affect more than one feature of the multi-step pathway of RPo formation (Galburt, 2018). In the model for TraR function, TraR acts on all promoters similarly.
TraR relieves kinetic barriers to accelerate RPo formation but at the same time likely stabilizes an intermediate prior to RPo formation (Galburt, 2018 RPo formation, providing molecular insight into activation and inhibition.

Structural mechanism for TraR-mediated activation
While the transcription output of activated promoters is limited by the slow rate of RPo formation, deletion of s 70 1.1 greatly enhances basal activity (32-fold on pthrABC; Figure 4D). This suggests that the presence of s Our results suggest that the presence of s 70 1.1 in the DNA channel is a significant obstacle to RPo formation at activated promoters and this step is targeted by TraR, but this is not the only mechanistic determinant of TraR-mediated activation. We found that the formation of the TraRG-Si3(II) interface ( Figure 3B, C) plays a role in activation, but not inhibition, since deletion of Si3 or mutation of TraR to disrupt the TraRG-Si3 interface decreased activation ~2-fold (compare Figure 3D with 3E) without affecting inhibition.
TraR binding also has a significant effect on RNAP clamp dynamics; all three major components of clamp motion in Es 70 were significantly restricted in TraR-Es 70 (Figures 6, 7). We propose that the restriction of clamp motions in TraR-Es 70 could contribute to activation by facilitating transcription bubble nucleation (Feklistov et al., 2017), likely a separate and earlier kinetic step than s 70 1.1 ejection. The TraRG-Si3(SBHMb) interface important for full activation forms in the rotated conformation of Si3 [Si3(II)] in which Si3(SBHMa) contacts with the b'jaw also form. In this way, Si3(II) forms bridging contacts across the RNAP cleft, which may encourage a clamp conformation conducive to more efficient transcription bubble nucleation. Alternatively, the TraRG-Si3(SBHMb) interaction may help stabilize the TraRC-blobe-Si1 interaction.

Structural mechanism for TraR-mediated inhibition
TraR-inhibited promoters have an intrinsically unstable RPo, with earlier intermediates significantly populated at equilibrium (Gopalkrishnan et al., 2017;Rutherford et al., 2009). TraR likely stabilizes one or more of these intermediates relative to RPo, further shifting the equilibrium away from RPo to the intermediate(s) and depopulating RPo.
TraR binding induces two distinct conformational changes in the RNAP that we propose disfavor RPo formation, the blobe-Si1 rotation ( Figure 4A) and the BH-kink ( Figures 5B, C). Consistent with this hypothesis, DSi1-RNAP shows reduced capacity to respond to TraR-mediated effects on inhibition ( Figure 4B).
The TraR-mediated blobe-Si1 rotation ( Figure 4A) alters the shape of the RNAP channel, which not only weakens contacts with s 70 1.1 to help activate positively regulated promoters, but we propose may stabilize DNA contacts in an intermediate prior to RPo. TraR binding also induces a kinked BH which sterically clashes with the proper positioning of the t-strand DNA near the active site ( Figure 5). Precise positioning of the t-strand DNA at the active site is critical for efficient catalysis of phosphodiester bond formation by RNAP in the SN2 mechanism (Yee et al., 2002). On the basis of TraR-BH contacts observed in the TraR-Es 70 crystal structure, Molodtsov et al. (2018) proposed that TraR-induced BH distortion might affect RPo formation, but other changes inducted by TraR that also contribute to inhibition were not seen in the crystal structure.
Upon the formation of RPo, TraR must dissociate before RNAP can catalyse the first phosphodiester bond because the presence of TraR in the secondary channel sterically blocks NTP binding and TL-folding ( Figure 5). We propose that the stable RPo formed at activated promoters is better able to compete wtih TraR binding than the unstable RPo at inhibited promoters , explaining how TraR-induced BH-kinking could inhibit some promoters but not others.

TraR manipulates Eco RNAP lineage-specific insertions to modulate transcription initiation
The large b and b' subunits of the bacterial RNAP are conserved throughout evolution, containing 16 and 11 shared sequence regions, respectively, common to all bacterial RNAPs (Lane and Darst, 2010a). These shared sequence regions are separated by relatively nonconserved spacer regions in which large LSIs can occur (Lane and Darst, 2010a). These are typically independently-folded domains, ranging in size from 50 to 500 amino acids, located on the surface of the RNAP and often highly mobile. A key feature of the TraR functional mechanism is modulation of Eco RNAP transcription initiation through conformational changes brought about by interactions with two of the Eco RNAP LSIs, bSi1 ( Figure 4A) and b'Si3 ( Figures 3A-C).

Deletions of Eco bSi1 supported basic in vitro transcription function and normal
in vivo cell growth, leading to its original designation as 'dispensable region I' (Severinov et al., 1994). Later studies revealed that in vivo, the DbSi1-RNAP was unable to support cell growth at 42°C and could only support slow growth at 30°C (Artsimovitch, 2003).
Thus, bSi1 may serve as a binding determinant for unknown transcription regulators that modulate Eco RNAP function during unusual growth conditions. Indeed, TraR interacts with Si1 as well as the nearby blobe to distort the RNAP active site cleft ( Figure 4A), effecting both inhibition ( Figure 4B) and activation ( Figure 4C) by TraR.
Eco b'Si3 is an unusual LSI as it is inserted in the middle of the TL, a key structural element in the RNAP nucleotide addition cycle that is conserved in all multisubunit RNAPs (Lane and Darst, 2010a). As a consequence, Si3 plays a central role in Eco RNAP function and deletion of Si3 is not viable (Artsimovitch, 2003;Zakharova et al., 1998). Si3 is known to be highly mobile, moving to accommodate folding and unfolding of the TL at each RNAP nucleotide addition cycle (Malinen et al., 2012;Zuo and Steitz, 2015); the movement corresponds to a rotation of Si3 by about 33°, resulting in a shift of the Si3 center-of-mass by 15 Å (Kang et al., 2018). Si3 was often disordered in Eco RNAP crystal structures [for example, see (Molodtsov et al., 2018)]. In our cryo-EM analysis, TraR engages with Si3, stabilizing a previously unseen conformation of Si3 that plays a role in TraR activation function (Figure 3). Si3 has been implicated previously in RPo formation since the Db'Si3-RNAP forms an unstable RPo (Artsimovitch, 2003).

Conclusion
TraR-like proteins are widespread in proteobacteria and related bacteriophage and plasmids (Gopalkrishnan et al., 2017;Gourse et al., 2018). While the in vivo function of TraR is incompletely understood, TraR engages with RNAP in much the same way as DksA/ppGpp, utilizing the same residues in the b'rim-helices that contribute to ppGpp site 2 in the DksA-ppGpp-RNAP complex, and uses its N-terminal a-helix to bind in the RNAP secondary channel near the RNAP active site (Gopalkrishnan et al., 2017;Ross et al., 2016). These general features of TraR binding were confirmed in an X-ray crystal structure of the TraR-Es 70 complex (Molodtsov et al., 2018), but crystal packing constraints prevented this structure from revealing the RNAP conformational changes induced by TraR binding, which are the keys to TraR function. Our structural and functional analyses described here greatly extend previous work (Gopalkrishnan et al., 2017) by identifying the RNAP conformational changes responsible for the effects of TraR on transcription. In so doing, our analysis dissects the complex, multifaceted mechanism that distinguishes activation from inhibition by TraR.
The RPo formation pathway proceeds through multiple steps. TraR binding to RNAP alters the RNAP conformation and conformational dynamics in multiple, complex ways. The complex interplay between TraR binding and RNAP conformation and conformational dynamics allows TraR to modulate multiple features of the energy landscape of RPo formation, which is key to allowing TraR to effect differential regulation across promoter space without direct TraR-promoter interactions.

Strains, Plasmids and Primer sequences
Plasmids are listed in Supplementary file 3 and oligonucleotide and geneblock sequences are in Supplementary file 4. Bacteria were grown in LB Lennox media or on LB agar plates. Media was supplemented with ampicillin (100 µg/ml) or kanamycin (30 µg/ml) if needed. TraR was made by cloning the traR gene in a pET28-based His10-SUMO vector which allowed removal of the cleavable N-terminal His10-SUMO tag with Ulp1 protease. ESI-Mass Spectrometry revealed that the molecular mass of purified TraR corresponded to that of a monomer lacking the N-terminal methionine [ Figure S6 of (Gopalkrishnan et al., 2017)], hence traR without the initial M was cloned into the SUMO vector. This tag-less version of TraR exhibited the same level of activity as a previous TraR construct with 4 additional residues (LVPR) at the C-terminal end leftover after His6 tag cleavage in the TraR-thrombin site-His6 construct (Gopalkrishnan et al., 2017).

Expression and purification of TraR for cryo-EM
The His10-SUMO-TraR plasmid was transformed into competent Eco BL21(DE3) by heat shock. The cells were grown in the presence of 25 µg/mL kanamycin to an

Preparation of TraR-Es 70 for cryo-EM
Es 70 was formed by mixing purified RNAP and a 2-fold molar excess of σ 70 and incubating for 15 minutes at room temperature. Es 70 was purified over a Superose 6 Increase 10/300 GL column in gel filtration buffer. The eluted Es 70 was concentrated to ~5.0 mg/mL (~10 μM) by centrifugal filtration. Purified TraR was added (5-fold molar excess over RNAP) and the sample was incubated for 15 min at room temperature. An rrnB P1 promoter fragment (Integrated DNA Technologies, Coralville, IA) was was added (2-fold molar excess over RNAP) and the sample was incubated for a further 15 minutes at room temperature. The rrnB P1 promoter fragment did not bind to TraR-Es 70 under the cryo-EM grid preparation conditions -the subsequent structural analyses did not reveal any evidence of promoter binding.

Preparation of rpsT P2-RPo for cryo-EM
Es 70 was prepared as described for TraR-Es 70 , but after the size exclusion chromatography the complex was concentrated to ~10 mg/mL (~20 μM) by centrifugal filtration. Duplex rpsT P2 promoter fragment (-60 to +25, Figure 3A, IDT) was added to the concentrated Es 70 to 3-fold molar excess. The sample was incubated for 20 mins at room temperature prior to cryo-EM grid preparation.

Acquisition and processing of TraR-Eσ 70 cryo-EM dataset
Grids were imaged using a 300 keV Krios (FEI) equipped with a K2 Summit direct electron detector (Gatan, Pleasanton, CA). Datasets were recorded with Serial EM (Mastronarde, 2005) with a pixel size of 1.3 Å over a defocus range of 0.8 μm to 2.4 μm.
Movies were recorded in counting mode at 8 electrons/physical pixel/second in dosefractionation mode with subframes of 0.3 sec over a 15 sec exposure (50 frames) to give a total dose of 120 electrons/physical pixel. Dose-fractionated movies were gainnormalized, drifted-corrected, summed, and dose-weighted using MotionCor2 (Grant and Grigorieff, 2015;Zheng et al., 2017). CTFFIND4 (Rohou and Grigorieff, 2015) was used for contrast transfer function estimation. Particles were picked using Gautomatch were aligned using RELION 3D auto-refinement resulting in a consensus map with nominal resolution of 3.62 Å. Using the refinement parameters, subtractive 3D classification (N=3) was performed on the particles by subtracting density outside of β'Si3 and classifying in a mask around β'Si3. Classification revealed three distinct β'Si3 dispositions ( Figure S1D). Local refinement metadata (highlighted in red dotted box, Figure S1D) for TraR-Eσ 70 (I) and TraR-Eσ 70 (II) were used for RELION multi-body refinements to examine clamp motions . Local resolution calculations were performed using blocres and blocfilt from the Bsoft package (Cardone et al., 2013).

Acquisition and processing of Eσ 70 cryo-EM dataset
The Es 70 image acquisition and processing was the same as for TraR-Es 70 except with the following differences. Grids were imaged using a 200 keV Talos  Curated particles were combined and a consensus refinement was performed in RELION using the cryoSPARC generated initial model resulting in a map with nominal resolution of 4.54 Å (without post-processing). Particles from this refinement (highlighted in red dotted box, Figure 1 -figure supplement 3) were further analyzed using RELION multi-body refinement as described in the text .
Additionally, particles were further curated using RELION 3D classification (N=3) without alignment. Classification revealed two lower resolution class and a higher resolution class. The higher resolution class containing 358,725 particles was RELION 3D auto-refined and subjected to RELION CTF refinement and RELION Bayesian Polishing (Zivanov et al., 2018). After polishing, particles were refined to a nominal resolution of 4.05 Å after RELION post-processing.

Acquisition and processing of rpsT P2-RPo cryo-EM dataset
The rpsT P2-RPo cryo-EM image acquisition and processing was the same as for TraR-Es 70 except with the following differences. The imaging defocus range was 0.5 μm to 2.5 μm. Movies were recorded in super-resolution mode at 8 electrons/physical pixel/second in dose-fractionation mode with subframes of 0.2 sec over a 10 sec exposure (50 frames) to give a total dose of 80 electrons/physical pixel. The rpsT P2-RPo dataset consisted of 6,912 motion-corrected images with 973,481 particles. In RELION, a consensus refinement was performed using the extracted particles and a cryoSPARC generated initial model resulting in a 4.62 Å resolution map. Using the refinement parameters, 3D classification (N=2) was performed on the particles without alignment. Classification revealed a lower resolution class and a higher resolution class of with 370,965 particles with nominal resolution of 4.38 Å after RELION 3D autorefinement. Refinement metadata and post-processing were used as inputs for RELION CTF refinement and RELION Bayesian Polishing (Zivanov et al., 2018). Subsequent 3D classification (N=3) was used to further classify the polished particles resulting in one junk class and two high resolution classes (Figure 2 -figure supplement 1). The highest resolution reconstruction (3.43 Å) contained 289,679 particles.

Model building and refinement of cryo-EM structures
To build initial models of the protein components of the complexes, a crystal structure of Eco Es 70 [PDB ID 4LJZ, with s 70 1.1 from 4LK1; (Bae et al., 2013)] was manually fit into the cryo-EM density maps using Chimera (Pettersen et al., 2004) and manually adjusted using Coot (Emsley and Cowtan, 2004). For TraR-Es 70 , s 70 1.1 from 4LK1 (Bae et al., 2013) and TraR from 5W1S (Molodtsov et al., 2018) were also added. For rpsT P2-RPo, the promoter DNA was manually added. Appropriate domains of each complex were rigid-body refined, then subsequently refined with secondary structure and nucleic acid restraints using PHENIX real space refinement (Adams et al., 2010).

IPTG (1 mM final) was used to induce expression of TraR (WT or variant) from
Eco BL21 DE3 dksA::Tn10 (RLG7075) host cells. TraR and variants were purified as described (Gopalkrishnan et al., 2017), either from His6-TraR overexpression plasmids with removal of the His6-tag with thrombin, or from His10-SUMO-TraR constructs with removal of the His10-SUMO-tag with Ulp1 protease, resulting in a 72 amino acid TraR lacking the N-terminal Met. WT-TraR purified by the two methods gave comparable results. WT and variant RNAPs were purified as described previously (Ross et al., 2016). The D1.1s 70 was expressed and purified as described previously . ED1.1s 70 was reconstituted with a 4:1 molar ratio of D1.1s 70 to core RNAP. The purified core RNAP lacked detectable WT-s 70 activity.

In Vitro transcription assays, site-directed mutagenesis, and TraR-RNAP binding assays
All of these procedures were carried out exactly as previously described (Gopalkrishnan et al., 2017).
(G) TraR C. (C) Transcription in the absence of TraR is plotted, relative to the same reactions without CHAPSO. Although it had no effect on the concentration of TraR required for half-maximal inhibition ( Figure S1B), CHAPSO reduced transcription slightly. Averages with range from two independent experiments are plotted.
C. Angular distribution for Es 70 particle projections.
D. The 4.1-Å resolution cryo-EM density map of Es 70 is colored according to the key. The right view is a cross-section of the left view.
E. Same views as (D) but colored by local resolution (Cardone et al., 2013).
F.  . The gold-standard FSC was calculated by comparing the two independently determined half-maps from RELION. The dotted line represents the 0.143 FSC cutoff, which indicates a nominal resolution of 4.1 Å.
G. FSC calculated between the refined structure and the half map used for refinement (work), the other half map (free), and the full map. (B) rpsT P2-RPo cryo-EM density map (3.4 Å nominal resolution, low-pass filtered to the local resolution) is shown as a transparent surface and colored according to the key. The final model is superimposed. The DNA was modeled from -45 to +21. The t-strand DNA from -10 to -2, and the nt-strand DNA from -3 to +2 were disordered.   E. FSC calculated between the refined structure and the half map used for refinement (work), the other half map (free), and the full map.  (Chlenov et al., 2005;Iyer et al., 2003), denoted SBHMa and SBHMb. The boxed region is magnified in (B).
(B) Magnified view of TraR-Es 70 (II) [same view as (A)]. The position of Si3(II) is outlined in magenta but the rest of Si3 is removed, revealing TraR behind. Three residues central to the TraR-Si3(II) interface (TraR-E46, R49, and K50) are colored yellow. at a range of concentrations of WT or variant TraR (2 nM -2 µM). Transcripts were quantified and plotted relative to values in the absence of any factor (n=2). For (F), IC50 for inhibition by WT-TraR was ~50 nM, by E46A TraR was ~115 nM, R49A TraR was ~85 nM and by K50A TraR was ~30 nM. at a range of concentrations of WT-or variant TraR (2 nM -2 µM). Transcripts were quantified and plotted relative to values in the absence of TraR. Error bars denote the standard deviation of three independent measurements. For (D), the IC50 for inhibition by WT-TraR was ~50 nM, by P43A-TraR was ~80 nM, and by P45A-TraR was ~115 nM.
(B) Basal level of transcription from rrnB P1 is only slightly affected by D1.1s 70 . Error bars denote standard deviation of three independent measurements. (B -D) Histograms of Eigenvalue distributions (% of particles assigned each Eigenvalue from the dataset) for each of the three major principle components (Eigenvectors) from the multi-body analysis . Each set of particles were divided into three equal-sized bins (Eigenvalue ≤ -2, red; -1 ≤ Eigenvalue ≤ 1, gray; Eigenvalue ≥ 2, blue). The solid lines denote Gaussian fits to the histograms.

(E -G)
Three-dimensional reconstructions were calculated from the red and blue-binned particles for each principle component and models were generated by rigid body refinement. The models were superimposed using a-carbons of the RNAP structural core, revealing the alternate clamp positions shown (red and blue a-carbon ribbons with cylindrical helices). The s 70 NCR, attached to the clamp but not included in the clamp motion analyses, is shown in faded colors. For each component, the clamp rotation and the direction of the rotation axis were determined (rotation axes are shown in gray).  is shown as a molecular surface (a, w, light gray; b, light cyan; b', light pink; s 70 , light orange) except the clamp/s 70 2 module is shown schematically as blue or red outlines (the s 70 NCR is omitted for clarity) to illustrate the direction and approximate range of motion for the three major components of the clamp motions (left, opening/closing; middle, twisting; left, rolling).
(bottom) Histograms denote the range of clamp motions for Es 70 , TraR-Es 70 , and rpsT P2-RPo, as indicated. The black bars denote the range of motion defined by dividing the Eigenvalue histograms into three equal bins and determining the clamp position for the red and blue bins (-33%/+33% bin; see Figure 6). The gray bars denote the estimated range of motion to include 98% of the particles calculated from the Gaussian fits to the Eigenvalue histograms (1% of the particles excluded from each tail; see Figure 6).  (Bae et al., 2013) b (Molodtsov et al., 2018) c Refinement: PHENIX real_space_refine (Adams et al., 2010). Validation: MOLPROBITY (Chen et al., 2010).