Binding interface between the Salmonella σS/RpoS subunit of RNA polymerase and Crl: hints from bacterial species lacking crl

In many Gram-negative bacteria, including Salmonella enterica serovar Typhimurium (S. Typhimurium), the sigma factor RpoS/σS accumulates during stationary phase of growth, and associates with the core RNA polymerase enzyme (E) to promote transcription initiation of genes involved in general stress resistance and starvation survival. Whereas σ factors are usually inactivated upon interaction with anti-σ proteins, σS binding to the Crl protein increases σS activity by favouring its association to E. Taking advantage of evolution of the σS sequence in bacterial species that do not contain a crl gene, like Pseudomonas aeruginosa, we identified and assigned a critical arginine residue in σS to the S. Typhimurium σS-Crl binding interface. We solved the solution structure of S. Typhimurium Crl by NMR and used it for NMR binding assays with σS and to generate in silico models of the σS-Crl complex constrained by mutational analysis. The σS-Crl models suggest that the identified arginine in σS interacts with an aspartate of Crl that is required for σS binding and is located inside a cavity enclosed by flexible loops, which also contribute to the interface. This study provides the basis for further structural investigation of the σS-Crl complex.

, and the DPE motif, corresponding to residues 135 to 137 and initially identified in σ S from E. coli 21 . Consistently, a fragment of σ S STM domain 2 lacking this motif, σ S STM  , and σ S STM variants at position D135 and E137, were not able to interact with Crl in bacterial two hybrid (BACTH) assays ( Supplementary  Fig. S1), confirming that the DPE motif in σ S STM is involved in Crl binding. However, the DPE motif is conserved in σ S of P. aeruginosa (σ S PA ) that does not have crl 22 , and the sequence of the helix α 2 differs from σ S STM by only four residues (Fig. 1c,d and Supplementary Fig. S2), prompting us to examine whether σ S PA can be activated by Crl. σ S PA activity and its response to Crl were evaluated in a S. Typhimurium strain in which the native rpoS gene was replaced by the rpoS allele from P. aeruginosa (rpoS PA ). Development of the rdar morphotype of S. Typhimurium is highly dependent on σ S and Crl 7 (compare spots 1, 2 and 11, Fig. 2). The S. Typhimurium strain harbouring the rpoS PA allele was able to develop the rdar morphotype, in contrast to the Δ rpoS mutant of S. Typhimurium (compare spots 3 and 11, Fig. 2), indicating that σ S PA was expressed and functional in this strain. However, the rdar morphotype of the rpoS PA strain was similar to that of the S. Typhimurium Δ crl mutant and was not affected by a Δ crl mutation (compare spots 2, 3 and 4, Fig. 2), suggesting that σ S PA did not respond to Crl. We also followed the expression of the rpoS-dependent katN-lacZ transcriptional fusion 13 and of the σ S protein, in wild type and Δ crl S. Typhimurium strains harbouring the rpoS STM and rpoS PA alleles ( Fig. 3 and Supplementary Fig. S3). The growth kinetics of strains harbouring rpoS STM and rpoS PA were similar and expression of katN-lacZ and σ S was induced in stationary phase, as expected 13 . The lower expression level of katN-lacZ in the rpoS PA strain, compared to that in the rpoS STM strain, could be due to differences in the expression level or intrinsic activities of σ S PA with respect to σ S STM . Activation of katN-lacZ expression by Crl in Salmonella harbouring the rpoS STM allele was maximal at the entry to stationary phase when σ S begins to accumulate (Fig. 3b), as previously reported 13 . Indeed, at the entry to stationary phase, katN-lacZ expression was decreased and delayed by the Δ crl mutation. In contrast, no significant effect of the Δ crl mutation on katN-lacZ expression was detected in Salmonella containing the rpoS PA allele (Fig. 3a), even though Crl amounts were similar in the rpoS PA and rpoS STM strains ( Supplementary Fig. S3c).
To determine whether the failure of σ S PA to respond to Crl activation resulted from a lack of interaction between the two proteins, we used the BACTH in vivo assay (Fig. 4) and isothermal titration calorimetry (ITC) in vitro assay ( Supplementary Fig. S4a). Unlike σ S STM 16,18 , σ S PA did not interact with Crl, suggesting that unidentified σ S STM residues, not conserved in σ S PA , are crucial for Crl binding.  A single amino acid substitution renders σ S PA sensitive to Crl activation. To further refine our understanding of σ S 2 residues involved in Crl binding, the amino acid sequence of σ S 2 was compared in bacterial species harbouring crl in their genome and in those lacking crl ( Fig. 1c and Supplementary  Fig. S2). Whereas the DPE motif was well conserved, the sequence of the helix α 2 was more variable, especially in σ S from strains lacking crl. In this region, the sequence between σ S STM and σ S PA differs by four surface-exposed residues (Y78, R82, L84 and R85 in σ S STM that correspond to H83, L87, Q89 and K90 in σ S PA , respectively) (Fig. 1d). Two of these (R82 and L84 in σ S STM ) are conserved in all σ S proteins from species harbouring crl, but less conserved in σ S from species lacking crl. To determine whether  the non-conserved residues at position 83, 87, 89 and 90 in σ S PA were responsible for the defect in Crl binding, we constructed σ S PA variants in which the σ S STM sequence was restored at these positions, and assessed their ability to interact with Crl in the BACTH assay (Fig. 4). Expression levels of σ S PA wild type and variants were similar (Fig. 4b). Interestingly, one variant, σ S PA L87R, was able to interact with Crl (Fig. 4a), suggesting that an arginine at position 87 in σ S PA (corresponding to position 82 in σ S STM ) is of paramount importance for Crl binding. This finding was further confirmed in vitro by ITC ( Supplementary Fig. S4c,d). Interestingly, σ S PA L87R and wild-type σ S STM showed similar affinity for Crl (Supplementary Table S1). The major difference observed between σ S PA L87R and σ S STM was in the value of Δ b H, which was less negative for σ S PA L87R than for σ S STM . This suggests that the number and type of intermolecular interactions in Crl-σ S PA L87R and Crl-σ S STM complexes might be slightly different, due to non-conserved σ S residues affecting directly or indirectly the σ S -Crl binding interface. However, the Δ b S and Δ b G values were similar for σ S PA L87R and σ S STM , endorsing the key role of an arginine at position 87 in the σ S PA variant. To assess whether the interaction between σ S PA L87R and Crl was functional (i.e. whether Crl activates σ S PA L87R), we monitored the rdar morphotype of S. Typhimurium harbouring the chromosomal rpoS PA-L87R allele expressing σ S PA L87R. Morphotypes of the rpoS PA-L87R and wild type Salmonella strains were similar and dependent on crl (compare spots 1, 7 and 2, 8, Fig. 2), suggesting that σ S PA L87R was able to respond to Crl activation. Consistently, katN-lacZ expression level in the rpoS PA-L87R strain was decreased by the Δ crl mutation (Fig. 3a). Altogether these results demonstrated that substitution L87R renders σ S PA sensitive to Crl activation.

Residue R82 in σ S STM is required for Crl binding and activation. To evaluate the impact in Crl
binding of σ S STM residue R82, corresponding to L87 in σ S PA , (Fig. 1d), the ability of the σ S STM R82L variant to interact with Crl was assessed in BACTH and ITC assays ( Fig. 4 and Supplementary Fig. S4b). Both assays showed that σ S STM R82L does not interact with Crl. Consistently, development of the rdar morphotype and katN-lacZ expression levels were similar in the rpoS STM-R82L strain (whatever its crl status) and the Δ crl strain harbouring the wild-type rpoS STM allele (spots 5, 6 and 2, Fig. 2 and Fig. 3b), indicating that σ S STM R82L was not activated by Crl. Far-UV CD spectra showed a similar secondary and tertiary structure for σ S STM , σ S STM R82L, σ S PA and σ S PA L87R ( Supplementary Fig. S4e,f), indicating that the σ S conformation was similar in the four proteins. In the absence of Crl, katN-lacZ expression level was similar in the rpoS STM-R82L and rpoS STM strains (Fig. 3b), suggesting that σ S stability and its interaction with the core RNAP were not affected by the R82L substitution. To assess the effects of a more drastic amino acid substitution at position 82, the σ S STM R82E variant was characterized. Expression level and activity of this variant were similar to those of σ S STM R82L (spots 9, 10 and 5, 6, Fig. 2 and Supplementary Fig.  S5). Altogether, these findings suggested that σ S STM R82 plays a key role in Crl binding and activation.
Solution structure of Salmonella Crl. We previously reported the X-ray crystal structure of Crl from Proteus Mirabilis (Crl PM ) (PDB 4Q11 18 ), which suggested a high degree of flexibility of the protein.
To get more insights into the dynamics of Crl, we solved the solution structure of Crl STM by NMR 23 (Fig. 5, Supplementary Table S2). Structural alignment with Crl PM indicated that the fold of Crl STM is conserved with a core consisting of a five-stranded β -sheet flanked by two helices, α 1 and α 3, with a cavity on top, closed by loops 1 and 2 ( Supplementary Fig. S6). The electrostatic surface potential of Crl STM delimits two faces of the protein, corresponding to lateral entries of the cavity (Fig. 5b). One face is overall neutral with several basic patches, whereas the opposite face is predominantly negatively charged, like loop 3 and the inside of the cavity, which are also rather acidic. The NMR ensemble structure ( Fig. 5a) showed that several regions at the periphery of the core display structural disorder. NMR signals in loop 1 (L19-F33) were broad, possibly due to conformational exchange at the millisecond timescale. Still a number of NOE contacts were found with α 1 and β 2, showing that it is not completely disordered. Due to the absence of the small helix α 2 found in Crl PM , loop 1 of Crl STM explores a wider space and contributes to forming a deeper cavity than in Crl PM ( Supplementary  Fig. S6). Residues in loop 2 displayed sharp signals but only few NOE contacts, indicating that this region is disordered and flexible. The difference of dynamics in loop 2 as compared to the structured regions was also corroborated by 15 N relaxation experiments, where it displays lower R 2 rates that deviate from simulated R 2 values ( Supplementary Fig. S7). Finally the region corresponding to helix α 4 in Crl PM is not structured in Crl STM (Supplementary Fig. S6b). Indeed, only few inter-residue NOE contacts were found in the P120-P128 region, but they provided evidence of the proximity between the C-terminus and helix α 3. Strikingly, signals of several residues in the core β -sheet were broad, denoting conformational fluctuations that might be coupled to those in loop 1. The corresponding side chains could not be constrained during structure calculation, prominently that of W82 in strand β 4, which points towards the cavity in the crystal structure, but appears to flip out in the NMR structures (Fig. 6c).
Structural analysis of the Crl STM D36A mutant. As shown previously, the Crl STM D36A variant neither activates nor binds to σ S STM 18 , but it was not clear if this was due to structural alterations, since previous biophysical data suggested that the substitution could lead to partial loss of secondary and tertiary structure. Therefore we investigated the structural integrity of Crl STM D36A by analysing its backbone chemical shifts. Signal overlap between wild-type and D36A Crl STM allowed to partly transpose chemical shift assignments from wild type to D36A Crl STM (Fig. 6a). But chemical shift perturbations (CSPs) were not restricted to the region of the mutation (Fig. 6b) and de novo backbone assignment had to be carried out. The data showed that there is no major difference for 13 C' or 13 Cα chemical shifts, excepted for D36 and C37 (Fig. 6b), indicating that the secondary structure and overall fold are conserved in the mutant. In contrast, amide chemical shifts were significantly perturbed all over the sequence, even if the largest CSPs were also observed around the mutation. They seem to be relayed from D36 in strand β 1 to β 4, via β 2 and β 3, and to loops 1 and 3 (Fig. 6c). CSPs in loops 1 can be traced back to the salt bridge formed between the R24 guanidinium and the D36 carboxylate in the X-ray structure of Crl PM as well as in most NMR conformers of Crl STM (Fig. 6c). When Asp is replaced by Ala, this interaction is disrupted, allowing loop 1 more conformational freedom. Loop 3 could be affected by breaking the hydrogen bond between D36 and the W82 indole observed in the crystal structure of Crl PM (Fig. 6c). This hydrogen bond is not present in the wild-type Crl STM NMR structure, but it cannot be ruled out that it is transiently formed in solution. Amide CSPs inside the β -sheet, but far from position D36, could be explained by a slight reorganization of the hydrogen bond network. Altogether, these results endorse the role of residue D36 in σ S binding. NMR analysis of the Crl STM binding interface for σ S STM . We next characterized the influence of σ S STM on Crl STM NMR spectra. 1 H-13 C HSQC spectra displayed line broadening, i.e. a decrease of intensities, in particular in the methyl region on addition of σ S (Supplementary Fig. S8). Differential broadening was observed in loop 2 and helix α 3. However, since methyl groups are mainly pointing to the inside of the structure and are not homogeneously distributed throughout the sequence, they may not be very sensitive probes for the Crl-σ S interaction, which was suggested to rely on electrostatic interactions 18 . σ S STM also induced overall line broadening in Crl STM 1 H-15 N HSQC spectra, as a consequence of faster transverse relaxation in the Crl-σ S complex than in free Crl, and additional line broadening for several residues (Fig. 7a,b, e.g. residue N43), due to exchange between free and complexed Crl. These are mainly clustered in loop 2 (Fig. 7b,c) which contains R51, one of the key residues for σ S binding 17,18 . Since this region appears to be flexible in free Crl, the dynamics of loop 2 certainly plays a role in the formation of the Crl STM -σ S STM complex. Helix α 1 and loop 1 also seem to be affected by σ S STM (Fig. 7b,c).  Modeling of the σ S -Crl complex based on mutational analysis. Charged residues in Crl, D36 and R51, were previously found to be essential for σ S binding 18 . The results above corroborate the hypothesis that the Crl-σ S complex formation is likely driven by electrostatic interactions by demonstrating that one positively charged residue, R82 in σ S STM , is of paramount importance for Crl binding and activation. In addition, two acidic residues in σ S , D135 and E137, were spotted as likely candidates for interaction with Crl ( Supplementary Fig. S1 and 21 ).
To integrate these data, we modelled the Crl-σ S complex from a structural model of S. Typhimurium σ S 2 16 and the Crl STM NMR structure. In a first step we performed normal mode analysis (NMA) on both Crl and σ S to detect collective low-frequency motions that could provide conformations more favourable for complex formation than the starting structures of isolated proteins. In the case of σ S , although shearing movements take place between helix α 2 and the DPE loop, residues R82 and E137 do not move wide apart and belong to a common interaction surface (Supplementary Fig. S9). In the case of Crl, collective motions of the three loops remodel the cavity either by closing it or widening it, which would help accommodating σ S 2 ( Supplementary Fig. S10). In a second step, Crl-σ S complexes were obtained in silico, using the information of critical binding residues and two different docking strategies. In the first strategy, we used the ZDock server 24 in combination with refinement on the RosettaDock server 25 , that do not take into account conformational changes and flexibility of proteins (Models A to E, Supplementary Fig. S11). The second strategy used the Haddock Webserver 26,27 to integrate the high degree of flexibility of the NMR structure of Crl ( Supplementary Fig. S12).
In models A and B ( Supplementary Fig. S11), σ S R82 interacts with the Crl residues E25 or E102, respectively. These residues are not conserved in Crl family members 22 and the Crl E25A and E102A variants interacted with σ S STM in the same manner as wild type Crl ( Supplementary Fig. S1) indicating that E25 and E102 are not required for σ S binding. It is noteworthy that in these two models the DPE motif does not have any interacting partner. Altogether these data suggest that models A and B do not represent the Crl-σ S interface. Models C and D are also unlikely since E137 in σ S interacts with R24 in Crl, a residue dispensible for σ S binding 18 . Furthermore, in model C, R82 in σ S interacts with E25 in Crl, which is not involved in σ S binding ( Supplementary Fig. S1) and in both models R51 in Crl does not have any possible charged interacting partner in σ S .
From the five models generated by the first docking strategies, model E appears the more likely. In this model two salt bridges are formed involving the critical binding residues R82 and E137 in σ S and D36 and R51 in Crl (Fig. 8). Moreover, several van der Waals and hydrogen bond interactions between the σ S helix α 2 and both loop 1 and β 1 of Crl, and the σ S loop containing the DPE motif and loop 2 of Crl, can further contribute to the σ S -Crl complex ( Supplementary Fig. S13), in agreement with NMR data which suggest that also loop 1 in Crl is affected upon σ S STM binding. In the second series of docking experiments using Haddock Webserver 26,27 , pairs of active residues with complementary charges straightforwardly formed salt bridges ( Supplementary Fig. S12), most often Crl-D36/σ S -R82. It was not possible to restrain the Crl-R51/σ S -E137 pair to form a salt bridge, but in a number of clusters the two loops that contain these two residues were in close contact, corroborating the relevance of model E for the σ S -Crl interface and in agreement with the NMR interaction experiments that pointed to the role of loop 2 for complex formation.
Residue D87 was previously pointed as important for Crl binding in E. coli 21 . The amino acid sequence of σ 2 is identical in σ S STM and σ S from E. coli and, in the σ S STM structural model, D87 is located at the edge of helix α 2, with its side chain directed on the opposite face with respect to residue R82, as imposed by the geometry of an α -helix ( Supplementary Fig. S14). Therefore D87 is unlikely to interact directly with Crl. Consistent with this hypothesis, in the study by Banta et al. 21 , some amino acid substitutions of D87, such as D87C, did not drastically affect the σ S interaction with Crl.

Discussion
In many Gram-negative bacteria, σ S /RpoS is the master regulator of gene expression in stress conditions and during stationary phase. σ S is exquisitely and tightly regulated by many mechanisms that keep its production level and activity under strict control [3][4][5] . Crl is a unique regulatory factor, specifically dedicated to σ S , which enhances its activity, helping the association of σ S with E 15 . Nevertheless, there are some rpoS-containing species, including P. aeruginosa, that do not harbour a crl gene 16 and in which σ S activity may be controlled by alternative mechanisms or functional homologs of Crl.
The strong sequence conservation of σ S 2 , the only σ S domain that binds Crl 16,18,21 , prompted us to assess possible activation of σ S PA by Crl. We show here that σ S PA is not activated by Crl due to its inability to interact with Crl. Taking advantage of the evolution of the σ S sequence in P. aeruginosa and other species lacking crl, we identified residues conserved in σ S sequences from crl proficient species, and potentially implicated in Crl recognition. Among these, a surface-exposed arginine in σ S STM , R82, was assigned to the σ S -Crl interface. This residue is not conserved in σ S PA , which instead contains a leucine. Importantly, substitution of this leucine by an arginine rendered σ S PA sensitive to Crl activation. It is noteworthy that, in some σ S proteins from species that do not harbour crl, the arginine residue is conserved ( Supplementary Fig. S2). It would be interesting to determine whether these σ S proteins interact with Crl, and if not, whether they could be used to identify additional σ S residues involved in Crl binding by the strategy described in this study for σ S PA . The in silico models of the σ S -Crl complex show that salt bridges can indeed be formed for the two pairs of residues Crl-D36/σ S -R82 and Crl-R51/σ S -E137. In some models they can be formed simultaneously. This leads to a picture of an ideal binding interface in which helix α 2 of σ S , containing R82, would dock into the cavity of Crl containing D36, disrupting the intermolecular R24-D36 contact, and the DPE motif and loop 2 of Crl would make contact on the outside, driven by electrostatic interactions between Crl-R51 and σ S -D135/E137 (Fig. 8).
What renders the σ S -Crl system very intriguing is its transitory and dynamic binding mechanism, which is unclear so far. Our NMR data together with the in silico modelling shed some light on how σ S and Crl may interact and form a transient complex. The chemical shift perturbations in the NMR spectrum of Crl in the presence of σ S indicate that loop 2 senses the presence of σ S , but extend beyond the region directly involved in σ S binding, including helix α 1, loop 1 and helix α 3. These findings suggest that local structural rearrangements might take place in the flexible loops that allow breathing of the cavity as indicated by normal mode analysis of the Crl structure. Such rearrangements might contribute not only to the formation of the σ S -Crl complex, but also to its dissociation, once Crl has accomplished its work. Moreover, in free Crl, residue D36 is involved in an intramolecular interaction with R24. To form a new salt bridge with σ S -R82, the first one has to be broken. The perturbations observed in the NMR spectra of the Crl D36A variant show how the disruption of this network is sensed by the whole Crl structure, in particular by loop 1. It is tempting to speculate that this variant mimics the molecular processes that Crl undergoes upon σ S binding, as we previously hypothesized 18 .
How does Crl binding to σ S increase the σ S association rate with E? Why is the σ S -Crl interaction so transient? These questions are still open. One possibility is that Crl triggers a conformational change in σ S favouring its association with E. There is no high resolution 3D structure for free σ factors, but several biochemical and structural studies using the housekeeping σ 70 have shown that σ factors undergo pronounced conformational changes upon E binding, allowing domains σ 2 and σ 4 to be spaced correctly for promoter binding 2,20 . These findings have led to the proposal that σ factors must be in a more compact conformation when free in the cell than in the Eσ complex. Consistent with this hypothesis, free σ are not able to bind promoters efficiently. This concept was further supported by the results obtained with engineered cysteine mutants of σ 28 , which showed that this σ factor has a compact conformation when free in solution 28 .
Modulation of the free σ S conformation might be a common way to regulate both the stability and activity of σ S . σ S is degraded by the ATP-dependent complex ClpXP protease [3][4][5] . However, σ S binding by the RssB protein is required for delivery to ClpXP [3][4][5] . It has been postulated that RssB binding triggers a conformational opening of σ S that exposes a ClpXP binding site, that is otherwise occluded in a closed conformation of free σ S5 . Therefore, if the conformation of σ S in the cell is rather compact, Crl binding to σ S 2 may alleviate intramolecular interactions between σ S 2 and other σ S domains, favouring an open conformation for just the time required for σ S to bind E, but transiently enough to avoid σ S degradation by ClpXP. Further investigation of the structure of the σ S -Crl complex, for which a starting base is provided in the present study, and of the free σ S conformation will assess the relevance of this scenario.

Methods
Bacterial strains, bacteriophage, plasmids and growth conditions. Strains and plasmids used for this work are listed in Supplementary Table S5. Bacteriophage P22HT105/1int was used to transfer mutations and the katN-lacZ fusion between Salmonella strains by transduction 29 . Green plates, for P22-infected cells or lysogens screening, were prepared as described previously 30 . Strains were grown in Luria-Bertani (LB) medium 31 at 37°C under aeration. Development of the rdar morphotype was monitored on CR plates (LB agar without NaCl supplemented with Congo red 40 μ g/ml and Coomassie brilliant blue R250 20 μ g/ml), at 28°C as described 7 . Antibiotics were used at the following concentrations: ampicillin (Ap) 100 μ g/mL; carbenicillin (Cb) 100 μ g/mL; chloramphenicol (Cm) 15 μ g/mL Scientific RepoRts | 5:13564 | DOi: 10.1038/srep13564 for the chromosomal resistance gene and 30 μ g/mL for the plasmid resistance gene; kanamycin (Km) 50 μ g/mL; and tetracycline (Tet) 20 μ g/mL. rpoS allelic exchange in Salmonella. Allelic exchange of rpoS in S. Typhimurium ATCC14028 was achieved with a two-step Red-recombinase-based recombineering procedure [32][33][34][35] . The procedure involves replacement of the tetRA module of strain VFC326 by PCR-amplified DNA fragments of the rpoS allele from pVFC629, pVFD410, pVFD412 and pVFD399 (Supplementary Table S5 and S6) through positive selection of tetracycline-sensitive recombinants. All strains were confirmed to contain the expected mutation by DNA sequencing.
Protein production and BACTH assays. The N-terminal (his) 6 -tagged σ S PA wild type and variant L87R, σ S STM R82L variant and Crl STM were produced in E. coli strain BL21 (DE3) harbouring plasmid derivatives of pETM11 (Supplementary Table S5). Production and purification of the proteins were carried out as previously described 18 . 15 N-, 13 C 15 N-or 15 N 2 H-labeled wild type (his) 6 -Crl STM and 15 N 13 C-labeled Crl STM (his) 6 -D36A protein samples for NMR experiments were produced in minimum M9 medium 31 supplemented with 15 NH 4 Cl and unlabelled or 13 C-or 2 H-labeled glucose following the same protocol as 18 . Samples were subsequently dialyzed into NMR buffer (50 mM sodium or potassium phosphate, 300 mM NaCl or KCl, 2 mM dithiotreitol, at pH 8 or 7.5).
For bacterial adenylate cyclase-based two hybrid assay, the E. coli cya strain DHT1 was transformed with derivatives of plasmids pKT25 and pUT18 encoding σ S and Crl proteins fused to the C-terminal part of T25 and the N-terminal part of T18, respectively (Supplementary Table S5). Co-transformants were plated onto MacConkey maltose plates supplemented with Cb, Km, and 0.5 mM IPTG to assess the Mal phenotype and on LB plates supplemented with X-Gal (40 μ g/ml) Cb, Km, and IPTG (0.5 mM) to assess the Lac phenotype. Plates were incubated at 30 °C for 2 days and then isolated colonies were grown in LB supplemented with Cb, Km, and IPTG, at 30 °C for 20 hours. β -galactosidase activities were measured as described by Miller  The scaling factor 1/10 corresponds to the gyromagnetic ratio difference between 15 N and 1 H.
NMR structure calculation. NMR structures of wild-type Crl STM were calculated using torsion angle dynamics in CYANA 2.2 39 . Backbone torsion angle restraints were generated with TALOS-N 40 using Crl STM backbone chemical shifts. Ambiguous distance restraints were collected from three sets of NOESY spectra and purged from 3D peaks without possible assignments in the 1 H dimension bound to a heteroatom. The disordered N-terminal His-tag (His(− 20)-His(0)) was excluded from structure calculation. Structure statistics were obtained from the Protein Structure Validation Server, version 1.5 (http://psvs-1_5-dev.nesg.org/) (Supplementary Table S2). Normal Mode Analysis was performed on single conformers on the ElNémo webserver 41 .
Protein-protein docking. Rigid-body docking was carried out first on the ZDock server 24 , which employs a fast Fourier transform (FFT) algorithm, to generate the initial models (about 100) for the σ S -Crl complex with Crl STM as the receptor and a homology model of S. Typhimurium σ S 2 as the ligand 16 . Five models selected from ZDock were further refined using RosettaDock 25 , which performs a searching for the lowest-energy binding interface structures giving as ouput the 10 best-scoring models from 1000 total models. The presence of a 'docking funnel' was verified, considering that at least three of the first five lowest-energy binding interface models have a value of I_rmsd < 4 Å 42 (Supplementary Table S3).
Flexible docking was carried out on the guru interface of the Haddock Webserver 26,27 using single conformers from the NMR structure ensemble of Crl STM and the homology model of S. Typhimurium σ S 2 . D36 and R51 in Crl and R82, D135 and E137 in σ S were defined as active residues. Passive residues were automatically defined around active residues. Loops 1 (E25-R32) and 2 (N43-E52) in Crl and the DPE loop (K133-F142) in σ S were defined as fully flexible. 1000 initial structures were generated. 200 Scientific RepoRts | 5:13564 | DOi: 10.1038/srep13564 final structures were refined in water and clustered according to RMSD criterion. Statistics for clusters obtained for the conformers 1 and 2 are given in Supplementary Table S4. Illustrations. Visualization and graphic rendering of protein structures were prepared with PyMOL 43 .
Other methods. Methods for DNA manipulation, immunoblot analysis of proteins and CD and ITC experiments are described in Supplementary Methods.