Many bacteria encode multiple toxin-antitoxin (TA) systems targeting separate, but closely related, cellular functions. The toxin of the E. coli hipBA system, HipA, is a kinase that inhibits translation via phosphorylation of glutamyl-tRNA synthetase. Enteropathogenic E. coli (EPEC) O127:H6 encodes the hipBA-like, tripartite TA system; hipBST, in which the HipT toxin specifically targets the tryptophanyl-tRNA synthetase, TrpS. Notably, in the tripartite system, the function as antitoxin has been taken over by the third protein, HipS, but the molecular details of how activity of HipT is inhibited remain poorly understood. Here, we show that HipBST is structurally different from E. coli HipBA and that the unique HipS protein, which is homologous to the N-terminal subdomain of HipA, inhibits the kinase through insertion of a conserved Trp residue into the active site. We also show how auto-phosphorylation at two conserved sites in the kinase toxin serve different roles and affect the ability of HipS to neutralize HipT. Finally, solution structural studies show how phosphorylation affects overall TA complex flexibility.
This important study presents an exhaustive structural analysis of a complete tripartite HipBST toxin-antitoxin system of the Enteropathogenic E. coli O127:H6, which represents a fascinating variation on the well-studied HipAB toxin-antitoxin system. The convincing data show that major features of the canonical HipAB system have been rerouted to form the tripartite HipBST, revealing a new mode of inhibition of a toxin kinase.
Bacteria employ a wide range of mechanisms to adapt to changes in their environments and imminent threats, such as exposure to antibiotics and phages, including activation of toxin-antitoxin (TA) systems that encode intracellular toxins able to rapidly reduce growth (LeRoux & Laub, 2022). Canonical (type II) TA systems are small, bicistronic loci that encode a protein toxin and its cognate antidote (antitoxin) that tightly interact to form an inactive higher-order complex, usually also capable of controlling transcription via a DNA-binding domain on the antitoxin (Harms et al., 2018). However, there are also examples of tricistronic TA loci, in which the third gene either encodes a chaperone required for folding of the antitoxin and thus, toxin inhibition (Bordes et al., 2016) or an additional, transcriptional regulator (Hallez et al., 2010, Zielenkiewicz & Ceglowski, 2005, Jurenas et al., 2021).
In the widespread and diverse hipBA (high persister) system, the HipA toxin is a serine-threonine kinase (STK) (Hanks et al., 1988, Stancik et al., 2018), while the cognate antitoxin, HipB, contains a helix-turn-helix (HTH) motif (Gerdes et al., 2021, Schumacher et al., 2009). For the hipBA system from Escherichia coli K-12, the toxin (HipAEc) specifically targets glutamyl-tRNA synthetase (GltX) by phosphorylation of an ATP-binding motif conserved in type I aminoacyl-tRNA synthetases (Eriani et al., 1990, Sekine et al., 2003), thereby inhibiting its activity and blocking translation (Germain et al., 2013). Accumulation of uncharged tRNAGlt subsequently induces the stringent response and cellular dormancy via RelA-mediated (p)ppGpp synthesis on starved ribosomes (Haseltine & Block, 1973, Pacios et al., 2020, Winther et al., 2018). In contrast to most classical type II TA systems, the antitoxin, HipB, does not directly block the HipA active site so inhibition has been proposed to occur by several alternative mechanisms, including inhibition of conformational changes in the HipA kinase required for catalysis, sequestration of HipA on DNA via the HipB DNA binding domain (Schumacher et al., 2009), and allosterically via placement of a C-terminal Trp residue of HipB into a pocket on HipA (Evdokimov et al., 2009). HipAEc is also regulated by trans autophosphorylation at a conserved serine (Ser150) situated in a loop near the active site termed the "Gly-rich loop", which is required for ATP binding and functionally (but not structurally) similar to the P loop of eukaryotic kinases (Huse & Kuriyan, 2002, Schumacher et al., 2012). Two discrete conformations of the Gly-rich loop are observed depending on the phosphorylation state of Ser150 and whether substrate (ATP) is bound or not (Schumacher et al., 2015, Schumacher et al., 2012, Schumacher et al., 2009). In the unphosphorylated state and in the presence of ATP, the flexible Gly-rich loop is found in an inward conformation in which main chain amino groups coordinate several of the ATP phosphate groups and the kinase is active (Schumacher et al., 2009). Phosphorylation of Ser150 results in strong interactions of the phosphate group with several active sites residues, which causes ejection of the Gly-rich loop and prevents binding of ATP, thus inactivating the kinase (Schumacher et al., 2012, Wen et al., 2014).
In addition to the canonical hipBA locus, enteropathogenic E. coli (EPEC) O127:H6 contains a partially homologous, tricistronic Hip system; hipBST, encoding three separate proteins. HipB, by analogy to HipBA, contains a putative DNA-binding HTH domain while HipT is a small HipA-like kinase the specifically targets tryptophanyl-tRNA synthetase, TrpS (Gerdes et al., 2021, Vang Nielsen et al., 2019). Surprisingly, the third protein, HipS, which at the sequence level corresponds to the N-terminal subdomain 1 of HipA (N-subdomain 1), was shown to function as the antitoxin of HipT but the molecular mechanism by which HipS neutralizes HipT is currently not known (Vang Nielsen et al., 2019). Phylogenetic analysis has demonstrated that HipT and HipA toxins are closely related within a very broad superfamily of HipA-homologous kinases (Gerdes et al., 2021). HipT has two phosphoserine positions, Ser57 and Ser59, in its Gly-rich loop, which are both modified by trans auto-phosphorylation in vivo suggesting that they might play a role in regulating the activity of HipT (Vang Nielsen et al., 2019).
Here, we determine a 2.9 Å crystal structure of a kinase-inactive HipBSTD233Q complex, revealing that the overall structure is markedly different from E. coli HipBA but similar to a HipBA complex from S. oneidensis. In this structure, HipT interacts tightly with HipS and that despite the lack of phosphorylation we find the Gly-rich loop in an outward, inactive confirmation similar to phosphorylated HipA. We show that this is likely a result of HipS neutralization of HipT through insertion of a bulky Trp residue into the kinase active site. Next, we show that both of the serine residues in the Gly-loop are essential for HipT toxicity in vivo and that auto-phosphorylation of Ser59 prevents neutralization by HipS. We next determine crystal structures of two kinase active HipBST variants revealing that autophosphorylation takes place at both Ser57 and Ser59 despite the lack of in vivo toxicity. Surprisingly, phosphorylation of Ser57 or Ser59 does not affect the conformation of HipT inside the HipBST complex, which remains in an inactive state with the Gly-rich loop ejected, which we attribute to the presence of HipS. Finally, we show that the flexibility of the complex in solution varies in a phosphorylation-dependent manner and use a phylogenetic analysis to demonstrate that variations of the HipT Gly-rich loop serine residues correlate with separate clades of proteins.
E. coli HipBST forms an heterohexameric complex with HipT in a canonical inactive conformation
To understand the molecular and structural basis for the functional differences between the HipBA and HipBST toxin-antitoxin systems, we determined the crystal structure of HipBSTD233Q, containing the inactive HipT D233Q kinase variant, to 2.9 Å by molecular replacement using HipAEc as search model (Figure 1a and 1b). The refined structure (R=19.5%, Rfree=22.8%) has two copies of each of the three proteins in the asymmetric unit and is generally well-defined for HipB and HipT, while HipS appears to be more flexible. The higher-order HipBST complex structure consists of two HipST toxin:antitoxin complexes separated by a dimer of HipB proteins. This differs markedly from the HipBAEc complex, in which two HipA toxins are closely packed head-to-tail around two HipB antitoxins (Figure 1c, top) but is reminiscent of the structure of a HipBA complex from Shewanella oneidensis (HipBASo), which was crystallised in complex with DNA (Figure 1c, bottom) (Wen et al., 2014).
HipB consists of an α-helical bundle of four helices, including the central, DNA-binding HTH motif (residues 62-83), which is exposed at the bridge of the hetero-hexamer (Figure 1b, green). The helical bundle is followed by a small β strand that forms an antiparallel sheet by pairing with the corresponding region in the neighbouring HipB protein, thus forming a homodimer. The N-terminus of HipB (residues 1-35) is not visible in the electron density, but AlphaFold2 prediction indicates that it forms a single α-helix (Figure 1 supplement 1a) (Jumper et al., 2021). For HipBAEc, a hydrophobic residue at the C-terminus of HipB was proposed to reach a cleft between the N-subdomain 1 and the core kinase domain of HipA and function as antitoxin by restricting domain movements required for catalysis (Figure 1c) (Schumacher et al., 2009). HipB from HipBST does not have such a terminal residue (Figure 1 supplement 1a), suggesting a different mechanism of toxin neutralisation. With respect to DNA binding, superposition of HipBST with HipBASo in complex with DNA (Figure 1 supplement 2a) shows that HipT has several conserved, positively charged residues near the expected location of the DNA phosphodiester backbone, suggesting that HipBST might bind DNA in a similar way.
HipS interacts directly with HipT and consists of a solvent-facing, five-stranded anti-parallel β sheet of which four strands are located in the N terminus (residues 1-52). A small domain of three helices (residues 53-94) forms the interface to HipT and is followed by the fifth β strand (residues 95-102, Figure 1b). Overall, the fold of the HipST heterodimer is very structurally similar to the longer HipA kinase (Figure 1 supplement 2b), as expected from sequence analysis (Figure 1 supplement 1b). Thus, HipS is structurally homologous to the N-subdomain 1 of HipA (Schumacher et al., 2012), both with respect to its overall fold and to its orientation with respect to HipT (Figure 1 supplement 2b). As a result of this, the HipS C-terminus overlaps with an extended linker that bridges N-subdomains 1 and 2 in HipA (Figure 1d and figure 1 supplement 2b). HipT corresponds structurally to the core kinase domain of HipA, except for an additional N-terminal mini-domain of unknown significance (residues 1-41, Figure 1a and 1e). In this structure, the Gly-rich loop of HipT is found in an outward-facing conformation despite the absence of Gly-rich loop phosphorylation, suggesting that HipT is inactive (Figure 1 supplement 2c).
In conclusion, we find that the overall structure of the tripartite HipBST complex is markedly different from the canonical E. coli HipBA complex, but structurally similar to a HipBA complex from S. oneidensis. Key features of HipB proposed to be involved in the kinase toxin inhibition mechanism in both HipBAEc and HipBASo are missing. Finally, HipT appears in a canonical inactive conformation despite being unphosphorylated, suggesting a different mechanism of toxin inactivation and inhibition.
HipS inhibits HipT through insertion of a hydrophobic residue near the active site
The observation that HipT is found in a canonical, inactive conformation despite being unphosphorylated led us to hypothesize that HipS might neutralize HipT by direct interaction. This would be consistent with the observation that HipS functionally acts as the antitoxin of the HipBST system in vivo (Vang Nielsen et al., 2019). There are three major areas of contact between the two proteins in the HipBST complex, involving both hydrogen bonds, hydrophobic and charged interactions (Figure 2a). Importantly, HipT and HipS were found to interact directly very close to the Gly-rich loop, where Glu63 from HipS points directly into the HipT active site while HipS Trp65 is wedged into a deep cavity formed by the Gly-rich loop in its outward conformation (Figure 2a, right and 2b). Of these, Glu63 is conserved in the N-subdomain 1 of the longer HipA toxins (Figure 1 supplement 1b), and therefore less likely contribute to functional differences between the two systems. Trp65, on the other hand, is found in many HipS orthologues (Figure 2 supplement 1a) and moreover, the region surrounding it is structurally different from the corresponding region in HipAEc (Figure 2b). Based on this, we hypothesized that insertion of the bulky Trp residue near the HipT active site might sterically block transition of the Gly-rich loop from its outward to the inward conformation, likely required for ATP binding and kinase activity. To test this, we generated a HipSW65A variant and probed its ability to neutralize HipT toxicity upon ectopic expression of the two proteins using a two-plasmid system in E. coli MG1655. Dilution spot assays confirmed that wt HipS, but not HipSW65A, can neutralize toxicity of HipT in vivo (Figure 2c). This result was reproduced by growth assays in liquid culture (Figure 2 supplement 1b). In conclusion, we show that HipS interacts directly with the HipT active site and neutralizes the kinase by insertion of a conserved tryptophan residue, likely preventing structural transition from the inactive to the active conformation. This mode of inhibition where the toxin is sterically and physically blocked by the antitoxin thus closely resembles that observed in other toxin-antitoxin systems.
The two phosphoserine positions in HipT are important for toxicity
To understand the functional implications of the two phosphoserine positions in HipT, Ser57 and Ser59 (Vang Nielsen et al., 2019), we initially decided to investigate their role in toxicity. For this, we designed a set of HipT variants with the serine residues individually substituted for either alanine (A, phosphoablative) to prevent autophosphorylation, or aspartic acid (D, phosphomimetic) to mimic the phosphorylated form. Growth assays in liquid culture showed that both wild-type HipT (S57IS59; SIS), as well as HipT D57IS59 (DIS) and S57ID59 (SID), caused growth inhibition upon overexpression in E. coli MG1655 (Figure 3a), demonstrating that the phosphomimetic mutations do not prevent toxicity, and therefore, kinase activity of HipT towards its target. Moreover, since it has previously been shown that only the HipT Gly-rich loop never is observed in doubly phosphorylated form with both Ser57 and Ser59 modified simultaneously, it is unlikely that the effects are due to autophosphorylation of the remaining serine residue in either case (Vang Nielsen et al., 2019). In addition, while co-expression of HipS neutralized the toxicity of HipT SIS and SID variants, we could not make the antitoxin neutralise the toxicity of the HipT DIS variant, suggesting that Ser57, but not Ser59, negatively impacts the function of HipS as antitoxin. Finally, alanine substitutions at either Ser position almost (S57IA59 ; SIA) or completely (A57IS59; AIS) abolished HipT toxicity (Figure 3a). Thus, both serine residues (in either phosphorylation state) are important for HipT toxicity. In summary, we find that introduction of a phosphomimetic mutation at Ser57 negatively affects the function of HipS as antitoxin under conditions where the HipT kinase is toxic. This represents a clear difference from the regulation of HipA, where phosphorylation of Ser150 inactivates the toxin (Schumacher et al., 2012).
The active site conformation of HipT is independent of phosphorylation
To understand how phosphorylation of the two serine residues affects the structure of the HipBST complex, we determined two additional crystal structures in context of the HipT AIS (2.4 Å) and SIA (3.4 Å) variants (Table 1), which are non-toxic in vivo, but in principle maintain an intact kinase active site. Inspection of the difference maps in the vicinity of the active site revealed that in both cases, additional electron density was present close to the non-mutated serine positions in each structure (Figure 3 supplement 1), none of which were not observed for HipBSTD233Q. For the HipBST AIS variant, we found a strong signal corresponding to phosphorylation of Ser59 in both copies of HipT in the asymmetric unit (Figure 3 supplement 1a), while for the HipBST SIA variant, phosphorylation of Ser57 was present but incomplete and could only be confidently modelled in one of the two HipT molecules in the asymmetric unit (Figure 3 supplement 1b). In its phosphorylated state, Ser59 (P-Ser59) forms strong interactions to Lys161, His212, and the catalytic Asp210 inside the active site (Figure 3b, top), while P-Ser57 is located further away and forms hydrogen bonds to Tyr162 and Asp210 in its phosphorylated state (Figure 3b, bottom). In both structures, however, the Gly-rich loop maintains a similar outward conformation to the one observed before, likely incompatible with ATP binding. Taken together, the structures suggest that HipT is inactive in the context of the HipBST complex regardless of the phosphorylation state of the serine residues in the Gly-rich loop. This is distinctly different from HipA, where phosphorylation of Ser150 was shown to convert the kinase from an active to an inactive form by causing ejection of the Gly-rich loop (Schumacher et al., 2012). To support these observations, we analysed the purified HipBST SIA and AIS variants on Coomassie-stained Phos-tag SDS-PAGE gels, which can separate different phosphoprotein species, using the inactive HipTD210A variant as control for unphosphorylated HipT. This experiment was carried out both before and after separation on a Heparin column which we found is able to separate complexes based on the phosphorylation state of HipT. This confirmed that HipT AIS primarily is found on the phosphorylated form (Figure 3c, upper band), while HipT SIA is mostly on the unphosphorylated form (Figure 3c, lower band).
In summary, we conclude that phosphorylation of neither of the two phosphoserine positions in HipT affects the conformation of the Gly-rich loop in the context of the HipBST complex, and in all cases the loop remains in an outward conformation suggesting the kinase is not in its canonical, activated state. This supports that phosphorylation itself is not responsible for inactivation of the kinase and is consistent with a specific role for HipS in this process. Surprisingly, autophosphorylation still occurs in both the SIA and AIS variants demonstrating that the kinase is still active and consequently, that the lack of toxicity observed for these variants might relate to target binding rather than kinase inactivation. Finally, we note that the position of the Ser59 phosphate group in the AIS variant (HipTS57A) corresponds closely to that of the gamma-phosphate of ATP in the active site of HipA, suggesting that phosphorylation in fact may stabilise the active site in a similar way to ATP (Figure 3d).
HipBST dynamics in solution is affected by the phosphorylation state
To finally understand if the two phosphoserines affect HipBST complex stability, we constructed a kinase-inactive version of the complex containing the HipTD210A active site mutant and analysed three different, phosphomimetic states of the Gly-rich loop (HipT SIS, DIS, and SID) in this context by purifying the HipBST complex using a C-terminal His-tag on HipT. Intriguingly, for the complex purified with the wildtype SIS Gly-rich loop, the band corresponding to HipS was missing (Figure 4 supplement 1a). As Asp210 interacts directly with Ser59, this supports that the serine residues are important for stabilisation of the active site and the Gly-rich loop in a conformation compatible with HipS binding as well as activity towards the target. Analysis of the purified samples of the three variants by small-angle X-ray scattering (SAXS) gave very similar intensity curves suggesting that the complexes have similar sizes and shapes in solution, despite the missing HipS protein in the SIS variant (Figure 4a). All Guinier plots show a linear behaviour at low angles demonstrating the absence of large aggregates (Figure 4 supplement 1b) while the indirect Fourier transformation (Figure 4 supplement 1c) and pair distance distribution functions (Figure 4 supplement 1d) support similar, overall structures. Interestingly, the maximum diameter of 200 Å is significantly larger than the crystal structure (max. diameter 135 Å), consistent with analysis of the forward scattering that revealed partial oligomerisation of the samples with an average mass corresponding to roughly a dimer of the HipBST heterohexamer. Finally, the normalised Kratky plot agrees with a relatively compact structure without significant random coil content (Figure 4 supplement 1e).
We next attempted to model the SAXS data using the crystal structure of HipBST, by including a structural prediction of the missing HipB N-terminus and a hydration layer (Figure 4). As expected from the predicted oligomerisation, this model was in relatively poor agreement with the measured data, so we allowed individual domains of the complex to undergo rigid-body motion while imposing C2 symmetry while at the same time refining the degree of oligomerization (Steiner et al., 2018). For each data set, the model with the best fit (lowest reduced χ2) to the data out of 10 runs was selected as the representative model. These three models (Figure 4b) all gave good fits to the data (Figure 4a) with varying degrees of oligomerisation from 1.7-1.8 and an average distance between the heterohexamers ∼80 Å. Consistent with the biochemical data, the SAXS data from the HipT SIS variant were matched very well by a model in which HipS was omitted but the disordered N-terminus of HipB was included as a helical element (Figure 4, blue). In this structure, the two HipT modules have rotated slightly to open up the cleft between them as compared to the crystal structure (Figure 4b, blue, arrows). On the contrary, data measured for both the SID (Figure 4a and b, red) and DIS (Figure 4a and b, orange) variants both matched the complete HipBST crystal structure (with the N-terminal HipB extension) with only minor adjustments of the domains. Taken together, SAXS analysis suggests that the HipBST complex is dynamic in solution in a phosphoserine-dependent way and that the phosphorylation state of HipT influences the inhibitory function of HipS. This dynamic behaviour likely explains how it is possible for the inactivated HipBST TA complex to undergo auto-phosphorylation in vivo.
The dual autophosphorylation sites are conserved among HipT kinases
To understand if Gly-rich loops with two phosphoserine positions are common among HipT-like kinases, we generated a phylogenetic tree of the 48 known HipT orthologues in bacterial genomes (Gerdes et al., 2021) (Figure 5a). Interestingly, investigation of this tree revealed that it largely separates the kinases by the configuration of the Gly-rich loop. The largest branch, which we named the SΨS group, contains HipT kinases with two serine residues separated by a small aliphatic amino acid, corresponding to the known autophosphorylation sites in HipT of E. coli O127:H6, Ser57 and Ser59 (Figure 5a). In many of the remaining HipT orthologues, this region contains various Ser/Thr motifs, including (S/T)ΨT, (S/T)ΨP, SΨQ, and ΨΨT, but invariably with a central, small aliphatic residue (Ψ: Ile, Leu, or Val). Overall, the second Ser/Thr position appears less conserved than the first, and some kinases even contain a proline (P) or glutamine (Q) here, suggesting different regulation patterns for different HipT kinases with Ser57 playing a more important role. This supports the functional analysis showing that Ser57 affects neutralization by HipS in vivo and is less prone to autophosphorylation in the SIA variant. The remaining sequence of the Gly-rich loop across a range of HipT kinases is highly similar to HipAEc (Figure 5b and 5c). Together, this allowed us to generate a consensus motif for the HipT Gly-rich loop, which includes several conserved hydrophobic positions in addition to the two Ser/Thr residues (Figure 5d). In summary, phylogenetic analysis reveals that a large fraction of HipT kinases contain a conserved SΨS motif with two potential phosphorylation positions in their Gly-rich loops, supporting that both Ser59, and in particular Ser57, play an important role in the regulation of the HipBST TA system.
In this study, we present a detailed structural and functional analysis of the tripartite HipBST toxin-antitoxin system from enteropathogenic E. coli O127:H6, which elucidates the neutralising mechanism of the unusual HipS antitoxin and dissects the functional roles of the two phosphoserine residues in HipT, Ser57 and Ser59. The crystal structure of the kinase-inactive HipBSTD233Q complex shows that HipS interacts in a similar way to the corresponding N-subdomain 1 in HipA but neutralises HipT by direction interaction involving insertion a large, bulky residue, Trp65, into the kinase active site. Trp65 is conserved in some, but not all HipS proteins, and is located in a motif that differs both structurally and at the sequence level from the corresponding region in HipA, which is otherwise very similar to HipS (Gerdes et al., 2021). Using multiple experimental approaches, we go on to show that both phosphomimetic and phosphoablative states of HipT impact toxicity and the interaction with the antitoxin HipS. Together, this suggests that the structural basis for the role of HipS as antitoxin in the HipBST system involves preventing the Gly-rich loop from transitioning from the outward, inactive state, to a conformation compatible with ATP binding. In some organisms, the residue corresponding to Trp65 is a proline, which may point to another way of preventing activation via steric hindrance (Figure 2 figure supplement 1a).
Recently, structures of a HipT orthologue from Legionella pneumophila (HipTLp) in complex with the non-hydrolyzable ATP analogue (AMPPNP), HipSLp in complex with HipTLp, and a HipS from Haemophilus influenzae have been determined (Lin et al., 2022, Zhen et al., 2022, Koo et al., 2022). HipTLp belongs to the SΨQ group, and phosphorylation of the Gly-rich loop can thus take place only at the position corresponding to Ser57 in HipTEc. This allows us to further dissect the functions of the two phosphorylation sites in HipTEc. Interestingly, and despite being phosphorylated, HipTLp is found with its active site loop in an inward and active conformation with nucleotide bound. In this conformation, the phosphoserine moiety is stabilized by two conserved arginine residues in a RxDR motif, which is found both in HipTEc and HipAEc (Figure 1 supplement 1). In HipA, however, the Gly-rich loop does not contact the RxDR motif in either phosphorylation state, suggesting that this interaction is a unique feature of HipT kinases (Schumacher et al., 2012). Gln56 (SΨQ), which corresponds to Ser59 in HipTEc and arguably in some respects mimics a phosphorylated serine, interacts with two residues, Asp145 and Lys201. These have chemically similar counterparts in HipTEc (Gln151 and Arg214) and thus might interact with the phosphorylated Ser59 when the Gly-rich loop is in its inward state, i.e., in the absence of HipS. Interactions at both sites (Ser57 to the RxDR motif, Ser59 to Gln151/Arg214) would stabilize the loop in its inward state and possibly affect the function of HipS as antitoxin. Moreover, since interaction of the phosphoserine with two arginine side chains at Ser57 would be much stronger than to a Gln and an Arg, due to the ionic character of the side chains, this could also explain why P-Ser57 interaction is stronger and thus potentially also why HipS is unable to inactivate the S57D phosphomimetic HipT mutant (DIS).
Finally, we show that the serine residues in the HipT Gly-rich loop (Ser57 and Ser59) are important for toxicity but apparently not for auto-phosphorylation as we observe auto-phosphorylation in both the AIS and SIA variants of HipT. This is compatible with a model in which auto-phosphorylation serves to both control target (TrpS) phosphorylation and HipS binding, which could explain the need for two sites. The molecular mechanism controlling at which site auto-phosphorylation takes places is still not known, but the structures of HipBST AIS and HipBST SIA suggest that subtle changes in the loop can affect this (Figure 3b).
Together, our data support a complex network model for the interplay between the active site and the Gly-rich loop depicted in Figure 6. In this model, residues involved in both ATP binding and catalysis, the phosphoserine residues in the Gly-rich loop, and parts of HipS form a tightly interconnected network of interactions. The result of this complexity is that posttranslational modification and/or mutation of any single residue can have multiple downstream effects. For example, inactivation of the HipT active site through the D210A mutation appears to also affects the position of the Gly-rich loop as well as HipS binding, via direct interactions to the serine residues. Likewise, the position of the Gly-rich loop and binding of HipS are mutually interconnected so that HipS controls whether the loop can transition from an outward to an inward state and conversely, the serine residues and their phosphorylation state likely also control with what affinity HipS binds. This shifts the role of HipS from that of a classical antitoxin that simply blocks and inhibits the toxin active site to that of an allosteric enzyme modulator. We do not yet understand whether auto-phosphorylation can take place with HipS bound or requires its release but note that our model is consistent with both these scenarios. The complexity of the network therefore makes it difficult to associate single roles to specific residues and/or modifications and requires that the full system is considered for each functional state.
The authors are indebted to the beamline staff at P14 in EMBL Hamburg, and BioMAX in MaxIV Lund for help during data collection. This project was funded by grants from the Novo Nordisk Foundation (NNF18OC0030646 to D.E.B.) and Danish Natural Research Foundation’s Centre of Excellence for Bacterial Stress Response and Persistence (grant number DNRF120).
The authors confirm that there are no competing interests.
Conceptualization, K.G., M.A.S, and D.E.B.; Investigation, R.L.B., S.V.N., F. B., J. S.P. and J.L.; Formal Analysis, R.L.B, S.V.N., R. B. S., F.B., and J.S.P; Writing – Original Draft, R.L.B., S.V.N., and D.E.B.; Writing – Review & Editing, R.L.B., S.V.N., R. B. S., J.L., J.S.P., K.G., M.A.S., and D.E.B; Visualization, R.L.B., S.V.N., M.A.S., and D.E.B.; Funding Acquisition, D.E.B. and K.G.; Resources, J.S.P., K.G., M.A.S. and D.E.B; Supervision, K.G., J.S.P., M.A.S., and D.E.B.
Materials and methods
Strains and plasmids
Strains and plasmids used or generated in this study are listed in Table S2, and DNA oligonucleotides in Table S3. The construct expressing the HipTS57A variant was constructed by PCR mutagenesis (Table S3) as previously described (Vang Nielsen et al., 2019).
Protein purification and structure determination
Expression of the E. coli O127:H6 HipBST complex was done by introduction of the constructs pSVN78 (HipBSTS57A), pSVN96 (HipBSTD233Q), and pMME3 (HipBSTS59A), into E. coli BL21 DE3. All constructs encode HipB, HipS, and C-terminal hexa-histidine tagged version of HipT, with genes separated and including individual, optimised Shine-Dalgarno sequences. All constructs contained an isopropyl-D-1 thio-galactopyranoside (IPTG)-inducible promoter and ampicillin resistance gene for selection. For each construct, 2 L cultures of E. coli BL21 DE3 grown in LB medium to a cell density of OD600 = 0.6 were induced with a final concentration of 1 mM IPTG and left to express overnight at 20°C with vigorous shaking. Cell pellets were resuspended in lysis buffer (50 mM Tris-HCl, pH 7.5, 300 mM NaCl, 20 mM imidazole, 5% glycerol, 5 mM BME, 1 mM PMSF) and lysed by sonication. The lysate was cleared by centrifugation and applied to a 5 mL HisTrap HP column (Cytiva) equilibrated in lysis buffer and subsequently washed in wash buffer (50 mM Tris-HCl, pH 7.5, 300 mM NaCl, 40 mM imidazole, 5 mM BME), before eluting with 50 mM Tris-HCl, pH 7.5, 300 mM NaCl, 400 mM imidazole, 5 mM BME. The eluate was applied directly to a 5 mL Heparin HF column equilibrated in 70% Buffer A (50 mM Tris-HCl, pH 7.5, 5 mM BME) and 30% Buffer B (50 mM Tris-HCl, pH 7.5, 1M NaCl, 5 mM BME) attached to an ÄKTA Pure system (Cytiva). This step separated the population into fully formed HipBST complexes that bound to the column, and various partial complexes that were washed off. Final separation was achieved following concentrating the Heparin elution to approximately 8 mg/mL using a 30 kDa MWCO Vivaspin filter (Sartorius), by applying the sample to a Superdex 200 10/300 GL (Cytiva) column equilibrated in gel filtration buffer (20 mM Tris-HCl, pH 7.5, 300 mM NaCl, 5 mM BME). Finally, the sample was concentrated to a protein concentration of 5 mg/mL by spin filtration before crystallisation. Crystals of HipBST mutants grew as clusters of thin plates in 2 μL drops containing a 1:1 ratio of protein to the crystallisation buffer, which consisted of 0.1 M Bicine, pH 9, 8% MPD, set up against a reservoir of 200 μL of the same condition. Crystals were cryoprotected in crystallisation buffer supplemented with 25% MPD before freezing in liquid N2.
Data collection for HipBSTS57A was carried out at the P14 beamline at EMBL/DESY, Hamburg and for HipBSTS59A and HipBSTD233Q at the BioMAX beamline at MAX IV in Lund, Sweden. For HipBSTS59A and HipBSTD233Q, 7,200 images were collected with an oscillation of 0.1° and a transmission of 100% while for HipBSTS57A, 3,600 images were collected with an oscillation of 0.1° and a transmission of 70%. All data were processed in XDS (Kabsch, 2010), using the CC½ value after scaling to set the diffraction limit (Karplus & Diederichs, 2012). The space group was confirmed using Pointless (Evans, 2006), and all structures were determined using molecular replacement with Phaser (McCoy et al., 2007) from the CCP4 suite (Winn et al., 2011). For HipBSTS57A, a heavily truncated structure of E. coli HipA was used as search model, while for HipBSTS59A and HipBSTD233Q, the preliminary structure of HipBSTS57A was used. Iterative refinement was carried out in Buster (Smart et al., 2012) using one big cycle of 20 small cycles of refinement during model building, and 5 big cycles of 100 small cycles of refinement for the subsequent polishing. Non-crystallographic symmetry (NCS) restraints were used throughout since the asymmetric unit (ASU) contained a dimer, but some refinement rounds without NCS were also included to allow the model to adapt to differences between the molecules in the ASU. Automatic water placement was performed in Buster and water molecules were manually pruned by inspecting the electron density. Model building was performed in Coot (Emsley et al., 2010). All structures were validated by the MolProbity server, and Rama-Z scores as calculated by Phenix (Sobolev et al., 2020). Final R-work/R-free for HipBSTD233Q, HipBSTS57A, and HipBSTS59A, were 0.19/0.23, 0.21/0.24, and 0.20/0.24, respectively.
Spot assays and growth curves
Cultures of relevant strains of E. coli were grown in liquid YT medium or MOPS minimal medium including 0.2% glucose at 37°C with shaking at 160 rpm. YT agar plates were used as solid medium and were incubated at 37°C for approximately 16 h. For selection, media was supplemented with 25 µg/mL chloramphenicol, 30 µg/mL ampicillin, and/or 25 µg/mL kanamycin. For spot assays, E. coli cells were grown as overnight cultures, diluted to obtain identical OD600, centrifuged at 5,000 rpm for 5 min, washed in phosphate-buffered saline (PBS), and serially diluted before being spotted onto YT agar plates containing the indicated amount of inducer or repressor. Gene expression from plasmids carrying the pBAD promoter was induced by a final concentration of 0.2% arabinose and repressed by 0.2% glucose. Gene expression from plasmids carrying the PA1/O4/O3 promoter was induced by a final concentration of 200 or 500 µM IPTG, as indicated. Growth experiments were done in YT medium with addition of relevant antibiotics, diluted from overnight cultures and grown exponentially for at least 4 h until a constant doubling time was observed. At OD600 ∼0.2, 0.2% arabinose was added to induce expression of wildtype HipT or autophosphorylation mutants. After another 1.5 h, 200 µM IPTG was added to induce expression of HipS. For each repetition, an independent colony from the strain was used to start separate cultures. No outliers were rejected.
15% Phos-tag acrylamide gels (Wako) were cast according to the manufacturer’s guidelines, except that 100 µM Phos-tag acrylamide was added to ensure proper separation between phosphorylated and unphosphorylated HipT. The unphosphorylated inactive HipTS57A+D210A control was expressed from pSNN2. The gel was run at 4 °C until the loading dye reached the bottom of the gel and visualized using standard Coomassie Blue staining as for normal SDS-PAGE gels.
Small-angle X-ray scattering measurements and analysis
HipBSTD210A, HipBSTS57D,D210A, and HipBSTS59D,D210A were expressed from pRBS1, pRBS2 and pRBS3, and purified as described above. SAXS measurements were performed using an optimized NanoSTAR instrument (Bruker AXS), which uses a high brilliance Ga metal-jet X-ray source (Excillum), special long optics, a two-pinhole collimation with a custom ‘scatterless’ slit pinhole, and a VÅNTEC-2000 (Bruker AXS) microgap 2D gas proportional detector (Lyngsø & Pedersen, 2021). Samples and buffer standards were measured in the same flow-through quartz capillary, and the scattering from the buffer was subtracted before the data were converted to absolute scale using the scattering from water. SAXS intensity data, I(q), were analysed as a function of the modulus of the scattering vector q = 4π sin(θ) /λ, where 2θ is the scattering angle and λ is the wavelength of the X-rays. Guinier analysis, giving radius of gyration Rg and the forward scattering I(0) the forward scattering, and an indirect Fourier transformation were then calculated (Glatter, 1977, Pedersen et al., 1994), giving the same parameters and also the pair distance distribution function p(r) and the maximum diameter Dmax of the particles. The predicted molar mass was then calculated by , where NA is Avogadro’s number, c is the mass concentration, and Δρm = 2.00 × 1010 cm g-1 is the typical excess scattering length per unit mass for proteins. The data was also plotted in a normalized Kratky plot of (qRg)2 I(q)/I(0) versus qRg to assess compactness and check for contributions from random coil parts. For structural modelling, a AlpheFold2-predicted structure for N-terminus of HipB was added to initially have a molecular model with the correct molecular mass. The predicted SAXS curve of the model was then calculated as described in (Steiner et al., 2018) where the hydration layer is described by dummy atoms. For the calculation, the Debye equation is used (Debye, 1915) using an average form factor for all non-hydrogen atoms:
which defines the form factor P(q), and where σ = 1.0 Å, b;i is, respectively, equal to the average excess scattering length of a non-hydrogen atoms for the protein and equal to an average excess scattering length for a hydration dummy atom. The parameter dij is the distance between the i’th and the j’th atom. Poor agreement was observed for all the data sets, so to improve the modelling, rigid-body refinement (RBR) of the structure was performed using three bodies for HipB (residues 1-31, 32-43, and 44-107, respectively), one body for the HipS, and three bodies for HipT (residues 2-59, 60-169, and 170-331, respectively). Distance restraints were then added at the domain boundaries with additional restraints between the domains (Glu A107-Val D101, Pro B55-Gly C154, Trp B 65-Gly C60, Gly D100-Aap F188, Leu D6-Arg F291, Val C147-Gly B94) based on analysis of the crystal structure, and finally a restraint of excluded volume was added. The Guinier and IFT analysis indicated some oligomerization in the samples, and therefore, a structure factor S(q) = 1 + (N – 1) sin(qD)/qD (Larsen et al., 2020) for a dimer was included. In the expression, where N is the average of the number of proteins in the oligomer, and D is the distance between the two copies in the dimer. The structure factor was included in the decoupling approximation (Kotlarchyk & Chen, 1983):
with corresponding amplitude,
where di,CM is the distance of the i’th atom from the scattering center of mass. During the optimization of the structure, the two parameters of the structure factor, N and D, were also varied.
The previously identified set of 48 HipT orthologues (Gerdes et al., 2021) was used for sequence alignment by Clustal Omega (Sievers & Higgins, 2018) at www.ebi.ac.uk and imported into Jalview (Waterhouse et al., 2009). The phylogenetic tree was visualized using iTOL (Letunic & Bork, 2019). Reconstruction of the phylogenetic tree was accomplished using IQ-TREE that uses the Maximum Likelihood approach and Ultrafast bootstrapping via the CIPRES module in Genious Prime (Minh et al., 2020).
Data and software availability
The accession numbers for the structures reported in this paper are PDB: 7AB3 (HipBST SIA), 7AB4 (HipBST AIS), and 7AB5 (HipBSTD233Q).
- Chaperone addiction of toxin-antitoxin systemsNat Commun 7
- Zerstreuung von RöntgenstrahlenAnnalen der Physik 351:809–823
- Features and development of CootActa Crystallogr D Biol Crystallogr 66:486–501
- Partition of tRNA synthetases into two classes based on mutually exclusive sets of sequence motifsNature 347:203–206
- Scaling and assessment of data qualityActa Crystallogr D Biol Crystallogr 62:72–82
- New kinase regulation mechanism found in HipBA: a bacterial persistence switchActa Crystallogr D Biol Crystallogr 65:875–879
- Phylogeny Reveals Novel HipA-Homologous Kinase Families and Toxin-Antitoxin Gene OrganizationsmBio 12
- Molecular mechanism of bacterial persistence by HipAMol. Cell 52:248–254
- A new method for the evaluation of small-angle scattering dataJournal of applied crystallography 10:415–421
- New toxins homologous to ParE belonging to three-component toxin-antitoxin systems in Escherichia coli O157:H7Mol. Microbiol 76:719–732
- The protein kinase family: conserved features and deduced phylogeny of the catalytic domainsScience 241:42–52
- Toxins, Targets, and Triggers: An Overview of Toxin-Antitoxin BiologyMol. Cell 70:768–784
- Synthesis of guanosine tetra- and pentaphosphate requires the presence of a codon-specific, uncharged transfer ribonucleic acid in the acceptor site of ribosomesProc. Natl. Acad. Sci. U S A 70:1564–1568
- The conformational plasticity of protein kinasesCell 109:275–282
- Highly accurate protein structure prediction with AlphaFoldNature 596:583–589
- Bistable Expression of a Toxin-Antitoxin System Located in a Cryptic Prophage of Escherichia coli O157:H7mBio 12
- XdsActa Crystallogr D Biol Crystallogr 66:125–132
- Linking crystallographic model and data qualityScience 336:1030–1033
- The Haemophilus influenzae HipBA toxin–antitoxin system adopts an unusual three-component regulatory mechanismIUCrJ 9:625–631
- Analysis of small angle neutron scattering spectra from polydisperse interacting colloidsThe Journal of Chemical Physics 79:2461–2469
- Assessment of structure factors for analysis of small-angle scattering data from desired or undesired aggregatesJournal of applied crystallography 53:991–1005
- Toxin-Antitoxin Systems as Phage Defense ElementsAnnu. Rev. Microbiol
- Interactive Tree Of Life (iTOL) v4: recent updates and new developmentsNucleic Acids Res 47:W256–W259
- Lin, J.D., P.J. Stogios, K.T. Abe, A. Wang, J. MacPherson, T. Skarina, A.-C. Gingras, A. Savchenko & A.W. Ensminger, (2022).
- A high-flux automated laboratory small-angle X-ray scattering instrument optimized for solution scatteringJournal of applied crystallography 54:295–305
- Phaser crystallographic softwareJ Appl Crystallogr 40:658–674
- IQ-TREE 2: New Models and Efficient Methods for Phylogenetic Inference in the Genomic EraMol. Biol. Evol 37:1530–1534
- (p)ppGpp and Its Role in Bacterial Persistence: New ChallengesAntimicrob. Agents Chemother 64
- The aggregation behavior of zinc-free insulin studied by small-angle neutron scatteringEur. Biophys. J 22:379–389
- HipBA-promoter structures reveal the basis of heritable multidrug toleranceNature 524:59–64
- Role of unusual P loop ejection and autophosphorylation in HipA-mediated persistence and multidrug toleranceCell Rep 2:518–525
- Molecular mechanisms of HipA-mediated multidrug tolerance and its neutralization by HipBScience 323:396–401
- ATP binding by glutamyl-tRNA synthetase is switched to the productive mode by tRNA bindingEMBO J 22:676–688
- Clustal Omega for making accurate alignments of many protein sequencesProtein Sci 27:135–145
- Exploiting structure similarity in refinement: automated NCS and target-structure restraints in BUSTERActa Crystallogr D Biol Crystallogr 68:368–380
- A Global Ramachandran Score Identifies Protein Structures with Unlikely StereochemistryStructure 28:1249–1258
- Serine/Threonine Protein Kinases from Bacteria, Archaea and Eukarya Share a Common Evolutionary Origin Deeply Rooted in the Tree of LifeJ. Mol. Biol 430:27–32
- The structure of the N-terminal module of the cell wall hydrolase RipA and its role in regulating catalytic activityProteins 86:912–923
- Serine-Threonine Kinases Encoded by Split hipA Homologs Inhibit Tryptophanyl-tRNA SynthetasemBio 10
- Jalview Version 2-- a multiple sequence alignment editor and analysis workbenchBioinformatics 25:1189–1191
- The bacterial antitoxin HipB establishes a ternary complex with operator DNA and phosphorylated toxin HipA to regulate bacterial persistenceNucleic Acids Res 42:10134–10147
- Overview of the CCP4 suite and current developmentsActa Crystallogr D Biol Crystallogr 67:235–242
- Activation of the Stringent Response by Loading of RelA-tRNA Complexes at the Ribosomal A-SiteMol. Cell 70:95–105
- Molecular mechanism of toxin neutralization in the HipBST toxin-antitoxin system of Legionella pneumophilaNat Commun 13
- The toxin-antitoxin system of the streptococcal plasmid pSM19035J. Bacteriol 187:6094–6105