The antibiotic sorangicin A inhibits promoter DNA unwinding in a Mycobacterium tuberculosis rifampicin-resistant RNA polymerase

Significance

Rifampicin (Rif) is an antibiotic that targets bacterial RNA polymerase (RNAP). The search for new antibiotics against proven targets is one approach to combat multidrug-resistant tuberculosis (TB), a global threat. Here, we characterized the effects of the antibiotic sorangicin A (Sor) on a Rif-resistant (Rif^R) RNAP found in clinically isolated Rif^R strains. Sor binds in the same pocket and inhibits wild-type RNAP by the same mechanism as Rif. However, Sor inhibits Rif^R RNAPs by a distinct mechanism, providing a new target for drug design. Intriguingly, Sor displays a better pharmacokinetic profile compared to rifamycins, which is important for development of inhibitors to treat TB patients suffering from comorbidities. These results inform approaches toward drug development against antibiotic-resistant targets.

Abstract

Rifampicin (Rif) is a first-line therapeutic used to treat the infectious disease tuberculosis (TB), which is caused by the pathogen Mycobacterium tuberculosis (Mtb). The emergence of Rif-resistant (Rif^R) Mtb presents a need for new antibiotics. Rif targets the enzyme RNA polymerase (RNAP). Sorangicin A (Sor) is an unrelated inhibitor that binds in the Rif-binding pocket of RNAP. Sor inhibits a subset of Rif^R RNAPs, including the most prevalent clinical Rif^R RNAP substitution found in Mtb infected patients (S456>L of the β subunit). Here, we present structural and biochemical data demonstrating that Sor inhibits the wild-type Mtb RNAP by a similar mechanism as Rif: by preventing the translocation of very short RNAs. By contrast, Sor inhibits the Rif^R S456L enzyme at an earlier step, preventing the transition of a partially unwound promoter DNA intermediate to the fully opened DNA and blocking the template-strand DNA from reaching the active site in the RNAP catalytic center. By defining template-strand blocking as a mechanism for inhibition, we provide a mechanistic drug target in RNAP. Our finding that Sor inhibits the wild-type and mutant RNAPs through different mechanisms prompts future considerations for designing antibiotics against resistant targets. Also, we show that Sor has a better pharmacokinetic profile than Rif, making it a suitable starting molecule to design drugs to be used for the treatment of TB patients with comorbidities who require multiple medications.

Mycobacterium tuberculosis (Mtb) is the causative agent of the disease tuberculosis (TB), which kills almost 2 million people annually (1). Efforts to eradicate TB are threatened by the increase of multidrug-resistant (MDR) strains of Mtb. Rifampicin (Rif) is a first-line treatment for TB. Resistance to either isoniazid (another first-line antibiotic) or Rif increases the therapeutic regimen from 6 mo to 2 y or more and requires more expensive drugs, often with greater side effects (2). Therefore, there is an urgent need to find new antimicrobials to treat Rif-resistant (Rif^R) TB. Rif targets RNA polymerase (RNAP), the enzyme responsible for all transcription in bacteria. Thus, RNAP is a proven antimicrobial target, supporting the development of new antibiotics that inhibit this enzyme. Especially compelling is the discovery and optimization of antibiotics that inhibit Rif^R Mtb RNAP (reviewed in ref. 3).

Drug–drug interactions (DDIs) with Rif are quite common, which complicates clinical management and treatment strategies for TB patients receiving multiple medications (4). Thus, combination chemotherapy for patients with comorbidities, especially those coinfected with Mtb and HIV, require close monitoring (5). DDIs with Rif are complex and they are mainly driven by the induction of cytochrome P450 3A4 (CYP3A4), which is mediated through activation of the nuclear pregnane X receptor (PXR) (4).

In bacteria, transcription initiation occurs when the RNAP catalytic core enzyme (E, subunit composition α₂ββ′ω, ∼400 kDa) associates with an additional subunit, σ^A, creating the holoenzyme (Eσ^A), which is capable of promoter-specific initiation (6). The holoenzyme performs a number of steps before transitioning to an elongating complex (Fig. 1A). The first step is binding the double-stranded promoter DNA through sequence-specific recognition of promoter sequence elements, most importantly the −35 and −10 elements. Following initial Eσ^A recognition of the double-stranded promoter DNA, the complex isomerizes through a series of intermediates as it unwinds 12 to 14 base pairs (bp) of DNA and positions the template (T)-strand in the RNAP active site to form the transcriptionally competent open promoter complex (RPo) (7–10). Recently, the structure of an Mtb RNAP promoter-melting intermediate containing a partially unwound DNA bubble (8-bp transcription bubble) was determined (11). This structure was proposed to correspond to RP2, a previously observed kinetic intermediate of Mtb RNAP on the same promoter (9, 11). Upon completion of the full transcription bubble, RNAP transitions into an initial transcribing complex (RPitc), characterized by the synthesis of a short RNA transcript. Eventually, RNAP “escapes” the promoter and synthesizes the full-length RNA product in an elongating complex (RPec) (12).

Fig. 1.

Sor and Rif inhibition of WT and S>L mycobacterial RNAPs. (A) Schematic of the steps of bacterial transcription initiation. The cartoon shows steps of bacterial transcription initiation, starting from RP1 to the elongation complex. The core RNAP is a pink circle, the promoter DNA is colored dark gray, and the active site Mg²⁺ is depicted as a yellow sphere. (B) Chemical structures of Rif and Sor. (C) IC₅₀ values for Rif and Sor on Mtb and Msm WT- and S>L-RNAPs calculated from two experiments (example gels in D and SI Appendix, Fig. S1). All reactions contained the transcription factor RbpA. The sequence of the promoter used for these assays is shown in SI Appendix, Fig. S1. (D) Transcriptional profile of Mtb WT (Left) and S456L (Right) RNAPs as a function of antibiotic concentrations. The full-length, 71-nucleotide run-off transcript and the abortive product (pppGpUpU*) are indicated.

The structure of Rif in complex with core RNAP was initially determined by X-ray crystallography using the RNAP from Thermus aquaticus (Taq) (13). That work, in combination with previous biochemical findings, established that Rif inhibits transcription by physically blocking the path of the elongating RNA when the RNA is two to three nucleotides in length, preventing the transition from RPitc to RPec (13, 14).

Sorangicin A (Sor) (15, 16) is a bacterial RNAP inhibitor chemically unrelated to Rif (Fig. 1B). Genetic studies identified extensive overlap in cross-resistance between Rif and Sor, suggesting significant overlap of the Rif- and Sor-binding determinants on RNAP (17–20). Our previous structure of Sor in complex with Taq RNAP showed almost complete overlap; Rif and Sor bound in the identical RNAP β-subunit pocket and contacted essentially identical RNAP amino acid residues (17). However, a subset of Rif^R mutants was Sor-sensitive (Sor^S) (17–20). One of these substituted loci yielding Rif^R but Sor^S is a Ser in the Rif-binding pocket (β-subunit S447 in Mycobacterium smegmatis [Msm], S456 in Mtb, S531 in Escherichia coli [Eco], and S411 in Taq). This residue is frequently substituted with a leucine (S>L) and is the most prevalent substitution in Rif^R Mtb clinical isolates (21). Studies in Eco have shown substitutions of the equivalent S531 to bulkier residues (Y, F, W) also led to Rif^R/Sor^S RNAP (17). However, modeling based on the crystal structure of Sor/RNAP suggested that the S>L (or to any bulky amino acid) RNAP could not accommodate Sor in the observed conformation due to steric clash between the antibiotic and the bulky amino acid side chain. We, therefore, proposed that Sor can bind the S>L substituted RNAP due to Sor’s positional and torsional flexibility. Molecular dynamics simulations and physical molecular models confirmed that Sor has sufficient torsional freedom to adapt to the altered binding pocket caused by the S>L substitution (17). By contrast, Rif is a very rigid molecule with little torsional flexibility to adapt to an altered binding pocket shape (17).

In the present work to elucidate how Sor binds to the S>L-RNAPs, we first determined crystal structures of 1) Msm wild-type (WT) RNAP with Sor, 2) apo (without antibiotic) Msm S447L-RNAP, and 3) Msm S447L-RNAP with Sor. The crystal structures show that Sor can bind the S>L-RNAPs, not because of its torsional flexibility as we previously proposed (17), but because of flexibility in the RNAP fork loop 2 (FL2), a flexible loop that acts as a lid on the Rif pocket, which moves to accommodate the antibiotic.

Unexpectedly, we also observed that Sor inhibited the Mtb S>L-RNAP at an earlier step of transcription initiation than WT-RNAP. We used cryo-electron microscopy (cryo-EM) to visualize the effect of Sor on complexes of Mtb RNAP with a de novo-melted promoter. Cryo-EM structures revealed that Sor permits WT-RNAP to proceed to RPo, but Sor traps the S>L-RNAP at the promoter-melting intermediate RP2 (11). Thus, Sor inhibits the Mtb Rif^R S>L-RNAP by a different mechanism, at an earlier step, than it inhibits the WT-RNAP. We also present data showing Sor has a favorable DDI profile. Altogether, these findings suggest that Sor is a compelling pharmacokinetic candidate for drug design to be used for the treatment of TB against antibiotic-resistant Mtb.

Results

Sor Inhibits Mycobacteria RNAPs.

We performed transcription assays to determine the half-maximum inhibitory concentration (IC₅₀) of Rif and Sor against WT and S447/S456L Msm and Mtb RNAPs (Fig. 1 C and D and SI Appendix, Fig. S1 A and B). Both the Mtb and Msm WT-RNAPs displayed similar sensitivity to each antibiotic, within the low nanomolar range of inhibition (Fig. 1C). As expected, the IC₅₀ of Rif on the S>L-RNAPs was in the high micromolar (124 to 370 µM) range. The Sor IC₅₀ for the WT-RNAPs was very similar to the IC₅₀ for Rif. The Sor IC₅₀ for the S>L-RNAPs also increased compared to the WT-RNAPs, but the increase was orders of magnitude less than for Rif, so the Sor IC₅₀ for the S>L-RNAPs was still in the low micromolar range (Fig. 1C).

Crystal Structures Reveal Sor Binds to Mycobacteria RNAPs in the Rif Pocket.

Previous structures of Rif and related compounds identified the interactions between Rif and various bacterial RNAPs (at resolutions of): Taq (3.1Å) (13), T. thermophilus (2.5 Å) (22), and Eco (3.8 Å) (23). More recently, a series of Mtb RNAP (3.8 to 4.3 Å) (24) and Msm RNAP (3.1 Å) (25) structures in complex with Rif revealed the Rif interactions with RNAP in mycobacteria. In all of these structures, the overall binding of the Rif and the identified interactions are similar to the original structure (13) and explain the broad-spectrum activity of Rif. Here, we determined crystal structures of Msm WT- and S447L-holoenzyme RNAP with RbpA, bound to an upstream fork as previously described (9) (Fig. 2A), with Sor to 3.1- and 3.0-Å resolution, respectively (SI Appendix, Table S1). We compared the changes in the Rif pocket with that of our previous 3.1-Å structure of Msm RNAP with Rif obtained under identical crystallization conditions (25). To distinguish which changes in the structures were a result of the S447L substitution or Sor binding, we also determined the 3.2-Å crystal structure of apo Msm S447L-RNAP (SI Appendix, Table S1). The crystal structure of Msm WT-RNAP with Sor was similar to that of the previous structure of Taq RNAP with Sor (17), confirming that Sor binds in the Rif pocket (Fig. 2B and SI Appendix, Fig. S2A).

Fig. 2.

The structural basis of Sor binding and inhibition of Msm Rif^R RNAP β S447L. (A) Upstream fork-promoter DNA used for crystallization of Msm RNAP/Sor structures. The −10 and −35 promoter elements are colored yellow. The extended −10 element is colored green. (B and D) 2Fo-Fc density maps (blue mesh) of Msm WT (B) or S447L (C and D) RNAPs with superimposed atomic models. The RNAP β subunit is cyan, but FL2 is dark blue. Sor (when present) is green. All structures contained the essential transcription factor RbpA. (B) Msm WT-RNAP with Sor. Amino acids defining the range of the FL2 (E462 and G450) are labeled, as are Sor-interacting amino acids discussed in the Conformational Changes in the Rif Pocket Caused by the S447L Mutation. (C) Msm S447L-RNAP with Sor, illustrating a loss of density for FL2. Notable is the loss of density for L447 and R456, whose interactions with Sor are lost. (D) Msm S447L-RNAP without Sor, illustrating a loss of density for FL2, as well as neighboring loops containing Q484 and P480. (E–H) Msm antibiotic/RNAP crystal structures (E–G) or model (H). The RNAP is shown as a molecular surface, color-coded according to the key on the left, except FL2 is dark blue. The antibiotics are shown as CPK spheres (Sor, green carbon atoms; Rif, yellow carbon atoms). The boxed area is magnified in Insets. ESAs for the antibiotics are shown at the bottom (28). (E) Msm WT-RNAP with Sor. (F) Msm WT-RNAP with Rif (PDB ID code 6CCV) (25). (G) Msm S447L-RNAP with Sor. The ΔESA is the increase in the Sor ESA for the S447L-RNAP vs. WT-RNAP. (H) Model of Msm S447L-RNAP with Rif, generated by superimposing the Rif from 6CCV onto the RNAP structure from F. The ΔESA is the increase in the Rif ESA for the S447L-RNAP model vs. WT-RNAP.

Conformational Changes in the Rif Pocket Caused by the S447L Mutation.

To address why the S447L-RNAP is Rif^R but Sor^S despite both antibiotics binding in the Rif pocket and sharing similar IC₅₀ for WT-RNAPs, we analyzed the Rif pocket in our crystal structures (Fig. 2). The Rif pocket is partially made up of the mobile FL2 (β-subunit residues S447-E462 in Msm, S456-E471 in Mtb; colored dark blue in Fig. 2). In WT-RNAP, three FL2 residues (S447, L449, R456) make direct contacts with Rif (25). These three residues and an additional FL2 residue, G450, also contact Sor (Fig. 2B and SI Appendix, Fig. S2A). In the structure of S447L-RNAP with Sor, the FL2 was displaced and residues 451 to 460 were disordered, resulting in the loss of interactions between L449, G450, and R456 and the antibiotic (Fig. 2C and SI Appendix, Fig. S2A).

The observed disorder of the FL2 could be attributed to the displacement of the bulky S447L by Sor binding, or solely because of the S>L substitution. To determine the cause of the FL2 disorder, we also solved the structure of S447L-RNAP without antibiotic. We found that not only was the FL2 density absent (Fig. 2D) but so were several adjacent loops and domains (SI Appendix, Fig. S3). This observation suggests that Sor binding stabilizes the S447L-RNAP structure. Sor interacts with residues in a loop following FL2 (Q484 and P480; Fig. 2B) that we call the extended FL2 (β-subunit residues 447 to 486; SI Appendix, Fig. S3). That interaction may stabilize not only the Rif pocket but also the nearby domains, including the β′-rim helices, the β′ F-loop, the β′-bridge helix, the β-lobe, the β-protrusion, and a domain of β we now call β-lobe 3 (SI Appendix, Fig. S3). We currently do not have an explanation for how the S>L substitution causes such extensive disordering of these domains but note that these domains are physically connected by contacting each other and/or the FL2 directly.

Previous structures of Eco S531L-RNAP (equivalent to S447L in Msm RNAP) did not show the same disorder of FL2 or other parts of the RNAP (23). The authors also crystallized the Eco S531L-RNAP in complex with a very high concentration of Rif (1 mM). They found that the FL2 became disordered upon Rif binding (similar to our finding with Sor), presumably as a result of the clash between Rif and the leucine substitution. The crystal-packing environments of the Msm RNAP and Eco RNAP crystals are very different; crystal-packing constraints in the Eco RNAP crystals may have stabilized the parts of the RNAP that became disordered in the apo Msm S447L-RNAP structure (SI Appendix, Fig. S3). Alternatively, the Eco and Msm RNAP β and β′ subunits share only 57.8 and 51.4% sequence identity, respectively; the Eco RNAP structure may be intrinsically more stable than the equivalent regions of the Msm RNAP.

This result illustrates how the various modules of Msm RNAP are interconnected in a balanced interaction network that may coordinate movements necessary for the transcription cycle but can be disrupted by a single amino acid substitution.

Comparison of Interactions in Crystal Structures of WT- and S447L-RNAPs with Sor and Rif.

We did not solve the structure of Rif bound to the S447L-RNAP because this enzyme is Rif-resistant and would require extremely high concentrations of Rif (>2 mM to achieve full occupancy in Msm RNAP) to bind, possibly leading to artifacts. To compare contacts that would be lost between Rif vs. Sor in the Msm S447L-RNAP, we modeled Rif into the S447L-RNAP by superimposing our previous structure of Msm WT-RNAP/Rif (Protein Data Bank [PDB] ID code 6CCV) (25) onto the current Msm S447L-RNAP/Sor structure. These analyses reveal that in the context of FL2 disorder, contacts lost with Rif would include nonpolar interactions with S447, L449, and R456 and a hydrogen bond with S447. We note that these observations are largely supported by the structure of Eco S531L-RNAP with Rif (23). Contacts lost with Sor include nonpolar interactions with S447, L449, and G450, as well as a hydrogen bond with R456 (SI Appendix, Fig. S2A).

The effect of these changes in the Rif pocket leads to high Rif^R but maintains modest Sor^S (Fig. 1C). The following observations explain this difference: 1) S447 is involved in nonpolar contacts with both Sor and Rif but makes a hydrogen bond with a hydroxyl (O2) on the naphthalene ring of Rif, which contributes significantly to the binding energy of Rif to RNAP based on structure–function studies of Rif (26). However, the S447L substitution increases the number of nonpolar contacts with Sor due to the branched hydrophobic substitution, which would favor the binding of Sor. 2) A second contact located on FL2 is L449, which makes extensive nonpolar contacts with the polyketide backbone of Sor and the planar rings of Rif (Fig. 2B and SI Appendix, Fig. S2A). Contacts with each antibiotic change significantly in the S447L-RNAP, with L449 not interacting with either antibiotic due to FL2 disorder (Fig. 2 B and C and SI Appendix, Fig. S2A). This interaction may be more critical for Rif binding. In Eco, an L449P (Eco numbering L533P) substitution causes Rif^R, but the bacteria remain Sor^s (17), which is in line with the FL2 movement affecting Rif binding more severely than Sor. 3) A third interaction that is lost because of the FL2 disorder is between R456 and each antibiotic. R456 makes van der Waals and nonpolar interactions with Sor and nonpolar interactions with the five-membered ring of Rif (Fig. 2B and SI Appendix, Fig. S2A). This amino acid cannot be modeled in the structures of S447L-RNAP due to FL2 disorder and explains why the IC₅₀ for Sor increases >30-fold compared to WT mycobacteria RNAPs (Figs. 1C and 2C). The S447L substitution leads to additional losses and gains to the Sor interactions with RNAP, which are illustrated in the ligplots (27) in SI Appendix, Fig. S2A.

Crystal Structures Reveal that FL2 Increases the Exposed Surface Area of Antibiotics.

The interactions lost with Sor due to FL2 disorder are also predicted to be lost with Rif (Fig. 2H). However, the mycobacterial S>L-RNAPs are ∼30- to 120-fold more sensitive to Sor than Rif (Fig. 1C). In addition to the differences in the interactions noted in the previous section, we propose that changes in the exposed-surface area (ESA) of each antibiotic in the S447L-RNAP also contribute to the differences in the sensitivity. ESAs were calculated using the webserver www.ebi.ac.uk/pdbe/prot_int/pistart.html (28). To compare the ESA of Rif when bound to Msm S447L-RNAP vs. WT-RNAP, we modeled Rif into the S447L-RNAP by superimposing our previous structure of Msm WT-RNAP/Rif (PDB ID code 6CCV) (25) onto the current Msm S447L-RNAP/Sor structure. The loss of FL2 interactions increases the Sor ESA by about 32 Ų (Fig. 2 E and G), whereas the increase predicted for Rif is about 74 Ų (Fig. 2 F and H). Moreover, the loss of FL2 results in the exposure of mainly the planar hydrophobic rings of the Rif naphthol moiety (Fig. 2H), which is expected to be very unfavorable. We conclude that the S447L-RNAP was more sensitive to Sor than Rif largely due to 1) loss of a hydrogen bond between S447 and O2, as discussed in Comparison of Interactions in Crystal Structures of WT- and S447L-RNAPs with Sor and Rif, and 2) the entropic cost of exposing the hydrophobic planar rings of Rif caused by the FL2 disorder, a direct result of relative increased ESA/decreased buried-surface area of Rif to that of Sor (Fig. 2H). This analysis of the X-ray crystal structures is relevant for free RNAP binding to Rif or Sor and preengagement to the −10 promoter elements. The remainder of this paper uses cryo-EM to investigate the effects of Sor on WT and S>L Mtb RNAP during de novo promoter melting.

Sor Inhibits S>L-RNAP at an Earlier Step of the Transcription Cycle than WT-RNAP.

Rif binding introduces a steric block to the elongation of the RNA transcript beyond a length of two or three nucleotides (13, 14, 24, 25). Transcription assays in the presence of Rif show a decrease of run-off products (the full-length transcript to the end of the template) along with a concomitant build-up of short RNAs (abortive transcripts; Fig. 1D and SI Appendix, Fig. S1A). Sor binds in the Rif pocket and inhibits extension of the RNA similarly (17) (Fig. 1D). Both Mtb and Msm WT-RNAPs exhibited similar transcription inhibition profiles in response to both Sor and Rif as Eco and Taq WT-RNAPs (13, 17, 25). However, although 1 to 10 μM Sor inhibited run-off products in the S>L-RNAPs, Mtb S456L-RNAP did not produce the characteristic abortive products that were seen with WT-RNAP (Fig. 1D). This unexpected observation suggested that Sor inhibited the Mtb S456L-RNAP by a different mechanism than the WT-RNAP. We did not observe the same result with the Msm S>L-RNAP as we did with the Mtb S>L-RNAP (Fig. 1D and SI Appendix, Fig. S1A). The binding pocket and active site of Msm and Mtb RNAPs are conserved in sequence and structure, so we do not have a ready explanation for this difference. The equilibrium between intermediates and RPo may be different between Msm and Mtb. We also note that the amino-terminal insert of the σ factors is more divergent (29) and may explain this difference.

Possible steps of inhibition that could lead to the reduction of both run-off and abortive transcription include DNA binding, DNA unwinding, binding of one of the initiating nucleoside triphosphates (NTPs), or catalysis of the phosphodiester bond (Fig. 1A). We show that Sor did not inhibit phosphodiester bond synthesis by Mtb S456L-RNAP (SI Appendix, Fig. S1C). Also, we did not attribute the lack of abortive product formation by the S456L-RNAP to Sor inhibition of the incoming NTP (iNTP) binding. Unlike the Rif derivative, Kanglemycin A (Kang A) (25), three-dimensional (3D) modeling did not suggest that Sor would clash with the modeled iNTP (30) (SI Appendix, Fig. S1D).

Cryo-EM of de Novo Promoter-Melted Mtb Open Complexes Reveals that Sor Physically Blocks Template Strand Placement in the Active Site of the S>L-RNAP.

To address which step of transcription Sor inhibits Mtb S456L-RNAP, we switched to applying cryo-EM. Previously, we used cryo-EM for visualizing conformational changes and intermediates that may be important for understanding the basis of inhibition (11). First, we added Sor (500 µM) to Mtb S456L-RNAP holoenzyme with RbpA and CarD and then added fully duplex AP3 promoter DNA as previously described (11). Unlike the upstream fork DNA template used for the crystal structures (Fig. 2A) (9), this DNA contains double-stranded DNA downstream of the −10 promoter region to +30 (Fig. 3A) and can be used to observe unwinding intermediates (11, 31). The cryo-EM dataset gave rise to two distinct structural classes (SI Appendix, Fig. S4A). The first class contained ∼54% of the particles and resolved to a nominal resolution of 3.6 Å (SI Appendix, Fig. S5 A–C). It consisted of the S456L-RNAP engaged with the promoter DNA, which was melted into the complete 13-nucleotide transcription bubble (positions −11 to +2) as in previously determined RPo structures, and so we deem it S456L-RPo. The structure did not contain strong density for Sor (Fig. 3B), leading us to conclude little Sor was present in these particles. The second class contained 46% of the particles, resolved to a nominal resolution of 3.7 Å (SI Appendix, Figs. S4A and S5 D–F), and contained strong density for Sor. However, the DNA was only partially unwound to an eight-nucleotide bubble (Fig. 3C) and was configured as seen in the previous RP2 structure (11), and so we deem it S456L-RP2. These observations suggested that the DNA in the RPo state was not compatible with Sor binding in the context of the S456L substitution. We explain Sor's weak density in the S456L-RNAP/RPo complex as due to either Sor’s low occupancy or more likely, the lack of clean classification of RP2/RPo states. When we applied a different classification approach, we indeed did not see Sor in the S456L-RPo (SI Appendix, Fig. S8).

Fig. 3.

Cryo-EM experiments demonstrate Mtb RNAP S456L-RNAP/Sor can form RP2 but not RPo. (A) Duplex AP3 promoter DNA (11) used for de novo unwinding in cryo-EM structures of Mtb RNAP with Sor. (B–E) Cryo-EM structures of Mtb S456L (B and C) or WT (D and E) RNAP holoenzyme/RbpA/CarD/Sor/AP3 complexes. RNAPs and transcription factors were incubated first with Sor and then duplex AP3 promoter DNA (A). At the top of B–E, difference maps around the Sor binding site are shown as a blue mesh (normalized and contoured at 7σ). Sor is shown in stick format in green (the Sor in B is modeled to show the Sor position; Sor is not modeled in the structure). At the bottom of B–E, the cryo-EM structures are shown, with the RNAP shown as a transparent molecular surface. The RNAP active site Mg²⁺ is shown as a yellow sphere, and blue is the T-strand +1 base modeled in for reference. Difference cryo-EM density for Sor (green, normalized, and contoured at 7σ) and DNA (red) is shown. The maps are normalized to each other (PyMOL). (B) Mtb S456L-RPo class (SI Appendix, Fig. S4A); Sor difference density is fragmented and nearly absent. (C) Mtb S456L-RP2 class (SI Appendix, Fig. S4A) shows strong density for Sor. (D) Mtb WT-RPo class (SI Appendix, Fig. S4B) shows strong density for Sor. (E) Mtb WT-RP2 class (SI Appendix, Fig. S4B) shows strong density for Sor.

As a control, we performed the same cryo-EM experiment with Mtb WT-RNAP. The data were processed similarly, and two structural classes were extracted, WT-RPo (56% of the particles, nominal resolution of 3.4 Å; SI Appendix, Figs. S4B and S6 A–C) and WT-RP2 (44% of the particles, nominal resolution of 3.4 Å; SI Appendix, Figs. S4B and S6 D–F). However, unlike the S456L-RNAP with Sor, both classes contained strong densities for Sor (Fig. 3 D and E). These findings are consistent with the observation that the S456L-RNAP, unlike the WT-RNAP, does not produce abortive products in the presence of Sor. Interactions between the Mtb WT and S456L-RNAPs and Sor are annotated in SI Appendix, Fig. S2B.

Sor Binding and RPo Formation Are Mutually Exclusive in Mtb S456L-RNAP.

In the cryo-EM structures, we observed weak Sor density with RPo DNA in the Mtb S456L-RNAP, suggesting that Sor’s presence prevents RPo formation by the S456L- RNAP. Therefore, we analyzed the FL2 conformations in the context of the DNA and the S456L substitution. Here, we extended the definition of FL2 to include residues 450 to 479 for structural comparisons. We compared these structures to our previous structure of Mtb WT-RNAP RPo (PDB ID code 6EDT [apo]), which was determined under identical conditions but without antibiotics (11). The structural core of RNAP, which does not show significant conformational changes in bacterial RNAPs (32), was used to align the S456L-RPo incubated with Sor (but no Sor occupancy) to the WT-RPo/apo to compare the conformations of FL2. The rmsd values of the core module and the FL2, as well as the ratios of the rmsd values of the FL2/core modules, are listed in SI Appendix, Table S3. FL2 of the Mtb S456L-RPo/apo was similar to FL2 of the WT-RPo/apo (Fig. 4A and SI Appendix, Table S3), with an rmsd ratio of 1.5 of FL2/core module. The FL2 is a naturally mobile loop, which combined with the effects of the S456L mutation, explains its higher rmsd compared to the core module. We conclude that the S456L substitution in the RPo conformation modestly affects the FL2 structure. By contrast, in the crystal structures, FL2 is disordered in the S>L-RNAP. This difference is likely because the crystal structure does not contain downstream DNA, which constrains the conformation of FL2.

Fig. 4.

Positional and torsional changes of FL2 and Sor. The central core of WT-RPo/apo (PDB ID code 6EDT; yellow backbone tube in A–D) (11, 32) was used as a reference to align the Mtb WT- and S456L-RNAP cryo-EM structures with Sor to compare the positional and conformational changes of the FL2 relative to the rest of RNAP. SI Appendix, Table S3 lists the rmsd values for the superpositions. In A–D, the DNA (T-strand, dark gray; NT-strand, light gray) is shown as a molecular surface. The side chains of S/L456 and L463, which approach the +2 base in the NT-strand (NT +2), are shown. (A) The S456L substitution does not significantly affect the conformation of FL2 in RPo without Sor. DNA is from PDB ID code 6EDT (WT/apo). (B) The presence of Sor does not significantly affect the conformation of FL2 in WT-RNAP RPo or RP2 structures. DNA is from PDB ID code 6EDT. (C and D) The combination of the S456L substitution, Sor, and promoter DNA (in RP2) forces an alteration of FL2 (dark blue), causing L463 to clash with the NT +2 DNA in RPo. The ratio rmsd of S456L RP2/Sor was 2.2. This ratio rmsd indicates the FL2 of S456L RP2 with Sor has greater variability than the other structures compared to the core module. (C) Sor binding to the S>L-RNAP affects the conformation of the FL2. The FL2 of WT-RPo, WT-RPo/Sor, WT-RP2/Sor, and S456L-RP2/Sor are shown. The DNA shown is from S456L-RPo. This figure illustrates that in the presence of Sor, the position of the S456L FL2 (dark blue tube) changes such that L463 clashes with the NT +2 DNA in the RPo state. (D) The FL2 of WT-RPo, S456L-RPo, and S456L-RP2/Sor are shown. The DNA shown is from S456L RPo. The collective results shown in C and D illustrate that the combination of the S456L substitution, Sor, and promoter DNA (in RP2) combine to force FL2 into a conformation that hinders the formation of RPo. (E) Sor is repositioned with minimal torsional changes upon binding to Msm S447L-RNAP with no downstream DNA. The repositioning and torsional movements of Sor were calculated by aligning the core modules (as described above for the cryo-EM structure of the Mtb RNAPs) of Msm WT-RNAP/Sor and Msm S447L-RNAP/Sor. The centers of mass of Sor from the S447L-RNAP and WT-RNAP were then calculated using PyMOL. On the left in E, Sor in the S447L-RNAP has a change in the center of mass of 1.1 Å, compared to the WT enzyme, moving toward the disordered FL2. On the right in E, no significant torsional conformation changes observed for Sor bound to the WT and S447L (rmsd of 0.064 over 58 atoms). (F) Sor, upon binding to Mtb S456L-RNAP in the presence of promoter DNA, exhibits repositioning and torsional movements. L456 atoms are shown as spheres. The repositioning and torsional movements of Sor were calculated using similar alignments of the core modules of the cryo-EM structure of the Mtb RNAPs with Sor as described in A and C, using WT RPo/Sor as the reference molecule. The center of mass (calculated using PyMOL) of Sor in the WT-RP2 structure showed a repositioning of 0.2 Å, which increased to 0.7 Å in the S456L-RNAP (arrow). Aligning the Sor molecule using the “pair fit” command in PyMOL showed that the torsional differences of Sor between the WT-RPo/Sor and WT-RP2/Sor to be 0.438 Å (over 58 atoms). The torsional differences between WT-RPo/Sor and WT-RP2/Sor to S456L RP2/Sor were 0.790 and 0.822 Å, respectively. The change in position of C30 is 2.4 Å, attributed to the S456L mutation (shown in spheres).

Using the WT-RNAP/apo as a reference for alignment, we also noted that Sor or DNA binding in the RPo or RP2 states modestly affected the conformation of the FL2 of the WT-RNAP. The FL2/core rmsd ratios for WT-RPo/apo to WT-RPo/Sor and WT-RP2/Sor were 1.2 and 1.5, respectively (SI Appendix, Table S3). Fig. 4B shows that FL2 in all three WT-RNAP structures, whether or not Sor was bound, were compatible with the DNA in the RPo conformation. In sum, each of the following factors individually perturbed the FL2 with a FL2/core ratio of less than or equal to 1.5: the S>L mutation (in presence of DNA), Sor binding, or DNA in the RP2 or RPo state (Fig. 4 A–C).

By contrast, the combination of the S456L substitution and binding of Sor increased the FL2/core rmsd ratio to 2.2, the highest of all of the ratios (SI Appendix, Table S3). This increase in rmsd ratio is due to the combination of Sor binding and the S456L substitution, leading to FL2 being pushed away to avoid steric clash with the antibiotic (Fig. 4C). In turn, this movement introduces a steric clash between L463 and the nontemplate (NT)-strand +2 base in the RPo conformation (Fig. 4C) but not in the RP2 conformation (SI Appendix, Fig. S7A). Our cryo-EM results show that in the absence of Sor, FL2 of the S456L-RNAP is compatible with the formation of RPo, but Sor binding does not allow RPo formation, explaining our finding that the S456L-RNAP does not form abortive products in the presence of Sor (Fig. 1D).

Comparing the FL2 of S456L-RPo to S456L-RP2/Sor shows that the FL2 in the presence of Sor cannot adopt the conformation required to accommodate the RPo DNA, likely due to the clash with L463 (Fig. 4D). L463 cannot adopt other rotamers that would reduce the clash with the DNA due to steric crowding from A458 and A468 (SI Appendix, Fig. S7B). Therefore, the presence of Sor causes the S456L-FL2 (dark blue in Fig. 4D) to shift toward the DNA, causing a steric clash between L463 and the NT-strand at position +2, explaining the observation that Sor binding and RPo DNA are mutually exclusive in Mtb S456L-RNAP (Fig. 3B).

The contrast between the FL2 disorder of the Msm S>L-RNAP and the ordered FL2 in Mtb S>L-RNAP is striking. We attribute the ordering in Mtb S456L-RNAP to the presence of downstream DNA. However, in the absence of an Mtb S456L-RNAP structure without downstream DNA, we acknowledge that the FL2 disorder might be a peculiar property of Msm S447L-RNAP, which could explain the presence of abortives with Sor, unlike with Mtb S456L-RNAP.

Effects of Sor on the Distribution of RPo and Pre-RPo Complexes.

Because we propose that Sor binding to Mtb S456L-RNAP RP2 would prevent Mtb S456L-RNAP RPo formation, we expected a decrease in the ratio of RPo/RP2 particles relative to the WT-RNAP. However, we did not observe this in our initial processing using iterative classification. This method was employed to obtain a homogeneous set of particles to maximize the resolution of cryo-EM maps for model building (SI Appendix, Fig. S4). Therefore, we reanalyzed our datasets using a different approach. Particles were repicked and curated using “decoy” classification (an adaptation of random-phase 3D classification) (33) to minimize the loss of RNAP particles while removing junk particles. These curated particles were then classified using signal subtraction with a mask around the downstream DNA channel (SI Appendix, Fig. S8 A and B). This approach allowed us to better sort promoter-melting intermediates before RPo, since we reasoned that if Sor inhibits RPo formation, it might lead to an increase in these intermediates. Therefore, unlike the processing used to get the high-resolution homogeneous structures described in Fig. 3, this approach allowed us to curate more particles. This distribution analysis demonstrated that the S456L-RNAP and WT-RNAP, in Sor's presence, exhibit a shift in the equilibrium between RPo and RP2 particles, with the mutant having more RP2-like particles relative to RPo and the WT-RNAP having more RPo relative to RP2 (SI Appendix, Fig. S8C). We do not rule out that the S456L-RNAP might also experience the same shift without Sor, but we note that the S456L-RNAP RPo complexes do not display Sor density (SI Appendix, Fig. S8C, Right). These data support our conclusion that Sor prevents RPo formation in Mtb containing the S456L-RNAP.

Sor Molecule in the Crystal Structure of Msm S>L-RNAP Rif-Binding Pocket.

Modeling bulkier amino acids (Tyr, Leu, Phe) at the position corresponding to the S>L substitution in the Taq β subunit suggested that both Rif and Sor would clash with the substituted amino acids, conferring resistance (17). We, therefore, proposed that the conformational flexibility of Sor allowed it to bind to the altered shape of the Rif pocket. By contrast, we proposed that the rigid Rif structure would not be able to adapt to the altered shape of the mutant Rif pocket, explaining the Rif^R/Sor^S phenotype of the S>L substitutions (17). The premise of this proposal was inspired by studies of inhibitors against WT and drug-resistant variants of HIV-1 reverse transcriptase. In these studies, the authors described the repositioning or reorientation of an inhibitor in a binding pocket altered by mutation, as well as the inherent torsional or conformational flexibility of the inhibitor structure itself to adapt to the mutated pocket (34). Aligning the two Msm RNAP/Sor crystal structures by the RNAP structural core (32), we observe that the S447L- and WT-RNAPs align well (0.880 Å over 2,661 atoms). Unexpectedly, Sor repositions upward in the direction toward the S447L substitution, with a significant shift of the center of mass of the Sor molecule of 1.1 Å (structural alignment on the left in Fig. 4E). When aligning the atoms of the Sor molecule from each crystal structure, the rmsd was negligible (0.064 Å over 58 atoms; structural alignment on the right in Fig. 4E), indicating that Sor did not undergo major torsional changes. Therefore, our original model was incorrect in the context of the enzyme without DNA. Instead, the protein (FL2) moved in response to Sor binding as a result of the S>L substitution. Because the antibiotic was not constrained by interactions with the FL2, which in turn was not constrained by the presence of downstream DNA, it repositioned toward the substitution instead. Our conclusion from the crystal structures is that Sor is not required to change its backbone conformation to be accommodated in the mutated pocket. However, we hypothesized that if DNA constrained the FL2, the Sor would have to reposition or undergo torsional fluctuations as the FL2 would clash with Sor. The Sor Molecule in the Mtb S>L-RNAP Rif-Binding Pocket with Downstream DNA describes this analysis of Mtb RNAP cryo-EM structures with a full promoter template with downstream DNA (Fig. 3A).

The Sor Molecule in the Mtb S>L-RNAP Rif-Binding Pocket with Downstream DNA.

To compare the Sor molecules from the cryo-EM structures that contained the full complement of DNA, we aligned the structural core module of RNAP (32) of the Sor-bound structures and examined the disposition of Sor. The center of mass of Sor between RPo and RP2 in the WT-RNAP did not change significantly (0.2 Å; structural alignments on the left in Fig. 4F). However, in the mutant RP2 structure, Sor repositioned 0.7 Å away from the substituted amino acid. Significantly, the backbone of Sor around C30 showed torsional changes up to ∼2.4 Å, and the overall rmsd between Sor from the S>L mutant and WT-RNAPs was 0.790 Å (WT-RPo) and 0.822 Å (WT-RP2) across 58 atoms of Sor (structural alignments on the right in Fig. 4F). This rmsd was almost double that of Sor in the WT-RPo and WT-RP2 structures (0.438 over 58 atoms). The torsional movements of Sor in the S456L-RNAP relieve steric clashing with L456, which occurs mostly around C30 (Fig. 4F). Therefore, we conclude that in the context of the RP2 DNA, for Sor to remain bound and inhibit RPo formation, the antibiotic undergoes torsional fluctuations to avoid steric clash with the bulky L456.

Sor Displays a Lower Potential for DDIs Compared to Rifamycins.

Treatment of TB patients with Rif is complicated due to the high prevalence of Rif^R Mtb. In addition, Rif binds to PXR, thereby inducing the expression of CYP enzymes, with CYP3A4 being the most relevant isoform in the context of DDIs between Rif and, e.g., antiretroviral agents. We compared Sor, Rif, and its derivatives rifabutin and rifapentine (35) in a luciferase reporter-gene assay using DPX-2 cells (36). Intriguingly, Sor was much less effective in inducing CYP3A4 through PXR activation (Table 1 and SI Appendix, Fig. S9). Compared to Rif, the half-maximum effective concentration (EC₅₀) was increased by approximately fourfold (3.9 µM for Sor vs. 1.0 µM for Rif), and the level of CYP3A4 induction reached only 41% compared to the vehicle control, using Rif as a positive control (98% induction). If Sor could be delivered at similar human doses like Rif, this natural product antibiotic Sor could, in theory, overcome the issues related to antimicrobial resistance and may significantly reduce DDIs as described for Rif.

Table 1.

Result summary of PXR-CYP3A4 induction assay

Compound	EC50, µM	Maximum CYP induction, %*
Rif	1.0	98
Rifabutin	0.5	93
Rifapentine	3.3	97
Sor	3.9	41

^*Compared to vehicle control.

Conclusion

The combination of biochemical data with crystal structures and cryo-EM structures reveals the following key points. 1) Sor inhibits the Rif^R S>L-RNAPs at least two orders of magnitude better than Rif (Fig. 1C). 2) Sor inhibits transcription initiation by the Mtb S456L-RNAP at a step prior to chain elongation (Fig. 1D). 3) In the absence of promoter DNA in the active site, the crystal structure of Msm S>L-RNAP FL2 is disordered, increasing the ESA of Sor and Rif, explaining the increased IC₅₀ values. The increase in ESA for Rif is significantly greater than Sor, helping to explain the Sor^S-Rif^R phenotype of this substitution (Fig. 2). 4) Sor does not significantly affect the steps up to and including promoter melting on WT-RNAP, explaining its effects on postinitiation steps (Figs. 1C and 3 and SI Appendix, Fig. S1). 5) Sor binding to the Mtb S456L-RNAP is compatible with the RP2 DNA location but not with the RPo DNA location. This observation possibly explains how Sor inhibits abortive transcription by Mtb RNAP (Figs. 1D and 3). 6) In the presence of Sor, FL2 must rearrange due to the bulky S>L substitution. We observe in the cryo-EM structures with Mtb S456L-RNAP and Sor that the altered FL2 conformation would clash with the fully opened DNA in RPo but not RP2 (Fig. 4 A–D and SI Appendix, Fig. S7). 7) We observe in the cryo-EM structure of Mtb S456L-RNAP RP2 that the combination of the DNA in the cleft and the S>L mutation forces Sor to undergo torsional fluctuations (Fig. 4F).

These data lead to a model that explains the effects of Sor on Mtb WT- and S456L-RNAP (Fig. 5). The top row in Fig. 5 shows how nucleic acids and apo WT-RNAP rearrange in steps to form the productive elongation complex RPec (also true for the apo S>L-RNAP; not shown). The apo RNAPs form RP1 (the initial engagement with DNA complex), proceed to RP2 (partial bubble intermediate), then to RPo (the fully unwound bubble with the T-strand DNA in the active site), then to RPitc (containing short transcripts), and eventually to RPec. The middle row in Fig. 5 illustrates how WT-RNAP with Sor is inhibited at the transition from RPitc to RPec due to steric clashes between Sor and the elongating RNA transcript, as previously described (17). The bottom row of Fig. 5 shows how the S>L-RNAP in the presence of Sor can proceed to RP2, but the transition to RPo is inhibited. The position of the DNA in RPo imposes constraints on the FL2 position, which in turn is incompatible with the presence of Sor when the S>L substitution is present in FL2, likely explaining this finding. Therefore, unlike with the WT-RNAP, Sor inhibits the S>L-RNAP at a step upstream of RPitc (middle vs. bottom rows in Fig. 5). In sum, the mechanism of inhibition by Sor of the Mtb S>L-RNAP is nuanced. While our data favor the model presented in the bottom row of Fig. 5, we acknowledge that other mechanisms might contribute to our results. These include allosteric inhibition of the binding of the initiating NTP, or inhibition of phosphodiester-bond synthesis due to the altered conformation of the mutated FL2 in the presence of Sor. These alternative explanations remain to be tested. In addition, although the pathway we depict is linear, it also possible that Sor can bind the S>L-RNAP in the RP2 state, which would also inhibit initiation.

Fig. 5.

Model of Sor’s effect on the transcription of WT- and S>L-RNAPs. The effects of Sor and the S456L mutation on the transcription initiation steps to elongation of Mtb RNAP are shown in cartoon with the key below. (Top) WT-RNAP (and the S>L-RNAP; not shown) without Sor can proceed to the productive complex RPec. (Middle) The WT enzyme with Sor can bind double-stranded DNA (RP1) and proceed to RPitc, but Sor blocked the transition from RPitc to RPec (red X). (Bottom) The S>L-RNAP is inhibited at the transition from RP2 to RPo (red dashed X) because of the clash between the FL2 and the NT-strand DNA. The dashed X indicates that other explanations, which require additional investigations, may apply to why RPo is inhibited.

Our findings highlight a need to revise the design and optimization of new inhibitors to target antibiotic-resistant macromolecules. We note that previous studies show that the Rif derivative Kang A inhibits both WT- and S>L-RNAPs at a step earlier than RPitc (25, 37). Kang A, despite having the same core structure and binding pocket as Rif, inhibits binding of the initiating NTP. Combined with our finding of an altered mechanism for Sor inhibition of Rif^R RNAP, these studies emphasize the importance of detailed biochemical and mechanistic characterization of inhibitors. Even when inhibitors are chemically closely related and bind the same site on a target molecule—one should not assume the inhibitors necessarily inhibit the same mechanistic step. The results presented here also show that the same inhibitor can affect the WT and antibiotic-resistant enzyme differently; thus, stressing the need for structural and biochemical comparative studies. In sum, the mechanism of Sor inhibition is unexpectedly complex. Multiple factors need to be considered in drug-inhibition studies: the thermal fluctuations of inhibitors in drug-binding pockets, the movement of the protein composing the antibiotic pocket, and the possible differential mechanisms of inhibition on WT and Rif^R Mtb RNAPs by the same antibiotic.

Finally, we also studied whether Sor activates PXR, which is a common cause of DDIs with Rif. Importantly, Sor displays only moderate CYP3A4 induction following PXR activation. This favorable pharmacokinetic property is a potential consideration when designing new RNAP-targeting inhibitors for TB therapy.

Methods

Detailed descriptions of Msm and Mtb σ^A, RbpA, CarD and RNAPs protein purification, transcription assays, crystallization, structural determination and refinement of Msm RbpA/fork/Sor complexes, preparation of WT and S456L Mtb RNAP complexes for cryo-EM, cryo-EM grid preparation, cryo-EM data acquisition and processing, model building and refinement, and CYP3A4 induction reporter gene assay are provided in SI Appendix.

Data Availability.

The X-ray crystallographic coordinates and structure factors for the Msm RNAP/Sor, Msm RNAP β^S447L/Sor, and Msm RNAP β^S447L structures have been deposited in the PDB (PDB ID codes 6VVS [Msm RNAP/Sor], 6VVT [Msm RNAP β^S447L/Sor], and 6VVV [Msm RNAP β^S447L]). The cryo-EM density maps have been deposited in the Electron Microscopy Data Bank (EMDB) (entry nos. EMD-21407 [Mtb RNAP RPo/Sor], EMD-21406 [Mtb RNAP RP2/Sor], EMD-21408 [Mtb RNAP β^S456L RPo], and EMD-21409 [Mtb RNAP β^S456L RP2/Sor]). The atomic coordinates have been deposited in the PDB (PDB ID codes 6VVY [Mtb RNAP RPo/Sor], 6VVX [Mtb RNAP RP2/Sor], 6VWO [Mtb RNAP β^S456L RPo], and 6VVZ [Mtb RNAP β^S456L RP2/Sor]). The cryo-EM density maps from the distribution analysis have been deposited in the EMDB (entry nos. EMD-22573 [Mtb RNAP PreRP2], EMD-22575 [Mtb RNAP RP2], EMD-22577 [Mtb RNAP RPo], EMD-22578 [Mtb RNAP β^S456L PreRP2], EMD-22579 [Mtb RNAP β^S456L RP2], and EMD-22580 [Mtb RNAP β^S456L RPo]).