Fidaxomicin jams M. tuberculosis RNA polymerase motions needed for initiation via RbpA contacts

Fidaxomicin (Fdx) is an antimicrobial RNA polymerase (RNAP) inhibitor highly effective against Mycobacterium tuberculosis RNAP in vitro, but clinical use of Fdx is limited to treating Clostridium difficile intestinal infections due to poor absorption. To enable structure-guided optimization of Fdx to treat tuberculosis, we report the 3.4 Å cryo-electron microscopy structure of a complete M. tuberculosis RNAP holoenzyme in complex with Fdx. We find that the actinobacteria general transcription factor RbpA contacts fidaxomycin and explains its strong effect on M. tuberculosis. We present additional structures that define conformational states of M. tuberculosis RNAP between the free apo-holenzyme and the promoter-engaged open complex ready for transcription. The results establish that Fdx acts like a doorstop to jam the enzyme in an open state, preventing the motions necessary to secure promoter DNA in the active site. Our results provide a structural platform to guide development of anti-tuberculosis antimicrobials based on Fdx.


Introduction
The bacterial RNA polymerase (RNAP) is a proven target for antibiotics. The rifamycin (Rif) class of antibiotics, which inhibit RNAP function, is a lynchpin of modern tuberculosis (TB) treatment (Chakraborty and Rhee, 2015). TB, caused by the infectious agent Mycobacterium tuberculosis (Mtb), is responsible for almost 2 million deaths a year. It is estimated that one third of the world is infected. Mortality from TB is increasing, partly due to the emergence of strains resistant to Rifs (Rif R ) (Salinas et al., 2016). Hence, additional antibiotics against Rif R Mtb are needed. Fidaxomicin (Fdx; also known as Dificimicin, lipiarmycin, or tiacumicin), an antimicrobial in clinical use against Clostridium difficile (Cdf) infection (Venugopal and Johnson, 2012), functions by inhibiting the bacterial RNAP (Talpaert et al., 1975). Fdx targets the RNAP 'switch region', a determinant for RNAP inhibition that is distinct from the Rif binding pocket (Srivastava et al., 2011), and Fdx does not exhibit cross-resistance with Rif (Gualtieri et al., 2009;Kurabachew et al., 2008;O'Neill et al., 2000). The switch region sits at the base of the mobile RNAP clamp domain and, like a hinge, controls motions of the clamp crucial for DNA loading into the RNAP activesite cleft and maintaining the melted DNA in the channel (Chakraborty et al., 2012;Feklistov et al., 2017). Fdx is a narrow spectrum antibiotic that inhibits Gram-positive anaerobes and mycobacteria (including Mtb) much more potently than Gram-negative bacteria (Kurabachew et al., 2008;Srivastava et al., 2011), but the clinical use of Fdx is limited to intestinal infections due to poor bioavailability (Venugopal and Johnson, 2012). Modifying Fdx to address this limitation requires understanding the structural and mechanistic basis for Fdx inhibition, which is heretofore unknown. Here, we used singleparticle cryo-electron microscopy (cryo-EM) to determine structures of Mtb transcription initiation complexes in three distinct conformational states, including a complex with Fdx at an overall resolution of 3.4 Å. The results define the molecular interactions of Mtb RNAP with Fdx as well as the mechanistic basis of inhibition, and establish that RbpA, an Actinobacteria-specific general transcription factor (GTF), is crucial to the sensitivity of Mtb to Fdx.

Fdx potently inhibits mycobacterial TICs in vitro
Fdx has potent inhibitory activity against multi-drug-resistant Mtb cells and the in vivo target is the RNAP (Kurabachew et al., 2008). To our knowledge, the in vitro activity of Fdx against mycobacterial RNAPs have not been reported. RbpA, essential in Mtb, is a component of transcription initiation complexes (TICs) that tightly binds the primary promoter specificity s A subunit of the RNAP holoenzyme (holo) (Bortoluzzi et al., 2013;Forti et al., 2011;Hubin et al., 2017a;Tabib-Salazar et al., 2013). We therefore compared Fdx inhibition of mycobacterial RNAPs containing core RNAP combined with s A (s A -holo) and RbpA with inhibition of Escherichia coli (Eco) s 70 -holo using a quantitative abortive initiation assay (Davis et al., 2015). Fdx inhibited Mtb and

M. smegmatis (Msm) transcription at sub-µM concentrations, whereas inhibition of an
Mtb TIC containing Fdx-resistant (Fdx R ) RNAP (b Q1054H ) (Kurabachew et al., 2008) required a nearly two orders of magnitude higher concentration of Fdx. Eco RNAP was

Cryo-EM structure of the Fdx/RbpA/s A -holo complex
We used single-particle cryo-EM to examine the complex of Mtb RbpA/s A -holo with and without Fdx ( Figure 1B). Preliminary analyses revealed that the particles were prone to oligomerization, which was reduced upon addition of an upstream-fork (us-fork) junction promoter DNA fragment ( Figure 1C). We sorted nearly 600,000-cryo-EM images of individual particles into two distinct classes, each arising from approximately half of the particles (Figure 1 -figure supplement 2).
The first class comprised Mtb RbpA/s A -holo with one us-fork promoter fragment and bound to Fdx. The cryo-EM density map was computed to a nominal resolution of 3.4 Å ( Figure 1D, Figure 1 -figure supplement 3, Supplementary file 1). The us-fork promoter fragment was bound outside the RNAP active site cleft, as expected, with the -35 and -10 promoter elements engaged with the s A 4 and s A 2 domains, respectively ( Figure 1D). Local resolution calculations (Cardone et al., 2013) indicated that the central core of the structure, including the Fdx binding determinant and the bound Fdx, was determined to 2.9 -3.4 Å resolution ( Figure 1E).

Cryo-EM structure of a Mtb RPo mimic
The second class comprised Mtb RbpA/s A -holo bound to two us-fork promoter fragments but without Fdx to a nominal resolution of 3.3 Å ( with the 5-nucleotide 3'-overhang ( Figure 1C) engaged with the RNAP active site (as the template strand) like previously characterized 3'-tailed templates (Gnatt et al., 2001;Kadesch and Chamberlin, 1982). Local resolution calculations (Cardone et al., 2013) indicated that the central core of the structure was determined to between 2.8 -3.2 Å resolution ( Figure 2B). The overall conformation of this protein complex and its engagement with the upstream and downstream DNA fragments was very similar to the crystal structure of a full Msm open promoter complex (RPo) (Hubin et al., 2017b) with one exception (see below). We will therefore call this complex an Mtb RbpA/RPo mimic.

The RbpA N-terminal tail invades the RNAP active site cleft
RbpA comprises four structural elements, the N-terminal tail (NTT), the core domain (CD), the basic linker, and the sigma interacting domain (SID) (Bortoluzzi et al., 2013;Hubin et al., 2017a;Tabib-Salazar et al., 2013). Our previous crystal structures of Msm TICs containing RbpA showed that the RbpA SID interacts with the s A 2 domain, the RbpA BL establishes contacts with the promoter DNA phosphate backbone just upstream of the -10 element, and the RbpA CD interacts with the RNAP b' Zinc-Binding-Domain (ZBD) (Hubin et al., 2017a;2017b). Density for the RbpA NTT (RbpA residues 1-25) was never observed in the crystal structures and was presumed to be disordered. In striking contrast to the crystal structures, both cryo-EM structures reveal density for the RbpA NTT , which unexpectedly threads into the RNAP active site cleft between the ZBD and s A 4 domains, snakes through a narrow channel towards the RNAP active site Mg 2+ ( Figure 3). On its path, conserved residues of the RbpA NTT interact with conserved residues of the s-finger (s 3.2 -linker) on one wall of the channel, and with conserved residues of the ZBD and b'lid on the other wall ( Figure 3C).
The most N-terminal RbpA residues visible in the cryo-EM structures (A2 in the Fdx complex, R4 in the RPo) sit near the tip of the s-finger where it makes its closest approach to the RNAP active site, too far (25 Å) to play a direct role in RNAP catalytic activity or substrate binding. The s-finger plays an indirect role in transcription initiation, stimulating de novo phosphodiester bond formation by helping to position the t-strand DNA (Kulbachinskiy and Mustaev, 2006;Zhang et al., 2012). The s-finger is also a major determinant of abortive initiation, playing a direct role in initiation and promoter escape by physically blocking the path of the elongating RNA transcript before s release (Cashel et al., 2003;Murakami et al., 2002). The intimate association of the RbpA NTT with the s-finger ( Figure 3C) suggests that the RbpA NTT also plays a role in these processes of Mtb RNAP initiation. This is consistent with our findings that the RbpA NTT does not strongly affect RPo formation but plays a significant role in in vivo gene expression in Msm (Hubin et al., 2017a). This location of the RbpA NTT explains the high Fdx sensitivity of Mtb RNAP (see below).

The RbpA NTT is critical for Fdx potency against Mtb RNAP in vitro and in vivo
In addition to the b and b' subunits, Fdx interacts with residues of the s-finger (D424 and V445; Figures  RbpA is essential in Mtb and Msm, but strains carrying RbpA D NTT are viable (Hubin et al., 2017a), allowing us to test the role of the RbpA NTT in Fdx growth inhibition of Msm cells. We performed zone of inhibition assays on two Msm strains that are isogenic except one harbors wild-type RbpA (RbpA wt ) and the other RbpA D NTT (Hubin et al., 2017a). The Msm RbpA ∆NTT strain grew considerably slower on plates, taking approximately twice the time as the wild-type Msm to reach confluency. Despite the growth defect, the RbpA D NTT strain was significantly less sensitive to Fdx ( Figure 4D). Discs soaked with up to 250 µM Fdx did not produce inhibition zones with RbpA D NTT but inhibition zones were apparent with RbpA wt . At 500 µM Fdx, the inhibition zone for RbpA D NTT was significantly smaller than for RbpA wt . By contrast, 860 µM streptomycin, a protein synthesis inhibitor, produced equal inhibition zones for the RbpA wt and RbpA D NTT strains. We conclude that the essential role of RbpA in Mtb transcription is key to the relatively high sensitivity of Mtb cells to Fdx.

Fdx traps an open-clamp conformation
The RNAP switch regions are thought to act as hinges connecting the mobile clamp domain to the rest of the RNAP (Gnatt et al., 2001;Lane and Darst, 2010). Bacterial Alignment of the structures revealed that the clamp conformational changes could be characterized as rigid body rotations about a common rotation axis ( Figure 5B). Assigning a clamp rotation angle of 0° (closed clamp) to the RPo structure (blue, Figure 5B), the RbpA/s A -holo clamp is rotated open by about 12° (green, Figure 5B). Because this complex is not interacting with any other ligands that might be expected to alter the clamp conformation (such as Fdx or DNA), we will refer to this as the 'relaxed' clamp conformation. The two Fdx-bound complexes, with or without usfork DNA, show further opening of the clamp (14° and 15°, respectively; orange and red in Figure 5B).

Fdx acts like a doorstop to stabilize the open-clamp conformation
In the high-resolution Fdx/TIC structure ( Figure 1D

Fdx inhibits RNAP by trapping an open-clamp conformation
Clamp dynamics play multiple important roles in the transcription cycle. Motions of the clamp module and the role of the switch regions as hinges were first noted by comparing crystal structures of free RNAPs Zhang et al., 1999) with the crystal structure of an elongation complex containing template DNA and RNA transcript (Gnatt et al., 2001). Binding of the downstream duplex DNA and RNA/DNA hybrid in the RNAP active-site cleft was proposed to close the clamp around the nucleic acids, explaining the high processivity of the transcription elongation complex.

Summary
Our results establish the molecular details of Fdx interactions with the bacterial RNAP ( Figures 4A, B) and a mechanism of action for Fdx ( Figures 5C, D). Crucially, the essential actinobacterial GTF RbpA is responsible for the high sensitivity of Mtb RNAP to Fdx both in vitro ( Figure 4C) and in vivo ( Figure 4D). This new knowledge provides a structural platform for the development of antimicrobials based on Fdx and underscores the need to define structure-activity relationships of drug leads using near-native states, in this case using cryo-EM with the RbpA/s A -holo complex to guide development of effective Mtb treatments.
Samples for Cryo-EM particle preparation used Mtb His-tagged-σ A and RbpA coexpressed and purified as previously described (Hubin et al., 2015;2017a). To compare Fdx sensitivity of full-length RbpA and RbpA DNTT , these proteins, and Mtb His-tagged-σ A were expressed separately and purified as described previously (Hubin et al., 2015;2017a). Briefly, Rosetta-2 cells were co-transformed with pET plasmids expressing Mtb σ A (His-tagged) and RbpA and induced with 0.5 mM IPTG at 30°C for 4 hours. Clarified lysates was subjected to Ni 2+ affinity, removal of the His-tag, a second Ni 2+ affinity (collecting the flow through this time) and size exclusion chromatography.
Mtb RNAP was expressed and purified as previously described for Mbo and Msm RNAPs (

Cryo-EM grid preparation
C-flat CAu-1.2/1.3 400 mesh gold grids were glow-discharged for 20 s prior to the application of 3.5 µl of the sample (4.0-6.0 mg/ml protein concentration). After blotting for 3-4.5 s, the grids were plunge-frozen in liquid ethane using an FEI Vitrobot Mark IV (FEI, Hillsboro, OR) with 100% chamber humidity at 22 °C.

Cryo-EM data acquisition and processing
Fdx/RbpA/σ A -holo/us-fork. The grids were imaged using a 300 keV Titan Krios (FEI) equipped with a K2 Summit direct electron detector (Gatan). Images were recorded with Leginon (Nicholson et al., 2010) in counting mode with a pixel size of 1.1 Å and a defocus range of 0.8 μm to 1.8 μm. Data were collected with a dose of 8 electrons/px/s. Images were recorded over a 10 second exposure with 0.2 second frames (50 total frames) to give a total dose of 66 electrons/Å 2 . Dose-fractionated subframes were aligned and summed using MotionCor2 (Zheng et al., 2017) and subsequent doseweighting was applied to each image. The contrast transfer function was estimated for each summed image using Gctf (Zhang, 2016). From the summed images, Gautomatch (developed by K. Zhang, MRC Laboratory of Molecular Biology, Cambridge, UK, http://www.mrc-lmb.cam.ac.uk/kzhang/Gautomatch) was used to pick particles with an auto-generated template. Autopicked particles were manually inspected, then subjected to 2D classification in cryoSPARC (Davis et al., 2015) specifying 50 classes. Poorly populated and dimer classes were removed, resulting in a dataset of 582,169 particles.
A subset of the dataset was used to generate an initial model of the complex in cryoSPARC (ab-initio reconstruction). Using the ab-initio model (low-pass filtered to 30 Å-resolution), particles were 3D classified into two classes using cryoSPARC heterogenous refinement. CryoSPARC homogenous refinement was performed for each class using the class map and corresponding particles, yielding two structures with different clamp conformations: open (Fdx/RbpA/σ A -holo/us-fork; Figure 1D) and closed [RbpA/σ A -holo/(us-fork) 2 ; Figure 2A] Mtb RbpA/σ A -holo. The grids were imaged using a 200 keV Talos Arctica (FEI) equipped with a K2 Summit direct electron detector (Gatan). Images were recorded with Serial EM (Mastronarde, 2005) in super-resolution counting mode with a superresolution pixel size of 0.75 Å and a defocus range of 0.8 μm to 2.4 μm. Data were collected with a dose of 8 electrons/px/s. Images were recorded over a 15 second exposure using 0.3 second subframes (50 total frames) to give a total dose of 53 electrons/Å 2 . Dose-fractionated subframes were 2 x 2 binned (giving a pixel size of 1.5 Å), aligned and summed using Unblur (Grant and Grigorieff, 2015). The contrast transfer function was estimated for each summed image using Gctf (Zhang, 2016).
From the summed images, Gautomatch (developed by K. Zhang, MRC Laboratory of Molecular Biology, Cambridge, UK, http://www.mrc-lmb.cam.ac.uk/kzhang/Gautomatch) was used to pick particles with an auto-generated template. Autopicked particles were manually inspected, then subjected to 2D classification in RELION (Scheres, 2012) specifying 100 classes. Poorly populated classes were removed, resulting in a dataset of 289,154 particles. These particles were individually aligned across movie frames and dose-weighted using direct-detector-align_lmbfgs software to generate "polished" particles (Rubinstein and Brubaker, 2015). A subset of the dataset was used to generate an initial model of the complex in cryoSPARC (ab-initio reconstruction).
"Polished" particles were 3D auto-refined in RELION using this ab-initio 3D template (low-pass filtered to 60 Å-resolution). RELION 3D classification into 2 classes was performed on the particles using the refined map and alignment angles. Among the 3D classes, the best-resolved class, containing 87,657 particles, was 3D auto-refined and post-processed in RELION. The overall resolution of this class was 6.9Å (before postprocessing) and 5.2Å (after post-processing). Subsequent 3D classification did not improve resolution of this class.
Fdx/RbpA/σ A -holo. The same procedure as described above for Mtb RbpA/s A -holo was used. After RELION 2D classification, poorly populated classes were removed, resulting in a dataset of 63,839 particles. In the end, the best-resolved 3D class, containing 21,115 particles, was 3D auto-refined and post-processed in RELION. The overall resolution of this class was 8.1Å (before post-processing) and 6.5Å (after postprocessing).

Model building and refinement
To build initial models of the protein components of the complex, Msm RbpA/σ A -holo/usfork structure (PDB ID 5TWI) (Hubin et al., 2017a) was manually fit into the cryo-EM density maps using Chimera (Pettersen et al., 2004) and real-space refined using Phenix (Adams et al., 2010). In the real-space refinement, domains of RNAP were rigidbody refined. For the high-resolution structures, the rigid-body refined models were subsequently refined with secondary structure restraints. A model of Fdx was generated from a crystal structure (Serra et al., 2017), edited in Phenix REEL, and refined into the cryo-EM density. Refined models were inspected and modified in Coot (Emsley and Cowtan, 2004) according to cryo-EM maps, followed by further real-space refinement with PHENIX.

Accession Numbers
The   The boxed region is magnified on the right. Density for the RbpA NTT is outlined in red.
35   (D) The core RNAP from the 3.3 Å resolution RbpA/RPo structure is shown as a gray molecular surface but with the closed clamp colored blue. The structure is sliced at the