ATfaRel2 is a compact, well-structured antitoxin
We determined the crystal structure of ATfaRel2 (residues 1–73) to 1.2 Å resolution (Fig. 1a-b and Supplementary Table 1). The antitoxin protein is monomeric in the crystal (Supplementary Fig. 2a), in good agreement with molecular weight estimates by Size Exclusion Chromatography (SEC) (Supplementary Fig. 2b and Supplementary Table 2). At the core of its compact, well-folded structure is an antiparallel β-sheet (β-strands β1-β3), with β2 and β3 connected via the central α-helix α1. The C-terminal extension provides an additional β-strand to the β-sheet, β4, which folds parallel to β2, as well as a second α-helix, α2, that forms a part of the to the protein’s hydrophobic core (Fig. 1b). Despite the lack of sequence similarity, ATfaRel2 is structurally similar to pZFDCapRel, with the two proteins superimposing with an r.m.s.d. of 2.8 Å, and they both contain conserved tyrosine residues at the C terminus (Fig. 1c).
Interestingly, the predicted topologies of pZFDCapRel in the neutralizing state and that of ATfaRel2 are similar except for the order of structural elements at the termini. The C-terminal β/α extension of ATfaRel2 is structurally similar to the N-terminal β/α region of pZFDCapRel that connects the antitoxin with toxSYNTHCapRel via Anchor1. This is suggestive of circular permutation (compare Fig. 1b to Fig. 1d and Supplementary Fig. 2c-e). However, with similarities extending only to the topology level, these two small β/α elements may have evolved independently.
These changes in the connectivity of secondary structural elements likely reflect the different modes of toxin neutralization, in cis vs in trans. ATfaRel2 – presumably – fully dissociates from FaRel2 when the toxin becomes active. Conversely, in the case of fused CapRels, dissociation is not possible and therefore the activation is believed to be mediated by a conformational change facilitated by the two Anchor regions resulting in the disruption of the toxSYNTH-recognising interface9 (Supplementary Fig. 2d-e). These rearrangements are not necessary for the rigid-body disengagement of ATfaRel2-like domains acting in trans or pZFD N- or C-terminally fused to Panacea domains18. Despite the topological arrangement, the strong structural similarity between pZFDCapRel and ATfaRel2 suggests the two antitoxins neutralize toxSYNTH domains via a common mechanism.
The YXXY motif is crucial for toxSYNTH inhibition
In CapRelJS46 the YXXY sequence motif is the key structural element of the pZFDCapRel that locks the enzyme in the catalytically inactive state by means of the two Y-residues blocking the donor-pyrophosphate ATP binding site of the toxSYNTH (Supplementary Fig. 2c-e). Substitutions directly or indirectly targeting this motif rendered CapRelJS46 constitutively active9. The YXXY motif is also conserved in pZFDCapRel, and is located in a switch region. As captured by X-ray, in the catalytically active state of CapRelJS46 the motif assumes an extended conformation9. Conversely, in the AlphaFold-generated neutralised state the YXXY motif was predicted to fold into a 310-helix that locks into the donor site (Supplementary Fig. 2d-e).
The structure of the free ATfaRel2 reveals that the 51YXXY54 motif, indeed, folds into a 310-helix structure that was predicted to be key for toxSAS neutralisation, scaffolded by β-strand β1 and β3 (Fig. 1a and Supplementary Fig. 1b). This suggests that, even in the absence of the toxin, ATfaRel2 is primed for efficient FaRel2 neutralization. Guided by the structural similarities between ATfaRel2 and pZFDCapRel, we probed the role the individual residues of the 51YXXY54 motif in the neutralization of FaRel2 toxicity in vivo (Fig. 1e). While Y54A substitution fully ablates the neutralizing activity of ATfaRel2, I45A and Y51A result in modest, but clearly detectable, defects in FaRel2 neutralization. Our Isothermal Titration Calorimetry (ITC) assays lend further support to the in vivo neutralization results. To overcome the challenges of producing an otherwise highly toxic FaRel2, we used a catalytically-impaired version Y128F (FaRel2Y128F) – a well-established substitution used for to study RSH enzymes9,16,19. The substitution interferes with the accommodation of the acceptor nucleotide in the active site but is located far from the predicted pZFDCapRel-toxSYNTH interface. We characterised complex formation between FaRel2Y128F and wild-type ATfaRel2 or variants carrying substitutions in the 51YXXY54 motif (Fig. 1f and Supplementary Table 3). V43A and A50M substitutions have a negligible effect on complex stability (78.8 nM and 38.0 nM compared to the wild-type ATfaRel2–35.2 nM). The destabilising effect is more pronounced in the case of I45A and Y51A variants. While these substitutions result in a 25- and 43-fold drop in affinity, respectively, the remaining affinity is still sufficient for partial neutralisation in vivo (Fig. 1e). Finally, the Y54A variant which is unable to neutralize FaRel2 in vivo, has a 250-fold lower affinity to the toxin than the wild-type ATfaRel2.
It is instructive to compare the structural-energetic interplay of the fused CapRelJS46 TA with the bipartite ATfaRel2:FaRel2 TA. In CapRelJS46, the forced colocalization to the toxin and antitoxin elements likely offsets the entropic penalty associated with the antitoxin assuming a compact and structured neutralizing state. Conversely, the order-to-disorder transition in the disengaged antitoxin is entropically beneficial upon formation of the CapRelJS46:Gp57 complex. In the case of the bipartite ATfaRel2:FaRel2 TA system, the free antitoxin naturally assumes the optimal conformation for recognising the toxin, with a more robust pre-organised interface primed for the formation of a stable TA complex. The lack of disorder-to-order transition in the antitoxin upon complex formation reduces the entropic penalty.
FaRel2 engages ATP in a catalytically primed, partially folded state
Next, we determined the structure of the catalytically-impaired FaRel2Y128F bound to the non-hydrolysable ATP analogue Adenosine-5'-[(α,β)-methyleno]triphosphate (APCPP) at 2.6 Å resolution (Fig. 2a). The toxSYNTH domain of FaRel2 shares the overall topology of other nucleotide-pyrophophotransferases, retaining a presumably ancestral fold composed of: a central five-stranded β-sheet framed by two α-helices, α3 and α619,20 (Fig. 2b). The core toxSYNTH domain is very similar to that of the (pp)pApp alarmone synthetase Tas1 (PDBID 6OX6) and the tRNA-pyrophosphokinase CapRelJS46 (PDBID 7ZTB), superimposing with r.m.s.d. of 1.0 Å and 0.8 Å, respectively (Supplementary Fig. 3a-b).
The catalytic core of RSH enzymes is typically decorated by regulatory elements which exert allosteric control on the enzymatic domains16,21. The FaRel2 toxSYNTH domain has a well-resolved N-terminal α-helical extension comprised of helixes α1 and α2, with residues K28 and R29 from the α2α3 loop, being crucial for tRNA binding7 (Supplementary Fig. 1c). While in the non-toxic (p)ppGpp alarmone synthetases RelP21,22 and RelQ23 the N-terminus is unresolved (Fig. 2c), in FaRel2 the N-terminus folds back towards the toxSYNTH core and intercalates α1 and α2 between α3 and α5, providing and anchor point for tRNAs near the active site G-loop (Fig. 2a and c). The C-terminal region of SYNTH/toxSYNTH domains of pyrophosphotransferases have varied architectures composed predominantly of small α-helical domains6,9,19,20,24,25: the C-terminal four α-helical bundle of (p)ppGpp synthesising SAS RelQ and RelP acts as an oligomerization interface21,23, the toxSYNTH of the monomeric Tas1 is followed by a small α-helical domain6, and SYNTH domain of long RHSs is followed by a flexible Core domain with a high α-helical propensity24. In the case of FaRel2 complexed with APCPP, the C-terminus is disordered and not visible in the electron density (Fig. 2d).
Despite these differences, the FaRel2-bound APCPP superimposes remarkably well with the APCPP bound in the donor site of other (p)ppGpp-synthesising RSHs: SAS RelP and RelQ, as well as long RSH Rel19,21,23 (Fig. 2d). As in the other alarmone synthetases, the adenosine group is stacking the conserved R64 and R95 from β1 and β2, with β5 E145 hydrogen-bonding the adenosine NH2 group (Supplementary Fig. 1c). The R64 residue plays a crucial role providing Van der Waals contacts to accommodate the ribose while, together with K66, while directly coordinating the 5′ α- and β- phosphates. The strongly basic α3 that follows β1 further stabilizes the triphosphates via S70, K74 and R77 (Fig. 2a). At the pyrophosphate acceptor side of the active site, the conformation of the G-loop and the orientation of the base-coordinating F128 (Y128 in the WT FaRel2) deviate from what was observed in pre- and post-catalytic complexes of (p)ppGpp synthetases21,23 (Supplementary Fig. 3c-d). It is tempting to speculate that while the ground state of FaRel2 is primed to bind ATP, efficient tRNA binding would likely involve further conformational rearrangements such as folding of the C-terminus and alignment of the active site.
ATfaRel2 binds FaRel2 via β-sheet extension of the ATP binding site
To directly uncover the mechanism of toxin neutralisation we determined the structure of the ATfaRel2:FaRel2 complex. The structure of the complex revealed a ATfaRel22:FaRel22 heterotetrametric arrangement (Fig. 3a-c). The 2:2 stoichiometry was confirmed in solution by SEC (62 kDa vs the theoretical 68 kDa) (Fig. 3d) and was consistent with the ITC measurements (Fig. 1f).
The C-terminal region of FaRel2, disordered in the FaRel2:APCPP complex, serves as a dimerization interface of 670 Å2 per one ATfaRel2:FaRel2 unit. The primary interface between ATfaRel2 and FaRel2 is 1190.0 Å2-large, with ATfaRel2 sterically blocking the access of ATP substrate to the PP-donor binding site of the toxSYNTH domain (Fig. 3b). Through this large interface ATfaRel2 contacts several key functional regions of FaRel2: i) the long N-terminal α-helix α3 (a structural element often involved in allosteric crosstalk in many RSH enzymes which interacts with α1 from ATfaRel2), ii) the basic α-helix α4 (involved in the stabilization of the ATP triphosphate group, which is coordinated by ATfaRel2 via β3 and the β3-α1 loop), iii) the central β-sheet (via β1, β2 and β5 that harbour the catalytic center and adenine coordinating residues) and iv) the C-cap of the α-helix α6 (disordered in the FaRel2:APCPP complex and that folds into a dimerization α-helical bundle when bound to ATfaRel2). The only notable exception was the predicted tRNA-recognition site which remains solvent-accessible (Supplementary Fig. 3e).
The center of the neutralisation interface is formed by the anti-parallel β-strands interaction between FaRel2 β1 and ATfaRel2 β3 that connects the β-sheets of both proteins extending the core of the complex. Residues I42-T46 of β3 form multiple Van der Waals contacts with FaRel2 β1, further stabilizing the complex. Residues V43 and I45 serve as scaffold orienting the YXXY 310-helix that anchors ATfaRel2 (Fig. 3b). As predicted for pZFDCapRel (Ref.9), this hydrophobic tether projects Y51 and Y54 into the ATP binding site through a π-stacking arrangement with R64 and R95, which precludes adenine coordination to the PP-donor site of FaRel2. These results suggest that while this mechanism of neutralization is likely the same at the structural level to that proposed for CapRel SJ46, the energetics of neutralization are certainly different with the dynamic association of pZFDCapRel regulating toxSYNTHCapRelin cis, contrasting with the stable in trans neutralization in the ATfaRel2:FaRel2 complex. These differences could have important implications for triggering these systems.
Dimerization enhances the neutralising activity of ATfaRel2
Compared to pZFDCapRel, ATfaRel2 is considerably more tolerant to substitutions in the toxin binding interface (compare Fig. 1e with Extended Data Fig. 3j in Ref.9). The structure of the ATfaRel22:FaRel22 complex revealed that ATfaRel2 engages the neighbouring FaRel2 in the heterotetramer via a secondary interface which is half the size of the primary one (~ 550.0 Å2 vs 1190.0 Å2), thus effectively crosslinking the complex. Therefore, we hypothesised that the stable oligomeric nature of ATfaRel22:FaRel22, compensates for the colocalization of both domains typical of monomeric CapRelSJ46, and accounts for the strong functional plasticity of the ATfaRel2-neutralising interface.
To test this hypothesis, we subjected the secondary interface to single-residue substitutions and assessed the complex stability in vivo through toxicity neutralisation assays (Fig. 3e). The three targeted residues R47A, Y57A and F59A of ATfaRel2 are all distant from the main contact interface that blocks the access of ATP to the active site (Fig. 3c). R47 is located on the two-fold symmetry axis of the complex; the side chains of R47 from each ATfaRel2 interlock via π-π interactions. Y57 and F59 are part of a small hydrophobic core that defines the secondary oligomerization interface. The R47A substitution resulted in a modest defect, whereas Y57A and F59A compromised the neutralisation severely (Fig. 3d).
The direct interrogation of these interactions by ITC is in good agreement with the in vivo data. The R47A substitution efficiently perturbed the secondary interface and decoupled the highly cooperative tetramer formation observed in the wild-type protein (Fig. 3f and Supplementary Table 3). The first high affinity binding event (KD = 33 nM, molar ratio of 0.44) followed a lower affinity recognition event (KD = 750 nM with a 0.8 stoichiometry). This likely represents the initial neutralization of FaRel2 (with 2:1 toxin-to-antitoxin ratio) followed by the formation of a less stable 2:2 tetramer. The impact of F59A on the affinity was even stronger with a 60-fold decrease in affinity and a confirmed 1:1 binding molar ratio that indicates an interaction mediated by only the primary interface (Fig. 3g and Supplementary Table 3). These results are consistent with SEC experiments that revealed a decrease in size of the toxin-antitoxin complex from the estimated 70.8 KDa of the wild-type (consistent with a 2:2 A:T stoichiometry) to 51.3 KDa suggestive of a ATfaRel2R47A:FaRel22Y128F complex (1:2 A:T stoichiometry) (Fig. 3d and Supplementary Table 2). The observation of a stable ATfaRel2R47A:FaRel22Y128F complex in SEC, matches the high-affinity interaction observed by ITC with ATfaRel2R47A (Supplementary Table 2). It is thus likely that the C-terminal region of FaRel2 that folds upon TA complex formation and provides a large FaRel2:FaRel2 interface in the complex, is still capable of partially stabilising the oligomer against the effect of mild substitutions such R47A, but not against F59A which had a major effect on complex formation.
Collectively, these results suggest that while the main TA interface drives toxin neutralization, oligomerization further stabilises the interaction between ATfaRel2 and the toxSYNTH domain. Multiple contacts of the main interface YXXY motif with the folded C-terminal α-helical FaRel2:FaRel2 bundle interface, link directly oligomerization with toxin neutralization, providing a potential allosteric path for activation and toxin release.
tRNA-pyrophosphorylating toxins specifically bind tRNA
We used ITC to examine the tRNA-binding capacity of representative toxSASs and housekeeping SAS: tRNA-phosphorylating Coprobacillus sp. D7 FaRel2 and Mycobacterium phage Phrann PhRel, (pp)pApp-synthesising Cellulomonas marina FaRel, and (p)ppGpp-synthesising Staphylococcus aureus SAS RelQ. FaRel2 and PhRel bind deacylated initiator tRNAifMet with similar affinities (KD 483 nM and 825 nM) (Fig. 4a-b and Supplementary Table 3). FaRel has a 37-fold lower affinity to tRNAifMet (KD 17.8 µM), consistent with its documented enzymatic activity while no tRNAifMet binding was observed for RelQ SASs (Fig. 4c-d and Supplementary Table 3). Finally, Enterococcus faecalis RelQ has been shown to interact with a short single-stranded model mRNA(MF) coding for MF dipeptide26. Our ITC experiments demonstrate that S. aureus RelQ binds mRNA(MF) with a sub-µM affinity (KD 922 nM) (Fig. 4e and Supplementary Table 3). Given the low concentrations these toxins are typically found in the cell, tRNA-phosphorylating activity would likely depend on a strong and specific association with tRNAs.
ATfaRel2 interferes with APCPP binding by FaRel2 but not with tRNA recognition
While association with ATfaRel2 decreases the affinity (KD) to APCPP ~ 20-fold, from 2.1 to 41.3 µM (Fig. 4f-g and Supplementary Table 3), the affinity to deacylated initiator tRNA (tRNAifMet) is virtually the same for the free toxin and inactive toxin:antitoxin complex (KD 483 nM vs 460 nM) (Fig. 4h and Supplementary Table 3). Both in the case of free monomeric FaRel2 and the heterotetrametric ATfaRel2:FaRel2 complex, the tRNA binding has 1-to-1 stoichiometry with respect to FaRel2.
To further investigate the effect of ATfaRel2 binding on the interaction of FaRel2 with ATP we determined the structure of the ATfaRel2:FaRel2 complex bound to APCPP (Fig. 4i). The high protein concentration intrinsic of the crystal lattice combined high nucleotide concentration used for soaking, facilitated the binding of APCPP to a partially blocked active site. As predicted from the structure of ATfaRel2:FaRel2, the coordination sites for the adenine, ribose, and αP groups at the PP-donor site were blocked by ATfaRel2. Thus, APCPP is bound in the PP-acceptor site in a conformation incompatible with pyrophosphate transfer (Fig. 4j). The adenine base is observed coordinated by Y128 and R95 resembling the expected coordination of the terminal adenine of the CCA tRNA moiety. The βP and γP further anchor the nucleotide, however, they are observed in a reversed orientation compared to the FaRel2:APCPP complex (with βP in the γP and vice versa) (Fig. 4j-k).
It is instructive to compare the local charge distributions in the active sites of tRNA-modifying FaRel2 to that of (pp)pGpp/(p)ppApp alarmone synthetases: Rel, RelQ and Tas1. All alarmone synthetases have a large positive patch that accommodates di- and tri-phosphate nucleotide substrates, located on the acceptor site close to the conserved Y residue that interacts with the acceptor base (Supplementary Fig. 4a-c). This positive patch is considerably smaller in FaRel2 (Supplementary Fig. 4d), which explains in the miss-orientation of the βP and γP of APCPP bound in the acceptor site of FaRel2 and the lack of alarmone synthetase activity of the tRNA-targeting toxSAS.
Collectively, our results demonstrate that FaRel2 is neutralized by the ATfaRel2 antitoxin by compromising the accommodation of ATP in the toxSYNTH active site without affecting the interaction with uncharged tRNAs. This suggests that in the cell a ternary complex ATfaRel22:FaRel22:tRNA2 could be formed, and in this complex the toxSAS would remain neutralized until the activation is triggered.
tRNA-modifying toxSASs are neutralized differently from toxic nucleotide synthesising toxSASs
Prompted by the conceptual differences between Tis1-mediated neutralisation of Tas1 and ATfaRel2-mediated neutralisation of FaRel2, we next used AlphaFold227 to explore the general principles underlying the mechanisms of toxin neutralisation across known toxSAS functional diversity (Fig. 5 and Supplementary Fig. 5a-l). On the toxSAS toxin side, AlphaFold2 predicts a strong conservation of a core toxSYNTH fold decorated with a variety of insertions at the N- and C-terminus (Fig. 5a-e). On the antitoxin side, the pZFD fold (of which the PAD1 domain is a subtype10,11 is found either as a standalone neutralising domain or, as part of multidomain antitoxins, combined with either Tis1/Tis1-like domains (Fig. 5f-h) or PanA domains (Supplementary Fig. 5a-j).
Structural predictions of the different neutralised complexes uncover a general trend: while tRNA-pyrophosphokinases such as fused and split CapRels, FaRel2, PhRel and PhRel2 are all inhibited through the pyrophosphate donor site (Supplementary Fig. 5a-j), (pp)pApp alarmone synthetase toxins such as FaRel and Tas1 are neutralized through the pyrophosphate acceptor site (Fig. 5c and Supplementary Fig. 5k,l). The generality of this observation holds even in cases of multi-domain antitoxins. In the case of the tRNA pyrophosphorylating PhRel2 enzymes from bacteriophage Lily and B. subtilis Ia1a the toxins are neutralised by the pZFD fold subtype PAD1 domain of the ATphRel2 antitoxin analogously to pZDF-mediated neutralisation of CapRel and FaRel2 (Supplementary Fig. 5fg and Ref.18). By contrast, (pp)pApp-producing FaRel is inhibited by the Tis1-like domain of AT2faRel in a manner analogous to Tis1-mediated inhibition of Tas1 (Fig. 5h and Supplementary Fig. 5k,l).
We validated the structural predictions of Alphafold2 through mutagenesis and toxicity neutralisation assays. In the case of PhRel2, as we have shown earlier18, the isolated PAD1 domain of B. subtilis Ia1a ATphRel2 is sufficient to neutralize the toxin. In the case of FaRel:ATfaRel the N-terminal Tis1-like domain of ATfaRel (Fig. 5g) is sufficient to neutralize the (p)ppApp-synthetase FaRel, while the C-terminal pZFD domain (Fig. 5h) has no neutralising activity (Fig. 5i). Interestingly, this loss of neutralising activity by the CTD is accompanied by the degradation of the YXXY recognition motif of the pZFD (Fig. 5f). Collectively our results suggest a coupling between substrate specificity of toxSAS and the mechanism of neutralisation.