Mycobacterium tuberculosis FasR senses long fatty acyl-CoA through a tunnel, inducing DNA-dissociation via a transmission spine

Mycobacterium tuberculosis is a pathogen with a unique cell envelope including very long fatty acids, implicated in bacterial resistance and host immune modulation. FasR is a two-domain transcriptional activator that belongs to the TetR family of regulators, and plays a central role in mycobacterial long-chain fatty acyl-CoA sensing and lipid biosynthesis regulation. We now disclose crystal structures of M. tuberculosis FasR in complex with acyl effector ligands and with DNA, uncovering its sensory and switching mechanisms. A long tunnel traverses the entire effector-binding domain, enabling long fatty acyl effectors to bind. Only when the tunnel is entirely occupied, the protein dimer adopts a rigid configuration, with its DNA-binding domains in an open state that leads to DNA dissociation. Structure-guided point-mutations further support this effector-dependent mechanism. The protein-folding hydrophobic core, connecting the two domains, is completed by the effector ligand into a continuous spine, explaining the allosteric flexible-to-ordered transition. The transmission spine is conserved in all TetR-like transcription factors, offering new opportunities for anti-tuberculosis drug discovery.


INTRODUCTION
The composition and complexity of the mycobacterial cell envelope are two of the most distinctive features of this genus. The complex lipids present in the cell wall of Mycobacterium tuberculosis (Mtb) act as major effector molecules that interact with the host, playing key roles in pathogenicity and also providing a barrier against environmental stress, antibiotics, and the host's immune response (1).
Understanding the biogenesis of the mycobacterial cell wall will thus provide with relevant insights into the biology of this pathogen, also identifying potential targets for the development of new antimycobacterial compounds.
The outer membrane of Mtb comprises very long-chain fatty acids (mycolic acids), found in the inner leaflet covalently bonded to the arabinogalactan-peptidoglycan layer, and also in the outer leaflet as non-covalently associated lipids in the form of trehalose-mono-and di-mycolate (2). Mycolic acids, a hallmark of Mycobacterium, are synthesized by way of two fatty acid synthase systems, FAS I and FAS II. The multidomain single protein FAS I catalyses de novo biosynthesis of acyl-CoAs in a bimodal fashion rendering C [16][17][18] and C [24][25][26] derivatives (3). Long chain acyl-CoAs are used as primers by the FAS II multiprotein system, and iteratively condensed with malonyl-acyl carrier protein (malonyl-ACP) leading to very long-chain meromycolyl-ACPs (up to C 56 ). The latter are eventually condensed to FAS I-synthesized C [24][25][26] fatty acids (previously activated by the acyl-CoA carboxylase 4 complex (4)) to produce mycolic acids. The long-chain acyl-CoAs generated by FAS I are not only used as precursors of mycolic acids, but also for the biosynthesis of phospholipids, triacylglycerides, diverse polyketides and other complex lipids, relevant for Mtb pathogenicity (5,6). A complex regulatory network integrating all these pathways must exist in order to maintain lipid homeostasis.
However, and despite the biological relevance of lipid-derived molecules in Mtb's lifecycle, little is known about the environmental signals and the regulation cascades controlling lipid metabolism in this bacterium.
We had previously identified the transcription factor FasR, as a key activator of fas and acpS gene expression in mycobacteria (7). The fas and acpS genes form an operon, respectively coding for the FAS I synthase and the 4-phosphopantetheinyl transferase, the latter being an essential enzyme to produce functional ACP, central to all fatty-acid biosynthesis. To activate the Mtb fas-acpS operon, FasR binds to three inverted repeats in the promoter region, a binding that is regulated by long-chain acyl-CoAs, products of FAS I (7). More specifically, acyl-CoAs ≥ C 16

disrupt the interaction of Mtb
FasR with its cognate DNA. Additionally, FasR is essential for M. smegmatis viability (7), further highlighting its key role in mycobacterial biology.
FasR sequence reveals homology to members of the TetR family of regulators (TFRs), which are one-component sensory transduction proteins (8), typically dimeric and with each protomer displaying a 2-domain all-helical structure (9). By aligning FasR to orthologous TFRs with known 3D structures ( Supplementary Fig. S1A) (10), FasR is expected to comprise a helix-turn-helix DNA-binding domain towards the N-terminus (residues 1-80), and a larger C-terminal domain corresponding to the ligandor effector-binding domain (residues 82-228). In contrast to the DNA-binding domain, the effectorbinding region reveals no sequence homology with other TFR proteins, such sequence diversity being consistent with the large variety of effector signals sensed by different TFRs (11). The prototype of the TFRs is TetR from the Tn10 transposon of Escherichia coli, which regulates the expression of the tetracycline efflux pump in Gram-negative bacteria (12). However, TFR proteins are widely distributed among bacteria, and control a broad range of processes, including fatty acid biosynthesis (13). 3 Interestingly, the vast majority of TFR proteins are transcriptional repressors, with very few exceptions acting as activators (11).
We have now determined the 3D structures of FasR from Mtb, in three different states obtained from i-the protein crystallized alone, ii-co-crystallised in complex with the fatty acid C 20 -acyl-CoA, and iii-in complex with a double-stranded DNA oligonucleotide bearing the specific FasR-binding sequence. The comparison of these crystal structures, together with the functional characterisation of structure-guided FasR point-mutants and molecular dynamics computational simulations, uncovered the molecular mechanisms by which long-(C 16 -C 20 ) and very long-chain (C 22 -C 26 ) acyl-CoA molecules are sensed by FasR, as well as the means by which such signal disrupts cognate FasR-DNA binding and hence actuates fas-acpS transcriptional activation. World-wide efforts have disclosed hundreds of protein structures from Mtb corresponding to potential drug targets, a valuable input for a number of drug discovery projects (14,15). The uncovering of novel structural and mechanistic insights about a key Mtb metabolic regulator, contributes with solid molecular bases for target-based drug discovery, certainly one of the sensible strategies to combat tuberculosis.

Bacterial strains and plasmids
E. coli strain DH5α cells were used for DNA cloning purposes and transformed according to standard methods. For protein expression the E. coli BL21 λ(DE3) strain was used instead. The fasR gene (rv3208) was PCR-amplified from Mtb H37Rv genomic DNA using the oligonucleotides F-Rv3208 (5′-CCCTCCATATGGAAAACCTGTACTTCCAGGGTATGAGCGATCTCGCCAAG-3′) to introduce an NdeI site at the translational start codon and encode a fused Tobacco Etch Virus protease digestion site (TEV), and R-Rv3208 (5′-GAATTCCTACGAGCGGGTAAGCG-3′) to introduce an EcoRI site at the end of the ORF. To generate a FasR recombinant protein with a hexa-histidine-TEV-tagged site at the N-terminus, the corresponding PCR product was extracted from the agarose gel, cloned into the pCR BluntII TOPO vector (Invitrogen), digested with NdeI and EcoRI, and finally subcloned into the expression vector pET28a as described by the manufacturer (Merck). The recombinant plasmid (pET28_ fasR) was transformed into BL21 λ(DE3) cells, and transformants were selected on LB agar plates containing 50 μg/ml kanamycin. To produce the His-tagged FasR version used in all EMSA experiments (FasRwt), fasR was PCR-amplified from genomic DNA of Mtb H37Rv using the oligonucleotides F-Rv3208 (5′-CATATGAGCGATCTCGCCAAGACA-3′) to introduce a NdeI site at the translational start codon, and R-Rv3208 (5′-GAATTCCTACGAGCGGGTAAGCGG-3′) to introduce an EcoRI site at the end of the ORF. To generate a fasR His-tag fusion gene, the PCR product was cloned into the pCR BluntII TOPO vector, digested with NdeI and EcoRI and subcloned into NdeI/EcoRI-cleaved pET28a. The resulting plasmid (pET28_ fasR H ) was transformed into BL21 λ(DE3) cells. The plasmids to produce recombinant FasR Δ33 and FasR point mutants (FasR LVL , FasR L106F , FasR L98A and FasR F123A ) with N-terminal hexa-histidine-TEV-tagged sites, were synthesized (GenScript). NdeI/HindIII restriction sites were engineered in a pUC57-Am plasmid where genes of interest were introduced. Synthetic plasmids were digested with NdeI and HindIII, extracted from the agarose gel, and inserted into the expression vector pET28a, as described by the 4 manufacturer (Merck). The recombinant plasmids were transformed into BL21 λ(DE3) cells and transformants selected on LB agar plates with 50 μg/ml kanamycin. DNA sequences of all genes were verified by Sanger sequencing.

Expression and purification of proteins
Expressions of all N-terminally hexa-histidine-TEV-tagged or hexa-histidine-tagged proteins used in this work were carried out following isopropyl-β-thiogalactoside (IPTG) induction in BL21 λ(DE3) E.
coli. Bacteria were grown at 37 C in 500 ml LB broth to 0.6-0.7 absorbance at 600 nm. IPTG was then added to 0.3-0.5 mM and the culture was grown for 12 h at 23 C. Cells were harvested by centrifugation at 2800 g at 4ºC, resuspended in 30 ml of lysis buffer (50 mM Tris.HCl pH 8, 150 mM NaCl, 5 mM imidazole, 10 % glycerol, 10mM β-mercaptoethanol) and lysed by sonication.
After centrifugation (25000 g, 30 min, 4 C), the supernatant was recovered and FasR , FasR Δ33 or FasR point mutants were separated from whole-cell lysates by Ni-NTA agarose chromatography (Qiagen, Inc). After three washing steps with lysis buffer, His 6 -tagged FasR were eluted from the resin with 250 mM imidazole in lysis buffer, dialysed overnight against FasR Buffer 1 (10 mM Tris.HCl pH 8, 300 mM NaCl). In the case of hexa-histidine-TEV-tagged proteins (FasR, FasR Δ33 , FasR LVL , FasR L1069F , FasR L98A and FasR F123A ) after three washing steps with lysis buffer, 0.5 mg TEV protease and DTT to 1 mM final concentration were added. The mixture was incubated 2-3 hs at 23 C and 12 hs at 4 C. Proteins were eluted from the resin with 5 mM imidazole in lysis buffer and dialysed overnight against FasR Buffer 2 (10 mM Tris.HCl pH 8, 300 mM NaCl, 5% glycerol). A final sizeexclusion chromatography step (S200-Superdex, GE) was performed for all proteins, pre-equilibrating the column in FasR Buffer 2 and eluting isocratically at 0.5 ml/min with high-performance liquid chromatography (Akta Purifier, GE). FasR LVL was the only mutant that showed detectable levels of monomeric form eluting from SEC (all other proteins eluting as dimeric species), in which case the peak corresponding to the dimer was recovered for further functional analyses by EMSA. Protein purity was controlled by Coomassie blue staining after SDS-PAGE on a 15 % polyacrylamide gel.
Protein concentrations were determined by UV spectroscopy. Purified proteins were stored at 4 C.

Surface plasmon resonance
Surface plasmon resonance analyses were performed with a Biacore T100 instrument (GE Healthcare). Pure FasR and FasR F123A were dialysed against 10 mM sodium acetate pH 4.5 and coupled to a CM5 sensor chip using the Amine Coupling Kit (GE Healthcare). Micromolar concentrations of C 20 -CoA were dialysed against 25 mM Tris.HCl pH 8, 150 mM NaCl, 0.005 % Tween 20. Serial two-fold dilutions were made in the same buffer and injected over the chip surface.
Dissociation was then carried out by injecting buffer alone. Nonspecific binding was considered by injecting identical analyte concentrations over a control surface with no protein. Experiments were performed at 25 C by triplicate, producing standard deviations of less than 12 %. Data were analysed using Biacore T100 Evaluation software.
Electrophoretic mobility shift assays (EMSAs) 5 His 6 -tagged FasR, and FasR point mutant proteins (FasR LVL , FasR L106F , FasR L98A and FasR F123A ) were used to assess the protein binding to Pfas MT (398 bp) promoter fragment. The promoter DNA fragment for these assays was generated by PCR amplification from Mtb genomic DNA with primers N2_Fas1Mt-prom (5′-CATAACGATTTGATAACAAAACTGC-3′) and C_Fas1Mt-prom (5′-CACCCGGTCGTGCTCGTGGATCGTC-3′). N2_Fas1Mt-prom primer was end-labelled with [γ- The equilibrium dissociation constant (K D ) is a quantitative measurement to assess the affinity of biological interactions. For the FasR:DNA EMSA binding experiments described in this work we define the K D as the concentration of FasR for which 50% of the DNA is in complex with the protein.
The relationship between K D and affinity is reciprocal, lower K D s correspond to higher affinities. After performing binding reactions in which the protein is titrated, the fraction of DNA bound at each concentration of protein is calculated and the data are adjusted to a binding equation using nonlinear regression (16). Densitometry of DNA bands was performed with GelPro software considering background subtraction. The fraction of bound DNA was plotted as a function of protein concentration and fit to the following binding equation (Eq. 1) using Prism software to perform non-linear regression: EMSA experiments with FasR wt and selected FasR mutants were performed by triplicate, to express app K D and IC 50 as average values ± one standard error of the mean.
Apparent inhibition constants ( app K i ) were also calculated (17,18), to correct IC 50   Bragg diffraction intensities were integrated with XDS (22), and scaled and reduced to amplitudes with Aimless and Ctruncate (23).

Structure determination and refinement
The structure of FasR Δ33 -C 20 -CoA was solved ab initio with Arcimboldo (24,25) which uses Phaser (26) as molecular replacement (MR) engine to place α-helices, and ShelxE (27)  The FasR-DNA complex was solved by MR (26) using a portion (corresponding to the dimeric regulatory domains with no ligands nor HTH domains included) of the refined FasR Δ33 -C 20 -CoA model as search probe. Initial refinement (33) with this partial model reduced R-factors to <50 %, and produced Fourier difference maps that clearly revealed the presence of both HTH domains and double-stranded DNA. Limited resolution and model incompleteness resulted however in mediocre 2mF obs -DF calc maps at this point. ShelxE (27) was instrumental at improving electron density continuity, using the unrefined MR solution model and diffraction intensities as inputs, a larger than usual sphere of influence (5 Å) for density modification, the free-lunch option set at 3.85 Å to better handle data incompleteness at the higher resolution shells, and testing different solvent contents (from 0.4 to 0.6).
All output maps were visualized superposed, readily allowing for manual main-chain tracing (29) of HTH and DNA portions, only including residues clearly visible in electron density at each cycle. This procedure was cycled iteratively using the progressively more complete protein models, running 5 cycles of ShelxE density modification each time. For the manual model (re)building, recent developments for real-space fitting in very low-resolution maps within Coot (34) proved essential, using Prosmart-generated external restrains, map blurring and optimized Geman-McClure parameters (alpha=0.4 proved best) for optimal movement of DNA within electron density.
The orthorhombic C222 1 space group of the FasR-DNA crystals was confirmed by several standard procedures (23,35), notably including integration in the corresponding triclinic group and using molecular replacement as a means of deducing real symmetry. However, the DNA molecule Structural analyses were done with the CCP4 suite (37), PISA (38) and illustrations produced with Pymol (39). 8

Structural bioinformatics to define the hydrophobic spine
FasR structural homologues were searched with PDBeFold (40), thus retrieving a wide range of sequence similarities. Structural alignment of such hits was performed with T-Coffee Expresso (41).
The resulting 337 sequences were filtered keeping only 76 that had < 80% identity. This multiple sequence alignment (MSA) served to generate a profile hidden Markov model using HMMBuild (42).
The UNIPROT database was searched with this profile HMM, CD-HIT (43) (46).

Molecular Dynamics simulation of FasR Δ33 bound to C 26 -CoA
The FasR Δ33 -C 26 -CoA complex was built using the FasR Δ33 -C 20 -CoA (PDB 6O6N) model as template. The bound acyl-CoA was manually extended by six carbons using Pymol (39). C 26 -CoA was optimized and 10,000 rotamers were generated with RDKit (47). Energy minimization was performed with the Rosetta suite (48) using dimer symmetry constraints, harmonic restraints were used to preserve the ligands positions as observed in the crystal structure, and 10,000 models were generated. The best complex was selected based on Rosetta energy score, with optimal stereochemical geometry and no clashes. The selected model was used as starting structure for classical molecular dynamics simulations using Gromacs 2018_cuda8.0 and GROMOS96 43a1 force field (49). An octahedron box was solvated and charge-balancing counterions were included to neutralize charges (50). Initially, the system was relaxed by energy minimization, and then equilibrated for 200 ps using a reference temperature of 27 C. Simulations were performed for 10 ns with no constraints, recording snapshots every 5 ps for analysis. Visualization of protein models and structural analyses were performed with VMD (51) and Pymol (39).

Three-dimensional structures of FasR
Recombinant FasR eluted from size exclusion chromatography suggesting a dimeric structure (~52 kDa), similar to all TFRs. Attempts to crystallise full-length FasR alone failed. We hypothesised that the 33-amino acid segment at the N-terminus is likely flexible, based on multiple sequence alignments to close orthologues (≥80% identity) from mycobacteria ( Supplementary Fig. S1B). A particularly strong sequence variation of the short N-terminal extensions was revealed, also with secondary structure predicted to be absent. We thus generated a truncated form of FasR lacking the first 33 amino acids (FasR Δ33 ). 9 FasR Δ33 readily crystallised in the absence of added ligands and also in complex with acyl C 20 -CoA (arachinoyl-or arachidoyl-CoA). Both crystal forms diffracted X-rays at better than 1.7 Å resolution (Table 1), and their structures confirm the dimeric architecture of FasR Δ33 , with each protomer organized in two all-helical domains (Fig. 1A), similar to all TFRs (11).
FasR Δ33 -C 20 -CoA was solved first using ab initio methods (25), facilitated by high enough diffraction resolution and the all-helical nature of the protein. A typical TFR structural architecture was found, with a DNA-binding HTH domain comprising helices α1-α3 from the N-terminus to residue Ser 82 . And a regulatory effector-binding domain located immediately C-terminal to the HTH, from Lys 83 to the C-terminus, including helices α4-α9. The latter are roughly organized in two bundles, the α4-α7 core runs along the long axis of the ellipsoid regulatory domain, whereas α8-α9, roughly perpendicular to the core, mediate dimerization by forming a 4-helix bundle with the other protomer's α8'-α9' helices (Fig. 1A). The FasR Δ33 -C 20 -CoA dimer is strictly symmetric, with the crystallographic 2fold axis relating one protomer to the other.
One of the most striking features of FasR Δ33 -C 20 -CoA is a tunnel-like cavity, delimited by helices α4, α5, α7 and α8, with its two openings towards the 'bottom' and the 'top' of the regulatory domain. C 20 -CoA binds within this tunnel, in an overall parallel orientation with respect to core helices α4, α5 and α7 (Fig. 1A,B). The tunnel is ~28 Å long, with a predominance of hydrophobic residues on its wall pointing their side chains towards the lumen of the tunnel (Fig. 1B). The fatty acid is well defined all along the tunnel ( Supplementary Fig. S2A). Towards the upper entrance of the cavity, most of the 4'phosphopantetheine portion of the CoA cofactor is also observed, with the higher electron density sulphur atom being instrumental in positioning the whole C 20 -CoA moiety. Likely due to high mobility of the CoA portions exposed to the solvent, electron density is less clear towards the tip of the pantoic group, and eventually becomes indistinguishable from noise in the region corresponding to the 3'phosphoadenosine diphosphate group, explaining why they were not included in the final refined model.
FasR Δ33 crystals were also grown in the absence of acyl-CoA with the goal of solving the ligandfree structure. Unexpectedly, additional electron density not corresponding to protein, was clearly visible within the FasR Δ33 ligand-binding tunnel ( Fig. 2A, Supplementary Fig. S2B). The shape of this density is compatible with part of a polyethylene glycol molecule (PEG 400 is a component of the crystallisation mother liquor), or yet also consistent with myristic acid (C 14 ). PEG is less likely, as it bears bridging oxygens along the alkyl chain, which would require satisfying H-bonding interactions in a hydrophobic environment. We hypothesize that C 14

Structural bases of FasR acyl-CoA-binding and effect on DNA association
To test the structure-derived hypotheses about acyl-binding and its effect on DNA-association, pointmutants were designed to block the entrance of the ligand into the hydrophobic tunnel, selecting residues that appeared to play key roles based on crystallographic evidence. Such 'tunnel blocking' approach was expected to abolish ligand-triggered rigidification of the protein and consequent DNA- 1 1 binding hindrance. A single mutant (FasR L106F ) and a triple mutant (FasR LVL ) were thus constructed.
The former substitutes, Leu 106 by a phenylalanine, at a critical position that borders the entrance of the tunnel. The triple mutant adds bulky side chains not only on position 106, but also substituting Leu 185 and Val 163 by phenylalanines (Fig. 3A). Point-mutations did not affect the dimeric architecture of the single mutant FasR L106F (Supplementary Fig. S3), and while slightly less than 50% of the triple mutant FasR LVL eluted as a monomer, >50% behaved as the wild-type protein.
Electrophoretic mobility shift assays (EMSAs) were performed by pre-incubating FasR, FasR L106F and FasR LVL (its dimeric form) with C 16 S4B). Ligand entrance within the tunnel is thus needed in order to trigger the protein conformational change that precludes binding of the regulator to its cognate DNA site.
Signal transmission: allosteric mechanism connecting the ligand-binding pocket with the DNA-binding domain.
The differential shift of the bottom-half of helices α4 and α7, comparing FasR Δ33 -C 14 and FasR Δ33-C 20 -CoA structures (Fig. 2C), strongly suggested that long enough alkyl chains in the effector-binding tunnel are critical in triggering the rigidification rearrangement, which results in arm-opening. To understand the molecular bases for such effect, the protein region around the distal tip of the ligand acyl chains were analysed in detail. The very last carbon atoms of C 20 -CoA interact with mostly bulky hydrophobic residues towards the end of the tunnel, i.e. at the opening of the tunnel that leads to the space separating both protomers in the dimer. Among these hydrophobic residues Leu 98 (on helix α4), Phe 123 (on α5) and Phe 138 (on α6) might play relevant roles (Fig. 3C). The substitution of such voluminous hydrophobic residues by smaller alanine sidechains, could uncouple the HTH mobilityrestraining effect from ligand-binding. Two point-mutants were constructed, FasR L98A and FasR F123A , which maintained the dimeric architecture of the protein (Supplementary Fig. S2). Both mutants showed significant functional effects, uncoupling ligand-binding and DNA-association (Fig. 3D), with a clearer effect observed in the case of FasR F123A . To further dissect the underlying mechanisms, the DNA-binding affinities were first analysed, comparing wild-type vs FasR L98A and FasR F123A mutants ( Supplementary Fig. S5). Apparent dissociation constants ( app K D ) were quantitated from electrophoretic mobility shift data, all in the nanomolar range (  Fig. S6). Indeed, both mutants had significantly lower response to the ligand (Table 2), with a particularly pronounced effect observed for FasR F123A . The length of the acyl chain was also critical, with IC 50 s and app K i s all shifted to significantly higher values when C 16 -CoA was used ( Supplementary Fig. S7 and Table 2). These results strongly suggest that residues Leu 98 and Phe 123 , and especially the latter, are key to ensure allosteric signal transmission while not influencing DNA-binding. That these mutations uncouple signal transmission but do not alter the protein's affinity for the effector ligand was assessed for the FasR F123A mutant (Supplementary Fig. S8). Surface plasmon resonance showed comparable association kinetics of FasR F123A to C 20 -CoA as compared to FasR wt . In sum, specific residues that are not essential to bind acyl ligands nor DNA, play a key role in transmitting the signal between both domains of the protein once the effector-binding tunnel is fully occupied.

The singular tunnel in FasR is predicted to accommodate very long fatty acyl effectors, triggering allosteric rigidification.
Very long fatty acyl moieties are relevant in the biology of Mycobacteriaceae including Mtb (52).
Intermediates in the synthesis, and constitutive moieties, of mycolic acids (with acyl components that can reach 60-90 carbons), very long fatty acids are essential components of mycobacterial cell walls.  (Fig. 4). The available volume in the open form of acyl-bound FasR anticipates even longer acyl chains to be readily accommodated, considering that the protein shows very stable behaviour with ~2 Å average rmsd.
The structure of effector-free FasR in complex with DNA confirms the acyl-CoA-triggered allosteric mechanism. 1 3 The crystal structure of full-length FasR was eventually solved by co-crystallisation with a 25-bp double stranded oligonucleotide bearing the native FasR-binding sequence motif (Fig. 5). A number of crystals and cryo-protections methods were tested, consistently producing strongly anisotropic X-ray diffraction data, reaching 3.85 Å resolution in the best direction (Table 1) This conformational change hampers effector occupation within the tunnel, the latter seems to be constricted by shifted residues (e.g. Phe 123 ) which move their side chains towards the tunnel's lumen.
The FasR Δ33 -C 14 structure described above proved that acyl-containing compounds from E. coli are invariably associated within the effector-binding cavity, even if specific acyl/acyl-CoA molecules are not added during protein purification and crystallisation. That the FasR tunnel in the FasR-DNA complex is free of bound ligands is supported by unequivocal evidence from difference Fourier maps at both early and late stages of refinement (detailed in Methods). The association to DNA thus correlates to expelling ligands from the effector-binding tunnel of FasR, at least those that attach more loosely within the cavity.

DISCUSSION
FasR is a TFR (TetR family of regulators)  acyl effectors can often be longer than the protein's own physical boundaries. We now answer to this question by revealing a unique hydrophobic tunnel that cuts across the entire effector-binding domain of FasR (Fig. 1), a tunnel that is conspicuously opened on both ends. This singular solution has evolved to lodge the kind of acylated chains of 20 and more carbons that inhibit FasR binding to its cognate DNA (7).
In addition to the tunnel itself, and in continuity with the bottom opening of it, FasR possesses a cavity delimited by the two protomers. The volume of such cavity (55) (58), or engage a different set of residues running in a perpendicular direction as compared to FasR's cavity (54,57,59). EthR is yet another TFR from Mtb (60), intensively investigated as a target to develop anti-tuberculosis medicines.
However, EthR is substantially different from FasR (22 % sequence identity; ~4.5 Å rmsd after superposition of the effector-binding domains), with a shorter N-terminal extension before the first αhelix, and a tunnel displaying wider regions or bulges ( Supplementary Fig. S10). Such bulges seem to correlate with EthR's capacity to bind compounds that include 1 to 3 aromatic or aliphatic rings (61). Finite deformations physics theory seems attractive to highlight allosteric regulation pathways by measuring mechanical strain rather than pairwise atomic position deviations (69). Unexpectedly, one of the segments subjected to highest mechanical strain in all TFRs analysed, corresponds to the loop that connects helices α6 and α7 (Supplementary Fig. S11). A triangle defined by α5, α6 and α7, a conserved feature in all TFRs (9), harbours the ligand-binding core cavity (which can in some cases expand into tunnel-like architectures with top, bottom and/or lateral openings). Helix α6, associating to helix α8, is attached to the fixed core, upper-half of the effector-binding domain; but simultaneously, α6's C-terminal tip and the α6-α7 junction also associate to the moving HTH. In sum, α6-α7 loop is bound to fixed and to moving parts, eventually leading to local deformation. Residues that could explain this strain, in contact with helices α6, the α6-α7 loop and the HTH domain, led us to identifying an array of hydrophobic residues, highly conserved among TFRs ( Supplementary Fig. S1 and Supplementary Data 1) and configured in three-dimensions as a continuous spine connecting the two domains of FasR. This spine belongs to, and connects the hydrophobic protein-folding cores of both domains, being interrupted by the ligand-binding cavity in all TFRs analysed ( Fig. 6 and Supplementary Fig. S12). Only in the ligand-bound condition this hydrophobic spine is completed, by the ligand molecule itself at the effector-binding domain, stabilizing a rigid and open conformation.
Such a mechanism predicts a disordered (flexible) to ordered transition of the TFR protein, which is consistent with the evidence we provide for FasR as well as with available evidence from other TFRs (9,11,(66)(67)(68)70). In particular, fluorescent probes such as 1-anilino-naphthalene-8-sulfonate that bind to partially folded proteins in "molten globule" states (71), have been shown to bind promiscuously to apo TFRs (9). Also, effector-triggered appearance of folding cooperativity between both domains, as well as proteolysis-resistance, have been reported in wild-type but not in allosteric-uncoupled mutants of TetR (70,72). Taken together, the extensive body of evidence lends strong support to the transmission spine mechanism as the most consistent interpretation of the effector-mediated allosteric control of TFRs' DNA-binding function.
When the ligand leaves the site, or if it is too short to fully occupy it, the hydrophobic spine is broken, protein folding is sub-optimal, anticipating a multitude of conformations (with HTH wiggling, 1 6 illustrated schematically in Fig. 7), including those that are competent for DNA-binding (HTH-closed).
The disorder-to-order transition is not fully triggered, resulting in asymmetry and higher flexibility as observed in the FasR Δ33 -C 14 complex. This hypothesis also explains why mutating bulky residues that contribute to building and stabilising the spine (e.g. Leu 98 and Phe 123 in FasR), can uncouple the allosteric effect: the ligand is then insufficient to achieve a complete, compact fold in the mutated TFR ( Supplementary Fig. S13). Residues that are not directly involved in effector-binding, but that contribute to building and stabilising the spine, will also be able to exert notable effects on allosteric coupling, upholding reported results (66)(67)(68).
A second conformation, the one bound to DNA, is however compactly folded. In this case it is the polynucleotide that pulls on the flexible HTH domains of the dimer, bringing them closer together.  (73), a mechanism that has been successfully exploited to develop ePK inhibitor-based drugs against human cancer and inflammatory diseases (74,75). A closer example is provided by EthR, a TetR-like repressor that controls the expression of EthA (the enzyme that bioactivates ethionamide, a second-line antituberculosis drug that inhibits mycolic acid biosynthesis).
EthR inhibitors have been conceived after comparing EthR crystal structures obtained in complex with larger and smaller ligands (76). In that work, compounds bearing thienyl and piperidinyl pharmacophores were selected as the most potent EthR inhibitors among screening hits. Only the piperidinyl-, and not the thienyl-, interacting pocket engages EthR amino acids that compose the hydrophobic spine (PDBs 3G1O, 3G1M), the transmission mechanism thus explains why the piperidinyl-binding pocket turned out to be the crucial region determining inhibitory activities (76).
Structure-guided drug discovery strategies that exploit the allosteric hydrophobic-spine transmission mechanism of TFRs might thus prove successful in developing novel medicines against tuberculosis, including multi-and extensively drug-resistant strains.       The inset shows a projection of a cut section through protomer's B tunnel, to display the dynamic stability of the dimeric interface and the acyl chain (as opposed to the large flexibility of the nucleotide portion of coenzyme-A). Free volume is available at the inter-protomer space, predicting that even longer acyl moieties should be able to accommodate.