Novel polyadenylylation-dependent neutralization mechanism of the HEPN/MNT toxin/antitoxin system

Abstract The two-gene module HEPN/MNT is predicted to be the most abundant toxin/antitoxin (TA) system in prokaryotes. However, its physiological function and neutralization mechanism remains obscure. Here, we discovered that the MntA antitoxin (MNT-domain protein) acts as an adenylyltransferase and chemically modifies the HepT toxin (HEPN-domain protein) to block its toxicity as an RNase. Biochemical and structural studies revealed that MntA mediates the transfer of three AMPs to a tyrosine residue next to the RNase domain of HepT in Shewanella oneidensis. Furthermore, in vitro enzymatic assays showed that the three AMPs are transferred to HepT by MntA consecutively with ATP serving as the substrate, and this polyadenylylation is crucial for reducing HepT toxicity. Additionally, the GSX10DXD motif, which is conserved among MntA proteins, is the key active motif for polyadenylylating and neutralizing HepT. Thus, HepT/MntA represents a new type of TA system, and the polyadenylylation-dependent TA neutralization mechanism is prevalent in bacteria and archaea.


INTRODUCTION
Toxin-antitoxin (TA) systems are small genetic elements found on plasmids and in the chromosomes of many bac-teria and archaea. They serve to either reduce metabolism during stress (1,2) or to inhibit phage infection (3)(4)(5). Currently, at least six different types of TA systems have been identified based on the nature of antitoxins (protein or RNA) as well as the way an antitoxin interacts with a toxin to block its toxicity. In type I and type III TA systems, the antitoxin is RNA, and the antitoxin blocks the toxicity of the toxin either by RNA-RNA interactions or RNAprotein interactions. For the rest of the TA systems, the antitoxins are proteins, and the way they block the toxicity of the cognate toxin varies. In type II systems, the antitoxin binds to the toxin, and the neutralization of the toxin depends on direct protein-protein interactions with the toxin. In type IV systems, the antitoxin and the toxin do not interact but act on the same target (6,7). In type V systems, the antitoxin specifically cleaves the toxin mRNA to inhibit toxin production (8). In type VI systems, when the TA complex is formed, the antitoxin facilitates the degradation of the toxin (9). Recently, several novel TA systems have been reported in which the antitoxin functions as an enzyme: Hha/TomB (antitoxin oxidizes Cys of the toxin) (10), ToxSAS/antiToxSAS (antitoxin degrades the product of the toxin, (p)ppGpp) (11), and TglT/TakA (antitoxin phosphorylates Ser of the toxin) (12), expanding the ways antitoxins neutralize their toxins.
The two-gene module encoding the HEPN (higher eukaryotes and prokaryotes nucleotide-binding) domain and cognate MNT (minimal nucleotidyltransferase) do-main has been predicted to represent one of the most abundant TA systems in archaea and bacteria (13). The MNT domain-containing protein appeared to be the only active enzyme in the putative TA system, so it was predicted to be the toxin (14). Unexpectedly, a genome-wide screen for TA modules using shotgun cloning led to the characterization of the HEPN domain-containing protein HhalT as the toxin and the neighboring MNT domaincontaining protein HhalA as the antitoxin in the halophilic bacterium Halorhodospira halophila SL1 (15). Our subsequent work demonstrated that the HEPN protein with a RX [4][5][6] H motif is the toxin which cleaves mRNA in vitro in the HEPN/MNT (SO 3166/SO 3165) TA pair in the model psychrotrophic bacterium Shewanella oneidensis (16). Furthermore, antitoxin SO 3165 binds to toxin SO 3166 at a rare 2:6 ratio to form a hetero-octamer, and this binding is important for the antitoxin to block the toxicity of the toxin (17). These results demonstrate that the HEPN/MNT module constitutes a bona fide TA system and resembles a type II TA system as both components are proteins that interact with each other. Noticeably, unlike other typical type II antitoxins, MNTs adopt a minimal domain of the pol␤ nucleotidyltransferase superfamily instead of a DNA-binding domain (18). However, these MNTs seem to contain only the catalytic domain but lack the region for substrate recognition usually present in known nucleotidyltransferases due to the size limitations (∼80-150 aa) (19). Thus, whether these antitoxins retain nucleotidyltransferase activity is unclear.
Here, by studying the representative HEPN/MNT TA pairs in Proteobacteria and Euryarchaeota, we discovered that the MNT antitoxin adenylylates the cognate HEPN toxin to inactive HEPN. A distinctive feature of the HepT (HEPN toxin)/MntA (MNT antitoxin) TA pair is a conserved RX 4 HXY motif for HepT and a conserved GSX 10 DXD motif for MntA. Specifically, structural and biochemical studies of the HepT/MntA TA pair (SO 3166/SO 3165) in S. oneidensis show that the GSX 10 DXD motif of antitoxin MntA is responsible for transferring three AMPs to HepT and that Y104 in RX 4 HXY in HepT is the site that receives three AMPs from MntA. Amino acid substitution experiments demonstrate that mutating the signature motif in MntA not only abrogates the adenylyltransferase activity of MntA but also reduces the ability of MntA to neutralize HepT toxicity. in vitro enzymatic assays showed that the three AMPs are transferred to HepT by MntA consecutively using ATP as the substrate. Structural analysis of the HepT/MntA TA complex containing modified or un-modified HepT indicates that the three AMPs are located near the active RNase domain of HepT and can interfere with the substrate binding sites. Critically, the HepT/MntA TA pair with signature motifs represents a new type of TA system that is prevalent in both bacteria and archaea. Likewise, we show the HepT toxin homolog in the hyperthermophilic archaeon Thermococcus cleftensis is also adenylylated by the cognate MntA homolog. The prevalence of these HEPN/MNT modules in bacteria and archaea suggest that these TA systems may be involved in the environmental adaption to extreme habitats.

Bacterial strains, plasmids and growth conditions
The strains and plasmids used in this study are listed in Supplementary Table S1. Experiments were conducted at 30 • C for S. oneidensis and at 37 • C for Escherichia coli in Luria-Bertani (LB) medium. 2,6-diamino-pimelic acid was added to E. coli WM3064 at a final concentration of 0.3 mM. Chloramphenicol (30 g/ml) and kanamycin (50 g/ml) were added to cells harboring plasmids with the indicated resistance genes listed in Supplementary Table S1. When required, 0.1-1 mM IPTG (isopropyl-␤-D-thiogalactopyranoside) was added as an inducer.

Mutagenesis
The mutations in hepT or mntA were introduced by using a modified two-step PCR method via two inside primers and two outside primers (20). PCR products using one of the inside primers containing the mutated sites was combined with one of the outside primers covering the whole gene or operon to generate a template for the second round of PCR, and the pair of outside primers were used for the secondround PCR to fuse the two PCR fragments from the firstround PCR (primers listed in Supplementary Table S2). The two insider primers both have the mutations and they are of opposite directions. All the mutants were confirmed by DNA sequencing.

Protein purification and western blot assays
Escherichia coli BL21 (DE3) strains carrying pET28b-based plasmids were grown in LB supplemented with kanamycin and induced with 0.2 mM IPTG for 12 h at 25 • C. Specifically, in order to purify toxin HepT and HepT/MntA variants that are toxic, stationary cells (OD 600 ∼ 1.0) were induced with 0.5 mM IPTG for only 2 h at 37 • C (longer induction periods and IPTG addition at inoculation resulted in too much toxicity). Cells expressing the protein of interest were collected, and the His-tagged proteins were purified following the method described in a previous study (21). For the western blot assay, protein samples were transferred to a PVDF membrane (Millipore, Bedford, MA, USA), and performed with primary antibodies raised against a His-tag (Cell Signaling Technology, Danvers, MA, USA) and horseradish peroxidase-conjugated goat anti-mouse secondary antibodies (Bio-Rad, Richmond, CA, USA).

Phosphodiesterase (PDE) cleavage assay
To conduct the PDE cleavage assay, the HepT/MntA TA complex and mutants (HepT Y104A /MntA, HepT/MntA G27A, S28T ) were purified as described above. The reaction mixture for the cleavage assay contained 20 mM Tris-HCl (pH 8.0) and 10 g of purified protein and was incubated with PDE (2 U l −1 ) at 37 • C for 5-40 min. The small molecule products of these reaction mixes were collected using a 3K membrane for further identification.

HPLC/Q-TOF-MS analysis
Identification of the molecular mass was carried out by MS detection using a quadrupole time-of-flight instrument coupled to an HPLC (Agilent Technologies). The elution gradient employed for the separation of proteins in each fraction was as follows

Crystallization and data collection
The purified HepT and MntA proteins were concentrated in 0.5 ml and loaded onto a Superdex 200 column (GE Healthcare) equilibrated with 20 mM Tris (pH 8.0) and 150 mM NaCl. Fractions containing the target proteins were concentrated, and crystallization was performed using the sitting drop vapor diffusion method at 16 • C. Crystallization drops containing 0.5 l of the protein solution were mixed with 0.5 l of reservoir solution. The protein concentration of HepT/MntA used for crystallization was ∼20 mg/ml. Diffraction quality crystals of the HepT/MntA complex were grown in the presence of 0.1 M Tris-HCl (pH 7.8-8.2), 0.2 M potassium sodium tartrate, and 12-20% (v/v) PEG 3350. Crystals were harvested with 20% (v/v) glycol as a cryoprotectant before flash freezing them in liquid nitrogen. HepT/MntA D39E, D41E and HepT Y104A /MntA were purified as the same procedure and the crystals were obtained in the 0.1 M HEPES pH 7.5 0.15 M, 0.2 M lithium sulfate monohydrate and 20% PEG3350. To obtain the complex of HepT Y104A /MntA bound with the substrate, AMP-PNP was used for co-crystallization with HepT Y104A /MntA at a 10:1 molar ratio at 4 • C for 30 min, crystals were grown using the same conditions. X-ray diffraction data were collected on BL-17U1 at Shanghai Synchrotron Radiation Facility (SSRF). The diffraction images were processed with the HKL2000 program (22).

Structure determination and refinement
The crystal structures of HepT/MntA, HepT/MntA D39E,D41E , HepT Y104A /MntA and HepT Y104A /MntA-AMP-PNP were solved via the molecular replacement (MR) method with Phaser using HepT/MntA (PDB code 5YEP) as the search template. Model building and structure refinement were performed manually in Coot (23) and automatically in PHENIX (24). The stereochemical analysis of the final models was performed by MOLPROBITY (25). Data collection and refinement statistics are presented in Table 1.

In vitro enzymatic assays
The in vitro adenylylation assay using purified unmodified HepT and MntA was modified based on the in vitro assay of Truttmann et al. (26). The unmodified HepT and MntA proteins are very similar in size (133 aa and 139 aa, respectively), making it difficult to distinguish the reaction products. Therefore, 24 extra aa (including the 6× His tag) were fused to the N terminus of MntA, a strategy we previously used that has potentially minimal effect on the activity of MntA (27). Purified MntA (10 g) was added to 40 l of reaction buffer (containing 20 mM Tris-HCl, 10 mM MgCl 2 and 5 mM DTT) supplemented with 0 to 12 M ATP and incubated at room temperature for 30 min. Unmodified HepT (20 g) was then added to the reaction and incubated at 37 • C for another 60 min. The reaction was stopped by adding SDS-PAGE loading buffer and the samples were analyzed by tricine-SDS-PAGE. To capture the intermediate products, a reaction with 3 M ATP serving as the substrate was stopped by adding EDTA, and the products were analyzed by HPLC/Q-TOF-MS as described above.
For the in vitro RNase cleavage assay, ompA mRNA was synthetized, and the assay was performed as previously described using various of concentrations of HepT (0-160 nM) (17). The reaction mixture was incubated at 37 • C for 30 min. To obtain the polyadenylylated HepT protein, the lysate from cells coexpressing hepT and mntA was shaken vigorously after incubating with Ni-NTA agarose, and polyadenylylated HepT with purity >90%, was obtained after a second elution (Supplementary Figure S1).

The bacterial two-hybrid (BACTH) assay
For the BACTH assays, the coding region of hepT was inserted into pKT25, and the coding region of mntA was inserted into pUT18C. The recombinant plasmids were cotransformed into E. coli BTH101(cya-99) competent cells and the quantitative analyses of the BACTH assay was performed as previously described (21).

Analysis of the distribution of HepT/MntA
HepT belongs to the Pfam family PF01934 (protein of unknown function with DUF86 belonging to clan CL0291) and MntA belongs to PF01909 (nucleotidyltransferase domain belonging to clan CL0260). Therefore, proteins belonging to CL0291 or CL0260 in 86 934 genomes (Supplementary Table S3) were retrieved using the function profile tool in the IMG/M system (Pfam v30) (28). Next, FIMO (29) was used to scan for proteins belonging to CL0291 that contain the RX 4 HXY motif or CL0260 that contain GSX 10 DXD. The initial hits were further curated by checking whether these two motifs were encoded by neighboring genes. This generated a total of 15 977 pairs, belonging to PF01934/PF01909 or PF08780/PF01909. This set was defined as Class I HEPN/MNT TA pairs. The Class II HEPN/MNT TA pairs contain the HEPN toxins from PF05168 but they do not contain the RX 4 HXY motif. In order to extensively evaluate the distribution of the HEPN/MNT pairs, both Class I and Class II proteins were recovered from the IMG/M database by the function profile tool (Supplementary Table S4). The motif of the each of two classes was generated by MEME (30).

Antitoxin MntA adenylylates toxin HepT in S. oneidensis
Previously, we showed that SO 3165 (here, we suggest the name MntA) and SO 3166 (we suggest the name HepT) of the HEPN/MNT module in S. oneidensis, form a TA pair (16) ( Figure 1A). However, during purification of the toxin protein HepT, we found that the mobility of HepT varied with and without MntA during SDS-PAGE fractionation. The vector pET28b-mntA-hepT-His 6 was constructed to co-express a C-terminal His 6 -tagged HepT and untagged MntA, while pET28b-hepT-His 6 was constructed to only express His 6 -tagged HepT. To exclude any influence of the resident HEPN/MNT module on our experiments, we typically used E. coli strains that did not have endogenous HEPN/MNT modules. Noticeably, HepT migrated slower when co-produced with MntA compared to when it was produced alone ( Figure 1B). The molecular mass of the unmodified His 6 -tagged HepT was 16.14 kDa ( Figure 1B and Supplementary Table S5), and the size of the unmodified and untagged MntA, which was pulled down due to its interaction with HepT, was 15.62 kDa ( Figure 1B, lane 3).
The presence of MntA led to the production of HepT with a molecular mass shift of ∼1 kDa higher than expected, suggesting that HepT might be modified in the presence of MntA. Thus, we also reasoned that this modification may involve the addition of a more complex chemical group than phosphate or acetyl, which are usually not detected by SDS-PAGE analysis.
MntA is a small protein of 139 aa and displays a minimal nucleotidyltransferase domain previously found by Aravind and Koonin (31). Since nucleotidyltransferase catalyzes the transfer of a nucleotide to an acceptor hydroxyl group, we hypothesized that MntA might transfer nucleotides to the side chains of HepT with hydroxyl groups. When a nucleotidyltransferase acts on its amino acid substrate, a phosphodiester bond is formed that is susceptible to cleavage by PDEs. To test this hypothesis, we purified the S. oneidensis HepT/MntA TA complex and treated the purified TA complex with snake venom PDE. As expected, HepT shifted back to the expected and unmodified size when treated with PDE ( Figure 1C), suggesting that PDE cleaved the modified moiety from HepT. Next, we sought to identify the modified nucleotide in the cleaved products. HPLC/MS analysis of the cleaved products revealed that the major product had an MW of 347.6 Da with a retention time of 5.7 min, which was the same as the control adenosine 5 monophosphate (5 -AMP) ( Figure 1D and Supplementary Figure S2). Together, these results show that toxin HepT is adenylylated by antitoxin MntA.

Key motif for adenylylation in HepT/MntA TA pair
Since many proteins with nucleotidyltransferase domains are adjacent to proteins with HEPN domains, we investigated the conserved motif of MntA along with a diverse set of other HEPN/MNT-associated proteins. MntA and HepT belong to the PF01909 and PF01934 families (Pfam models: NTP transf 2 and DUF86), respectively. Sequence analysis revealed that these MNT proteins included in the PF01909/PF01934 pairs have a highly conserved  Figure 2B). These results demonstrate that this motif is critical for the modification of HepT.
Next, we identified the adenylylated site in HepT. Adenylylation involves the formation of a phosphodiester bond between the hydroxyl group of the side chain of the amino acid and the phosphate group of AMP. The covalent addition of an AMP moiety to Ser, Thr or Tyr has been previously reported (32). More importantly, we also noticed that the HEPNs neighboring the MNTs in the PFAM database have a signature motif of RX 4 HXY (where X is any amino acid) (Figure 2A). In addition to the reported active RNase RX 4 H motif of HEPNs, we found that Y104 in the RX 4 HXY motif is highly conserved, which has been also previously reported (13). To further check whether the Y104 residue in HepT is adenylylated in vivo, we performed site-directed mutagenesis on hepT of S. oneidensis at the codon for Y104 to yield the substitution Y104A. We found that the mobility of the substituted HepT toxin (HepT Y104A ) produced by pET28b-mntA-hepT Y104A -His 6 was the same as that produced by pET28b-hepT-His 6 during SDS-PAGE fractionation and western blot analysis in an E. coli host ( Figure 2C and Supplementary Figure S4). The results clearly show that substitution of the Y104 residue of HepT prevents adenylylation by MntA. As expected, no AMP moiety in the HepT Y104A /MntA complex or the HepT/MntA G27A, S28T complex was found after PDE treatment by LC-MS analysis ( Figure 2D). To investigate whether MntA can modify HepT in its original host S. oneidensis, a FLAG-tag was fused to the C-terminus of the toxin, and western blot was performed to check the status of HepT in the following constructs: HepT/MntA, HepT, HepT/MntA G27A,S28T , HepT/MntA D39E,D41E and HepT Y104A /MntA. In agreement with the results in E. coli, HepT toxin is modified by native MntA in S. oneidensis, and substitutions in the MntA adenylylation motif eliminate HepT modification in S. oneidensis ( Figure 2E and Supplementary Figure S4). The molecular masses of the native HepT/MntA TA complex, HepT alone, and the mutated HepT/MntA G27A, S28T TA complex were analyzed by HPLC/Q-TOF-MS. The results revealed that the MW of adenylylated HepT increased by 987.45 Da compared to unmodified HepT, and this change in MW is equal to three AMPs (one AMP is 329 Da) ( Figure 2F and Supplementary Figure S5). This result agrees with the SDS-PAGE analysis, where the MW of adenylylated HepT increase by  Supplementary Table S5. At least three independent protein purifications were performed for B and C, and only one representative image is shown here.

Structural analysis revealed that Y104 of HepT is polyadenylylated
To further verify the chemical modification of HepT directly by MntA, protein samples of the S. oneidensis wild-type (WT) HepT/MntA complex and its derivative complex were prepared by co-expression of the corresponding proteins. We again observed a shift in HepT in the WT HepT/MntA complex when compared with the HepT Y104A /MntA complex in the SDS-PAGE gel (Figure 1BC), suggesting that HepT was modified automatically when co-expressed in E. coli. We determined the crystal structures of HepT/MntA and HepT Y104A /MntA at 3.08 and 2.6Å, respectively (Table 1). There were four molecules composed of one MntA polypeptide and three HepT polypeptides in one asymmetric unit (ASU), and the complex formed a hetero-octamer based on symmetry analysis ( Figure 3A); this result is consistent with our previous report (17). Similarly, a (HepT) 6 -(MntA) 2 hetero-octamer was found in the ASU of HepT Y104A /MntA (Supplementary Figure S6). During structure determination, three regions of the unambiguous electron map connected to the hydroxyl group of Y104 in HepT were exclusively observed in the structure of WT HepT/MntA. In view of the above mentioned results, antitoxin MntA polyadenylylates toxin HepT, and in the final structure, a total of three AMP moieties linked by 3 -5 phosphodiester bonds were accommodated in every region of the additional electron density map ( Figure 3B). In contrast, this additional electron density map was no longer present in the HepT Y104A /MntA complex (Supplementary Figure S6). Y104 of HepT was polyadenylylated, and a phosphodiester bond was formed by the ␣-phosphate of the first AMP (AMP1) and the hydroxyl group of Y104 ( Figure 3B). The MntA molecule in HepT/MntA was almost identical to its counterpart in HepT Y104A /MntA, with a maximal root mean square deviation (RMSD) of 0.26Å over 98 C␣ and 0.29396 C␣, respectively. Comparison of the polyadenylylated HepT and unmodified HepT Y104A revealed that they were nearly identical with an RMSD of 0.396Å over the 953 C␣ main chain, except for the loop containing Y104, which was rotated clockwise nearly 180 • ( Figure 3C). Interestingly, detailed analysis of the AMP triplet moieties of the three HepT molecules in the ASU revealed that the orientation on Chains A or B was different from the one on Chain C. For clarity, the polyadenylylated structures of chain A of HepT and chain C of HepT were extracted ( Figure 3D). The extraction showed that the core of their structures aligned well, except for where the loops of the AMP chains were located. The AMP moiety in Chain A was twisted toward the cleft formed by ␣4 and ␣5 of HepT. In contrast, the AMP moieties in chain C of HepT were oriented to MntA ( Figure 3D). Collectively, our structural analysis confirm that Y104 of HepT is polyadenylylated.

MntA transfers three AMPs to HepT consecutively using ATP as the substrate
To elucidate the molecular mechanism of polyadenylylation mediated by MntA, we performed an in vitro enzymatic assay using purified MntA and HepT (unmodified). The result showed that MntA mediated the modification of HepT using ATP as a substrate (left panel, Figure 4A), and the size of the HepT obtained in this reaction fits well with size found for modified HepT in vivo (right panel, Figure 4A). As expected, when the key active motif GSX 10 Figure S7A). Moreover, previous work indicated the Mg 2+ ions are needed for adenylyltransferase activity, so we tested this requirement and found that the in vitro adenylylation assay requires Mg 2+ ions (Supplementary Figure S7B). Therefore, MntA modifies HepT via ATP through its GSX 10 DXD motif and requires Mg 2+ as is typical for adenylyltransferase activity.
Since few adenylyltransferases can mediate the transfer of more than one AMP to the target, we thus investigated whether the three AMPs were added to HepT consecutively by analyzing the intermediate products of HepT modification by MntA. To obtain intermediates, the reaction was stopped by adding EDTA, and the products were analyzed by HPLC/Q-TOF-MS. The MS analysis showed that four types of HepT proteins were identified, including unmodified HepT and HepT proteins with one, two or three AMPs ( Figure 4B). These results clearly show that MntA mediates the transfer of the three AMPs to HepT in a consecutive fashion using ATP as the substrate.
In addition, to understand the adenylylation reaction at the atomic scale, we determined the crystal structure of HepT Y104A /MntA in complex again but this time with the non-hydrolysable ATP analogue adenylyl imidodiphosphate (AMP-PNP) at 2.77Å. The structure reveals that AMP-PNP is in the cavity formed between MntA and HepT Y104A , with the phosphate group of AMP-PNP inserted into the GSX 10 DXD motif ( Figure 4C). Multiple interactions facilitate AMP-PNP binding. Specifically, the ␥ -phosphate of AMP-PNP bonds with S38, D39 and S28 of MntA, and the ␤-phosphate of AMP-PNP forms hydrogen bonds with R59 of HepT and G27, S28 of MntA. In addition, the adenosine engages in hydrogen-bonding with N7, D103 of HepT and Q85, Q87 of MntA. There are two Mg 2+ ions in every catalytic center of MntA, with one Mg 2+ ion is coordinated by the oxygen atoms of the ␣, ␤ and ␥ phosphate group and D39, D41 of MntA and the other Mg 2+ ion is coordinated by oxygen atom of the ␣ phosphate group and D41, D39 and D71 of MntA. Therefore, the HepT Y104A /MntA structure with AMP-PNP shows how MntA utilizes ATP and Mg 2+ for the adenylylation reaction with HepT.
To further confirm the role of the conserved MntA GSX 10 DXD motif in mediating adenylylation, the structure of the HepT/MntA D39E, D41E hetero-octamer was determined at 2.35Å (Supplementary Figure S6). The structural analysis confirmed that HepT was not adenylylated even though Y104 was available for receiving AMP. Taken together, we demonstrated that MntA transfers three AMPs to the Y104 residue of HepT by employing a highly conserved GSX 10 DXD motif.

Neutralization of HepT by MntA relies on polyadenylylation
Based on the conserved signature RX 4 HXY motif in the toxin component and the signature GSX 10 DXD motif in the antitoxin component, we reasoned that polyadenylylation likely plays an important role in the neutralization mechanism of this TA pair. We previously demonstrated that HepT is an RNase that cleaves mRNA substrates in vitro (17). Compared to the type II TA pair ParE SO /CopA SO , MntA and HepT have a weaker interaction in vivo (Supplementary Figure S8). Based on this weaker interaction, the modified HepT protein was successfully purified from the lysate of cells co-expressing the TA complex (Supplementary Figure S1). The in vitro RNase activity of the modified HepT and unmodified HepT proteins was measured with ompA mRNA as the substrate. The results clearly revealed that polyadenylylated HepT did not cleave mRNA as effectively as unmodified HepT ( Figure 5A). Since the toxicity of HepT can be measured by cell growth (OD 600 ) and cell viability (CFU/ml)(17), we performed a toxicity test using HepT/MntA native to S. oneidensis and HepT and MntA variants. As shown previously, overexpressing MntA did not induce toxicity (17). Although substitution of Y104 in the RX 4 HXY motif did not affect the toxicity of HepT (16), this substitution clearly reduced the ability of MntA to neutralize HepT ( Figure 5B). The substitutions of DXD with EXE in the GSX 10 DXD motif also reduced the ability of MntA to neutralize HepT ( Figure 5B). Thus, both in vitro and in vivo assays demonstrate that polyadenylylation is important for reducing HepT toxicity.
Since MntA can bind to HepT and polyadenylylate HepT, we wanted to check the importance of proteinprotein interactions and/or adenylylation in terms of neutralizing HepT toxicity. The substitutions of MntA (G27/S28 or D39/D41) showed a reduced ability to neutralize HepT toxin ( Figure 5B and Supplementary Figure S9). However, our structural analysis clearly showed that the MntA D39E, D41E variant binds to HepT to form a hetero-octamer in a way that is like native MntA (Supplementary Figure S6). Additionally, the MntA variants MntA D39E, D41E are still pulled down due to their strong protein-protein interaction with HepT during SDS-PAGE fractionation of the TA complex ( Figure 5C). These results collectively suggest that the binding to HepT is not sufficient for MntA to neutralize the toxicity of HepT. In our previous study, we found that MntA binds to HepT mainly via the insertion of the ␣4 helix into the HepT RX 4 H motif, and the deletion of the ␣4 helix of MntA (MntA ␣4 98-113 ) reduced the ability to neutralize the toxicity of HepT ( Figure 5B) (17). We then tested whether the deletion of the ␣4 helix of MntA affects adenylylation using HepT/MntA ␣4 98-113 (Supplementary Figure S10). Noticeably, unmodified HepT was observed in the presence of MntA ␣4 98-113 but not in the presence of native MntA, suggesting that the ␣4 helix of MntA is needed for the full adenylyltransferase activity. To probe the molecular mechanism of how polyadenylylation affects the toxicity of HepT, we analyzed the polyadenylylated and unmodified dimeric forms of HepT. The analysis of the interaction of HepT and the AMP moieties revealed that AMP fits near the cavities along the cleft access to the RX 4 H RNase activity site (Figure 5D). Furthermore, the highly conserved R70 that was located in the positively charged region of HepT forms hydrogen bonds with the first AMP moiety ( Figure 5E). This result is in agreement with our previous study that substitution of R70 with H70 or substitution of L107 with A107 in HepT abolished HepT toxicity (16). Collectively, these results reveal that the neutralization of MntA mainly depends on polyadenylylation.

HepT/MntA TA systems are ubiquitously distributed in prokaryotes
Since HEPN domain-containing and MNT domaincontaining proteins are widespread in bacteria and archaea (13), we searched for HEPNs encoded next to MNTs using the conserved GSX 10 DXD and RX 4 HXY motifs by FIMO (29). The majority (99%) of neighboring HEPN/MNT domains belonged to the HEPN homologous domains (PF01934: DUF86 or PF08780: NTase sub bind) and the MNT domain (PF01909: NTP transf 2) according to the Pfam database. Because another important HEPN homologous domain (PF05168: HEPN) was also found to flank the MNT-domain PF01909 family (14), these two classes of HEPN/MNT pairs were recovered from the IMG/M database with functional profile tools (Supplementary Tables S3 and S4). We found that the Class II HEPN/MNT module (PF05168/PF01909) was more prevalent than the Class I HEPN/MNT module (PF01934 or PF08780)/PF01909 in archaea, similar to the report by Makarova and Koonin (14). Moreover, we found that Class I pairs were much more prevalent than Class II pairs in bacteria, with Class I pairs found in nearly 15% of Proteobacteria genomes ( Figure 6A and Supplementary Figure S11). Interestingly, the antitoxin components of the Class I and II modules both contain the conserved motif GSX 10 DXD, suggesting that nucleotidyltransferase activity is conserved in these antitoxins. In contrast, the conserved motif RX 4 HXY in HepT was only found in the toxin component of the Class I module ( Figure 6BC). Furthermore, a conserved tyrosine, which can be adenylylated, was also found in the toxin of the Class II TA module, although no active RNase motif RX 4 H was found ( Figure 6C). Collectively, these results suggest that these HEPN/MNT modules represent two different TA families that may differ in the catalytic function of the toxin component. To further test whether the HEPN domain-containing proteins are toxins in the Class I TA modules, four putative TA pairs from Class I were selected: from the deep-sea bacterium Shewanella putrefaciens W3-18-1; from the archaeon T. cleftensis CL1, which was isolated from a deep-sea hydrothermal sulfide chimney; from the methanogenic archaeon Methanosarcina mazei TUC01, a strain isolated from an Amazon flooded area; and from Bifidobacterium merycicum, which was isolated from the rumen of cattle (Supplementary Table S6 and Figure S3). Consistently, in these HEPN/MNT modules, the HEPNs serve as the toxin, and the upstream MNT domain-containing proteins completely neutralize the toxicity of the cognate toxins ( Figure  6D and Supplementary Figure S12). Collectively, these results demonstrate that HEPNs with the RX 4 HXY motif and neighboring MNTs with the GSX 10 DXD motif represent a new type of TA system that is prevalent in both bacteria and archaea.

HepT is adenylylated by MntA in the archaeon T. cleftensis
To further investigate whether the antitoxin component of the Class I HEPN/MNT TA module can adenylylate the toxin in archaea, the HepT/MntA pair in T. cleftensis CL1 was chosen ( Figure 7A). T. cleftensis CL1 is a hyperthermophilic archaeon isolated from a Paralvinella sp. polychaete worm living on the deep-sea hydrothermal sulfide chimney of the Juan de Fuca Ridge (33). Overexpressing HepT CL1 in E. coli led to severe growth inhibition, and co-expressing neighboring MntA CL1 completely neutralized the toxicity of HepT CL1 ( Figure 6D and Supplementary Figure S12). Similarly, site-directed mutagenesis was utilized to generate two MntA CL1 variants with the conserved GSX 10 DXD motif substituted with ATX 10 DXD or GSX 10 EXE. One HepT CL1 variant with a substitution at Y82 in the RX 4 HXY signature motif was also constructed. As expected, substitution of Y82 with F82 in the RX 4 HY motif of HepT CL1 or substitution of G33S34 with A33T34 in the GSX 10 DXD motif of MntA CL1 eliminated the ability of MntA CL1 to modify HepT CL1 ( Figure 7B). Additionally, HPLC/MS analysis showed that the AMP peak only existed in the cleaved products of HepT/MntA CL1 ( Figure  7BC and Supplementary Figure S13). Substitution of the GSX 10 DXD motif to ATX 10 DXD in MntA CL1 or the substitution of Y to F in the RX 4 HXY motif of HepT CL1 both reduced the ability of the antitoxin to rescue the growth inhibition of HepT ( Figure 7D). Taken together, these results suggest that HEPN proteins with the RX 4 HXY motif and the neighboring MNT-containing protein with the GSX 10 DXD motif form a TA pair in archaea, and the adenylylation of the toxin by the GSX 10 DXD signature motif of the antitoxin might represent a common mechanism for this TA pair.

DISCUSSION
The HEPN/MNT module is predicted to be one of the most abundant TA pairs and is prevalent in various bacterial and archaeal strains (34). Here, we provide evidence that the HEPN/MNT module is a new type of TA system (HepT/MntA) in both bacteria and archaea. Currently, six different types of TA systems have been reported and recognized. Recently, three examples of antitoxins inactivating toxins as enzymes have been found. In the newly proposed type VII Hha/TomB system in E. coli, the antitoxin TomB oxidizes the cysteine residues of toxin Hha to neutralize Hha toxicity (10). In the TglT/TakA system in Mycobacterium tuberculosis, the antitoxin TakA functions as a serine protein kinase that inactivates toxin TglT by phosphorylating it at residue Ser78 (12). We found here that MntA antitoxin acts as an adenylyltransferase and catalyzes the transfer of three AMP moieties onto toxin HepT to block its toxicity, representing a new type of interaction between the toxin and the antitoxin (Figure 8). Thus, three enzymatic antitoxins TomB, TakA and MntA chemically modify the toxin to neutralize its toxicity. Therefore, we propose to classify HepT/MntA TA pair as a type VII TA system in which the neutralization mechanism relies on the enzymatic activity of the antitoxins in a similar way as Hha/TomB (35).
A typical feature of type II TA systems is the tight affinity between the toxin and the antitoxin component, which key to the neutralization mechanism. In the Hha/TomB TA pair, there was only a transient affinity between the toxin and antitoxin (10). Although a hetero-octamer of HepT/MntA was found in S. oneidensis, we found that the affinity between HepT and MntA was lower than that of a typical type II TA system. Here, we also explored the importance of the enzymatic activity of MntA and the involvement of protein-protein interactions in terms of the neutralization mechanism. Both the in vitro and in vivo assays demonstrated that polyadenylylation is more important for reducing HepT toxicity than the affinity between HepT and MntA. Structural analysis of HepT Y104A /MntA in complex with AMP-PNP also revealed that the AMP-PNP substrate is not in the proximity of Tyr104 in HepT, and it is possible that the HepT/MntA hetero-octamer is not the only state of this TA pair in the cell. We also noticed that the amount of MntA protein was equally expressed in general as the toxin; thus, free MntA is likely present in the cell and might be involved in HepT modification. Furthermore, our bioinformatics analysis of class I HEPN/MNT TA modules in different bacteria and archaea revealed a conserved motif with adenylyltransferase activity and a conserved Y critical for receiving AMPs, suggesting that the chemical modification of the toxin by its antitoxin is a common neutralization mechanism in these putative TA pairs. Previously, toxins with nucleotidyltransferase activity have been reported for the FicT/FicA (36) and AbiEii/AbiEi (37) TA systems. We report here that an antitoxin can also function as a nucleotidyltransferase. The conserved motif in MNT antitoxins is GSX 10 DXD which is in agreement of the conserved motif (GX 11 DXD) in the previously reported AMPylators with the adenylyltransferase domain (38). The first AMPylating protein with an adenylyltransferase domain reported was glutamine synthetase adenylyltransferase (GS-ATase) in E. coli that mediates the transfer of an AMP to the glutamine synthetase in response to changes in nitrogen (39)(40)(41)(42). Later, the Legionella pneumophila effector protein DrrA (or SidM) was reported to AMPylate the membrane traffic regulator GT-Pase Rab1 (43). Unlike GS-ATase and DrrA, which contain more than one domain (>300 aa), MNTs are <150 aa. To the best of our knowledge, MntA is the first adenylyltransferase that can mediate the transfer of three AMPs to the target consecutively. Mutagenesis experiments revealed that the GSX 10 DXD motif of MntA is the key for adding three AMP moieties to Y104 of HepT. The structural analysis showed that the Y104 residue near the active RX 4 H of HepT receives the three AMP moieties, thus this residue may hinder the access of HepT to its RNA substrate. This is consistent with previously reported AMPylators which can modulate the activity of modified targets by interfering with substrate binding (36,44).
The understanding of the regulatory role of adenylylation in prokaryotes is in its infancy, and many questions about the biological processes it participates in remain to be answered. Furthermore, two different classes of HEPN/MNT pairs have been found, and their abundance in bacteria and archaea is different. Class II HEPN/MNT modules were first predicted by Koonin's group and have a wide distribution in the hyperthermophilic bacteria and archaea; yet, no biochemical studies on this module have been conducted as far as we know (14). Our work here shows the Class I HepT/MntA pair in archaeon T. cleftensis CL1 works through adenylylation. However, the HEPNs in Class II do not have the RNase RX 4 HXY domain, but the neighboring MNTs still have the conserved motif GSX 10 DXD. Thus, further study is needed to validate whether they constitute a TA system and to investigate the adenylylation mechanism in archaea. In addition, understanding the conditions that trigger adenylylation and/or de-adenylylation of the toxin is a critical question that needs to be addressed in future studies of HEPN/MNT modules to understand the physiological role of these TA systems. Potential PDEs that can remove AMPs from the modified HepT to reactivate the toxin have yet to be identified. Nevertheless, the prevalence of these HEPN/MNT modules in bacteria and archaea isolated from diverse environments provides an opportunity to learn more about chemical modification and biological processes in different domains of life.