Structural basis of ECF-σ-factor-dependent transcription initiation.

Extracytoplasmic (ECF) σ factors, the largest class of alternative σ factors, are related to primary σ factors, but have simpler structures, comprising only two of six conserved functional modules in primary σ factors: region 2 (σR2) and region 4 (σR4). Here, we report crystal structures of transcription initiation complexes containing Mycobacterium tuberculosis RNA polymerase (RNAP), M. tuberculosis ECF σ factor σL, and promoter DNA. The structures show that σR2 and σR4 of the ECF σ factor occupy the same sites on RNAP as in primary σ factors, show that the connector between σR2 and σR4 of the ECF σ factor-although shorter and unrelated in sequence-follows the same path through RNAP as in primary σ factors, and show that the ECF σ factor uses the same strategy to bind and unwind promoter DNA as primary σ factors. The results define protein-protein and protein-DNA interactions involved in ECF-σ-factor-dependent transcription initiation.


Structures of Mtb RPo-σ L and Mtb RPitc-σ L
Structures were determined using recombinant Mtb RNAP core enzyme prepared by co-expression of Mtb RNAP subunit genes in E. coli, recombinant Mtb σ L , and synthetic nucleic-acid scaffolds based on the sequence of the σ L -dependent promoter P-sigL (the promoter responsible for expression of the gene encoding σ L ; Hahn et al., 2005;Dainese et al., 2006;Rodrigue et al. 2007) (Figs. S1-S2). Transcription experiments demonstrate that Mtb RNAP-σ A holoenzyme (containing the group-1 σ factor σ A ) does not efficiently perform transcription initiation at the P-sigL promoter, whereas Mtb RNAP-σ L holoenzyme (containing the ECF σ factor σ L ) does (Fig. S1E). We prepared "downstream-fork-junction" nucleic-acid scaffolds containing P-sigL sequences, analogous to the downstream-fork-junction nucleic-acid scaffolds containing consensus group-1-σ-factor-dependent promoter sequences used previously for structural analysis of group-1-σ-factor-dependent transcription initiation (Fig. S2, left panels). Because the P-sigL transcription start site (TSS) had been mapped only provisionally (Hahn et al., 2005;Dainese et al., 2006;Rodrigue et al., 2007), we prepared and analyzed a set of downstream-fork-junction nucleic-acid scaffolds having different lengths--4 nt, 5 nt, 6, or 7 nt--of the "spacer" between the P-sigL promoter -10 region and downstream dsDNA (Fig. S2, left panels). Transcription experiments indicated that all analyzed nucleic-acid scaffolds were functional in σ L -dependent de novo transcription initiation at the expected TSS (with the initiating nucleotide base-pairing to template-strand ssDNA 2 nt upstream of dsDNA), and σ L -dependent primer-dependent transcription initiation at the expected TSS (with the primer 3' nucleotide base-pairing to template-strand ssDNA 2 nt upstream of dsDNA), with highest levels of function observed for a spacer length of 6 nt ( Fig. S1F-G). Robotic crystallization trials identified crystallization conditions yielding high-quality crystals for spacer lengths of 4 nt, 5 nt, or 6 nt (Table 1; Fig. S2, center panels). X-ray datasets were collected at synchrotron beam sources, and structures were solved by molecular replacement and refined to 3.3 to 3.8 Å resolution (Table 1; Fig. S2, right panels).
Experimental electron-density maps showed clear density for RNAP, σ L , and nucleic acids (Fig. S2, right panels). The resulting structures were essentially identical for nucleic-acid scaffolds having spacer lengths of 4 nt, 5 nt, or 6 nt (Figs. S2, right panels). However, map quality was highest for the nucleic-acid scaffold having a spacer length of 6 nt, and, therefore subsequent analysis focussed on structures with a spacer length of 6 nt (Mtb RNAP-σ L RPitc5_sp6). For the nucleic-acid scaffold containing a 6 nt spacer, the translocational state of the transcription complex was experimentally verified by preparation of a scaffold having a single 5-bromo-dU substitution and collection of bromine anomalous diffraction data (Table 1; Fig. S2D). The fit of σ L separately was experimentally verified by preparation of a selenomethionine-labelled σ L derivative and collection of selenium anomalous diffraction data (Table 1; Fig. S2E).

Protein-protein interactions between ECF σ factor and RNAP
The structural organization of the ECF σ L -factor-dependent transcription initiation complex is unexpectedly similar to that of a group-1 σ A -factor-dependent transcription initiation complex (Figs. 1B,2). σR2 and σR4 of σ L occupy the same positions on RNAP, and make the same interactions with RNAP, as σR2 and σR4 of σ A factor (Fig. 2). Despite the much smaller size of the connector between σR2 and σR4 in σ L as compared to σ A (20 residues vs. 84 residues; Fig. S1A), the connector in σ L spans the full distance between the σR2 and σR4 binding positions on RNAP and follows a path through RNAP remarkably similar to that of the connector in σ A (Fig 2.). Thus, the σ L σR2/4 linker, like the σ A σR3/4 linker, first enters the RNAP active-center cleft and approaches the RNAP active center, and then makes a sharp turn and exits the RNAP active-center cleft through the RNAP RNA-exit channel.
Inside the RNAP active-center cleft, the σ L σR2/4 linker, like the σ A σR3/4 linker, makes direct interactions with template-strand ssDNA nucleotides of the unwound transcription bubble S3A). The interactions of the σ L σR2/4 linker with template-strand ssDNA include a direct H-bonded interaction of σ L Ser96 with a Watson-Crick H-bonding atom of the template-strand nucleotide at promoter bottom). The interactions of the σ L σR2/4 linker with template-strand ssDNA are similar to, but less extensive than, those of the σ A σR3/4 linker with template-strand ssDNA, which include direct H-bonded interactions of σ A Asp432 and Ser433 with Watson-Crick H-bonding atoms of template-strand ssDNA nucleotides at promoter positions -4 and -3 (Figs. 3A and S3A).
In the case of the group-1 σ factor, σ A , the interactions between this segment of the σR3/4 linker and template-strand ssDNA pre-organize template-strand ssDNA to adopt a helical conformation and to engage the RNAP active-center nucleotide-addition site, thereby facilitating initiating-nucleotide binding and de novo initiation (Zhang et al., 2012; see also Kulbachinskiy and Mustaev, 2006;Pupov et al., 2014).
The similarity of the interactions made by the ECF σ factor, σ L , suggests that ECF σ factors likewise pre-organize template-strand ssDNA and facilitate initiating-nucleotide binding and de novo initiation.
In the case of the group-1 σ factor, σ A , the interactions between this segment of the σR3/4 linker and template-strand ssDNA must be broken, and this segment of the σR3/4 linker must be displaced, when the nascent RNA reaches a length >4 nt during initial transcription, and this requirement for breakage of interactions and displacement is thought to impose an energy barrier that results in, or enhances, abortive initiation (Murakami et al., 2002;Kulbachinskiy and Mustaev, 2006;Zhang et al., 2012;Basu et al., 2014;Pupov et al., 2014) and initial-transcription pausing (Duchi et al., 2016;Lerner et al., 2016;Dulin et al., 2018). The similarity of the interactions made by the ECF σ factor, σ L , suggests that ECF σ factors likewise have a similar requirement for displacement of a linker segment during initial transcription--in this case, when the nascent RNA reaches a length of >5 nt (Fig. S3B)--and that this similar requirement imposes a energy barrier that results in, or enhances, abortive initiation and initialtranscription pausing. Consistent with this hypothesis, transcription and transcript-release experiments indicate that Mtb RNAP-σ L holoenzyme efficiently performs abortive initiation, efficiently producing and releasing short abortive RNA products (Fig. S3C).
The 5 C-terminal residues of the σ L σR2/4 linker, like the 10 C-terminal residues of the σ A σR3/4 linker, exit the RNAP active-center cleft and connect to σR4 by threading through the RNAP RNA-exit channel (Fig. 2). In the case of the group-1 σ factor, σ A , the C-terminal segment of the σR3/4 linker must be displaced from the RNA-exit channel when the nascent RNA reaches a length of 11 nt at the end of initial transcription and moves into the RNA-exit channel, and this displacement is thought to alter interactions between σR4 and RNAP, and thereby to trigger promoter escape and to transform the transcription initiation complex into a transcription elongation complex (Murakami et al., 2002;Vassylyev et al, 2002;Mekler et al., 2002). The similarity of the threading through the RNAP RNA-exit channel by the ECF σ factor, σ L , suggests that ECF σ factors have a similar requirement for displacement of a linker C-terminal segment and have a similar mechanism of promoter escape and transformation from transcription initiation complexes into transcription elongation complexes.
In its interactions with template-strand ssDNA and the RNAP RNA exit channel, the σ L σR2/4 linker, like the σ A σR3/4 linker, linker serves as a molecular mimic, or a molecular placeholder, for nascent RNA, making interactions with template-strand ssDNA and the RNAP-RNA exit channel in early stages of transcription initiation that subsequently, in late stages of transcription initiation and in transcription elongation, are made by nascent RNA. The σ L σR2/4 linker and the σ Α σR3/4 linker, both have net negative charge (Fig. S1A), and both employ extended conformations (fully extended for the σ L σR2/4 linker; largely extended for the σ Α σR3/4 linker; Fig. 2) to interact with template-strand ssDNA and the RNAP RNA exit channel, consistent with function as molecular mimics of a negatively charged, extended nascent RNA. Nevertheless, the σ L σR2/4 linker and the σ Α σR3/4 linker exhibit no detectable sequence similarity (Fig. S1A) and no detailed structural similarity (Fig. 2). We conclude that the σR2/4 linker of an ECF σ factor and the σR3/4 linker of a group-1 σ factor provide an example of functional analogy in the absence of structural homology.

Protein-DNA interactions between ECF σ factor and promoter: -10 element
The structure reveals the interactions between the ECF σ factor, σ L , and promoter DNA that mediate recognition of the promoter -10 element S4). The σ L conserved module σR2, like the σ A conserved module σR2, mediates recognition of the promoter -10 element through interactions with nontemplate-strand ssDNA in the unwound transcription bubble . In the case of the group-1 σ factor, σ A , a crucial aspect of recognition of the promoter -10 element is unstacking of nucleotides, flipping of nucleotides, and insertion of nucleotides into protein pockets at two positions of the σ A -dependent promoter: i.e., position -11 (referred to as the "master nucleotide," based on its especially RNAP σ L holoenzyme unstacks, flips, and inserts into a protein pocket a guanosine at position "-11" of the σ L -dependent promoter, making extensive interactions with the base moiety of the guanosine, including multiple direct H-bonded interactions with Watson-Crick H-bonding atoms (Figs. 3,4B,5,S4A). The interactions between σ L and guanosine at position "-11" of the σ L -dependent promoter are similar to the interactions between RNAP σ A holoenzyme and adenosine at position -11 of the σ A -dependent promoter -10 element, including, in particular, similar stacking interactions of σ L aromatic amino acid Trp48 with guanosine and of corresponding σ A aromatic amino acid Tyr436 with adenosine ( Fig. S4A). The different specificities--guanosine at position "-11" for σ L vs. adenosine at position -11 for σ A --arise from differences in H-bond-donor/H-bond-acceptor character of atoms forming the floors of the relevant protein pockets of σ L and σ A , with H-bonding complementarity to guanosine in σ L and H-bonding complementarity to adenosine in σ A (Fig. S4A).
RNAP σ L holoenzyme also unstacks, flips, and inserts into a pocket a guanosine at position "-7" of the σ L -dependent promoter, making extensive interactions with the base moiety of the guanosine, including a direct H-bonded interaction with a Watson-Crick H-bonding atom (Figs. 3, 4B, 5, S4B).
These interactions are similar in location to, but different in detail from the interactions made by RNAP σ A holoenzyme with thymidine at position -7 of the σ L -dependent promoter (Figs. 4,S4B). The differences in detail arise from the fact that σ L does not contain conserved module σR1.2. In the case of σ L , the interactions involve residues of σR2 and residues of RNAP β subunit, with the base moiety of the guanosine at position "-7" being inserted into a cleft between σR2 and β (Figs. 5, S4B). In contrast, in the case of σ A , the interactions involve residues of σR2 and residues σR1.2, with the base moiety of the thymidine at position -7 being inserted into a cleft between σR2 and σR1.2 ( Fig S4B).
RNAP σ L holoenzyme also appears to unstack and insert into a protein pocket a thymidine at position "-12" of the σ L -dependent promoter (Fig. 4B, 5), placing one face of the base moiety of the thymidine in a shallow surface pocket, in position to make a direct H-bonded interaction with a Watson-Crick atom (Fig. 5). The interaction with an unstacked nucleotide inserted into a protein pocket implies that position "-12" of the σ L -dependent promoter must be ssDNA in RPo-σ L and RPitc-σ L , and thus that the transcription bubble must extend to position "-12" in RPo-σ L and RPitc-σ L . This interaction does not have a counterpart in the σ A -dependent transcription initiation complex, in which position -12 of the σ Adependent promoter is dsDNA and in which the transcription bubble extends only to position -11 (Bae et al., 2015;Zuo and Steitz, 2015;Feng et al, 2016).
In addition to these potential specificity-determining interactions with unstacked nucleotides inserted into protein pockets, RNAP σ L holoenzyme makes potentially specificity-determining interactions with positions "-9" and "-8" of the σ L -dependent promoter (Fig. 5). RNAP σ L holoenzyme makes a direct H-bonded interaction, through RNAP β subunit, with a Watson-Crick atom of the base moiety of cytidine at position "-9" (Fig. 5) and makes two direct H-bonded interactions, through σR2 and RNAP β subunit, with a Watson-Crick atom of the base moiety of adenosine at position "-8" (Fig. 5).
" Alanine-scanning" experiments (Cunningham and Wells, 1989), in which residues of σ L that contact -10-element nucleotides in the crystal structure are substituted with alanine and effects on σ L -dependent transcription are quantified, confirm the functional significance of the observed interactions (Fig. 6C).
"Loss-of-contact" experiments (Ebright, 1985(Ebright, , 1986(Ebright, , 1991Zhang et al., 1990), in which residues of σ L that contact -10-element nucleotides in the crystal structure are substituted with alanine and effects on specificity at the contacted positions are quantified, confirm that σ L His54 determines specificity for thymidine at position "-12" (Fig. 6D) and that σ L Asp60 determines specificity for guanosine at position "-11" (Fig. 6E). In the crystal structure, σ L His54 makes a van der Waals interaction with the 5'-methyl group of the base moiety of thymidine at position "-12" (Fig. 5); in loss-of-contact experiments, substitution of His54 by alanine eliminates specificity for thymidine at position "-12" (Fig. 6D). In the crystal structure σ L Asp60 makes an H-bonded interaction with Watson-Crick atoms of the base moiety of guanosine G at position "-11" (Figs. 5, S4A; in loss-of-contact experiments, substitution of Asp60 by alanine eliminates specificity for thymidine at position "-11" (Fig. 6E).

Protein-DNA interactions between ECF σ factor and promoter: core recognition element (CRE)
The structure reveals the interactions between RNAP σ L holoenzyme and nontemplate-strand ssDNA downstream of the promoter -10 element in the ECF σ L -dependent transcription initiation complex (Figs. 3B, S5). In the case of group-1-σ-factor-dependent transcription initiation complexes, sequence-specific interactions occur between RNAP β subunit and a 6 nt segment of nontemplate-strand ssDNA downstream of the promoter -10 element referred to as the "core recognition element" (

DISCUSSION
Our structural results show that: (1) σR2 and σR4 of an ECF σ factor σ L adopt the same folds and interact with the same sites on RNAP as σR2 and σR4 of a group-1 σ factor (Figs. 1-2); (2) the connector between σR2 and σR4 of ECF σ factor σ L enters the RNAP active-center cleft to interact with template-strand ssDNA and then exits the RNAP active-center cleft by threading through the RNAP RNA-exit channel in a manner functionally analogous--but not structurally homologous--to the connector between σR2 and σR4 of a group-1 σ factor (Figs. 1-2; S3); (3) ECF σ factor σ L recognizes the -10 element of an σ L -dependent promoter by unstacking nucleotides and inserting nucleotides into protein pockets at three positions of the transcription-bubble nontemplate-strand ssDNA (positions "-12," "-11," and "-7"; Figs. 3-5, S4), and (4)  Our results regarding the connector between σR2 and σR4 of an ECF σ factor, in conjunction with previous results, indicate that all classes of bacterial σ factors contain structural modules that enter the RNAP active-center cleft to interact with template-strand ssDNA and then leave the RNAP active-center cleft by threading through the RNAP RNA-exit channel, providing mechanisms to facilitate de novo initiation, to coordinate extension of the nascent RNA with abortive initiation and initial-transcription pausing, and to coordinate entry of RNA into RNA-exit channel with promoter escape. For ECF σ factors, as shown here, the relevant structural module is the σR2/4 linker (Figs 3-4, S3); for group-1, group-2, and group-3 σ factors, the module is the functionally analogous--but not structurally homologous--σR3/4 linker (Murakami et al., 2002(Murakami et al., , 2013Vassylyev et al, 2002;Zhang et al., 2012Zhang et al., , 2014Bae et al., 2013Bae et al., , 2015Basu et al., 2014;Zuo and Steitz, 2015;Liu et al., 2016); and for group-σ 54 /σ N σ factors, the module is the functionally analogous--but not structurally homologous--region II.3 (RII.3; Yang et al., 2015;Glyde et al., 2018).
More broadly, our results, in conjunction with previous results, indicate that cellular transcription initiation complexes in all organisms--bacteria, archaea, and eukaryotes--contain structural modules that enter the RNAP active-center cleft to interact with template-strand ssDNA and then leave the RNAP active-center cleft by threading through the RNAP RNA exit channel. In different classes of bacterial transcription initiation complexes, as described in the preceding paragraph, these roles are performed by the functionally analogous--but not structurally homologous--σR2/4 linker, σR3/4 linker, and RII.3. In archaeal transcription initiation complexes, these roles are performed by the TFIIB zinc ribbon and CSB, which are unrelated to the σR2/4 linker, σR3/4 linker, and RII.3 (Renfrow et al., 2004). In eukaryotic RNAP-I-, RNAP-II-, and RNAP-III-dependent transcription initiation complexes, these roles are performed by the Rm7 zinc ribbon and B-reader, the TFIIB zinc ribbon and B-reader, and the Brf1 zinc ribbon, respectively, each of which is unrelated to the σR2/4 linker, σR3/4 linker; and RII.3 (Kostrewa et al., 2009;Liu et al., 2010;He et al., 2016;Ptaschka et al., 2016;Engel et al., 2017;Han et al., 2017;Sadian et al., 2017;Vorländer et al., 2018;Abascal-Palacios et al., 2018). It is extraordinary that non-homologous, structurally and phylogenetically unrelated, structural modules are used to perform the same roles in different transcription initiation complexes, and is unknown how or why this occurs.
Both our structural results and our biochemical results point to the special importance of the nontemplate-strand nucleotide at position "-11" ("master nucleotide"; Figs. 3-6, S4A). Our results regarding recognition of the "-11" "master nucleotide" by an ECF σ factor are consistent with the NMR structure of a complex comprising σR2 from the E. coli ECF σ factor σ E and a 5 nt oligodeoxyribonucleotide corresponding to part of the nontemplate strand of the -10 element of a σ E -dependent promoter (Campagne et al., 2014(Campagne et al., , 2015. The NMR structure showed unstacking, flipping, and insertion into a protein pocket of the "-11" "master-nucleotide" (a cytidine, rather than a guanosine, reflecting the different specificities of E. coli σ E vs. Mtb σ L ; Campagne et al., 2014Campagne et al., , 2015. The NMR structure did not show unstacking and flipping of the nucleotide at position "-7," reflecting the fact that the oligodeoxyribonucleotide in the NMR structure did not extend to position "-7" (Campagne et al., 2014(Campagne et al., , 2015. The NMR structure also did not show unstacking of the nucleotide at position "-12," (Campagne et al., 2014(Campagne et al., , 2015, possibly reflecting an uncertainty in the NMR structure, or possibly reflecting a difference between E. coli σ E and Mtb σ L in recognition of position "-12." Based on the NMR structure, Campagne et al. (2014Campagne et al. ( , 2015 hypothesized that the loop of σR2 that forms the protein pocket into which the "-11" "master nucleotide" is inserted--"loop L3" (residues 63-72 of E. coli σ E , which correspond to residues 56-67 of Mtb σ L )--serves as a functionally independent, functionally modular, determinant of specificity at the "master-nucleotide" position, such that different loop-L3 sequences confer different specificities at the "master-nucleotide" position, in each case, through interactions with an unstacked, flipped, and inserted "master nucleotide." Campagne et al. (2014Campagne et al. ( , 2015 supported this hypothesis by identifying examples of L3 -loop sequences that conferred specificity for cytidine, thymidine, and adenosine at the "master-nucleotide" position, and by providing evidence that swapping L3-loop sequences swaps specificity at the "master-nucleotide" position. Our results provide further support for the hypothesis by identifying an example of an L3-loop sequence, the Mtb σ L loop-L3 sequence, that confers specificity for guanosine at the "master-nucleotide" position, and by documenting that specificity for guanosine involves interactions with an unstacked, flipped, and inserted "master nucleotide" (Figs. 3-5, S4A).
In the crystal form identified and analyzed in this work, σR2 of each molecule of transcription initiation complex makes no interactions with other molecules of transcription initiation complex in the crystal lattice (Fig. S7A), and, therefore, with this crystal form, it should be possible to substitute σR2 without losing the ability to form crystals (Fig. S7A). This potentially provides a platform for systematic structural analysis of σR2 and σR2-DNA interactions for the thirteen Mtb σ factors, by determination of crystal structures of transcription initiation complexes containing "chimeric σ factors" (see Kumar et al., 1995;Rhodius et al., 2013) comprising σR2 of a Mtb σ factor of interest fused to the σR2/4 linker through σR4 of Mtb σ L (Fig. S7B; left red arrow) and containing the promoter sequence for the Mtb σ factor of interest. In the crystal form identified and analyzed in this work there also are no lattice interactions for the connector between σR2 and σR4, and there likely would be no lattice interactions even if that connector were to contain σR3 and a σR3/4 linker, as in group-1, group-2, and group-3 σ factors ( Figure S7A). Accordingly, this crystal form potentially provides a platform for systematic structural analysis not only of σR2 and its protein-DNA interactions, but also of the connector between Engel, C., Gubbey, T., Neyer, S., Sainsbury, S., Oberthuer, C., Baejen, C., Bernecky, C., and Cramer, P.        (Ebright, 1985(Ebright, , 1986(Ebright, , 1991Zhang et al., 1990) indicating that σ L residues His54 and Asp60 determine specificity at position "-12" and "-11," respectively. Left: transcriptional activity with wild-type σ L for all possible single base-pair-substitutions at indicated position (strong specificity for consensus base pair). Right: transcriptional activity of σ L derivatives having alanine substitutions (no specificity for consensus base pair).

M. tuberculosis RNAP core enzyme
Mtb RNAP core enzyme was prepared by co-expression of genes for Mtb RNAP β' subunit, β subunit, N-terminally decahistidine-tagged α subunit, and ω subunit in E. coli, followed by cell lysis, polyethylenimine precipitation, ammonium sulfate precipitation, immobilized-metal-ion affinity chromatography on Ni-NTA agarose (Qiagen), and anion-exchange chromatography on Mono Q (GE Healthcare), as in Lin et al., 2018.
Alanine-substituted σ L derivatives were prepared as described for preparation of σ L , but using plasmid pSR32 derivatives constructed using site-directed mutagenesis (QuikChange Site-Directed Mutagenesis Kit; Agilent).
Selenomethionine-substituted σ L was prepared as described for preparation of σ L , but using culture media and culture procedures as in Stols et al., 2004.

M. tuberculosis RNAP σ L holoenzyme
Mtb RNAP core enzyme and Mtb σ L or σ L derivative were incubated in a 1:4 ratio in 20 mM Tris-HCl, pH 8.0, 100 mM NaCl, 5 mM MgCl 2 , and 1 mM 2-mercaptoethanol for 12 h at 4°C. The reaction mixture was applied to a HiLoad 16/60 Superdex S200 column (GE Healthcare) equilibrated in the same buffer, and the column was eluted with 120 ml of the same buffer. Fractions containing Mtb RNAP σ L holoenzyme were pooled, concentrated to ~10 mg/ml using 30 kDa MWCO Amicon Ultra-15 centrifugal ultrafilters (EMD Millipore), and stored in aliquots at -80°C.

Structure determination: structure solution and refinement
The structure of Mtb RPitc5-σ L _sp6 was solved by molecular replacement with MOLREP (Collaborative Computational Project, 1994) using the structure of Mtb RPo (PDB 5UHA; Lin et al., 2017), omitting σ A and nucleic acids, as the search model. One molecule of RNAP was present in the asymmetric unit. Early-stage refinement included rigid-body refinement of RNAP core enzyme, followed by rigid-body refinement of each subunit of RNAP core enzyme, followed by rigid-body refinement of 38 domains of RNAP core enzyme (methods as in Zhang et al., 2012). Electron density for σ L and nucleic acids was unambiguous, but was not included in models in early-stage refinement. Cycles of iterative model building with Coot (Emsley et al., 2010) and refinement with Phenix (Adams et al., 2010)) then were performed. Improvement of the coordinate model resulted in improvement of phasing, and electron density maps for σ L and nucleic acids, which were not included in models at this stage, improved over successive cycles. σ L and nucleic acids then were built into the model and refined in stepwise fashion.
The final model was generated by XYZ-coordinate refinement with secondary-structure restraints, followed by group B-factor and individual B-factor refinement. The final model, refined to R work and R free of 0.19 and 0.23, respectively, was deposited in the PDB with accession code 6DVC (Table 1).
Analogous procedures were used to solve and preliminarily refine structures of Mtb RPitc5−σ L _sp4, RPitc5-σ L _sp5, RPitc5-σ L _sp6, and [BrU]RPo-σ L _sp6; models of σ L and nucleic acids then were built into mF o -DF c difference maps, and additional cycles of refinement and model building electrophoresed in TBE (90 mM Tris-borate, pH 8.0, and 0.2 mM EDTA), and analyzed by storage−phosphor scanning (Typhoon: GE Healthcare).

Transcript-release assays
Transcript-release assays (Fig. S3B,  Supernatants were mixed with 10 µl loading buffer, heated 5 min at 95°C, cooled 5 min on ice, and analyzed by urea-PAGE and storage-phosphor imaging as in the preceding section. Pellets were washed with 3x200 µl transcription buffer at 22°C; mixed with 50 µl loading buffer, heated 5 min at 95°C, cooled 5 min on ice, and analyzed by urea-PAGE and storage-phosphor imaging as in the preceding section.

Data analysis
Data for transcription activities are means of at least two technical replicates.

Data availability
Atomic coordinates and structure factors for the crystal structures of Mtb RPitc5-σ L _sp4,  (E) Transcription experiments demonstrating Mtb RNAP σ A holoenzyme selectively recognizes σ A -dependent promoter P-N25, and Mtb RNAP σ L holoenzyme selectively recognizes σ L -dependent promoter P-sigL.
(G) Transcription experiments demonstrating primer-dependent transcription initiation by Mtb RNAP-σ L holoenzyme on P-sigL derivatives having spacer ("sp") lengths of 4, 5, 6 and 7 bp (sequences in Fig. S2).  ApApUpU; identities confirmed by reference to products of parallel reactions omitting ATP and GTP; identities further confirmed by reference to products of parallel reactions with E. coli RNAP σ 70 (see Borowiec and Gralla, 1985)].