Induced DNA bending by unique dimerization of HigA antitoxin

The crystal structures of HigA3 antitoxin and DNA-bound HigA3 allow detailed structural insights into the HigBA toxin–antitoxin system of Mycobacterium tuberculosis. The unique dimerization of HigA3 induces DNA bending.


Introduction
Tuberculosis (TB) is one of the major infectious diseases caused by the human pathogenic bacterium Mycobacterium tuberculosis, affecting 23% of the global population (Lö nnroth & Raviglione, 2008). It is one of the top ten risk factors for death and the leading cause of death from a single infectious agent (above HIV/AIDS). The WHO estimated that 10 million people were infected with TB in 2017, among which 5-15% of this population developed active TB, resulting in a mortality rate of 1.3 million people worldwide in a single year (WHO 2017 data;WHO, 2018). However, the current regimens of antibiotic combinations have led to resistance, and concerns over tolerance are increasing rapidly. The emergence of multidrug-resistant TB (MDR-TB), which is the cause of TB treatment failure, has been a public health crisis for the last decade. In 2017, approximately 450 000 people developed MDR-TB, and the treatment success rate for MDR-TB was only 55% globally (WHO, 2018). During the treatment, TB enters a dormancy state that is linked to the antibiotic tolerance of TB due to the capability of persister cell formation from active cells (Torrey et al., 2016). These persister cells make up a small fraction of the total cells in the initial infection, but become a significant fraction during the dormancy state (Torrey et al., 2016;Wood et al., 2013). However, studies on the persister cell formation for the effective treatment of TB have been poorly performed. It is well known that toxinantitoxin (TA) systems play an important role in both antibiotic resistance via multi-resistant plasmid maintenance and antibiotic tolerance via persister cell formation (Page & Peti, 2016). In addition, it has been reported that M. tuberculosis has approximately 88 TA modules, while non-pathogenic bacteria such as Mycobacterium smegmatis have only 4 TA modules (Sala et al., 2014;Ramage et al., 2009). Therefore, the importance of TA modules in TB pathogenicity should be emphasized.
TA modules are autoregulated operons that are abundant in the plasmids and chromosomes of many pathogenic bacteria. TA systems regulate the growth of cells responding to a wide variety of stressful environments such as nutrient starvation, antibiotic exposure, heat shock and DNA damage (Pandey & Gerdes, 2005;Yamaguchi et al., 2011). TA operons consist of two linked genes in which one encodes a stable toxin protein that inhibits cell growth and the other encodes a cognate antitoxin, such as an antisense RNA or protein that neutralizes the toxin activity (Greenfield et al., 2000). Toxins show enzymatic functions that interfere with various vital processes of bacteria. Toxins have diverse targets, such as DNA replication, mRNA stability, protein translation, cell wall biosynthesis and ATP synthesis (Yamaguchi et al., 2011). Activation of toxins is prevented by the binding of antitoxin under normal conditions. However, under stressful conditions, antitoxins are degraded by proteases, and the free toxin is activated (Brzozowska & Zielenkiewicz, 2013). Broadly, TA systems are classified into six types depending upon the type of antitoxin and how the antitoxin inhibits toxin activity (Page & Peti, 2016). Among these systems, the type II TA system is one of the well established TA systems; it encodes a stable toxin protein and a labile antitoxin protein, which form a stable TA complex to inhibit the function of the toxin in normal cells (Yamaguchi & Inouye, 2009). Antitoxins are usually flexible and unstructured in free form but are continuously produced to neutralize the toxicity of toxins (Aizenman et al., 1996). In most cases, antitoxins bind to the promoters located upstream of the corresponding TA operon to repress transcription. In addition, the TA complex itself acts as a repressor of TA operon transcription by binding to the promoter region, resulting in autoregulation of the TA module (Marianovsky et al., 2001;Kedzierska et al., 2007). Postsegregational killing (PSK), abortive infection and bacterial persistence have been reported as the major biological functions of TA modules (Harms et al., 2018). Among these functions, bacterial persistence is strongly associated with the type II TA system (Page & Peti, 2016). Upon antitoxin degradation, the free toxin induces the formation of persister cells, leading to the dormancy state of cells. The unusually wide distribution of the type II TA system in M. tuberculosis and dormancy of TB may be related. Therefore, the TA system could be a candidate for a new antibiotics target for TB treatment.
The representative type II TA families of M. tuberculosis are VapBC, MazEF, ParDE, RelBE and HigBA. To date, diverse structural studies of the type II TA system have been reported, but the structural studies of HigBA in M. tuberculosis are limited. The first higBA module discovered was on the Rts1 plasmid from Proteus vulgaris (Tian et al., 1996). The higBA module differs from other type II TA systems in that the toxin gene higB is located upstream of the antitoxin gene higA. There are three higBA loci (higBA1-3) in M. tuberculosis, and these loci are highly expressed under multiple stress conditions, such as drug-induced and environmental stress (Gupta et al., 2017). The HigB toxin is a ribosome-dependent ribonuclease that cleaves adenosine-rich mRNA (Hurley & Woychik, 2009). This ribonuclease activity inhibits protein transcription following cell growth arrest, which has been reported for several bacteria, including M. tuberculosis (Schuessler et al., 2013). The HigA antitoxin of M. tuberculosis (MtHigA) has a Cro/C1-type helix-turn-helix (HTH) domain, which is a common DNA-binding domain existing in prokaryotes and eukaryotes (Luscombe et al., 2000). This domain is named after the transcription repressors Cro and C1 of bacteriophages 434 and lambda (Aggarwal et al., 1988). The Cro/C1-type HTH domain-containing proteins function mainly as homodimers and consist of two domains: a bundle of five N-terminal helices that serves a stabilizing hydrophobic core and a C-terminal domain mediating dimerization (Luscombe et al., 2000). The 2and 3-helices of the Nterminal helical bundle compose the HTH motif, which is typically defined as a 20-amino-acid segment with two perpendicular -helices. The 3-helix recognizes the DNA helix.
In this study, we determined the crystal structure of the MtHigA3 antitoxin alone, and the MtHigA3 bound to the operator DNA. Compared with previously known structures of HigA from other species, MtHigA3 shows unique characteristics, including a dimerization interface. Upon forming a complex with promoter DNA, the quaternary structure of MtHigA3 and the curvature of DNA bending seem to be closely related with each other. The interaction between MtHigA3 and the operator DNA in the solution state was investigated by NMR titration experiments which provided additional information about the binding mechanism between DNA and MtHigA3. Upon binding to DNA, residues responsible for dimerization showed major changes along with the HTH motif. Based on this study, we could conclude that the unique dimerization mode of MtHigA3 determines the arrangement of HTH motifs. The position of HTH motifs for each monomer facilitates the sensing of the promoter DNA, resulting in DNA bending. This work could contribute to the understanding of the HigBA system of M. tuberculosis on the atomic level, and moreover, may contribute to the development of new antibiotics for TB treatment. The genes encoding M. tuberculosis HigA3 antitoxin (MtHigA3) and 34 residues truncated from the N-terminus of MtHigA3 (MtHigA3 35-117 ) were amplified from the M. tuberculosis strain H37Rv genomic DNA by polymerase chain reaction (PCR). The restriction enzymes used for PCR were Nde1 and Xho1, and the oligonucleotide primers used in this study are listed in Table S1 of the supporting information. The amplified DNA of MtHigA3 and MtHigA3 35-117 digested by NdeI and XhoI were ligated to the predigested expression vector pET-21a(+) (Novagen) containing a C-terminal hexahistidine tag (LEHHHHHH). Each recombinant plasmid was transformed into Escherichia coli DH5 competent cells. The resulting recombinant plasmids were verified by DNA sequencing.
MtHigA3 and MtHigA3 35-117 cloned into recombinant plasmids were transformed into E. coli Rosetta (DE3) and E. coli BL21 (DE3) competent cells, respectively. The cells were allowed to grow in Luria broth (LB) medium at 37 C with 50 mg ml À1 ampicillin until the OD 600 reached 0.5. Then, protein expression was induced by the addition of 0.5 mM isopropyl -d-1-thiogalactopyranoside (IPTG). The cells were grown at 37 C for an additional 4 h after induction and were harvested by centrifugation at 5600g and 4 C. The cells were resuspended and lysed on ice by sonication in lysis buffer A [50 mM Tris-HCl, pH 7.5 and 500 mM NaCl containing 10%(v/v) glycerol]. The cell lysate was centrifuged at 20 000g for 1 h at 4 C. The supernatant was applied to an affinity chromatography nickel-nitrilotriacetic acid (Ni-NTA) column (Novagen) previously equilibrated with buffer A. The column was washed with buffer A containing 40 mM imidazole, and the protein of interest was eluted with buffer A containing 300 mM imidazole. The protein was further purified by size-exclusion chromatography (SEC) using a Superdex 75 (10/300GL) column (GE Healthcare) that was previously equilibrated with 20 mM citric acid, pH 5.0 and 150 mM NaCl containing 2 mM -mercaptoethanol. The purity of the recombinant MtHigA3 protein was verified by SDS-PAGE, and the protein was concentrated to 10 mg ml À1 using an Amicon Ultra centrifugal filter unit (Millipore) for crystallization.
For co-crystallization of MtHigA3 bound to DNA, the protein was incubated with a 20 bp double-stranded DNA (dsDNA; 5 0 -CCACGAGATATAACCTAGAG-3 0 ) in a 1:4 molar ratio (protein:dsDNA) for 1 h at 20 C before being subjected to size-exclusion column chromatography. For selenomethionine (SeMet) labelling of MtHigA3, the cells were allowed to grow in M9 minimal medium supplemented with SeMet (50 mg l À1 ). The E. coli cells containing the recombinant plasmid were grown overnight at 37 C in 10 ml of M9 medium that contained uniformly labelled 15 N ammonium chloride [U-15 N] (Cambridge Isotope Laboratory) as a nitrogen source and/or 13 C glucose [U-13 C] (Cambridge Isotope Laboratory) as a carbon source. This culture was then inoculated into 1 l of fresh M9 medium supplemented with [U-15 N] for 15 N-labelled MtHigA3 35-117 or with [U-15 N] and [U-13 C] for 15 N, 13 C-labelled MtHigA3 35-117 . The process for protein isolation and purification was the same as described above for non-labelled protein. The final buffer for all NMR samples contained 20 mM citric acid, pH 5.0, 150 mM NaCl and 1 mM DTT. 10%(v/v) D 2 O was added to the protein samples for NMR to provide the NMR internal lock signal.
MtHigA3 and the matrix solution (5 mg ml À1 -cyano-4hydroxycinnamic acid in 50% acetonitrile/0.1% TFA) were loaded onto a Teflon-masked MALDI-TOF target. MALDI-TOF MS was performed on a Voyager DETM Biospectrometry Workstation (Applied Biosystems) with a 337 nm nitrogen laser and operating in positive mode at the National Center for Inter-University Research Facilities, Seoul National University (NCIRF).

Crystallization, X-ray data collection and structure determination
Crystals of MtHigA3, SeMet-labelled MtHigA3 and MtHigA3 bound to DNA were grown by the sitting-drop vapour diffusion method at 293 K using 96-well crystallization plates. Initial crystallization conditions were established using screening kits from Hampton Research (Crystal Screens I and II, Index, PEG/Ion and Natrix), Molecular Dimensions (Memgold I and II, Structure Screen I and II and Proplex) and Emerald BioSystems (Wizard I, II, III and IV) by mixing 0.5 ml of protein solution with 0.5 ml of reservoir solution. The best crystals of MtHigA3 and SeMet-labelled MtHigA3 were obtained from a reservoir solution containing 1 M ammonium sulfate, 0.1 M HEPES and 0.5%(w/v) PEG8000. The best crystals of MtHigA3 bound to the DNA complex were obtained from a reservoir solution containing 23%(w/v) PEG3350 and 0.2 M sodium acetate trihydrate pH 7.0. For crystal freezing, the crystals were transferred to a cryoprotectant solution with 20%(v/v) glycerol under crystallization conditions prior to freezing and mounting. The X-ray diffraction data for the crystals of MtHigA3 and MtHigA3 bound to DNA were collected at 100 K using an ADSC Quantum 315r CCD detector system (Area Detector Systems Corporation) at the BL-5C experimental station of Pohang Light Source, Korea. X-ray diffraction data for the crystals of SeMet-labelled MtHigA3 were collected at 100 K using a PILATUS 6M detector system (Dectris) at the BL-11C experimental station of Pohang Light Source, Korea. For each image, the crystal was rotated by 1 and the raw data were processed and scaled using the program suite HKL2000 (Otwinowski & Minor, 1997). Further data analysis was carried out using the CCP4 suite and Phenix Adams et al., 2010).
To determine the MtHigA3 structure, the SAD dataset was used and the Autosol and Autobuild Wizards in the Phenix package were employed (Adams et al., 2010). The four selenium sites were identified and the initial phases calculated from these sites were further improved by density modification. The structure of MtHigA3 bound to DNA was research papers 750 Park et al. Induced DNA bending by unique dimerization of HigA antitoxin determined using molecular replacement with the program PhaserMR within the Phenix package (Adams et al., 2010;Vagin & Teplyakov, 2010). Iterative cycles of model building were performed using Coot (Emsley et al., 2010), followed by refinement with REFMAC5 . The MtHigA3 crystal structure belongs to space group I4 and contains two molecules per asymmetric unit. The crystal structure of MtHigA3 bound to DNA was determined to be C2 with dimeric MtHigA3 and pseudo-palindromic DNA. A portion of the data (5%) was set aside for the refinement calculations of R free . The data collection and final crystallographic statistics are summarized in Table S2. The protein interface area was calculated using the PISA server (Krissinel & Henrick, 2007), and sequence alignments were performed using ClustalX (Larkin et al., 2007) and visualized by ESPript (Robert & Gouet, 2014). The electrostatic surface was calculated using the program APBS (Baker et al., 2001), and all figures were generated using PyMol version 1.8 (Schrodinger LLC) and Chimera (Pettersen et al., 2004).

Isothermal titration calorimetry measurements
Isothermal titration calorimetry (ITC) experiments were performed using a MicroCal 200 (GE Healthcare) at 25 C, and all buffers, reagents and proteins were degassed prior to use. MtHigA3 and dsDNAs were prepared in a buffer consisting of 50 mM sodium acetate, pH 5.2 and 150 mM NaCl. MtHigA3 (50 mM) was titrated into the experimental chamber containing 3.3 mM DNA. Measurements were conducted with an injection volume of 0.4 ml as an initial injection volume followed by 2 ml for the remaining 19 injections with 150 s spacing and a stirring speed of 750 rev min À1 at 25 C. The heat signals obtained from the raw ITC data were analysed with the Origin software, supplied by MicroCal Inc., fitting the data into a single-site binding isotherm to obtain the binding affinity (K d ), the change in enthalpy (ÁH) and the change in entropy (ÁS). The Gibbs free energy (ÁG) was calculated using the standard equation ÁG = ÁH À TÁS.

NMR spectroscopy and chemical shift perturbation experiments
All NMR spectra of 15 N-labelled MtHigA3 35-117 and 15 N, 13 C-labelled MtHigA3 35-117 were obtained on a Bruker Avance 800 MHz spectrometer equipped with a cryoprobe. All NMR samples were dissolved in the final buffer (20 mM citric acid, pH 5.0 and 150 mM NaCl containing 1 mM DTT), and all experiments were performed at 298 K. Data for the carbonyl carbon were obtained using the HNCO and HNCACO spectra, and data for the / carbon were acquired using the HNCA, HNCOCA, HNCACB and CBCACONH spectra. The p.p.m. values of the backbone N and HN resonances of MtHigA3 35-117 were assigned using these datasets. The chemical shift perturbation experiments were carried out by titrating U-15 N labelled MtHigA3 35-117 with promoter DNA. The phase-sensitive 2D-[ 1 H-15 N] TROSY-HSQC spectrum with sensitivity improvement was acquired for U-15 N labelled MtHigA3 35-117 alone and with increasing concentrations of dsDNA (protein:dsDNA in 1:0.05, 1:0.1, 1:0.2 and 1:0.4 molar ratios). Upon titration, the disappearance or shifting of the NMR signal was monitored. All spectra were processed and analysed using NMRPipe/NMRDraw (Delaglio et al., 1995) and NMRView (Johnson & Blevins, 1994). The average chemical shift perturbation (CSP) values of 15 N and 1 H were calculated using the following equation (Schumann et al., 2007): where Á N and Á NH represent the CSP values of the amide nitrogen and proton, respectively.

Crystal structure of MtHigA3
There are three HigBA systems in M. tuberculosis, but structural information is not yet available on the antitoxin HigA. We first determined the crystal structure of the HigA3 antitoxin from M. tuberculosis (MtHigA3) to 1.97 Å resolution by single anomalous dispersion using selenomethioninesubstituted protein. MtHigA3 was crystallized in the tetragonal space group I4 with one MtHigA3 homodimer per asymmetric unit. Although we tried to crystallize MtHigA3 with the full sequence (MALDI-TOF MS confirmed the molecular weight of purified MtHigA3 to be 12.85 kDa, which corresponds to the molecular weight of full-length MtHigA3, see proteolytically degraded (Hadži et al., 2017(Hadži et al., , 2013. Considering that MtHigA3 is a protease-labile antitoxin and is structurally similar to HigA2 from V. cholerae, it is possible that the N-terminus of MtHigA3 is proteolytically degraded during crystallization. The monomer structure of MtHigA3 consists of either two or three -strands, four -helices and one 3 10 -helix () [ Fig.  1(a)]. The topology follows the order 1 (residues 36-38), 1 (residues 41-48), 2 (residues 53-60), 3 (residues 64-72), 1 (residues 75-77), 4 (residues 80-89), 2 (residues 93-100) and 3 (residues 103-106). Four -helices and one 3 10 -helix together form the -helix bundle that includes the HTH DNAbinding motif. While forming a dimer, the 1-strand from one monomer is aligned parallel to two antiparallel -strands (2 and 3) from the other monomer to form a mixed -sheet. This intermolecular interaction (mixed -sheet) forms a '-lid' on one side of the dimer, while the two HTH motifs responsible for DNA binding flank outwards [ Fig. 1(b)].
The SEC results (Fig. S2) show a single peak corresponding to a molecular weight of 26 kDa, suggesting that MtHigA3 forms a dimer in solution. In the crystal structure, each monomer forms a tight dimer with a large interface area (1459.2 Å 2 ) accounting for 26% of the total solvent-accessible surface. Both hydrophilic and hydrophobic interactions were included in the dimerization. Hydrophilic interactions can be observed at the loop between 1 and the 4-helix, including residues Ser76, His77, Thr78, Glu79, Leu80 and Gly81, and at the solvent-exposed surface of the -lid, including residues Val36, Ala38, His93, Arg95, Val97, Glu99, Thr104, Glu106 and Thr108 [ Fig. 1(a)]. Notably, the -lid is highly responsible for the hydrophilic interactions required for dimerization. The hydrophobic interactions involve a dimeric hydrophobic core through Ile44, Ala47 and Leu48 on 1, Leu75 and His77 on 1, Leu83, Val87 and Leu90 on the 4-helix, and the internal residues of the -lid. This dimerization mode corresponds to typical characteristics of the Cro/C1-type HTH domain: 2and 3-helices form the HTH motif, and 4and 5-helices are involved in dimerization, while the -helix bundle composes a hydrophobic core (Luscombe et al., 2000). Additionally, residues Arg45 and Glu71 in the -helix bundle form a salt bridge that improves the stability of the hydrophobic core, including the HTH motif. A well ordered HTH motif is located on the surface of MtHigA3 and exhibits a distinct positive charge on the overall negatively charged surface, facilitating the binding of DNA with MtHigA3.

Structural comparison of MtHigA3 with other homologues
To date, structures of HigA have been reported from five species, namely Coxiella burnetii (sequence homology 8.6%), E. coli (sequence homology 11.5%), P. vulgaris   Fig. 2(a)]. All the known HigA structures showed only -helices, while MtHigA3 had an additional two -strands that participated in dimerization. HigA of E. coli (EcHigA) and HigA of S. flexneri (Sf HigA) show complete conservation of 141 amino acids, composing 9 -helices. Their entire length is longer than that of MtHigA3, and in contrast to MtHigA3, the HTH motif of EcHigA and Sf HigA is located on the Cterminus. The N-terminal 1-4-helices are responsible for dimerization of EcHigA and Sf HigA, and the 3-6-helices are involved in the formation of complex structures with the corresponding toxins. The structures of HigB toxin-bound HigA from P. vulgaris (PvHigA) and from V. cholerae (VcHigA2) reveal that the N-termini of both PvHigA and VcHigA2 interact with the corresponding toxins. The structure of VcHigA2 was determined in the HigBA2 complex in addition to the structure of the HigA2 antitoxin alone. Interestingly, 36 residues at the N-terminus in the VcHigA2 antitoxin crystal structure were missing, as observed in MtHigA3,  whereas in the V. cholerae HigBA2 complex, the whole HigA structure was observed. The N-terminal 36 residues, which were missing in the VcHigA2 solitary structure, contained a single -helix, a single -strand and a long loop. This component connects the -helix bundle and simultaneously interacts with HigB2 of V. cholerae.
Since the proteins that have Cro/C1-type HTH domains function mainly as dimers to bind to DNA, their antitoxin quaternary structures are compared. Although HigA antitoxins from different species share the same fold in the -helix bundle, their dimerization mode varies due to the individual N-and C-terminal secondary structures. Comparison of the distances between the two HTH motifs of each monomer in the HigA dimer structures showed that the distances varied from 30 to 75 Å [ Fig. 2(a)]. Additionally, the angles between the central axes of the HigA antitoxin molecules in the dimers varied from 63 to 78 (Fig. S3). The distances of the HTH motifs and the dimer angles determine the orientations of the two HTH motifs from each chain. Therefore, the orientation of the two HTH motifs from each chain is unique, indicating their distinct DNA binding patterns.
To compare the structural similarity, MtHigA3 was analysed using the DALI server (Holm & Rosenstrom, 2010). The DALI search revealed the structure of E. coli HipB (EcHipB, PDB 4z5h, DALI score 7.2, sequence homology 14.6%, r.m.s.d. 2.0) as the highest match to the structure of MtHigA3. High-persistence A (hipA) was first discovered in the mutant E. coli K-12 in 1983 (Moyed & Bertrand, 1983). The toxin HipA is an EF-Tu kinase that inhibits protein synthesis, and the antitoxin protein HipB neutralizes toxin activity or is degraded by Lon protease (Correia et al., 2006;Hansen et al., 2012). From a structural perspective, the -helix bundle of EcHipB is more similar to that of MtHigA3 than those of other HigA homologues. EcHipB and MtHigA3 share similar characteristics in the hydrophobic core [ Fig. 2(a)]. The Cterminal -lid, a unique characteristic of MtHigA3, was only observed in EcHipB, but only one -strand from each domain forms the -lid for EcHipB. Instead of an additional -strand, EcHipB has a C-terminal stretch that consists of 16 residues and is unstructured. This C-terminus is recognized as a degradation signal by Lon protease (Hansen et al., 2012).
Despite the low sequence homology shared among the members of the HigA family, the HTH motif is well conserved [ Fig. 2(b)]. Although the sequences of the HTH motifs are not identical, similarities between residues in HTH motifs can be observed in HigA family proteins. Several proteins of this family share positively charged residues that can contribute to DNA recognition, whereas others possess aromatic residues that can contribute to the conformational stability of the DNA-protein complex structure (Anjana et al., 2012;Cherstvy, 2009) [Fig. 2(c)]. In the cases of EcHigA and Sf HigA, the 2-helix is shorter than those of other homologues, which makes the HTH motif insecure. Another difference is that the aromatic residues are occasionally found in -helices [ Fig. 2(c)]. The aromatic residues are found in Phe27 in the 2-helix of PvHigA, Phe45 in the 3-helix of EcHigA, Phe59 in the 2-helix of VcHigA2 and Trp73 in the 3-helix of VcHigA2. This sequential variation in the HTH motif aids in the recognition of diverse DNA sequences and the stabilization of the DNA-protein complex structure.

Analysis of the HigBA operon of M. tuberculosis and identification of the operator site
In most type II TA systems, the antitoxin is located upstream of the toxin. Both the antitoxin and TA complex could bind to the operator DNA of TA modules and autoregulate the TA operon. However, MtHigBA3 has a reverse TA system, and MtHigB3, the cognate toxin of MtHigA3, is located directly upstream of the MtHigA3 antitoxin. We hypothesized that, in this reverse gene order, an additional promoter for the antitoxin could regulate the gene transcription. Therefore, we searched for the putative promoter regions of MtHigBA3 and MtHigA3 using the BPROM tool (Solovyev, 2011). However, unlike the Acinetobacter baumannii higBA2 operon (Armalyte et al., 2018), we found only the predicted operator regions of MtHigBA3. In the A. baumannii higBA2 operon, Armalyte et al. found two distinct promoter regions located on the DNA: one for the complete TA system (higBA2) and the other for the higA2 antitoxin (Armalytė et al., 2018). The higBA3 promoter region, 230 bp of the MtHigB3 upstream DNA, includes two putative operator sites: a À35 box (DNA1) and a À10 box (DNA2). We expanded our search to look for any additional DNA interacting regions, and found two palindromic sequences in the promoter region. One of these sequences is located upstream of the À35 box (DNA3), and the other is located around the À10 box (DNA4) [ Fig. 3(a)]. To confirm the interaction of DNA with the MtHigA3 antitoxin, we performed an EMSA. The DNA was resolved on a gel in the absence or in the presence of the protein. The DNA concentration was maintained at 0.01 mM, and the protein concentration was increased from 0 to 0.6 mM, resulting in 1:0, 1:10, 1:20, 1:30, 1:40, 1:50 and 1:60 molar ratios (DNA:protein). We observed a band upshift for DNA2 only, confirming the binding of DNA2 with MtHigA3 [ Fig. 3(b)]. The result indicates that the À10 box could act as the higA3 operator, and MtHigA3 binds specifically to this sequence.
We additionally conducted ITC experiments to quantify the interaction between MtHigA3 and DNA2 (À10 box). The binding stoichiometry (n) was calculated as 0.910 AE0.004 using a one-site model. This indicates that one dimer of MtHigA3 binds to one DNA2 fragment. The binding affinity (K d ) was determined to be 11.3 AE1.8 nM, indicating high affinity. The interaction between DNA2 and MtHigA3 is exothermic, with a negative change in enthalpy (ÁH = À8.83 AE0.09 kcal mol À1 ) and a favourable entropy (TÁS = 2.03 kcal mol À1 ) [ Fig. 3(c)].

Overall structure of MtHigA3 bound to DNA
Since DNA2 showed clear binding in the EMSA and ITC experiments, we tried to co-crystallize MtHigA3 with DNA2. The dsDNA was incubated for 1 h at room temperature with dimeric MtHigA3 before crystallization. MtHigA3 bound to DNA was found to crystallize in space group C2 with the MtHigA3 homodimer and 20 bp dsDNA in the asymmetric research papers 754 Park et al. Induced DNA bending by unique dimerization of HigA antitoxin unit. Since the DNA duplex used for co-crystallization was pseudo-palindromic, it was not in a strict twofold symmetry. However, the homodimer and dsDNA dyad forms (dyad of the MtHigA3 monomer and single-stranded DNA) correspond to crystallographic dyads. The same dataset was successfully reintegrated and determined in space group C2221 which lacks the crystallographic dyad. This is probably a consequence of the random bimodal orientation of the complex around the dyad (Becker et al., 1998). Detailed DNA structural analysis was difficult, but we could identify the structures of MtHigA3 residues and DNA nucleotides. The electron-density map of the N-terminal 35 amino acids of MtHigA3 bound to DNA was missing, probably due to protease activity (Hadži et al., 2017;Hadži et al., 2013). The whole topology of MtHigA3 bound to DNA is similar to that of the MtHigA3 dimer. The overall structures of the MtHigA3 dimer and MtHigA3 bound to DNA are very similar (0.81 Å r.m.s.d. for 78 corresponding C atoms of chain A) [ Fig. 4(a)]. However, the N-terminal residues of chain A of the MtHigA3 homodimer (Asp34 and Ala35) show a relatively high r.m.s.d.
(1.45-2.48 Å ). These residues are missing in the structure of MtHigA3 bound to DNA. Another region that shows high r.m.s.d. is the loop region (Ser76 and His77) which is compactly positioned in the DNA minor groove, located at the centre of the DNA [ Fig. 4(a)].
The interactions between DNA and the MtHigA3 dimer occur mainly through HTH motifs. Similar to other HTH motifs, the major groove of DNA directly interacts with HTH motifs of the MtHigA3 dimer. The recognition 3-helix is inserted in the major groove, and the 2 and the loop that links the 2and 3-helices support the contacts with the DNA backbone. In the structure of MtHigA3 bound to DNA, the MtHigA3 homodimer and DNA show an average interface area of 341.8 Å , which is relatively small. Since HTH motifs are positively charged, this motif binds to negatively charged DNA stably via an overall charge-charge interaction [ Fig.  4(c)]. Fig. 4(b) shows the overall structure and conformation of DNA with the dimeric protein. The major groove of the DNA is largely responsible for binding and/or interacting with the HTH motifs of the dimeric protein [ Fig. 4(b)]. The DNA- MtHigA3 binds to the higBA3 promoter region at a specific sequence. (a) Organization of the higBA3 operon. The higB3 toxin gene is located upstream of the higA3 antitoxin gene. Two putative operator regions predicted by the BPROM web interface (Solovyev, 2011) located at the À35 and À10 positions are shown in red letters. DNA1, which includes the À35 box, is highlighted as grey, and DNA2, which includes the À10 box, is highlighted in blue. Predicted palindromic sequences (DNA3 and DNA4) for the higBA3 gene are coloured grey. The locations of the higB and higA genes are indicated by red and blue arrows, respectively. bound complex structure indicates that the N-terminal region does not mediate DNA contact. The data suggest that the Nterminal end (34 residues), although susceptible to protease activity, is not involved in DNA interaction. Interestingly, unusual bending of DNA was observed in MtHigA3 bound to DNA. The DNA fragment bends towards the minor groove with an overall bending angle of 46.5 , as calculated by the CURVES server (Lavery et al., 2009) [Fig.  4(b)]. The average width of the minor groove in canonical B-DNA is 6 Å , but in the structure of MtHigA3 bound to DNA, the width of the minor groove at the A-tract region (bases 9-12) located at the centre of the DNA is narrow ($3.7 Å ) (Fig.  S4). The A-tract region in the DNA core induces the bending of DNA through the interaction between the proteins and the phosphodiester backbone in DNA, causing compression of the minor groove (Ball et al., 2012;Hizver et al., 2001). The minorgroove width at bases 16 to 18 ($2.5 Å ) is also compressed, which stablizes the structure of MtHigA3 bound to DNA.
When the structure of MtHigA3 bound to DNA was compared with that of the MtHigA3 dimer, we found that the distance between the HTH motifs of each monomer was changed. The distance measured from Q64 of each monomer on the 3-helix was increased by 1.9 Å in the MtHigA3 bound to DNA compared with the MtHigA3 dimer (from 37.3 Å to 39.2 Å ). This increase is due to a shift of the MtHigA3 monomer in MtHigA3 bound to DNA, which has been reported previously in PvHigA (Schureck et al., 2019). The results suggest that rearrangement of the dimer structure is required for DNA binding, providing a wider space between the HTH motifs of each monomer. Overall structure of MtHigA3 bound to DNA. (a) Superimposed cartoon representation of MtHigA3 bound to DNA and the MtHigA3 dimer structure (produced by matching the MtHigA3 dimer of both models). The MtHigA3 dimer is coloured cyan and blue. MtHigA3 bound to DNA is coloured grey. The DNA fragment of MtHigA3 bound to DNA is coloured yellow. Plot of C r.m.s.d. of the MtHigA3 dimer and MtHigA3 bound to DNA to compare C movement after DNA binding. N-terminal (Val36-His39), loop region (Ser76-His77) and C-terminal (Thr108-His113) residues show relatively large deviations. The region showing more than 1 Å r.m.s.d. is drawn in the black box and highlighted in green. (b) Cartoon and surface representation of MtHigA3 bound to the DNA structure in diverse orientations. The regions of MtHigA3 and DNA structure, which involve contact with each other, are coloured blue (MtHigA3) and yellow (DNA). Bent DNA in MtHigA3 bound to the DNA structure is marked as a red line. Promoter DNA is dramatically bent by the binding of MtHigA3 (46.5 ). (c) Superimposed MtHigA3 dimer and DNA structure of MtHigA3 bound to DNA. The electrostatic surface potential of the MtHigA3 dimer is coloured between À10 kT e À1 (red) and 10 kT e À1 (blue) in two orientations rotated by 90 . The DNA structure of MtHigA3 bound to DNA is presented as a cartoon representation. DNA binds to the positively charged region of the MtHigA3 dimer. solution state, which is close to the natural state, we performed NMR titration experiments. Initially, we tried to assign the full sequence of MtHigA3, but the NMR resonances of full-length MtHigA3 were not well dispersed. This might be caused by the flexibility of the N-terminus of MtHigA3. Based on our MtHigA3 crystal structure, we eliminated 34 amino acids at the N-terminus of MtHigA3 from the construct. N-terminustruncated MtHigA3, including the His 6 -tagged C-terminus (MtHigA3 35-117 ), shows a well dispersed 2D-[ 1 H-15 N] TROSY-HSQC spectrum. Backbone assignment of MtHigA3 35-117 (with the exception of the residues Ala35-Leu48, Gln53, Ala54, Lys69, Gly73, Leu75-His77, Tyr86, Asn103, Leu107 and His113-His117) could be performed and was used for the NMR titration experiment of 0.4 mM 15 Nlabelled MtHigA3 35-117 with promoter DNA [ Fig. 5(a)]. By monitoring the HSQC spectra of MtHigA3 35-117 with increasing DNA concentration, line broadening and chemical shift changes were observed. The signal broadening of the NMR spectrum implies slow-to-intermediate exchange on the NMR time scale, a characteristic of moderate-to-high binding affinity. Most NMR signals upon the addition of DNA show weakening or disappearance. These signals correspond to residues that are in close contact with DNA in the crystal structure of MtHigA3 bound to DNA [Figs. 5(b) and 5(c)]. The peaks of MtHigA3 35-117 which disappeared in the titration experiment, namely, the peaks for His50, Arg52, Ala57, Leu59, Met60, Ser63, Ala65, Arg66, Ser68, Glu71, Thr78, Glu79, Ile96 and Thr108, were located around the HTH motif consisting of the 2and 3-helices (53-72), which interact with the major groove of the DNA. The residues Thr78 and Glu79, located at the loop between 1 and 4, interact with the minor groove of the A-tract region in the DNA. The residues His50 and Arg52 are located in the loop between the 1and 2-helices, which is in front of the HTH motif. Their tight binding with both ends of the DNA could be related to the curvature of the whole DNA molecule.
The most highly perturbed peaks presenting higher CSP values than the average CSP value (0.075) were for Ala51, Val56, Val67, His93, Leu94, Ala98, Val105, Glu111 and His112. These residues are located in the 2and 3-helices of the HTH motif and in the 2and 3-strands that form the -lid. Generally, peak shifting in the NMR spectrum can be attributed to the fast exchange phenomenon on the NMR time scale, which indicates low-to-moderate binding affinity in solution state. Although the 2and 3-strands participating in the dimerization of MtHigA3 are not directly involved in DNA binding in the crystal structure, the NMR titration result indicates that this region is associated with DNA interaction. This implies that the dimerization mode was affected by DNA binding.

Discussion
Tuberculosis caused by M. tuberculosis is a prevalent disease worldwide, and unsuccessful treatment resulting from antibiotic resistance and tolerance of M. tuberculosis is a risk factor for global public health care. TA systems related to antibiotic resistance and tolerance are emerging as new antibiotic drug targets of TB. Among the six types of TA systems classified to date, M. tuberculosis has numerous type II TA systems. We studied the MtHigA3 type II TA system protein, a cognate antitoxin of the MtHigB3 ribonuclease toxin.
MtHigA3 is predicted to be a transcriptional regulator that autoregulates the higBA3 operon by binding to the operator as an antitoxin itself or via complex formation with the toxin MtHigB3, similar to other type II TA systems. We firstly present the structure of HigA3 of M. tuberculosis, possessing unique characteristics compared with other HigA homologues, with the exception of the HTH motif (Figs. 1 and 2). In particular, the presence of a -lid at the C-terminus of MtHigA3 is an unusual structural feature of HigA homologues. From a structural perspective, EcHipB, which has a small -lid at the C-terminus, presents a rather similar structure to MtHigA3. This structural difference in the same TA family may indicate structural diversity within the TA system. HigBA homologue complexes do not share similar complex interfaces. Briefly, the N-terminal region, being intrinsically disordered, is not observed in VcHigA2 and in MtHigA3 as a solitary antitoxin structure, whereas the structure of VcHigA2 in the VcHigBA2 complex has a long loop, one -helix and one -strand at the N-terminus forming a complex interface.
We have shown that the HTH motif mainly contributes to the DNA binding of MtHigA3 by determining the complex structure of MtHigA with operator DNA and by NMR titration experiments of MtHigA3 35-117 with DNA. The HTH motifs of dimeric proteins are responsible for binding to DNA and are associated with the characteristic bending of DNA. Residues with large chemical shift changes or residues that disappeared in the NMR titration, corresponding to the DNAbinding residues, are located at the HTH motif (Ala51, Arg52, Ser63, Ala65, Arg66 and Glu71). The structure of the Structural comparison of MtHigA3 bound to DNA and its homologues. (a) Cartoon representation of MtHigA3 bound to DNA (yellow), PvHigA bound to DNA (blue) and EcHipB bound to DNA (cyan). The C-terminus of the antitoxin which differs markedly in comparison is marked with a black dotted circle. Only the protein structures in these DNA-protein complexes are shown for clarity. (b) The superimposed cartoon represents the DNA structure of MtHigA3 bound to DNA (yellow), PvHigA bound to DNA (blue) and EcHipB bound to DNA (cyan). Only the DNA structures in these DNAprotein complexes are shown for clarity. (c) Structure of MtHigA3 bound to DNA, PvHigA bound to DNA and EcHipB bound to DNA. The antitoxins are presented as cartoon diagrams in surface view. Dimer angles between central stalks are indicated.
MtHigA3 dimer in the complex structure is similar to that of the MtHigA3 dimer. However, the distance between each monomer is increased in the complex structure, and the angle of the MtHigA3 dimer is changed.
By comparing the structure of MtHigA3 bound to DNA with those of PvHigA bound to DNA and EcHipB bound to DNA, we found that the quaternary structure of the antitoxin affects the curve of DNA bending. When the structures of MtHigA3 bound to DNA, PvHigA bound to DNA and EcHipB bound to DNA were aligned based on the antitoxin monomer, the curvature of the DNA increased as the dimer angles decreased [ Fig. 6(b)]. The major structural difference between these DNA-protein complexes was located at the Cterminus of the antitoxins, which take part in protein dimerization. MtHigA3, PvHigA and EcHipB show different tertiary structures at the C-terminus, located on the opposite side of the DNA-binding interface. PvHigA lacks a -strand at the C-terminus and is dimerized only through a hydrophobic core composed of -helices. While EcHipB has two -stranded -lids at the C-terminus, MtHigA3 shows a broad -lid formed by four -strands at the C-terminus and one -strand at the Nterminus [Figs. 6(a) and 6(c)]. The difference in the three structures revealed that the higher the amount of secondary structure at the C-terminus, the tighter the binding network of the antitoxin dimer. This difference induces a change in the quaternary structure of antitoxin dimers by regulating the dimer angle of the antitoxin structure. This change in the quaternary structure of the antitoxin could induce the bending of DNA.
In this study, we obtained the first structural data for M. tuberculosis HigA3. From this result, we found that, even though HigA homologues have structurally similar HTH motifs, the overall structures are different from each other. We also obtained the crystal structure of MtHigA3 bound to DNA and information for the binding of MtHigA3 with DNA by NMR titration experiments. From EMSA experiments, we confirmed that MtHigA3 recognizes DNA sequences specifically. Our research could provide structural information for understanding higBA3 operon regulation. Based on these structural data, the development of peptides mimicking and/or inhibiting the MtHigA3-DNA binding interface may be possible. These results may provide useful information for structure-based antibiotic development.