The structure and function of Mycobacterium tuberculosis MazF-mt6 toxin provide insights into conserved features of MazF endonucleases

Toxin-antitoxin systems are ubiquitous in prokaryotic and archaeal genomes and regulate growth in response to stress. Escherichia coli contains at least 36 putative toxin-antitoxin gene pairs, and some pathogens such as Mycobacterium tuberculosis have over 90 toxin-antitoxin operons. E. coli MazF cleaves free mRNA after encountering stress, and nine M. tuberculosis MazF family members cleave mRNA, tRNA, or rRNA. Moreover, M. tuberculosis MazF-mt6 cleaves 23S rRNA Helix 70 to inhibit protein synthesis. The overall tertiary folds of these MazFs are predicted to be similar, and therefore, it is unclear how they recognize structurally distinct RNAs. Here we report the 2.7-Å X-ray crystal structure of MazF-mt6. MazF-mt6 adopts a PemK-like fold but lacks an elongated β1-β2 linker, a region that typically acts as a gate to direct RNA or antitoxin binding. In the absence of an elongated β1-β2 linker, MazF-mt6 is unable to transition between open and closed states, suggesting that the regulation of RNA or antitoxin selection may be distinct from other canonical MazFs. Additionally, a shortened β1-β2 linker allows for the formation of a deep, solvent-accessible, active-site pocket, which may allow recognition of specific, structured RNAs like Helix 70. Structure-based mutagenesis and bacterial growth assays demonstrate that MazF-mt6 residues Asp-10, Arg-13, and Thr-36 are critical for RNase activity and likely catalyze the proton-relay mechanism for RNA cleavage. These results provide further critical insights into how MazF secondary structural elements adapt to recognize diverse RNA substrates.

Toxin-antitoxin complexes are two-component operons distributed widely among prokaryotic and archaeal genomes where they have diverse roles in bacterial physiology. These small gene pairs are critical for postsegregational killing, adaptive responses to environmental stress, antibiotic tolerance, and host-pathogen virulence responses (1,2). Toxin-antitoxins are organized into six different classes based on how the antitoxin component regulates toxin activity (3). In all classes, the toxin is proteinaceous and inhibits an essential cellular function such as replication, translation, or an assembly pathway including ribosome and membrane biogenesis in response to stress (2,4). The antitoxin is either RNA or protein that inhibits toxin activity by direct binding (Types II, III, and VI), preventing toxin protein expression (Types I and V), or competing with the toxin for substrate binding (Type IV) (3).
MazF-mt6 toxin cleaves 23S rRNA at Helix 70 (H70) 2 of the large 50S subunit at a consensus sequence of  UU2 CCU 1943 -3Ј (13). H70 is at the interface between the 30S and 50S subunits and adopts a 10-nucleotide (nt) stem-loop with the consensus sequence located in an internal bulged region. Here, we solved the 2.7-Å X-ray crystal structure of MtbMazF-mt6 and show that MazF-mt6 adopts a plasmid emergency maintenance K (PemK)-like fold, an architecture commonly adopted by several bacterial toxins including PemK and CcdB (15). MazF-mt6 contains a significantly shorter ␤1-␤2 linker, the region predicted to interact with RNA (13). We found that MazF-mt6 cleaves a 35-mer H70 RNA, strongly suggesting that the entire 23S rRNA is not required for MazF-mt6 activity. We further determined that MazF-mt6 residues Asp-10, Arg-13, and Thr-36 are critical for activity and likely perform mRNA cleavage via a proton relay mechanism. These data provide a structural and functional framework for how MazF endoribonucleases recognize diverse RNA substrates by the remodeling of loops to mediate RNA recognition.

MazF-mt6 cleaves both 23S rRNA and H70
MazF-mt6 inhibits protein synthesis by cleaving the 23S rRNA at H70 ( 1939 UU2CCU 1943 ), an RNA sequence conserved between M. tuberculosis and E. coli 70S (13). H70 forms interactions with the A-site tRNA and ribosome recycling factor during elongation and interacts directly with 16S rRNA decoding helix 44 ( Fig. 1A) (16,17). To test for activity, we incubated MazF-mt6 with purified E. coli 50S and monitored 23S rRNA cleavage product formation using reverse transcription (RT) from a complementary DNA oligo annealed to 23S rRNA nt 1959 -1945 (Fig. 1B). We used a poisoned RT reaction that yields full-length and cleaved products of 19 and 25 nt, respectively (18). The extension is halted by incorporating dideoxy-CTP at position 1935, 6 nt after the cleavage site. Using this experimental setup, we confirmed that MazF-mt6 primarily cleaves between residues 1940 and 1941 ( Fig. 1C and supplemental Fig. S1). We next tested whether MazF-mt6 required an intact 50S by incubating MazF-mt6 with a 35-nt RNA containing the H70 sequence (nt 1933-1967) (Fig. 1B). The major RT product is 11 nt, corresponding to cleavage between 1940 and 1941 (64% of the product formed), as demonstrated previously (13). However, there are additional minor RT products corresponding to cleavage after C1942 and U1943 in the same bulged loop (8 and 23%, respectively; Fig. 1C). Although MazF-mt6 cleaves H70 almost to completion, MazF-mt6 only cleaves 23S rRNA in the 50S to ϳ30% completion.

Structure determination of MazF-mt6
MazF-mt6 was overexpressed and purified fused to an N-terminal hexahistidine (His 6 )-trigger factor (TF) as described previously (13). In addition to immobilized metal affinity chromatography purification, the His 6 -TF-MazF-mt6 complex was incubated with thrombin protease to release TF. MazF-mt6 was further purified by immobilized metal affinity chromatography and size exclusion chromatography (Superdex S200, GE Healthcare) and concentrated to 10 mg/ml. Crystals formed in the P6 4 22 space group with two identical, symmetry-related MazF monomers in the asymmetric unit. The X-ray crystal structure was solved to 2.7-Å resolution by molecular replacement using B. subtilis MazF (BsMazF) as a search model in the Phaser program of the PHENIX suite (Table 1 and Fig. 2; Protein Data Bank (PDB) code 4MDX) (19,20). The MazF-mt6 model was manually built in the Crystallographic Object-Oriented Toolkit program (Coot), and geometry and B-factor refinements in PHENIX and rebuilding in Coot were iteratively performed (20,21). The final model contains residues 2-99 (of 103 total residues).

Structure of M. tuberculosis MazF-mt6 toxin
ϳ2,000 Å 2 of hydrophobic and electrostatic interactions between ␤6 and ␣4 in comparison with other MazFs that have an additional 500 Å 2 buried.
Sequence alignments of MazF-mt6 with BsMazF and EcMazF (referred to as MazF hereafter) reveal RNA-binding regions that differ in length (19). Comparison of the structure of MazF-mt6 with MazFs that interact with mRNA reveals that MazF-mt6 contains a significantly shorter ␤1-␤2 linker, the region that interacts with mRNA (19,29) (Fig. 2B). The MazF ␤1-␤2 linker is 13 residues in length, whereas in MazF-mt6, ␤1-␤2 linker is only five residues. The MazF ␤1-␤2 linker extends ϳ10 Å from one monomer to the other monomer, forming part of the dimer interface that packs against and covers a net positive charge where mRNA binds (19). Because there is a shortened MazF-mt6 ␤1-␤2 linker, this region does not extend toward the dimer interface, leaving the positively charged cleft exposed. An additional positive region is exposed upon MazE antitoxin binding (19,29). This second positively charged region is missing in MazF-mt6, and instead a negative solvent-exposed patch is present.

MazF-mt6 residues Asp-10, Arg-13, and Thr-36 are important for activity and inhibition of bacterial growth
The MazF-mt6 structure reveals that residues Asp-10, Lys-11, Arg-13, Thr-36, Thr-37, and Thr-38 are adjacent to the putative active site, which also contains a bound sulfate ion ( Fig.  3A and supplemental Fig. S3). The negatively charged sulfate ion can mimic the phosphate backbone of RNA and provides further support that this location is likely the active site. To identify which MazF-mt6 residues are important for toxin activity, we used an established cell-based assay whereby overexpression of wild-type (WT) MazF-mt6 inhibits growth (11,30). MazF-mt6 variants that restore growth are interpreted as disrupting canonical MazF-mt6 function. As demonstrated previously, overexpression of WT MazF-mt6 inhibits cell growth in E. coli, whereas uninduced or vector control only exhibits normal growth (supplemental Fig. S4) (19). MazF residues previously identified as important for activity include an arginine that functions as both an acid and a base and either a serine or threonine that stabilizes the transition state during mRNA cleavage (19,29,31). MazF-mt6 residue Arg-13 is located on the shortened ␤1-␤2 linker in a similar location as catalytic residues EcMazF Arg-29 and BsMazF Arg-25. Expression of MazF-mt6 R13A variant allows for normal growth, demonstrating that this residue is important for activity (Fig. 3B). Proximal residues to the MazF-mt6 active site include Asp-10 and Lys-11. Expression of the MazF-mt6 D10A and K11A variants allow for growth, suggesting that these residues have a critical role in activity (Fig. 3B). MazF-mt6 contains three threonine residues adjacent to its active site, suggesting that each could potentially stabilize the transition state. We found that the T38A variant has no effect on MazF-mt6 inhibition of growth, whereas the T37A variant has a modest growth effect (Fig. 3C). In contrast, the T36A variant causes a reversal of growth inhibition, suggesting that Thr-36 plays a critical role in MazF-mt6 activity.

MazF-mt6 residues Asp-10, Arg-13, and Thr-36 have the largest impact on rRNA degradation
Bacterial growth assays suggested that MazF-mt6 residues Asp-10, Arg-13, Thr-36, and possibly Lys-11 are important for endoribonuclease activity (Fig. 3). To assess the quantitative  impact each MazF-mt6 residue has on mRNA cleavage, we performed single turnover assays where 5Ј-32 P-labeled H70 RNA was incubated with a 10-fold molar excess of MazF-mt6 over a time course of 45 min ( Fig. 4 and supplemental Fig. S5). The cleavage of H70 by WT MazF-mt6 was monitored by the accumulation of RNA products corresponding to 8 -11 nt. The major product for WT MazF-mt6 cleavage was an RNA product corresponding to cleavage between U1940 and C1941 (3.7 ϫ 10 Ϫ3 s Ϫ1 ) ( Table 2). MazF-mt6 variants R13A and T36A had the most significant impact on activity with minimal RNA cleavage products detected (Fig. 4A). In contrast, MazF-mt6 variants K11A, T37A, and T38A had minor effects on RNA cleavage as compared with WT, suggesting that although these residues are proximal to the active site they do not contribute substantially to catalysis. Although the K11A variant appeared to have a large effect on activity in the bacterial growth assays (Fig. 3B), the variant has a minimal effect on activity in single turnover experiments. These data suggest that the K11A variant either has impaired RNA binding or lower protein solubility, which could reduce its expression in the cell-based assays. MazF-mt6 residue Asp-10 is adjacent to critical residue Arg-13. The MazF-mt6 D10A variant decreased the rate of RNA cleavage by 10-fold (4 ϫ 10 Ϫ4 s Ϫ1 ), suggesting that this residue is important for catalysis (Fig. 4A). In addition to the canonical 8-nt product produced by all MazF-mt6 variants, RNA products of 9 -11 nt were also seen but at varying extents depending on the variant (Fig. 4A). For example, MazF-mt6 variants D10A, R13A, and T36A that had large impacts on RNA degradation produced additional RNA products in similar amounts as compared with the canonical 8-nt product. In contrast, MazF-mt6 variants T37A and T38A that had a modest impact on overall activity produced products at the canonical position to the largest extent (40 and 60%, respectively, of overall product production), and 9 -11-nt RNAs were only seen at low levels (2-13%). The MazF-mt6 K11A variant impacts cleavage to a minimal extent ( Table 2; 3-fold); however, the major RNA product was 11 nt, corresponding to cleavage between U1943 and U1944. Thus, in addition to these MazF variants altering activity at the major cleavage site, they also seem to impact specificity. Another possibility is that in the absence of the 50S subunit the MazF-mt6 dimer may be more flexible in how it engages the bulged loop containing the consensus sequence, thus resulting in nucleotide promiscuity.

Discussion
Bacterial toxins regulate growth in response to environmental stress including antibiotic treatment (32). In E. coli, there is a single MazF toxin family member that cleaves free mRNA to inhibit translation (31). The MazF family is expanded to nine members in M. tuberculosis concurrent with the expansion of different target RNAs including tRNAs and rRNAs (12)(13)(14). This expansion suggests that M. tuberculosis MazFs may con-    (19). MazF ␤1-␤2 linkers typically contain 13 residues that surround the single-stranded mRNA substrate. The MazF-mt6 ␤1-␤2 linker is truncated to five residues, which we predict would alter how it interacts with H70. H70 adopts a stem-loop structure with an internal bulged loop region where the MazF-mt6 cleavage site is located. Therefore, the length of the ␤1-␤2 linker may be a consequence of the rigidity of the RNA it encounters. For example, MazFs that recognize flexible RNAs such as single-stranded mRNAs may require the loop to confine and orient the substrate. The H70 RNA likely requires less remodeling by the ␤1-␤2 linker, and therefore alteration of this MazF-mt6 region may allow for different MazF homologues to distinguish among diverse RNAs.
In addition to the ␤1-␤2 linker interacting directly with RNA substrates, this region also modulates an open to a closed MazF state, effectively functioning as a gate to allow either RNA or its cognate antitoxin MazE to bind (29). In the context of a MazF dimer, which forms the closed state, the ␤1-␤2 linker extends across the face of the monomer to interact with the adjacent MazF monomer (19). These interactions are mediated by sym-metry-related electrostatic contacts between the distal end of the ␤1-␤2 linker (Glu-24) and ␤6/␣3 of the other monomer (Lys-79 and Arg-86) (Fig. 5A). This ␤1-␤2 linker extension additionally establishes a positively charged RNA-binding pocket and obscures an adjacent and slightly overlapping positively charged region where the antitoxin MazE binds (29). Therefore, this closed state appears to direct mRNA binding while preventing MazE binding. Antitoxin MazE binding induces an open state by displacing the ␤1-␤2 linker and thus disrupting interactions between the MazF monomers, causing a reorganization of MazF active-site residues (29). ␤1-␤2 linker residue Glu-24 is critical for MazF activity (29), suggesting that maintenance of interactions between the monomers is important for MazF homologues that recognize mRNA.
The structure of MazF-mt6 reveals a truncated ␤1-␤2 linker, suggesting that it likely does not undergo conformational changes to convert from a closed to an open state as seen in other MazF homologues. The shortening of ␤1-␤2 allows for a permanently accessible, positively charged channel where RNA binds, whereas at the dimer interface, a negatively charged rather than a positively charged surface forms (Fig. 5B). A C-terminal MazE ␣-helix typically binds at the positively charged MazF dimer interface (19). However, MazF-mt6 lacks this positively charged surface (Fig. 5B). This suggests that MazE-mt6 antitoxin binding required to neutralize MazF-mt6 may be distinct from other MazEs solved to date. Homology modeling of MazE-mt6 predicts a C-terminal ␣-helix similar to other structures of MazE and ParD (19, 29, 33, 34). Two posi- tively charged residues align on one side of the ␣-helix, which could bind the negatively charged dimer interface (Fig. 5C). As mentioned, MazE binding disrupts the MazF active site and in particular the catalytic arginine (29). Modeling of the MazE-mt6 C-terminal ␣-helix bound to MazF-mt6 indicates that MazE-mt6 may be able to interact with catalytic Arg-13, suggesting that active-site disruption could occur.
MazF endoribonucleases cleave RNA substrates via a proton-charge relay mechanism whereby an arginine functions as both a general acid and base and a threonine or serine stabilizes the transition state (29,35) (Fig. 6A). In the absence of MazE and therefore containing a longer ␤1-␤2 linker in a closed state, the catalytic arginine of MazF forms a hydrogen bond with the backbone (Gly-22), which is located at the distal end of the ␤1-␤2 linker (Fig. 6A), a region missing in MazF-mt6. Instead, the shorter MazF-mt6 ␤1-␤2 linker forms an electrostatic interaction with Arg-13 using the negatively charged carboxyl group of Asp-10 (Fig. 6A). In MazF, a threonine is located adjacent to the scissile phosphate and is predicted to stabilize the transition state, and in MazF-mt6, Thr-36 is in an equivalent location. We show that both Asp-10 and Thr-36 are critical for MazF-mt6 activity by either stabilizing the catalytic Arg-13 or the RNA, respectively. Other nearby MazF-mt6 residues including Lys-11, Thr-37, and to a lesser extent Thr-38 potentially contribute to RNA substrate stabilization by the formation of electrostatic interactions with the phosphate backbone or ribose as these residues are close to Arg-13 and Thr-36 (Fig.  2B). Although Thr-38 is adjacent to the active site, our data suggest that this residue has no significant role in RNA catalysis. Therefore, despite the significantly shorter ␤1-␤2 linker, MazF-mt6 active-site residues form functionally equivalent interactions seen in other MazF orthologues.
MazFs cleave single-stranded mRNA at a consensus sequence of 5Ј-ACA-3Ј with BsMazF requiring at least one uridine flanking the 5Ј-and 3Ј-ends of the consensus (27,31). MazF-mt6 differs by requiring an additional uridine upstream of the consensus 5Ј-UU2CCU-3Ј sequence of the cleavage site (13). Structurally aligning the MazF-RNA complex to MazF-mt6 suggests several potential residues that could contribute to H70 RNA specificity. For upstream specificity, MazF-mt6 Arg-75 (located on ␤7) is located in a similar position as BsMazF Phe-10 and Asp-91, which recognize the Ϫ2 uridine position (Fig. 6B). A conserved BsMazF serine (Ser-73) is located in a pocket formed by the ␤4-␤5 linker and interacts with the Watson-Crick face of the Ϫ1 uridine. MazF-mt6 Ser-58 likely fulfils an analogous role (Fig. 6B). In previously solved structures of MazF, the ␤1-␤2, ␤3-␤4, and ␤5-␤6 linker residues recognize the downstream nucleotides ACAU (19). Because MazF-mt6 has a significantly shorter ␤1-␤2 linker, the ␤5-␤6 and ␤3-␤4 regions are likely the only linkers responsible for base-specific recognition. The ϩ1 position is recognized by the elongated ␤1-␤2 linker, but in the case of MazF-mt6, the ϩ1 cytidine may be recognized by ␤5-␤6 linker residue Gln-66 because it is located in an equivalent position (Fig. 6B). MazF-mt6 residues Asp-63 and Asn-64 align with the MazF residues Glu-78 and Gln-79, suggesting that these residues play similar roles in base-specific nucleotide interactions of the ϩ2 and ϩ3 regions. Lastly, by containing a truncated ␤1-␤2 linker, the RNA-binding path is deeper, potentially allowing a structured RNA like the bulged loop region containing the H70 consensus sequence to better fit on the surface of MazF-mt6.
MazF homologues are part of the PemK-like endonuclease family that cleave single-stranded mRNA transcripts (27). The one exception is the E. coli F-plasmid CcdB toxin that inhibits
MazF-mt6 variants were constructed using site-directed mutagenesis, and sequences were verified by DNA sequencing (oligos listed in supplemental Table S1). MazF-mt6 variant purifications were performed similarly to wild type. Each variant eluted as a single, symmetrical peak at the expected volume during the gel-filtration step as wild type, suggesting that the variants are properly folded.

MazF-mt6 crystallization, data collection, and structure determination
MazF-mt6 (10 mg/ml) crystallized by hanging-drop vapor diffusion in 50 mM sodium cacodylate, pH 6.0, 10 mM MgCl 2 , and 1.05 M LiSO 4 at 20°C in 3 weeks to dimensions of ϳ100 m 3 . The crystals were cryoprotected in 50% (w/v) lithium sulfate, 30% glycerol, and 150 mM ammonium tartrate for ϳ60 s and flash frozen in liquid nitrogen. Native X-ray diffraction data sets were collected at the Northeastern Collaborative Access Team 24-IDC beamline at the Advanced Photon Source. The data sets were collected under cryogenic conditions (100 K) using 0.979-Å radiation. The data were integrated and scaled to 2.9 Å using the X-ray Detector Software (XDS) package (39). The BsMazF-mRNA complex (PDB code 4MDX) was used as a search model (with the mRNA removed), and all MazF residues were changed to alanine during molecular replacement using AutoMR from the PHENIX software suite (20). Multiple rounds of positional, B-factor, and energy minimization refinement in PHENIX were followed by manual model rebuilding in Coot performed with R work and R free converging to 23.9/27.1% (21). The final MazF-mt6 model consisted of residues 1-99 of 103 total. The model quality was determined using MolProbity of the PHENIX validation suite, and figures were generated using PyMOL (40).
The reverse transcription primer (Integrated DNA Technologies; 5Ј-GTCGGAACTTACCCGAC-3Ј) was 5Ј-32 P-labeled using T4 polynucleotide kinase (New England Biolabs), and 4 pmol of the primer was annealed to 4 (50S) or 7 (H70) pmols of RNA in 50 mM Tris-Cl, pH 8.0, and 50 mM NaCl to 90°C or 65°C, respectively. These reactions were slowly cooled to 42°C by removing the heat block and placing on the benchtop at room temperature. The reverse transcription reactions were initiated with the addition of 1 M dATP, 1 M dGTP, and 1 M dTTP, limiting amounts of 0.1 M ddCTP and Protoscript II reverse transcriptase (New England Biolabs) as described previously (18). Reactions were incubated at 42°C for 1 h, then halted with the addition of 1 volume of formamide dye (100% (v/v) deionized formamide, 0.01% (w/v) bromphenol blue, and 0.01% (w/v) xylene cyanol), and heated to 90°C for 2 min. Dideoxy sequencing reactions were performed as described previously (18). Samples were run on a preheated 50% urea and 20% polyacrylamide gel at 37 mA (limiting) for 90 min. The gel was fixed in 20% (v/v) methanol, 20% (v/v) acetic acid, and 3% glycerol (v/v) for 60 min; dried for 1 h, and exposed to a phosphor screen for 1 hr. Exposure was visualized using a Typhoon FLA 7000 gel imager (GE Healthcare), with the band density determined by ImageQuant (GE Healthcare).

RNA cleavage assays
H70 was 5Ј-32 P-labeled as described above and incubated (0.2 M RNA) with 2 M wild-type MazF-mt6 or MazF-mt6 variants for 45 min at 37°C. Aliquots of the reactions were halted at 1-, 3-, 5-, 10-, 25-, and 45-min time points with the addition of 1 volume of formamide gel loading buffer and heated to 70°C for 2 min. Samples were run on a 50% urea and 20% polyacrylamide gel at 37 mA (limiting) for 2 h. The gel was fixed in 40% ethanol, 20% acetic acid, and 3% glycerol for 1 h; dried for 90 min; then exposed to a GE Healthcare phosphor screen for 30 min. Exposure was visualized using a Typhoon FLA 7000 gel imager (GE Healthcare), with the band density determined by ImageQuant (GE Healthcare). Background density was subtracted using default settings. The amount of cleavage product was determined by determining the amount of full-length RNA in each lane. The amount of full-length RNA was divided by the total lane density for each sample, and the reciprocal of the resulting reaction was recorded as the amount of RNA cleaved. The amount of cleaved product against time were fit by GraphPad Prism 5 using the following single-exponential equation.
Product ϭ P max ͑1 Ϫ e Ϫkt ͒ (Eq. 1) where P max is the RNA cleavage product plateau (pmol of mRNA cleaved), k is the observed rate constant (min Ϫ1 ), and t is the time the reaction progressed (min). The rates of both MazF-mt6 R13A and T36A were too low to fit the data points to the single-turnover equation with an R 2 Ͼ80%.