A structure-guided fragment-based approach was used to target the lipophilic allosteric binding site of Mycobacterium tuberculosis EthR. This elongated channel has many hydrophobic residues lining the binding site, with few opportunities for hydrogen bonding. We demonstrate that a fragment-based approach involving the inclusion of flexible fragments in the library leads to an efficient exploration of chemical space, that fragment binding can lead to an extension of the cavity, and that fragments are able to identify hydrogen-bonding opportunities in this hydrophobic environment that are not exploited in Nature. In the present paper, we report the identification of a 1 μM affinity ligand obtained by structure-guided fragment linking.
Structure-guided fragment-based approaches have provided efficient routes to the design of selective chemical tools and novel therapeutics, not only for conventional ligandable targets , but also for more challenging protein–protein interaction sites . In the present study we explore the use of this approach to target the lipophilic allosteric binding site of EthR.
EthR has been implicated in ethionamide drug action [3–5] and is regulated allosterically by hexadecyl octanoate, a highly lipophilic molecule . EthR, a TetR transcription family member , has an elongated ligand-binding site; many hydrophobic residues line the binding site, leaving very few opportunities for hydrogen bonding . Nevertheless, previous work on EthR has reported the discovery of drug-like high-affinity EthR inhibitors [9–11]. In the present study we demonstrate that a fragment-based approach can also be used to identify novel chemotypes, forming hydrogen bonds within the largely lipophilic binding site. We describe the identification of new chemical scaffolds that bind EthR and trigger dissociation from its cognate DNA-binding site. We demonstrate successful fragment linking to yield a 1 μM compound as characterized by a functional SPR (surface plasmon resonance) assay. Early evaluation suggests that this compound can pass through the Mycobacterium tuberculosis cell membrane and shows a functional effect by decreasing the MIC (minimum inhibitory concentration) of ethionamide when used in combination.
The campaign described in the present paper emphasizes several conclusions likely to be of wider application. First the inclusion of flexible fragments in the library can lead to a more efficient exploration of chemical space; secondly, that fragments can lead to conformational changes in the protein that create an extension of the binding cavity, thus providing a new opportunity to discover novel ligands; and thirdly, that even in a very hydrophobic environment, fragments are able to identify hydrogen-bonding opportunities which are not exploited in Nature.
EthR: cloning, expression and purification
The EthR gene was cloned in a pHAT5 vector  with BamHI and EcoRI restriction sites. For expression, the Escherichia coli BL21 (DE3) (Novagen) strain was used. For protein expression, LB media was inoculated with fresh overnight liquid culture (25 ml per 1 litre) and grown to exponential phase (37°C, 230 rev./min). The cultures were then induced with IPTG (0.5–1 mM). After 3 h, the cells were harvested by centrifugation (4200 g for 15 min at 4°C). Cell pellets from 1 litre of culture were re-suspended in 30 ml of lysis buffer [50 mM Hepes (pH 7.5) and 150 mM NaCl] supplemented with EDTA-free complete protease inhibitor cocktail (Roche). The cells were lysed by sonication (10 pulses of 30 s each). Debris was removed by centrifugation (35000 g for 1 h at 4°C) and the supernatant was passed through a 5 ml HiTrap IMAC Fast Flow column (GE Healthcare) charged with Ni2+. After washing with 50 ml of wash buffer [50 mM Hepes (pH 7.5), 150 mM NaCl and 20 mM imidazole], the protein was eluted with 50 mM Hepes (pH 7.5), 150 mM NaCl and 250 mM imidazole. The protein was further purified by size-exclusion chromatography (Superdex 200) and concentrated (4500 g at 4°C) using 10 kDa Amicon® Ultra concentrators.
Screening with a thermal-shift assay
For identification of hits, a thermal-shift assay  was used. Each 100 μl reaction contained 20 μM EthR, 150 mM NaCl, 20 mM Tris/HCl (pH 8.0), 2.5×Sypro Orange dye (Invitrogen) and 10 mM test fragment with 10% DMSO (fragment stock solutions were prepared in DMSO). The temperature of the sample was raised from 25°C to 90°C in 0.5°C increments. The fluorescence of the sample was measured at each temperature step, with excitation/emission wavelengths of 490/575 nm. These experiments were performed using the iCycler iQ Real-Time PCR machine (Bio-Rad Laboratories).
The SPR assay was carried out on a BIAcore T100 machine, designed to measure the interaction of EthR with ethA promoter DNA (106 bp) immobilized, via a biotin–streptavidin linkage, on to a CM5 Sensor Chip (BIAcore). DNA from pUC19 (113 bp) was used as the control against non-specific binding. These DNA fragments were produced as described previously [14,15]. Streptavidin was attached to the surface of the CM5 chip according to the instructions provided with the amine coupling kit (BIAcore). Biotinylated control and promoter DNA fragments were then flowed over different channels of the chip to achieve stable fixation levels of 170 and 174 RUs (resonance units) for promoter and control DNA respectively.
For screening, EthR/fragment solution [2 μM EthR and the indicated concentration of fragment made up in running buffer (2 mM MgCl2, 10 mM Tris/HCl, pH 7.5, 0.1 mM EDTA, 200 mM NaCl and 2% DMSO)] was flowed over the chip at 20 μl/min for 120 s. The dissociation time was 150 s. To determine binding levels, the response of the control channel at steady-state was subtracted from that of the experiment channel. The chip was re-generated between samples by passing 20 μl/min 0.03% SDS in running buffer for 60 s over binding channels.
For IC50 calculations, the response of EthR binding to the immobilized DNA was measured at various concentrations of compounds. The resulting RUs were used to fit the data in XLfit software (ID Business Solutions) and concentrations necessary to inhibit 50% of maximal interaction were calculated. A representative example of data fitting is shown for fragment hit 1 (Supplementary Figure S4 at http://www.biochemj.org/bj/458/bj4580387add.htm).
EthR crystallization and crystal structure solution
Crystallization of EthR was carried out by the hanging-drop vapour-diffusion method using conditions based on those described previously . The best crystals were obtained by mixing 2 μl of protein solution [>20 mg/ml EthR, 500 mM NaCl, 20 mM Tris/HCl (pH 8.0) and 10% (v/v) glycerol] with 4 μl of reservoir [1.8–2.2 M ammonium sulfate, 100 mM Mes-Na (pH 6–7), 5–10% (v/v) glycerol and 7–10% 1,4-dioxane] at 16°C.
Fragments (100 mM in DMSO) were mixed with mother liquor [1.8 M ammonium sulfate, 100 mM Mes-Na (pH 6.75) and 12.5% (v/v) glycerol] at 1–10 mM concentrations to make crystal-soaking solutions. The crystals were then washed in order to remove the dioxane from the binding site by placing them in mother liquor devoid of dioxane for a few hours. The washed EthR crystals were then placed in fragment-containing solutions for soaking for between 1 and 16 h.
Crystals were cryoprotected by passing them briefly through mother liquor supplemented with 20% (v/v) ethylene glycol, before freezing in liquid nitrogen. Data collection was carried out at the European Synchrotron Radiation Facility (Grenoble, France), Diamond Light Source (Harwell, U.K.), Swiss Light Source (Villigen, Switzerland) and at the in-house source (×8 Proteum, Bruker AXS). Diffraction data were analysed using programs within the CCP4 suite  run with its graphical user interface . Diffraction images were indexed and integrated using Mosflm  and then scaled with Scala . Molecular replacement was carried out using Phaser  with PDB structure 1T56  as the initial model and structure refinement was carried out using Refmac5  for maximum-likelihood restrained refinement. Model fitting, and water and ligand fitting into the difference map were done manually with Coot . Ligand topology and parameter files were generated either by the Dundee PRODRG2 server  or with libcheck . Final Figures were prepared using PyMOL (http://www.pymol.org).
Determination of the ethionamide MIC boosting effect
Determination of ethionamide MIC boosting was performed using the REMA (resazurin reduction microplate assay) as described previously . Serial dilutions (2-fold) of ethionamide were prepared in 96-well plates, alone or in combination with the EthR inhibitors at a fixed concentration of 1 μM. Frozen aliquots of M. tuberculosis H37Rv in mid-exponential cultures were thawed and diluted to a D600 of 0.0025 in liquid 7H9 medium (Difco) to obtain a total volume of 100 μl. Plates were incubated for 6 days at 37°C before the addition of resazurin (0.025%, 10 μl). After overnight incubation, fluorescence of the resazurin metabolite resorufin was determined (excitation at 560 nm and emission at 590 nm, measured using a Tecan infinite M200 microplate reader). The MIC was defined when the level of fluorescence is equivalent to the highest concentration of ethionamide where all cell growth is inhibited.
RESULTS AND DISCUSSION
A thermal-shift screen  was carried out using a library of 1250 fragment molecules. Any fragment displaying thermal stabilization (ΔTm) of more than 1°C was classified as a hit. In total, 86 fragment hits were identified giving a hit rate of 7%. Of these hits, 22 raised the melting temperature by more than 3°C, and five increased the melting temperature by more than 5°C. Temperature shifts of approximately 1–2°C are routinely observed for fragment-sized ligands. In the case of EthR, the thermal shifts are quite pronounced with a significant number of hits obtained displaying thermal shifts of more than 2°C. These large thermal shifts probably have resulted from the nature of the ligand-binding site, which is enclosed by the protein and flanked by a number of hydrophobic residues and some hydrogen-bonding interactions.
All 86 hits from the thermal-shift screening were then tested using a functional assay measuring their ability to inhibit the DNA–EthR interaction, using the SPR technique as described previously . In addition to the 86 hits, 45 fragments that showed no increase in melting temperature were included in the SPR assay as negative controls. In order to rank the compounds, a single concentration screen was carried out using fragments at 500 μM. The resultant relative decrease in the SPR RU value was then used to calculate the percentage inhibition. Any compound inhibiting the EthR–DNA interaction by more than 10% was classified as a validated hit. Of the 86 fragment hits from the thermal-shift assay, 45 compounds showed inhibition of more than 10%, giving a final hit rate of 3.6%. Out of 45 negative controls from the thermal-shift assay, all except one exhibited inhibition of less than 10% in the SPR assay, validating the use of thermal shift as a primary screening technique (see the Supplementary Online Data at http://www.biochemj.org/bj/458/bj4580387add.htm). Hits with novel chemical scaffolds were then further characterized by performing dose–response titrations in the SPR assay and the resultant percentage inhibitions were used to calculate the IC50 values.
Table 1 shows four examples of molecules that are capable of binding to EthR and inhibit the EthR–DNA interaction, three of which, 2, 3 and 4, have a similar scaffold. These four fragment hits were selected for further characterization using X-ray crystallography by soaking into preformed crystals of EthR, as described in the Experimental section of the Supplementary Online Data. Binding of a fragment was judged by the presence of difference electron density (Fo−Fc) corresponding to the fragment molecules in the hexadecyl octanoate-binding site. All four fragments showed excellent electron density allowing unambiguous determination of the binding mode of the fragment molecules. The detailed crystallographic information and refinement statistics are presented in Supplementary Table S2 at http://www.biochemj.org/bj/458/bj4580387add.htm.
As shown in Figure 1, the EthR ligand-binding site is an elongated hydrophobic tunnel flanked by a number of hydrophobic residues (Phe110, Phe114 and Phe184; Trp103, Trp138, Trp145 and Trp207; Leu87 and Leu183; Ile107; Met142) with only three polar, but uncharged, amino acids (Asn176, Asn179 and Thr149). Two dioxane molecules, an important component of the crystallization conditions, occupy the allosteric site. The binding channel can be roughly divided into three parts, a larger central tunnel, and two relatively small cavities that are termed here ‘cavity 1’, ‘cavity 2’ and ‘cavity 3’ respectively.
The EthR physiological dimer structure shown in complex with two dioxane molecules, PDB code 1T56 (the dioxane is shown in stick representation)
All four fragments described in the present study were able to bind Asn179 in the central binding pocket of EthR. The same region of the binding pocket, cavity 1, has been targeted by a previous study , and has been shown to exhibit conformational plasticity . Fragment hit 1 was observed to bind in cavity 2 of the channel, in addition to cavity 1, with associated side-chain movements. Fragments 3 and 4 were also found to bind the central cavity 1 twice in the same crystal structure (see Supplementary Online Data), whereas only one molecule of fragment 2 was found bound in cavity 1
Fragments 2, 3 and 4, all containing a sulfonyl group, displayed slightly different activities. Analyses of crystal structures of these fragments bound to EthR indicated clear hydrogen bonding and hydrophobic interactions. As shown in Figure 2, key interacting partners for hydrogen bonding in this series of compounds were Asn176 and Asn179, whereas Phe110 exhibited π-stacking interactions with the aromatic component. Fragment 2 formed hydrogen bonds with Asn176 and Asn179, and fragment 3 formed a π-stacking interaction with Phe110. Among these three, fragment 4 formed the most interactions. It interacted with Asn176 and Asn179 through hydrogen bonding, as well as π-stacking with Phe110. The p-chloro substituents in fragments 3 and 4 were positioned in a hydrophobic environment. Interestingly the thioamide of fragments 2 and 4 formed an additional hydrogen bond with the backbone of Asn176. A second molecule of fragment 4 in the same crystal structure formed additional hydrogen bonds with Tyr148, Leu90 and a water molecule. Comparing the modes of interaction of these three fragments and their ability to inhibit the DNA–protein interaction, it appears that the thioamide group and the p-chloro-phenyl substituent both make important contributions.
Interaction of sulfonyl-containing fragments with EthR
The crystal structure of fragment 1 in complex with EthR reveals an interesting mode of interaction and novel insights into the conformational plasticity of EthR. Fragment 1 occupies the hydrophobic channel at two positions, even though the preformed central binding region of EthR, cavity 1, is large enough to accommodate only a single molecule of fragment 1. The second molecule is able to bind to the previously unexploited cavity 2 by bringing about changes in the conformation of side chains of Phe184 and Gln125. A significant movement of the Phe184 side chain allows access to the smaller second cavity (Figure 3B). This induction of a conformational change was necessary to bind the two molecules of fragment 1 that together are able to span the entire binding pocket of EthR.
Conformation change upon fragment binding and retention of the initial conformation and interactions during fragment linking
Interestingly the two molecules of fragment 1 adopt slightly different conformations in the same crystal structure. Fragment 1 binds in both cavity 1 and cavity 2, close to the DNA-binding HTH (helix–turn–helix) motif. It has an IC50 of 280 μM and a unique binding mode. In addition, the high sp3 content and number of rotatable bonds distinguishes it from planar aromatic starting points of many fragment-based campaigns.
For chemical elaboration of fragment hit 1, fragment linking as well as fragment growing were considered. As the two molecules of fragment 1 together could span the entire binding channel, the fragment-linking approach was the more attractive option. However, before fragment linking, steps were taken to modify fragment 1 with the aim of improving the binding affinity and making the subsequent chemical synthesis more straightforward. The compounds made and their respective activities are reported in Table 2. Analogues 5 and 6 of fragment 1 were synthesized by varying ring A and ring B, keeping intact the important carbonyl oxygen that forms key hydrogen bonds. Although they showed a slight reduction in activity, they offered better prospects for subsequent chemical elaboration. The crystal structure of compounds 5 (Figure 3C) and 6 in complex with EthR showed that they maintained the binding mode of the initial fragment hit 1, including occupying two sites in the binding channel and maintaining the hydrogen-bonding pattern, i.e. ability to interact with Asn179 and to form a hydrogen bond with a water molecule (Supplementary Figure S3 at http://www.biochemj.org/bj/458/bj4580387add.htm).
In order to link the two modified compounds 5 and 6, several different linkers were explored in silico. The designed molecules were docked into the fragment hit 1-bound crystal structure of EthR using GOLD [28,29]. The two separate linkers, a disulfide for compound 5 and amide for compound 6, were found to offer a suitable geometry allowing individual ligands to be able to adopt the starting fragment conformation. The linked compounds were synthesized (see Supplementary Online Data) and tested using the SPR functional assay. The structure and activities of these linked compounds are also reported in Table 2.
As can be seen from Table 2, the compounds with the disulfide linker showed significantly higher activity than the amide-linked compounds. This could be due to the structural rigidity of the amide bond preventing individual arms of the molecule from adopting a conformation that is able to exploit the hydrophobic and hydrogen-bond interactions. In addition, an energy cost in burying the polar amide hydrogen atoms in the binding channel may have contributed to its reduced activity.
Crystal structures of the disulfide-linked compounds 7, 8 and 9 bound to EthR were obtained by soaking the compounds into EthR crystals. Compounds 7, 8 and 9, the most potent ligands, maintain the same ligand-binding mode as that of the starting fragment and are able to form a key hydrogen bond with Asn179 as well as with a water molecule. The retention of binding mode and interactions of compounds 1, 5 and 9 during the fragment linking is shown in Figure 3.
EthR ligands, post-binding, are thought to bring about a change in the EthR conformation, notably leading to changes in the distance between the HTH motifs of the EthR dimer . The changes in the cell dimensions and the separation of the HTH domain of EthR are suggestive of the mechanism of EthR inhibition. Ligands from the present study were able to induce changes in conformation of proteins as measured by the distance between HTH motifs of the two protomers within the crystallographic dimers. The distance ranged between 49.5 Å (1 Å=0.1 nm) for the apo-form crystal structure to 52.3 Å for the compound 9-bound EthR structure (Supplementary Figure S5 at http://www.biochemj.org/bj/458/bj4580387add.htm). Crystals treated similarly with soaking and cryo-solutions that are devoid of any ligands did not reveal any changes in cell dimensions. This indicates that the changes in cell dimensions were due to ligand interaction and not because of backwashing of crystallization reagent, dioxane or other experimental conditions. The analysis of all of the X-ray structures with or without ligands did not reveal any noticeable changes, other than the side-chain conformations as described above.
The appreciable activity of the final linked compound 9 in the in vitro SPR assay was encouraging, prompting evaluation of its effectiveness in combination with ethionamide in live M. tuberculosis cultures. In addition, the starting fragment hit 1 was also selected to check whether it had any EthR boosting activity. These compounds were tested using the REMA (Figure 4)  in combination with ethionamide at a fixed concentration of 1 μM since these compounds were found to display some bactericidal activity at 10 μM but none at 1 μM. The MIC for ethionamide on its own was found to be 15 μM.
Evaluation of the ethionamide boosting effect by REMA 
Compound 9 reduced the MIC of ethionamide from 15 μM to 1.9 μM at 1 μM concentration, an effective 8-fold reduction (Figure 4). It is worth noting that the IC50 of compound 9 is approximately 280-fold that of fragment 1, whereas at a 1 μM concentration the ethionamide MIC boosting activity of compound 9 is similar. The large discrepancy between the IC50 and the boosting activity of fragment 1 and compound 9 could be attributed to the relative permeability of individual compounds. It is also possible that compound 9 serves as a prodrug and is reduced in M. tuberculosis to the thiol, which is the active species. This possibility will be explored in future studies by testing the reduced compound in the SPR assay as well as by REMA.
Transcription factors have previously been considered to be undruggable . In the present study we have targeted EthR as an example of a challenging drug target, and several observations concerning the fragment-based approach are likely to be of wider application. Using the thermal-shift assay as a screening technique allowed rapid screening of a library of compounds. The subsequent SPR assay was used as a functional assay to assess the ability of compounds to inhibit the EthR–DNA interaction. By using a cascade of two complementary techniques a final set of fragments was obtained, which functionally inhibited the EthR–DNA interaction by specifically interacting with EthR.
The crystal structures of EthR in complex with fragment hits 2, 3 and 4 demonstrated that the fragments were able to form key hydrogen bonds, even though the highly lipophilic hexadecyl octanoate, a potential natural regulator of EthR, interacts with EthR primarily through hydrophobic interactions (Supplementary Figure S1 at http://www.biochemj.org/bj/458/bj4580387add.htm). As shown here, even in a very hydrophobic environment fragment-based methods are able to identify hydrogen-bonding opportunities that are not exploited in Nature. The observed conformations of the two molecules of fragment 1 (which has a number of rotatable bonds) in the the same crystal structure differ from each other (rmsd of 1.1 Å for a 14 atom fragment). The same observation was made for fragment 4, showing that conformational flexibility allows the fragment to achieve subtly different binding interactions/orientations, emphasizing the need to have the right balance between flat aromatic molecules and molecules with increased three-dimensionality and rotatable bonds in screening collections. The crystal structures of EthR bound to fragments demonstrated that the fragments were able to induce conformational changes in the protein, allowing them to occupy different pockets of the protein. In this structure-guided work it is shown that these fragment hits could be readily linked, resulting in higher activity molecules, while retaining the initial conformational changes and binding mode and preserving key interactions.
In summary, the crystal structures of EthR complexes show that fragments are able to exploit key hydrogen bonds in a hydrophobic environment that the natural ligand does not exploit. The newly discovered hydrogen-bonding interactions are maintained in inhibitors designed from linking the fragments and give rise to higher affinity and presumably more selective leads for drug discovery. Also, the observations in the present study argue for the use of flexible lipophilic fragments with hydrogen-bond donors or acceptors for targeting lipophilic sites. Such fragments may be valuable in probing difficult targets.
Sachin Surade and Tom Blundell envisaged the project; Sachin Surade established the early protocols in protein purification, crystallization, thermal-shift screening and the SPR assay, and designed and supervised the fragment screening; Narin Hengrung performed the fragment screening and determined the crystal structure of fragment hits; Nancy Ty and Chris Abell designed compounds 5–11; Nancy Ty carried out in silico prioritization and performed synthesis of compounds 5–11. Sachin Surade determined the crystal structures of EthR in complex with ligands. Benoit Lechartier perfomed the REMA under the supervision of Stewart Cole. Tom Blundell and Chris Abell supervised the operation of all of the project. All authors contributed to the development of the paper.
We are grateful to our many colleagues in the Department of Biochemistry, University of Cambridge including Dr Marko Hyvonen, Dr Marcio Dias, Dr Leo Silvestre, Dr Michal Blaszczyk and Dr Vitor Mendes for stimulating discussions. We would also like to acknowledge Dr Dima Chirgadze and Dr Katherine Stott for maintaining the X-ray Crystallographic and Biophysics Facility in the department.
This work was supported by the Bill and Melinda Gates Foundation, the European Community's Seventh Framework Programme [grant number 260872] and the University of Cambridge. B.L. is a recipient of a grant from Fondation Jacqueline Beytout.