An energy charge sensor for balancing RNA polymerase recycling and hibernation

Cellular RNA polymerases can become trapped on DNA or RNA, threatening genome stability and limiting free enzyme pools, or enter dormancy. How RNA polymerase recycling into active states is achieved and balanced with quiescence remains elusive. We structurally analyzed Bacillus subtilis RNA polymerase bound to the NTPase HelD. HelD has two long arms: a Gre cleavage factor-like coiled-coil inserts deep into the RNA polymerase secondary channel, dismantling the active site and displacing RNA; a unique helical protrusion inserts into the main channel, prying β and β’ subunits apart and dislodging DNA, aided by the δ subunit. HelD release depends on ATP, and a dimeric structure resembling hibernating RNA polymerase I suggests that HelD can induce dormancy at low energy levels. Our results reveal an ingenious mechanism by which active RNA polymerase pools are adjusted in response to the nutritional state. ID 6ALH)

recombination 21 and adaption to environmental changes 22 . B. subtilis HelD and RNAP directly interact 22 and are present at comparable levels during sporulation 23 . Together with δ, HelD enhances RNAP cycling 22 . We hypothesized that HelD is a general recycling factor and set out to elucidate its mechanism of action.
We assembled an RNAP-δ-HelD complex by supplementing stationary phase RNAP with δ, HelD, and a DNA/RNA scaffold with an arti cial transcription bubble (Extended Data Table 1), followed by size exclusion chromatography (SEC). RNAP bound HelD but not the nucleic acid scaffold, and ω was again underrepresented in the RNAP-δ-HelD fractions (Extended Data Fig. 1c). Cryo-electron microscopy (cryoEM) data were collected after vitrifying puri ed complexes without crosslinking in the presence of detergent to overcome preferred particle orientations (Extended Data Fig. 2). For structure analysis, we iteratively extracted ~1,000,000 particle images from ~9,100 micrographs for multi-particle 3D re nement (Extended Data Fig. 3a). Re nement led to two maps for monomeric RNAP-δ-HelD and dimeric (RNAP-δ-HelD) 2 complexes at global resolutions of 4.2 Å and 3.9 Å, respectively; local resolutions in both structures extended to well below 3.0 Å (Extended Data Fig. 3b; Extended Data Table 2).
In both monomeric and dimeric complexes, we observed well-de ned density for RNAP subunits α1/2 (Nterminal domains [NTDs]), β, β', δ, ε and HelD (Extended Data Fig. 4). Density for the ω subunit or nucleic acids was missing. Unless mentioned otherwise, the following descriptions refer to the monomeric complex.

Organization of RNAP in an RNAP-δ-HelD complex
In the RNAP-δ-HelD complex, RNAP adopts a conformation in which the main channel, where downstream DNA and the RNA:DNA hybrid are accommodated in an EC, is wide open, with a distance of 52 Å between the β2 lobe (P242) and the β' clamp helices (N283), compared to ~18 Å between the corresponding elements in the E. coli EC 24 (Fig. 1a,b; Extended Data Table 3; Supplementary Data 1), and a concomitant widening of the RNA exit tunnel by more than 17 Å (β ap R800 to β' lid D245 ). Comparison to the E. coli EC showed that RNAP opening leads to repositioning of the β' secondary channel elements, which would clash with ω at its canonical binding site, explaining loss of ω upon assembly of the RNAPδ-HelD complex. The α1/2 NTD s dimer remains bound at the closed end of the open β/β' crab claw.
The ε subunit is positioned in a cavity formed by the α1/2 NTDs, the C-terminal β clamp, and β' residues 492-655 that form part of the secondary channel (Fig. 1a), in contrast to previous mapping of ε at the β' jaw based on a low-resolution cryoEM analysis and structural similarity of ε to the phage T7 Gp2 19 . The ε subunit of B. subtilis RNAP occupies a position analogous to a small domain in archaeal and eukaryotic nuclear RNAPs from homologs of the bacterial α1 subunit (D, Rpb3 and AC40 of archaeal RNAP, eukaryotic RNAP II and eukaryotic RNAP I/III, respectively; Fig. 2). In some archaeal and eukaryotic RNAPs, these small domains bind an 4Fe-4S cluster 25 . B. subtilis RNAP, but not the E. coli enzyme, features a cavity that could accommodate an equivalent of the archaeal subunit N (Rpb10 in eukaryotic RNAP I, II and III), but remains unoccupied in the present structures. ε may support the structural integrity of RNAP, securing interactions between α, β and β' subunits when β and β' are forced apart by HelD (see below).
The δ subunit consists of a globular N-terminal domain (NTD; residues 1-90), and an intrinsically disordered, highly acidic C-terminal region (CTR; residues 91-173) 7 . δ NTD resides on the surface of RNAP between the β' shelf and jaw (Fig. 1a HelD resembles a two-pronged fork poking into RNAP. In perfect analogy to transcript cleavage factors 28 , one prong, HelD Pike , inserts deeply into the secondary channel, through which substrate NTPs enter the RNAP active site during elongation (Fig. 1a,c). D1/D2 reach around the β2 lobe, positioning the other prong, HelD Bumper , in the main channel where it pushes against the β' clamp, forcing β and β' apart ( Fig.   1a). In the course of HelD engaging RNAP, a large combined surface area (~11,500 Å 2 ; ~8,000 Å 2 with β'; 1,800 Å 2 with β; ~1,700 Å 2 with δ) is buried.
To con rm contacts and the dramatic structural rearrangements triggered by HelD binding, we used RNAP ΔδΔHelD and recombinant δ and HelD to assemble RNAP ΔδΔHelD -δ, RNAP ΔδΔHelD -HelD and RNAP ΔδΔHelD -δ-HelD, and mapped molecular neighborhoods in these complexes and RNAP ΔδΔHelD by CLMS with the heterobifunctional, photoactivatable crosslinker sulfosuccinimidyl 4,4′-azipentanoate (sulfo-SDA; Fig. 4a,b; Extended Data Table 4; Supplementary Table 1). Matching the δ NTD binding site deduced by cryoEM, a short stretch of δ residues crosslinked to the β' jaw in both RNAP ΔδΔHelD -δ (δ Y82,P83,Y85 -β' K1032 ) and RNAP ΔδΔHelD -δ-HelD (δ Y83,Y85,L87,E90 -β' K1032 ). Multiple crosslinks of HelD were identi ed for RNAP ΔδΔHelD -HelD and RNAP ΔδΔHelD -δ-HelD complexes inside the RNAP main channel, along the region connecting the main and secondary channels, and in the active site region, in excellent agreement with our cryoEM structures (Extended Data Fig. 5). RNAP ΔδΔHelD , RNAP ΔδΔHelD -δ and RNAP ΔδΔHelD -HelD yielded many over-length crosslinks when compared to the RNAP-δ-HelD structure (Fig.  4c,d). A speci c set of crosslinks between the β1/2 lobes (residues 146-248) and the β' shelf and jaw (residues 794-1141) represents a conformation in which β and β' approach each other across the main channel unless both δ and HelD are bound to RNAP (Fig. 4e,f). Together, our results demonstrate that HelD interacts with the main and the secondary channels of RNAP and that stable main channel opening depends on the presence of both δ and HelD.
HelD Pike dismantles the RNAP active site and competes with RNA Upon penetrating the secondary channel, HelD Pike locally disrupts the β' bridge helix (BH; between residues 780 and 787) and locks the β' trigger loop (TL; Fig. 5), i.e. key elements that rearrange for nucleotide addition during elongation 29 . While HelD Pike carries negatively charged side chains (D56, D57, E60) at its tip, these residues do not remodel the active site as observed with GreB 28 . Instead, the tip plows through the active site, thereby dismantling it. The β C-terminal clamp is pushed away from the nucleic acids, β switch region 3 (Sw3), which lines the hybrid in the EC, becomes disordered and the active site loop (ASL) is rearranged so that the catalytic Mg 2+ ion is lost (Fig. 5).
RNAP-RNA binary complexes are catalytically active, implying that RNA resides in the active site cavity 8 .
As seen by comparison with an E. coli EC 24  con guration observed in RNAP complexes to date, augmented by more than 20 Å relative to a Mycobacterium smegmatis σ A holoenzyme 30 (Fig. 6b,c). We observed cryoEM density around HelD Bumper that could only be interpreted by the intrinsically disordered, acidic δ CTR . We con rmed direct HelD-δ interaction via δ CTR by analytical SEC; while HelD co-migrated with δ and the complex eluted earlier than the individual proteins ( Fig. 6d), no such interaction was detected with δ NTD (Fig. 6e). Thus, δ CTR supports HelD in its push against the β' clamp by reaching across the main channel and encircling HelD Bumper (Fig.  1a), and occupies regions next to the β subunit where downstream DNA is accommodated in the EC (Fig.  6b). Clearly, binding of HelD Bumper and δ CTR in the main channel is incompatible with DNA occupying this site, explaining why a nucleic acid scaffold fails to bind the RNAP-δ-HelD complex (Extended Data Fig. 1c; Supplementary Data 2).
To further delineate the contributions of δ and HelD to DNA displacement, we conducted band shift assays. HelD displaced about 20 % of DNA from RNAP ΔδΔHelD , while δ led to about 80 % displacement in the absence of HelD (Fig. 6f, lanes 4-6). This nding is consistent with the observation that δ CTR alone can displace RNA or DNA from RNAP, albeit only if present in large excess 7 . Increasing amounts of δ titrated to DNA-bound RNAP ΔδΔHelD in the presence of HelD led to gradual reduction of bound DNA, with essentially all DNA displaced when equimolar amounts of δ relative to RNAP ΔδΔHelD -HelD were added (Fig. 6f, lanes 7-13). Under otherwise identical conditions only ~50 % of the DNA were displaced by addition of δ NTD (Fig. 6f, lane 14). Together, these results underscore the importance of δ in DNA displacement, show that HelD is required to achieve complete DNA release, and support the cooperation of δ CTR and HelD inferred from our structure and CLMS.
Notably, HelD/δ-mediated DNA displacement did not require ATP. Furthermore, comparison with DNAbound UvrD 27 revealed that the template strand would be continuous with a putative HelD-loaded strand, and that conformational changes would be required for HelD to accommodate a DNA strand at D1/D2 in a UvrD-like manner (Fig. 4d). However, as DNA displacement is supported by transcription bubble rewinding 31 , it is unlikely that HelD captures single-stranded DNA at the position revealed in our structure. Our analysis, therefore, indicates that neither DNA binding nor unwinding by HelD is required for RNAP recycling, consistent with lack of helicase activity in isolated HelD 22 .

ATP-dependent HelD release
As HelD completely incapacitates RNAP ( Fig. 5a,b), it has to be released to allow transcription to resume. σ A did not displace HelD in SEC (Expanded Data Fig. 6a). Comparison of UvrD bound to DNA and ADP-Mg 2 F 3 27 showed that the D1/D2 conformation of RNAP-bound HelD is incompatible with ATP binding (Fig. 7a). We thus surmised that ATP-bound HelD may have a lower a nity for RNAP than the apo factor. Consistent with this notion, ATPγS, AMPPNP and, to a somewhat lesser extent, ATP led to release of HelD from RNAP-δ-HelD during SEC, while ADP or AMP had minor effects ( Fig. 7b; Extended Data Fig. 6b).
AMPPNP and ATPγS mimic conditions of constantly high ATP supply, whereas ATP is likely hydrolyzed and separated from RNAP/HelD during SEC, reducing its effect. Although ATP and analogs lead to HelD release, ATP-bound HelD most likely retains physiologically relevant a nity for RNAP, as evident from its ATP-dependent stimulatory effect on transcription 22 .

Dimeric RNAP-δ-HelD
About two thirds of our particle images conformed to dimeric (RNAP-δ-HelD) 2 complexes ( Fig. 7c; Supplementary Data 3), which were not su ciently stable during SEC (Extended Data Fig. 1d). The protomers of the dimeric assembly closely resemble the monomeric RNAP-δ-HelD complex (root-mean-square deviation of 1.2-1.3 Å for 23,360-23,971 pairs of Cα atoms), but elements of the RNAP active site are further remodeled in the dimer (Fig. 5a,b). The HelD-repositioned clamp forms an essential contact region in the dimer, which also sequesters the initiation/elongation factor-binding β ap tip (FT; Fig. 7c).
The dimeric RNAP-δ-HelD complex shows striking resemblance to the hibernating dimeric eukaryotic RNAP I 32-34 , with analogous regions contributing to the dimer interfaces (Fig. 7d). These observations suggest that, like the RNAP I dimer, dimeric RNAP-δ-HelD represents a dormant state.

Discussion
Results of this and the accompanying reports (Newing et al., submitted; Kuba et al., submitted) show that HelD mounts a two-pronged attack at the RNAP main and secondary channels. Both B. subtilis and distantly related M. smegmatis HelD pinch RNAP around the BH, widen the main and RNA exit channels to provide escape routes for DNA and RNA, and displace the bound nucleic acids. However, the exact implementations of this conserved mechanism are distinct. B. subtilis HelD uses similarly sized arms to penetrate deeply into the channels, with δ playing a supporting role. δ NTD aids the main channel opening, whereas δ CTR may support HelD recruitment and guide HelD Bumper into the main channel to avoid topological trapping of DNA. In contrast, M. smegmatis HelD has evolved a branched main channel arm that functionally compensates for the absence of δ and for a rudimentary secondary channel arm, which merely helps HelD anchoring on RNAP. We presume that the large surface area buried upon RNAP-HelD complex formation, rather than HelD ATPase, provides the driving force for the dramatic RNAP opening.
To engage RNAP, HelD reaches around the β2 lobe, a mode of attack that is not possible with RNAPs containing a β' lineage-speci c insertion, SI3, stacked onto the β2 lobe, such as E. coli (Extended Data Fig.  7a). Consistently, E. coli does not encode HelD, and a distantly related ATPase, RapA, has been proposed to aid RNAP recycling 15 . Unlike HelD, RapA binds near the RNA exit tunnel and does not induce major conformational changes in the EC (Extended Data Fig. 7b). Instead, RapA is thought to rescue ECs by promoting backtracking 35 . Alternative recycling mechanisms likely exist in SI3-containing species. Indeed, E. coli DksA has recently been proposed to remove RNAP from nucleic acids 36 . DksA binds in the secondary channel using a Gre-like coiled-coil 37 , induces conformational changes in RNAP 38 , albeit less dramatic than HelD, and is present only in bacteria that have SI3 39 .
The HelD/δ-dependent recycling mechanism uncovered here represents a marvelously simple, direct and effective way of recovering RNAP from virtually any state trapped post-termination. However, RNAP is truly recycled only when (i) HelD is released and (ii) cellular conditions support robust RNA synthesis. We show that HelD is released by ATP ( Fig. 7b; Extended Data Fig. 6), suggesting that high levels of ATP could help prevent HelD from trapping RNAP in an inactivated complex during exponential growth. Noteworthy, both B. subtilis and M. smegmatis HelDs cannot bind ATP when fully engaged with RNAP, suggesting that intrinsically timed isomerization into a less engaged conformation must precede ATP binding and release from RNAP. With δ CTR destabilized after HelD release, σ could regenerate ready-to-act holoenzyme.
When cells sporulate during stationary phase, conversely, the levels of ATP are low 40 , transcription is limited, HelD levels match those of RNAP 23 , and HelD is thus expected to remain bound to RNAP. Given that HelD locks RNAP in an inactive state, could it be used to store RNAP until the conditions improve? Intriguingly, we observed RNAP-HelD dimers, suggesting that HelD/δ can promote RNAP hibernation that may be essential for fast RNAP recovery, in line with observations that overexpression of HelD enhances sporulation 41 and deletions of HelD, δ or both prolong the lag phase 22 . E. coli RNAP core also readily forms dimers 9 , and an increased propensity for dimerization and other structural aspects of the RNAP-δ-HelD complex resemble features of the hibernating eukaryotic RNAP I (Fig. 7c, Availability of active RNAPs and ribosomes is directly linked to cellular growth and their homeostasis is thus essential for optimal tness. Syntheses of RNA and protein consume large quantities of ATP and GTP, which fuel many cellular engines and serve as reporters of energy status. Our study suggests that the ribosome and RNAP use analogous strategies to decide whether to hibernate during famine or engage in active polymerization when nutrients are plentiful. Similarly to HelD-trapped RNAP dimers, 100S ribosome dimers are stabilized by hibernation promoting factors (HPF), which also occlude the binding sites for the mRNA template and A-and P-site tRNAs [42][43][44] . Ribosome revival is mediated by evolutionarily conserved GTPases, such as stress-induced H X 45,46 or housekeeping EF-G and ribosome recycling factor (RRF), which split the hibernating dimers into 70S monomers in a GTP-dependent fashion 47 , or recycle ribosomes after translation termination 45,48 . Although it is possible that another factor is involved in RNAP reactivation, our results hint that HelD possesses both the post-termination recycling and dimerpromoting activities (Fig. 8).
This one-step mode of regulation is more in line with direct sensing of nucleotides used by RNAP, whereas the ribosome instead relies on a set of translation GTPases. For example, transcription initiation in E. coli is adjusted to the growth rate by RNAP binding directly to ATP/GTP in the active site or to the stringent response alarmone (p)ppGpp in an allosteric site 49 . By contrast, the GTP/(p)ppGpp ratio is conveyed to the ribosome by translation initiation factor 2 (IF2), which acts as a metabolic sensor that switches between an active GTP-bound form and an inactive (p)ppGpp-bound form 50 . In either case, the synthetic output is feedback-controlled to ensure optimal tness and avoid waste of precious resources.
This and the accompanying studies present a hitherto unrecognized transcription recycling system that underpins genome integrity and persistence during periods of dormancy. In our model, reservoirs of active RNAP are controlled by HelD, which directly senses cellular energy charge and may rescue trapped RNAP during fast growth, promote RNAP hibernation during slow growth, and enable e cient RNAP recovery upon shift to a nutrient-rich environment (Fig. 8)  Crosslinking/mass spectrometry Sulfo-SDA predominantly establishes lysine-X crosslinks through a primary amine-reactive moiety on one side and a UV-activatable moiety on the other (theoretical crosslinking limit 25 Å). Sulfo-SDA was prepared at 3 mg/ml in 20 mM HEPES-NaOH, 5 mM Mg(OAc) 2 , 300 mM NaCl, 5 mM DTT, 5% (v/v) glycerol, pH 8.0 immediately prior to addition of RNAP ΔδΔHelD , RNAP ΔδΔHelD -δ, RNAP ΔδΔHelD -HelD or RNAP ΔδΔHelD -δ-HelD (protein:sulfo-SDA 1:3 [w/w]). Samples were incubated on ice for two hours and then irradiated in a thin lm using 365 nm UV irradiation (UVP CL-1000 UV Crosslinker, UVP Inc.) for 20 min on ice (5 cm distance from UV-A lamp). The crosslinked samples were separated by 4-12 % BIS-TRIS NuPAGE, gel bands corresponding to crosslinked monomeric complexes were excised and digested in-gel as described previously 52 Resulting peptides were desalted using C18 StageTips 53 .
10 % of each sample were analyzed by LC-MS/MS without fractionation, the remaining 90 % were fractionated using SEC on a Superdex Peptide 3.2/300 column (GE Healthcare) in 30 % (v/v) acetonitrile, 0.1 % (v/v) tri uoroacetic acid at a ow rate of 10 µl/min to enrich for crosslinked peptides 54 . The rst six peptide-containing fractions (50 μl each) were collected, solvent was removed using a vacuum concentrator and the fractions were analyzed by LC-MS/MS on an Orbitrap Fusion Lumos Tribrid mass spectrometer (Thermo Fisher Scienti c), connected to an Ultimate 3000 RSLCnano system (Dionex, Thermo Fisher Scienti c).
The non-fractionated samples were injected onto a 50 cm EASY-Spray C18 LC column (Thermo Fisher Scienti c) operated at 50 °C. Peptides were separated using a linear gradient going from 2 % mobile formic acid) at a ow rate of 0.3 μl/min over 110 minutes, followed by a linear increase from 40 % to 95 % mobile phase B in 11 minutes. Eluted peptides were ionized by an EASY-Spray source (Thermo Fisher Scienti c) and MS data were acquired in the data-dependent mode with the top-speed option. For each three-second acquisition cycle, the full scan mass spectrum was recorded in the Orbitrap with a resolution of 120,000. The ions with a charge state from 3+ to 7+ were isolated and fragmented using higher-energy collisional dissociation (HCD) with 30 % collision energy. The fragmentation spectra were then recorded in the Orbitrap with a resolution of 50,000. Dynamic exclusion was enabled with single repeat count and 60 s exclusion duration. SEC fractions were analyzed using an identical LC-MS/MS setup. Peptides were separated by applying a gradient ranging from 2 % to 45 % mobile phase B (optimized for each fraction) over 90 min, followed by ramping up mobile phase B to 55 % and 95 % within 2.5 min each. For each three-second data-dependent MS acquisition cycle, the full scan mass spectrum was recorded in the Orbitrap with a resolution of 120,000. The ions with a charge state from 3+ to 7+ were isolated and fragmented using HCD. For each isolated precursor, one of three collision energy settings (26 %, 28 % or 30 %) was selected for fragmentation using a data-dependent decision tree based on the m/z and charge of the precursor. The fragmentation spectra were recorded in the Orbitrap with a resolution of 50,000. Dynamic exclusion was enabled with single repeat count and 60 s exclusion duration.
LC-MS/MS data generated from the four complexes were processed separately. MS2 peak lists were generated from the raw MS data les using the MSConvert module in ProteoWizard (version 3.0.11729). The default parameters were applied, except that Top MS/MS Peaks per 100 Da was set to 20 and the denoising function was enabled. Precursor and fragment m/z values were recalibrated. Identi cation of crosslinked peptides was carried out using xiSEARCH software (https://www.rappsilberlab.org/software/xisearch; version 1.7.4) 55 . For RNAP ΔδΔHelD , peak lists were searched against the sequence and the reversed sequence of RNAP subunits (α, β, β' and ε) and two copuri ed proteins, σ A and σ B . For RNAP ΔδΔHelD -δ, RNAP ΔδΔHelD -HelD and RNAP ΔδΔHelD -δ-HelD samples, protein sequences of δ, HelD or both were included in the database. The following parameters were applied for the search: MS accuracy = 4 ppm; MS2 accuracy = 8 ppm; enzyme = trypsin (with full tryptic speci city); allowed number of missed cleavages = 2; missing monoisotopic peak = 2; crosslinker = sulfo-SDA (the reaction speci city for sulfo-SDA was assumed to be for lysine, serine, threonine, tyrosine and protein N termini on the NHS ester end, and any amino acid residue for the diazirine end); xed modi cations = carbamidomethylation on cysteine; variable modi cations = oxidation on methionine and sulfo-SDA loop link. Identi ed crosslinked peptide candidates were ltered using xiFDR 56 . A false discovery rate of 5 % on residue-pair level was applied with the "boost between" option selected.
Crosslinked residue pairs identi ed from the four complexes are summarized in Extended Data Table 4 and Supplementary Table 1 Fractions containing RNAP, δ and HelD were pooled and concentrated to approximately 5 mg/ml. Immediately before preparation of the grids, the sample was supplemented with 0.15 % (w/v) noctylglucoside. 3.8 µl of the nal mixture were spotted on plasma-treated Quantifoil R1/2 holey carbon grids at 10 °C/100 % humidity, and plunged into liquid ethane using a FEI Vitrobot Mark IV. Image acquisition was conducted on a FEI Titan Krios G3i (300 kV) with a Falcon 3EC camera at a nominal magni cation of 92,000 in counting mode using EPU software (Thermo Fisher Scienti c) with a calibrated pixel size of 0.832 Å. A total electron dose of 40 e/Å 2 was accumulated over an exposure time of 36 s. Movie alignment was done with MotionCor2 57 using 5x5 patches followed by ctf estimation with Gctf 58 .
All following image analysis steps were done with cryoSPARC 59 . Class averages of manually selected particles were used to generate an initial template for reference-based particle picking from 9,127 micrographs. Particle images were extracted with a box size of 440 and binned to 110 for initial analysis. Ab initio reconstruction using a small subset of particles was conducted to generate an initial 3D reference for 3D heterogeneous re nement. The dataset was iteratively classi ed into two well-resolved populations representing monomeric and dimeric RNAP-δ-HelD. Selected particles were re-extracted with a box of 220 and again classi ed in 3D to further clean the dataset. Finally, selected particle images were re-extracted with a box of 280 (1.3 Å/px) and subjected to local re nement using a generously enlarged soft-mask for monomeric or dimeric RNAP-δ-HelD. Local re nement of the dimer particles using the monomeric mask was conducted as a control to trace differences of RNAP-δ-HelD in the authentic monomer and dimer structures. After per-particle CTF correction, non-uniform re nement was applied to generate the nal reconstructions.

Model building and re nement
The nal cryoEM map for the dimeric RNAP-δ-HelD complex (Extended Data Fig. 3) was used for initial model building. Coordinates of M. smegmatis RNAP α, β and β' subunits (PDB ID 5VI8) 60 were docked into the cryoEM map using Coot 61 . Modeling of δ was based on the NMR structure of B. subtilis δ (PDB ID 2M4K) 62 . Modeling of ε was supported by the structure of YkzG from Geobacillus stearothermophilus (PDB ID 4NJC) 19 . Model building of HelD was supported by the structure of UvrD helicase from E. coli (PDB ID 3LFU) 63 as well as the C-terminal domain of a putative DNA helicase from Lactobacillus plantarun (PDB ID 3DMN). The subunits were manually rebuilt into the cryoEM map. The model was completed and manually adjusted residue-by-residue, supported by real space re nement in Coot. The manually built model was re ned against the cryoEM map using the real space re nement protocol in PHENIX 64 . Model building of the monomeric complex was done in the same way but starting with a model of half of the dimeric complex. The structures were evaluated with Molprobity 65 . Structure gures were prepared using PyMOL (Version 1.8 Schrödinger, LLC).

Structure comparisons
Structures were compared by global superposition of complex structures or by superposition of selected subunits in complexes using the "secondary structure matching" algorithm implemented in Coot or the "align" algorithm implemented in PyMOL.
Size exclusion chromatography/multi-angle light scattering SEC/MALS analysis was performed on an HPLC system (Agilent) coupled to mini DAWN TREOS multiangle light scattering and RefractoMax 520 refractive index detectors (Wyatt Technology). RNAP-δ-HelD complex was assembled as for cryoEM. 60 μl of the sample at 1 mg/ml were chromatographed on a Superose 6 Increase 10/300 column (GE Healthcare) in buffer H supplemented with 0.02 % (w/v) NaN 3 at 18 °C with a owrate of 0.6 ml/min. Data were analyzed with the ASTRA 6.1 software (Wyatt Technology) using monomeric bovine serum albumin (Sigma-Aldrich) as a reference.
Interaction assays HelD interactions with δ or δ NTD were analyzed by analytical SEC. 21 µM HelD and 42 µM δ or δ NTD were mixed in 20 mM HEPES-NaOH, 50 mM NaCl, 1 mM DTT, pH 7.5, and incubated for 10 min at room temperature. 50 µl of the samples were loaded on a Superdex S200 Increase PC 3.

Declarations
Reporting summary Further information on research design is available in the Nature Research Reporting Summary linked to this paper.

Data availability
CryoEM maps were deposited in the Electron Microscopy Data Bank (https://www.ebi.ac.uk/pdbe/emdb/) under accession codes EMD-11104 (monomeric RNAP-δ-HelD) and EMD-11105 (dimeric RNAP-δ-HelD), and will be released upon publication. Structure coordinates have been deposited in the RCSB Protein Data Bank (https://www.rcsb.org/) with accession codes 6ZCA (monomeric RNAP-δ-HelD) and 6ZFB (dimeric RNAP-δ-HelD), and will be released upon publication. CLMS data have been deposited in jPOST (https://jpostdb.org/) with accession code JPST000858/PXD019437 (https://repository.jpostdb.org/preview/11873847225ed4c61c43749; access key 3884) and will be released upon publication. All other data supporting the ndings of this study are available from the corresponding author on request. Color coding in all gures, unless otherwise noted: α1, dark gray; α2, gray; β, black; β', light gray; β' clamp, Page 23/35 violet; ε, lime green; δ, slate blue; HelD, red. b, Comparison to an E. coli EC (PDB ID 6ALH), illustrating dramatic widening of the main channel in RNAP-δ-HelD. ω, cyan; template (t) DNA, brown; non-template (nt) DNA, beige; RNA, gold. c, Comparison to an E. coli GreB-modi ed EC (PDB ID 6RIN), illustrating similar secondary channel invasion by coiled-coil elements in GreB and HelD.  HelD architecture. a, Cartoon plot of HelD colored by domains (for color-coding see legend). Numbers refer to domain borders. b, Comparison of HelDNTR to GreB (PDB ID 6RIN) reveals similar topology of the coiled-coils, which insert into the secondary channel, and the globular domains; in GreB, the latter is responsible for high-a nity binding to the RNAP β' rim helices. HelDNTR and GreB are rainbow-colored (blue, N-termini; red, C-termini) as indicated in the legend. Numbers refer to domain borders. c, Comparison of NTPase domains in HelD and in E. coli UvrD (PDB ID 2IS6). The D1-D2 regions are rainbow-colored (blue, N-termini; red, C-termini) as indicated in the legends. Neighboring and inserted regions (Ins), gray. Numbers refer to domain borders. d, UvrD-bound DNA (dark and light green; PDB ID 2IS6) and nucleic acid scaffold from the E. coli EC (PDB ID 6ALH) transferred onto the RNAP-δ-HelD complex by superpositioning of the UvrD NTPase domains on HelD and of the β subunits, respectively.     indicate that the respective factor may be released at the respective step. If the factors remain after termination, NusG will likely be displaced by HelD-induced main channel opening, while the NusA binding site is sequestered in hibernating RNAP-HelD.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.