Structure of FitAB from Neisseria gonorrhoeae Bound to DNA Reveals a Tetramer of Toxin-Antitoxin Heterodimers Containing Pin Domains and Ribbon-Helix-Helix Motifs*

Neisseria gonorrhoeae is a sexually transmitted pathogen that initiates infections in humans by adhering to the mucosal epithelium of the urogenital tract. The bacterium then enters the apical region of the cell and traffics across the cell to exit into the subepithelial matrix. Mutations in the fast intracellular trafficking (fitAB) locus cause the bacteria to transit a polarized epithelial monolayer more quickly than the wild-type parent and to replicate within cells at an accelerated rate. Here, we describe the crystal structure of the toxin-antitoxin heterodimer, FitAB, bound to a high affinity 36-bp DNA fragment from the fitAB promoter. FitA, the antitoxin, binds DNA through its ribbon-helix-helix motif and is tethered to FitB, the toxin, to form a heterodimer by the insertion of a four turn α-helix into an extensive FitB hydrophobic pocket. FitB is composed of a PIN (PilT N terminus) domain, with a central, twisted, 5-stranded parallel β-sheet that is open on one side and flanked by five α-helices. FitB in the context of the FitAB complex does not display nuclease activity against tested PIN substrates. The FitAB complex points to the mechanism by which antitoxins with RHH motifs can block the activity of toxins with PIN domains. Interactions between two FitB molecules result in the formation of a tetramer of FitAB heterodimers, which binds to the 36-bp DNA fragment and provides an explanation for how FitB enhances the DNA binding affinity of FitA.

Neisseria gonorrhoeae (GC) 3 is the agent of the sexually transmitted disease, gonorrhea. The mechanisms used by GC to initiate infection have been very well characterized. Gonococci adhere via a multistep cascade and subsequently enter cells forming the epithelial barrier of the urogenital tract, traffic across these cells and exit into the subepithelial matrix (1,2). Although studies have identified many of the molecular mechanisms used by GC to adhere to and enter cells, our knowledge of the mechanisms that operate in the later stages of infection is limited.
GC are able to survive and grow within epithelial cells (3); they also traverse the epithelial monolayer to infect the stromal tissue of the subepithelium (2). The immune response to bacteria in the subepithelium produces the inflammation and purulent discharge characteristic of gonorrhea (4,5). On occasion, GC establish a carrier state in which an asymptomatic individual harbors culturable and transmissible bacteria. These carriers are key to the spread of gonococcal disease, as humans are the only known reservoir for GC (6). The mechanisms by which GC maintains this persistent state are unknown. One hypothesis is that the organism resides within the epithelial cells instead of crossing into the subepithelium, thus evading the host immune response. The gene product(s) that affect GC intracellular growth and transcytosis are therefore important for the maintenance of gonococci in the human population.
The fitAB operon was identified in a screen for GC mutants with a fast intracellular trafficking phenotype across polarized epithelial monolayers (3). A GC mutant that lacks fitAB grows normally extracellularly, but has an accelerated rate of intracellular replication with a concomitant increase in the rate at which this mutant traverses a monolayer of polarized epithelial cells. Thus, either FitA or FitB, or their complex, is hypothesized to slow intracellular replication and intracellular trafficking of GC.
FitA is an 8.4-kDa protein with a predicted N-terminal ribbon-helix-helix (RHH) DNA binding motif (7,8). FitB is a 15.3-kDa protein with a predicted PIN (PilT-N terminus) domain according to the BLAST search tool (9). The function of the PIN domain is unknown; however many proteins that contain a PIN domain are thought to perform roles in nucleic acid metabolism including synthesis and remodeling. In genome studies on Archaea and thermophilic bacteria, sequences predicted to encode PIN domain-containing proteins are found in regions predicted to encode DNA polymerases, helicases, and nucleases (10). In addition, the Dis3p exonucleases from Saccharomyces cerevisiae and nonsense-mediated mRNA decay (NMD) proteins in Caenorhabditis elegans are predicted to have PIN domains (11,12). Bicistronic operons where an RHH DNA-binding protein is juxtaposed with a PIN domain have been proposed to form one family of toxin/antitoxin systems (13). The PIN domain containing protein is thus predicted to act as a toxin; this is in agreement with the role of FitB in slowing GC replication when the bacteria are within epithelial cells (3).
FitA and FitB form a heterodimer, and copurify after overexpression in Escherichia coli. The FitAB complex binds DNA from the fitAB upstream region with high affinity (8). In our current model the FitAB complex binds to the fitAB promoter when GC are in an extracellular environment. This results in both sequestration of FitAB and repression of fitAB transcription. Upon invasion, FitAB may be released from the DNA and subsequently dissociate to slow both GC replication and transcytosis by an as yet undefined mechanism.
To understand the structural basis of the DNA binding specificity and possible nuclease function of the FitAB complex, the x-ray crystallographic structure of a FitAB complex bound to a high affinity 36-base pair DNA fragment from the fitAB upstream region was determined. Four FitA and four FitB proteins form a unique tetramer of heterodimers structure that explains the ability of FitAB to bind DNA with higher affinity than FitA alone (8). Furthermore, the structure suggests that FitB could slow intracellular GC replication by acting as a "toxin" when the FitA ("the antitoxin")-mediated inhibition of FitB nuclease activity is relieved upon complex dissociation.

EXPERIMENTAL PROCEDURES
Protein Preparation, Crystallization, and X-ray Intensity Data Collection-FitA and FitB were overexpressed in E. coli using a pET28b vector (Invitrogen) that encodes the intact FitAB operon (8). This vector incorporates a T7 promoter at the 5Ј-end of fitA and sequences encoding six histidine residues at the 3Ј-end of fitB. The C-terminal amino acid sequence of FitB was also slightly altered by the addition of an XhoI restriction site, changing from . . . NPWHD to . . . NPWHLEHHHHHH. The overexpressed FitAB complex was purified using nickel affinity chromatography (Qiagen). Purified FitAB complex was concentrated to 5 mg/ml in 25 mM Tris, pH 7.5, 500 mM NaCl, 200 mM imidazole.
The intact FitAB complex did not crystallize despite numerous attempts. Therefore, limited proteolysis was done on the FitAB complex in order to generate a stabile core that might be more amenable to crystallization. Using 0.1 mg/ml trypsin (Sigma) the complex was digested for 30 min at 22°C before crystallization trials. Trypsin inhibitor cross-linked to agarose beads (Sigma) was used to remove trypsin from the FitAB solution after digestion. Polyacrylamide gel electrophoresis and mass spectrometry analysis revealed that this treatment removed the ribbon-helix-helix motif of FitA and the resulting complex, termed FitcAB, does not bind DNA (data not shown). Crystallization was carried out using hanging drop-vapor diffusion where FitcAB was mixed 1:1 (v:v) with a reservoir solution of 0.26 M sodium phosphate/citrate pH 4.7 and 2.0 M ammo- To generate selenomethionine (SeMet)-substituted FitcAB complex, the expression vector described above was used as a template for standard PCR mutagenesis (Stratagene) of FitB, which contains no methionines, to yield a construct encoding FitAB where residues Leu 43 , Leu 63 , and Leu 116 of FitB were substituted with methionines (FitAB3(LxM)). The FitAB3(LxM) protein was purified as described above. DNA binding assays confirmed that the FitAB3(LxM) complex has the same affinity for DNA as wild-type FitAB (data not shown). For overexpression of SeMet-substituted FitAB3(LxM), E. coli harboring the expression vector were grown in minimal medium lacking methionine with added SeMet as described (14). Using nickel affinity column chromatography, SeMet-containing FitA and FitB3(LxM) copurify as do the wild-type proteins. The SeMet-containing heterodimer was concentrated and trypsinized as described for wild-type FitAB. Crystallization of the SeMet-FitcAB3(LxM) complex employed 0.26 M sodium citrate pH 5.6 and 2.0 M ammonium sulfate. Crystals with dimensions 0.2 mm ϫ 0.2 mm ϫ 0.2 mm were obtained after 4 weeks.
To crystallize the FitAB-DNA complex, 5 mg/ml of purified native FitAB3(LxM) was mixed in a 4:1 molar ratio with IR36 DNA (8), where two of the thymine bases were replaced with 5-iodouracil (I) (top strand, 5Ј-AGATTGCTATCATTTTTT-TTATTTTGATAGCATITG; bottom strand, 5Ј-CAAATGC-TATCAAAAIAAAAAAAATGATAGCAATCT). The protein-DNA complex was then mixed 1:1 (v/v) with a reservoir solution of 0.1 M sodium acetate, pH 4.0, 0.27 M sodium acetate, pH 7.0, 7.2% PEG 20,000, 7.2% PEG monomethyl ether 550. In all panels, FitA is shown in magenta and FitB is in cyan. a, ribbon diagram of the FitAB heterodimer. The ␣-helices and ␤-sheets for both proteins are labeled as are the N and C termini. b, sequence alignment of FitA and FitB proteins with homologues. Secondary structural elements of the proteins are shown above the alignment. Y4jJK from Rhizobium sp. and StbCB from Pseudomonas syringae are highly homologous to FitAB by sequence analysis but no structural data are available for either of these systems. Arc and Mnt from Enterobacteria phage P22 are ribbon-helix-helix proteins with structural homology to FitA. PilT from Azotobacter vinelandii and PAE2754 from Pyrobaculum aerophilum are PIN domain containing proteins with structural homology to FitB. Alignments were generated with ClustalW (49). Residues involved in FitA-FitB heterodimerization are highlighted in yellow, those forming the FitA-FitA dimer interface are highlighted in green, and those making up the FitB-FitB dimer interface are highlighted in orange. Conserved residues involved in specific protein-DNA contacts are highlighted in blue and the conserved acidic residues forming the putative FitB active site are highlighted in red. c, stereoview of the FitA-FitB heterodimer interface. Residues involved in ionic interactions contributing to the stability of the interface are colored according to gray (carbon), red (oxygen), and blue (nitrogen). This extensive interface buries up to 1900 Å 2 accessible surface area.
Structure Determinations and Refinements-The structure of SeMet-FitcAB3(LxM) was solved by multiple wavelength anomalous diffraction (MAD) methods using the SeMet-FitcAB3(LxM) data collected at three wavelengths (Table 1). Seven selenium sites were located and initial phases were calculated with SOLVE (17) using data from 20.0 to 3.0 Å resolution and improved by solvent flipping (with 45% solvent content) as implemented in the crystallography and NMR system (CNS) (18). The handedness was determined by inspection of electron density maps where the initial phases were derived either from the selenium atom sites found by SOLVE or sites with the coordinates inverted. Electron density maps for the entire resolution range (58.0 -2.0 Å) were then calculated using CNS. FitB and amino acid residues 46 -65 of FitA were built into the map using O (19), and the location of the selenomethionine residues as reference points. 66 water molecules were added to the model and simulated annealing (SA), positional, and thermal parameter refinement using CNS were performed, followed by rebuilding of the model in O. When the R work and R free were 28.2 and 29.7%, respectively, the coordinates were used to solve the structure of the wild-type, native FitcAB molecule by molecular replacement using MOLREP (20). Multiple rounds of positional and thermal parameter refinement in CNS followed by model rebuilding in O were done using the native data to the limiting resolution of 1.8 Å until the R free converged to 22.4%.
The final model contains residues 1-139 of FitB, 46 -64 of FitA, 92 water molecules, 1 acetate ion, 2 sulfate ions, and 3 magnesium ions. The final model was verified by inspection of 2F o Ϫ F c simulated annealing-composite omit maps.
The high resolution FitcAB structure was used as a model to solve the structure of the FitAB-IR36 complex by molecular replacement using MOLREP (20). The remainder of the FitA sequence and the DNA was built into the map using O (19). After simulated annealing (SA) and extensive positional and thermal parameter refinement using CNS followed by model rebuilding in O, the R work and R free converged to 21.2% and 26.9%, respectively, at 3.0 Å resolution.

RESULTS AND DISCUSSION
Structure of the FitAB Heterodimer-The structure of the FitB protein complexed with a C-terminal fragment of FitA (FitcAB) was determined to 1.8 Å resolution by multiple wavelength anomalous diffraction using selenomethionine substituted proteins ("Experimental Procedures" and Table 1). This structure was used as a model to solve the structure of the full-length FitAB complex bound to a 36-bp DNA molecule to 3.0 Å resolution by molecular replacement ("Experimental Procedures" and Table 1).
The FitA monomer has an extended structure, with the topology ␤1 (residues 4 -7), ␣1 (residues 11-23), ␣2 (residues 28 -43), ␣3 (residues 48 -59) (Fig. 1a). Electron density is visible for the intact N terminus of FitA, beginning at Ala 2 . Met 1 is not present in our preparation, as determined by N-terminal sequencing of the protein (data not shown). At the C-terminal end, variable electron density is seen for the four molecules of the asymmetric unit, the final 9 -14 residues are not visible (depending upon the monomer). The first 45 residues of this protein (␤1-␣1-␣2) are highly homologous to the RHH class of DNA-binding proteins, which includes the bacteriophage P22 proteins Mnt and Arc (Fig. 1b) (7, 8, 22). An overlay of the FitA  DECEMBER 8, 2006 • VOLUME 281 • NUMBER 49

Structure of the FitAB Complex Bound to DNA
and Arc repressor structures results in a root mean-squared deviation (RMSD) of 1.1 Å over the first 45 residues (23). An arginine found in ␤1 of the RHH proteins is conserved in FitA (Fig. 1b, highlighted in blue).
In addition to the RHH and PIN domain-containing proteins, there is a group of prokaryotic proteins with a high level of sequence homology to FitAB (Fig. 1b). These typically consist of both a FitA and a FitB homologue in a conserved operon organization and little is understood about their biochemical function, although they are known to play a general role in plasmid stability and/or partition (27)(28)(29)(30)(31). These have been proposed to act as toxin/ antitoxin pairs, with the RHH protein acting as the antitoxin, repressing the toxic activity of the PIN domain-containing protein (32). The structure of the FitAB complex is likely to predict the structures of these toxin/antitoxin proteins. The biochemical function of this group of proteins within prokaryotic cells is likely to be similar as that performed by FitAB. Sequence alignments of FitA and FitB with two examples of such systems (Y4jJ/K and StbCB) are shown in Fig. 1b.
The FitA-FitB Interface-The FitA and FitB proteins form a tightly associated dimer and the FitAB structure reveals the heterodimerization interface, which is formed predominantly by contacts between ␣3 and the C-terminal extended coil region of FitA and helices ␣1, ␣2, and ␣4 of FitB (Fig. 1, a and c). The interface buries 1900 Å 2 accessible surface area in which the FitA helix fills a large exposed hydrophobic groove on FitB resulting in a globular heterodimeric domain (Fig. 1, a and c). FitA is in magenta and FitB is in cyan. Selected residues are colored according to gray (carbon), red (oxygen), and blue (nitrogen). a, FitA dimerization forms the DNA binding ␤ sheet, with ␤1 from each of the FitA subunits (I and IV or II and III). b, the two FitA RHH motifs form one globular domain, with interactions between ␣1 of one subunit and ␣2 of the dimer partner (I with IV and II with III). Two identical interfaces are formed in this manner per FitA dimer, which together with the ␤ sheet stabilize the domain and bury 2850 Å 2 accessible surface area. c, two FitB monomers have an extensive dimerization interface (I and II or III and IV), where ␣3 from one subunit contacts ␣5 from the other subunit. Half of this interface is shown here, as there is 2-fold symmetry around residue Phe 78 (F78) in the complete interface. FitB homodimerization buries 1870 Å 2 accessible surface area.
Much of the interface is hydrophobic. For example, the nonpolar side chain of residue Leu 52 of the FitA helix sits between Ile 9 and Pro 12 of FitB ␣1 (Fig. 1c). At the C terminus of FitA, outside of the helical region, the side chains of residue Ile 59 contacts the side chains of residues Val 58 and Leu 59 from ␣4 of FitB (Fig. 1c). Other residues of the ␣4 and ␣4-␤3 turn of FitB that are important components of the dimer interface are Ile 67 , Leu 70 , and Phe 71 , which contact Leu 48 , Met 51 , and Ile 55 of FitA (Fig. 1c).
FitA helix ␣3 binding to FitB is also stabilized by four ionic interactions, which serve to orient and buttress the two molecules, thereby facilitating a tight association between the complementary hydrophobic surfaces (Fig. 1c). Interestingly, two of these electrostatic interactions are analogous to the charge clamp described in structures of the human nuclear receptors bound to helices of coactivators (33,34). In the charge clamp, charged residues engage in favorable electrostatic interactions with the oppositely charged dipole of the helical termini. The guanidinium group of FitA residue Arg 47 interacts with the backbone carbonyl of residue Asp 26 at the C-terminal end of the FitB ␣2-helix. The N atom of FitB residue Lys 55 interacts with a backbone carbonyl oxygens of residues Glu 60 and Glu 61 at the C-terminal end of the FitA helix. Other stabilizing electrostatic interactions occur between the side chains of FitA Glu 58 and FitB Arg 62 and FitA Glu 63 and FitB Arg 14 (Fig. 1c).
Structure of the FitAB-IR36 Complex-Four FitAB heterodimers form a unique tetramer structure that binds to one 36-base pair DNA fragment (Fig. 2). The FitA subunits from complexes I and IV form a tightly associated domain that binds to one of the inverted repeat half-sites while the corresponding FitA portions of complexes II and III bind to the other inverted repeat (Fig. 2a). By contrast, the FitB-FitB dimer interfaces are formed between subunits from complexes I and II and complexes III and IV (Fig. 2a). Such mixing and matching of dimer interfaces is novel for the RHH family and results in the formation of the tetramer of heterodimers. Other toxin-antitoxin pairs have also been shown to form various higher order structures. For example, the MazE-MazF complex forms an extended heterohexamer (MazF 2 -MazE 2 -MazF 2 ), and the RelE-RelB complex is a single globular domain that assembles into a heterotetramer (RelE 2 -RelB 2 ) (35, 36).
The FitA homodimer buries an extensive accessible surface area (2850 Å 2 ). This compact dimeric domain is characteristic of the RHH DNA-binding proteins (22), and its formation involves nearly every amino acid residue of ␤1, ␣1, and ␣2 of FitA (Fig. 3, a and b). The ␤1-strands from each FitA monomer combine to form a two stranded antiparallel ␤-sheet with four intersubunit hydrogen bonds between the backbone carbonyl oxygen and amide nitrogen atoms and a van der Waals contact between the side chains of Val 5 and its dyadic mate (Fig. 3a). The two ␣-helices of the RHH domain contribute to the extensive interface whereby residues Thr 13 , Ala 16 , and Ile 17 of ␣1 from one subunit form hydrophobic contacts to residues Leu 36 , Ile 39 , and Ala 40 of ␣2 of the other subunit (Fig. 3b). This interface occurs twice in each RHH domain. In addition to these van der Waals contacts, an ionic interaction is formed between the N⑀ group of Arg 20 from ␣1 and the O⑀ atom of Gln 43 from ␣2Ј (Fig. 3b).
The FitB homodimer also buries a large accessible surface area (1870 Å 2 ). At this interface, ␣3 from one FitB monomer contacts ␣5 from the other FitB monomer (Fig. 3c). As above for FitA, there are two such dyad related interfaces per homodimer. Phe 78 in the ␣4-␣5 loop is a key residue for FitB-FitB dimerization. The backbone amide and carbonyl atoms of Phe 78 form hydrogen bonds to their counterparts in the adjacent monomer. In addition, the Phe 78 aromatic ring contributes to the hydrophobic interface by approaching the C␤ methyl group of residue Ala 37 from ␣3 (Fig. 3c). Other hydrophobic contacts involve the side chains of residues Ala 41 , Leu 45 , and Ala 48 from ␣3 of one FitB subunit and Tyr 86 , Ala 87 , and Ser 91 from ␣5 of the other FitB subunit (Fig. 3c). Arg 44 and Glu 80 form a salt bridge between the two helices that also serves to stabilize this extensive interface.
Together, the described interfaces (Figs. 1 and 3) create a stabile tetramer of FitAB heterodimers, which is in accord with the oligomerization state that was observed in previous solution studies (8). The biochemical significance of this unusual quaternary structure is underscored by the finding that FitA dimers bind the IR36 site with an affinity of ϳ180 nM, while the FitAB tetramer of heterodimers binds this site with a much improved affinity (K d , 4.5 nM (8)). The increase in stability provided by FitB to the tetrameric complex might explain part of this 40-fold increase in DNA binding affinity, even though FitB does not interact directly with the DNA molecule (Fig. 2b). However, an equal or more important contributor to the higher affinity displayed by the FitAB tetramer of heterodimers is the increase in the local concentration of FitA dimer that is brought about by the dimerization of the FitB proteins.
The IR36 FitAB binding site is found upstream of the fitAB operon in N. gonorrhoeae. The inverted repeat half-sites (Fig.  2c) were defined in biochemical studies as the specific bases required for FitAB binding to this region (8). In the FitAB-IR36 complex structure, the two FitA ␤-sheets bind on the same face of the IR36 DNA (Fig. 2b). However, this positioning is unnecessary for high affinity DNA binding, as inverted repeats with various spacer lengths between the half-sites, ranging from 14 to as little as 2 base pairs, bind FitAB with equally high affinity (8). A 4-residue flexible loop that connects FitA helices ␣2 and ␣3 would allow facile rotation of the tetramer complex and a The overall bend angle is the total angle of curvature between the first and last helical axis segments. This value was divided by the number of base steps in the sequence (length minus one) for comparative purposes. b The average value calculated for the global interbase pair twist (omega). c The terminal base pairs were deleted from IR36 prior to analysis, creating IR34.
The terminal bases are not tightly base paired, greatly changing the average values for the entire fragment when included.  thereby provides an explanation for the observed high affinity binding of FitAB to IR sites with different relative orientations.
The IR36 fragment is interesting and unusual because the two half-sites are separated by a long (14-base pair) spacer region of AT-rich DNA. Sequences like the central region of IR36, containing four or more consecutive A⅐T base pairs are known as A-tracts (37). These sequences adopt a structure different from that of typical B-form DNA in which A-tracts are essentially straight and rigid (38,39). In addition, these sequences deviate from B-form DNA by having a compressed minor groove and a shorter helical repeat of only 10 bp (37,40). The rigidity of A-tracts is predicted to allow for sharp bends at their edges (37), however there is no pronounced local bending in the structure of IR36 (Fig. 2, a and b and Table 2). Rather the DNA is smoothly curved such that the end-to-end bend angle is 44°. As expected, the central 14 base pairs are straighter, i.e. more rigid, than the rest of the DNA; there is a greater bend in the inverted repeat sequence (1.83°per base step) than in the A-tract region (1.08°per base step) ( Table 2). The rigidity of the A-tract region is also demonstrated by the thermal parameters of the structure; the average B factor for the entire DNA molecule is 51 Å 2 , while it is only 42 Å 2 for the isolated A tract ( Table  1). The average twist between base pairs is comparable between the A-tract and inverted repeat segments of the sequence with the inverted repeats having slightly less twist than the A-tracts ( Table 2). The central sequence shares another characteristic of canonical A-tract DNA, a significantly compressed minor groove, which is underscored when compared with the minor groove widths of the flanking sequences. Minor groove widths in the A-tract are 3.7 Å on average, while in the flanking sequences the width of the minor groove averages 7.1 Å ( Table  2). The functional significance of the IR36 A-tract is unknown. However, its complete solvent exposure would allow access to the replication and transcription machinery.
FitA-DNA Contacts-FitA makes few specific contacts to the DNA and all contacts to base pairs are mediated by its residues from the conserved ␤ sheet. As predicted, the highly conserved RHH residue Arg 7 is crucial for DNA recognition (Fig. 4). The guanidinium side chain of Arg 7 from each FitA subunit hydrogen bonds with the O6 and N7 atoms of the most 5Ј-guanine base of the inverted repeat sequence (Fig. 4b) as well as the thymine base on the 5Ј-side of this guanine. In subunits I and IV, this thymine base forms a water-mediated hydrogen bond to Asn 8 as well (Fig. 4). The only other specific contact is a van der Waals interaction between the side chain of Val 5 and the thymine from the T/A sequence central to each inverted repeat (Fig. 4a). In addition to these base contacts, a number of residues from the N-terminal end of FitA helix ␣2 contact the phosphate-sugar backbone of the DNA molecule. The side chains of Arg 33 and Ser 27 from all FitA subunits are involved in FIGURE 5. FitB is homologous to a protein with nuclease activity. a, superimposition of the structure of FitB (cyan) and PAE2754 from Pyrobaculum aerophilum (blue) (PDB accession 1V8P). The alignment is based on optimized superimposition of 93 C␣ atoms (FitB residues 1-11, 29 -51, 54 -71, 74 -80, 99 -126, 131-136), resulting in a final RMSD of 1.9 Å between the two structures. The two proteins share only 13% amino acid identity over these aligned regions. b, the conserved acidic residues that define the PIN domain cluster in a surface pocket at the C-terminal end of the central, 5-stranded ␤-sheet. FitB is in cyan and PAE2754 is in blue. c, in FitB, the putative active site for nuclease activity is blocked by the presence of Arg 68 from FitA. FitA is in magenta and FitB is in cyan. Residues are colored gray (carbon), red (oxygen), blue (nitrogen).
such interactions, as are the amide nitrogens of Thr 28 and Glu 29 (Fig. 4a).
FitB Contains a PIN Domain-FitB has a high degree of structural homology to the PIN domain containing protein PAE2754. 93 corresponding C␣ atoms of the two proteins overlay with an RMSD of 1.9 Å (Fig. 5a). PIN domains contain four highly conserved acidic residues that cluster at the C-terminal end of the ␤-sheet and form a negatively charged pocket near the center of the molecule (25,41). This acidic pocket of the PIN domain containing flap endonuclease-1 is the active site for its exonuclease activity (41) and that of PAE2754 has been proposed to carry out an exonuclease function as well (25). In FitB these conserved residues are Asp 5 , Glu 42 , Asp 104 , and Asp 122 and they cluster to form an acidic pocket just as those from PAE2754 (Fig. 5, b and c).
In an initial attempt to identify a nuclease activity for FitB, several in vitro assays were carried out using a variety of nucleic acid substrates, including single and double-stranded RNA and DNA and Flap structures. However no nucleic acid cleavage was detected in the context of the FitAB complex. Unlike some homologous PIN domain containing exonucleases, FitB has a dimerization partner, FitA. Intriguingly, an arginine residue, Arg 68 , at the C terminus of FitA is located in the FitB acidic pocket, potentially blocking access of potential substrates and thus, inhibiting any enzymatic function (Fig. 5c). The guanidinium group of Arg 68 interacts with the carboxyl groups of residues Asp 5 , Glu 42 , and Asp 104 from FitB, forming strong electrostatic interactions that would not be easily displaced by a competing nucleic acid substrate. These negatively charged residues form the Mg 2ϩ binding pocket in PAE2754 (29) and may similarly bind Mg 2ϩ in FitB when Arg 68 from FitA is not present. Indeed, the trypsinized FitcAB molecule, the structure of which was solved as a part of this work, bound two solvent molecules in this acidic pocket that are very likely magnesium ions ("Experimental Procedures" and data not shown). A FitB(Mg 2ϩ )-dependent nuclease might be activated in the cell when the FitA-FitB complex dissociates in response to some unknown signal, allowing Mg 2ϩ to bind in the place of Arg 68 from FitA. This dissociation event is proposed to release the FitA antitoxin and allow the induction of a bacteriostatic action by the FitB toxin, by analogy to other proposed toxin-antitoxin systems (42)(43)(44). A common feature of toxinantitoxin structures is that an otherwise unstructured peptide from the antitoxin blocks the activity of the toxin (35,36,45).
Preliminary experiments with FitcAB or FitB alone, which it should be noted is highly insoluble in vitro, have also failed to reveal any nuclease activity. However, at this point these studies have been limited to use of the IR36 sequence and there is a strong possibility that the FitB nuclease requires a specific DNA or RNA substrate that meets certain structural requirements. While the YoeB and RelE toxins have been shown to act as nucleases, their structure and active site residues are very different from those seen in FitB and thus they are not suitable as models for the putative FitB nuclease activity (35,45). The identity of the putative FitB substrate(s) is the subject of ongoing biochemical and biological studies.

CONCLUSION
In conclusion, FitA and FitB form a heterodimer in which FitA is the DNA binding subunit and FitB contains a nuclease activity that is blocked by the presence of FitA and activated by an as yet unidentified intracellular signal, which dissociates the FitAB complex. Four such FitAB heterodimers associate into a novel tetrameric structure that binds to the IR36 sequence from the fitAB promoter region with high affinity. Many PIN domain-containing proteins are involved in nucleic acid metabolism and/or remodeling, with the prokaryotic FitAB and its homologues responsible for controlling rates of DNA replication and/or plasmid maintenance (3,27,28). This structure illustrates the mechanism by which antitoxins with RHH motifs are able to block the activity of PIN domain toxins in prokaryotes (46). Future studies on the activity of FitB will help us understand both generally how PIN domains control such diverse processes as replication and nonsense-mediated mRNA decay and specifically the role of FitAB in GC virulence (3,11).