The L. pneumophila effector Lpg2370 is a Ser/Thr kinase
We first verified whether Lpg2370 is translocated into host cells via the type IV secretion system (T4SS) Dot/Icm as an effector protein. To this end, we performed TEM-1 β-lactamase translocation assay by infecting RAW264.7 macrophages with the fusion protein-expressing L. pneumophila cells grown to post-exponential phase. Vectors expressing TEM-1-RaIF (positive control), TEM-1-FabI (negative control) or TEM-1-Lpg2370 fusion proteins were introduced into the T4SS-competent L. pneumophila strain Lp02 or the Dot/Icm-deficient strain Lp03, which were then assessed for the delivery of the β-lactamase fusions into the host macrophage cells by visual inspection under a fluorescence microscope. Cells infected by Lp02 cells expressing the TEM-RalF fusion protein emitted blue fluorescence, whereas infection with the TEM-FabI-expressing cells did not result in any emission of blue fluorescence by host cells (Fig. 1a). In addition, none of the fusion proteins were detectably translocated upon infection with the Dot/Icm-deficient strain Lp03 (Fig. 1a). Consistent with previous studies24, the Lpg2370-overexpressing L. pneumophila strain Lp02 can be secreted into host cells, though at very low translocation efficiencies, suggesting that Lpg2370 is indeed a L. pneumophila effector protein.
Although previous studies implied that Lpg2370 is an E3 ubiquitin ligase based on sequence similarity with the RING-type E3 ubiquitin ligase FANCL25,26,27, we repeated primary sequence analysis and did not find any significant similarity between the two proteins. However, we found notable sequence identity (~20%) between Lpg2370 and the residues 64–440 of E. coli HipA, an atypical Ser/Thr kinase from the E. coli K-12 strain (Supplementary Fig. 1)28. The results particularly indicated conservation of the sequences corresponding to the P-loop (RISVAGAQ), the signature motif responsible for ATP binding, and the catalytic loop, which contains the catalytic residue D310 required for the kinase activity of E. coli HipA29,30,31 (Fig. 1b). Moreover, comparing the Lpg2370 sequence to the NCBI database using basic local alignment search tool (BLAST) identified Lpg2370 as a HipA-like Ser/Thr kinase. These findings led us to hypothesize that Lpg2370 could be a kinase.
Kinases frequently undergo autophosphorylation on an invariable Ser or Thr residue in the P-loop32. For instance, autophosphorylation of Ser150 on the P-loop has been observed in HipA kinases from E. coli (HipAEc) and the proteobacterium Shewanella oneidensis (HipASo)30,31. To investigate whether Lpg2370 is also autophosphorylated, we incubated purified recombinant Lpg2370 expressed in E. coli with the N6-benzyladenosine-5’-O-(3-thiotriphosphate) (N6-Bn-ATPγS) (Fig. 1c)33. Immunoblotting with anti-N6-Bn-ATPγS antibody detected a protein band corresponding to Lpg2370 (35 kDa) (Fig. 1d), which clearly indicated that Lpg2370 can be autophosphorylated. We next performed LC-MS/MS to identify the autophosphorylation site on Lpg2370. A mass shift of 79.97 Da (m/z = 684.85, z = 2) was consistently observed in the putative P-loop (53-MSVQGVQKK-61), revealing that the residue Ser54 within the P-loop is the autophosphorylation site (Fig. 1e). Taken together, these results suggest that Lpg2370 is a Ser/Thr kinase, though its substrates are currently unknown.
Lpg2370 adopts a Ser/Thr kinase-like fold
To gain deeper insight into the molecular function of Lpg2370, we set out to determine its crystal structure. Diffraction phases for the SeMet-labeled Lpg2370 were determined using the single-wavelength anomalous diffraction method and the final structural mode was refined at 1.46 Å (Table 1).
Like other members of the protein kinase superfamily, Lpg2370 has a globular kinase fold that can be further divided into N-lobe and C-lobe. The N-lobe, which contains the P-loop, is composed of sheets β1–5 sandwiched by helices α1 and α2, whereas the C-lobe is predominantly α-helical and consists of helix bundles α3–α6 and α8–α11 and a short β-sheet β6–8 (Fig. 2a). In line with the results of primary sequence analysis, there is no apparent structural similarity between Lpg2370 and the E3 ligase FANCL (Supplementary Fig. 2a), and residues 4–28 of Lpg2370, which are relatively well aligned with FANCL25, are a part of the typical kinase N-lobe. In agreement with our LC-MS/MS results, we can observe a phosphate group covalently attached to the Ser54 residue (Fig. 2b). However, unlike HipAEc and HipASo whose P-loops are disordered upon serine autophosphorylation29, the electron density of the Lpg2370 P-loop is well defined in the present structure (Fig. 2b). Interestingly, the P-loop of Lpg2370 differs from the counterparts of typical protein kinases in that it contains a single glycine residue34. The positioning of the phosphorylated Ser54 (pSer54) is stabilized by the positively charged side chains of K40, R131, and R134, as well as hydrogen bonds between side chain of Gln56 and side chains of Asp145 and Lys201 (Supplementary Fig. 2b).
Dali search suggested that Lpg2370 shares the highest structural homology score with HipAEc and HipASo (Supplementary Table 1)35. Lpg2370 and HipAEc superimpose with a relatively large root-mean-square deviation (RMSD) value of 3.846 Å over 221 Cα atoms (Fig. 2c, d). In addition to lacking a counterpart to the N-terminal region of HipAEc (i.e., helices α1–α4 and strands β1–β3 of HipA), Lpg2370 differs from HipAEc mainly in the configuration of the N-terminal lobe (Fig. 2c). One prominent difference is that the Lpg2370 P-loop with phosphorylated Ser54 is exposed to solvent in an orientation similar to that of unphosphorylated P-loop of HipAEc, which upon serine autophosphorylation rotates by ~180° and bends away from the N-terminal lobe by 17.3 Å (Fig. 2e, f). Conversely, the six C-terminal α-helices of Lpg2370 and HipAEc, including the catalytic residues and some ATP-binding residues, are almost perfectly aligned (Fig. 2e).
Lpg2368–Lpg2369–Lpg2370 constitute the tripartite HipBST TA system of L. pneumophila
Further analysis of Lpg2370 showed that proteins containing the C-terminal domain of HipA are widespread in bacteria (Fig. 3a). Moreover, the structural similarity between Lpg2370 and E. coli HipA and the fact that E. coli HipA along with HipB from the same genomic locus composes a type II TA system prompted us to examine the locus of lpg2370. Indeed, we found that lpg2370 is preceded by open reading frames (ORFs) of lpg2368 and lpg2369. Analogously to TA systems such as HipBA28, lpg2368, and lpg2369 as well as lpg2369 and lpg2370 overlap by 4 bp. Further analysis suggested that lpg2369 encodes a 102-aa protein similar to the N-terminal region of E. coli HipA and that lpg2368 encodes a 72-aa protein homologous to the helix–turn–helix (HTH) domain of HipB (Fig. 3b and Supplementary Fig. 3a, b). Such locus organization is reminiscent of the HipBST TA module in E. coli O127:H617,18, suggesting that the lpg2368–lpg2369-lpg2370 locus is a potential tricistronic operon encoding component of a HipBST TA system.
In the HipBST TA system of E. coli O127:H6, the toxin HipT (denoted HipTO127) can form a heterotrimeric complex with the antitoxin HipSO127 and the HTH domain protein HipBO12718. We therefore performed size-exclusion chromatography, pull-down assays and isothermal titration calorimetry (ITC) to analyze interactions between Lpg2370, Lpg2369, and Lpg2368. The co-expressed 6×His-tag Lpg2369 and untagged Lpg2370 were co-eluted using Ni affinity chromatography, and the size-exclusion chromatography analysis revealed that the peak is shifted forward by 0.4 mL compared to the peak of Lpg2370 alone, suggesting Lpg2369 can interact with Lpg2370 (Fig. 3c). Moreover, size-exclusion chromatography indicated that Lpg2368 co-elutes with the co-expressed 6×His-tagged Lpg2369–Lpg2370 complex and binds to the Lpg2369–Lpg2370 complex assembled in vitro (Fig. 3c), which was then further confirmed by the pull-down assays (Supplementary Fig. 4). These results suggest that Lpg2370 directly interacts with Lpg2369, whereas Lpg2368 binds to a stable Lpg2369–Lpg2370 complex. Moreover, the results of ITC assays demonstrated that the dissociation constants between Lpg2370 and Lpg2369 and Lpg2370-Lpg2369 complex and Lpg2368 are 42 nM and 1.5 µM, respectively (Supplementary Fig. 5a, b), which is in agreement with the previously published data on HipBSTO12718.
Given the established analogy between the L. pneumophila Lpg2368–Lpg2369–Lpg2370 operon and the HipBST TA system, we next aimed to functionally characterize Lpg2370 by investigating its potential toxicity to host bacteria. Heterogeneous expression of the recombinant Lpg2370 in E. coli had no observable effect on cell growth (Supplementary Fig. 6). The toxicity assays were performed in the L. pneumophila strain Lp02. To avoid undesirable effects from endogenous expression, we prepared deletion strain ∆lpg2368-∆lpg2369-∆lpg2370 (∆3) and examined bacterial growth upon overexpression of recombinant Lpg2370. Overexpression of Lpg2370 significantly inhibited the growth of L. pneumophila, both on plates and in liquid medium (Fig. 3d). Moreover, the catalytically inactive H199A (H/A) mutant failed to inhibit bacterial growth (Fig. 3d), indicating that the kinase activity of Lpg2370 is strictly required for its toxicity (Fig. 3d).
To assess the impacts of Lpg2368 and Lpg2369 on bacterial growth, we inserted lpg2370 into the low-copy-number IPTG-inducible vector pZL507, and lpg2368, lpg2369, or lpg2368–lpg2369 were separately inserted into the plasmid pJL03 with the arabinose-inducible pBAD promoter. Growth and viability of the L. pneumophila Δ3 strain carrying combinations of these plasmids was then monitored. Growth inhibition caused by the expression of Lpg2370 was counteracted by co-expression of Lpg2369, suggesting that Lpg2369 functions as the antitoxin (Fig. 3e). Co-expression of Lpg2368 and Lpg2369 was also found to counteract Lpg2370-dependent growth inhibition, whereas the expression of Lpg2368 without Lpg2369 could not prevent the growth inhibition (Fig. 3e). Taken together, these results are consistent with the findings on the E. coli O127:H6 HipBST module17,18 and demonstrate that Lpg2368, Lpg2369, and Lpg2370 from L. pneumophila constitute the tripartite HipBST TA system18 and will thus hereafter be referred to as HipTLp, HipSLp, and HipBLp, respectively17,18.
The kinase activity of HipTLp is likely independent of P-loop serine autophosphorylation
A comparison of the crystal structure of pHipTLp and the structures deposited in the PDB revealed that the autophosphorylated P-loop in HipTLp adopts an orientation similar to that of the P-loop in the crystal structure of E. coli HipA S150A mutant (Fig. 2f)30. This observation led us to speculate that pHipTLp can bind ATP. Thermal shift assays performed with the purified wild-type HipTLp revealed a 2.5 °C-increase in the melting temperature (Tm) in the presence of non-hydrolysable ATP analogue adenylyl-imidodiphosphate (AMP–PNP), suggesting that pHipTLp indeed binds ATP (Supplementary Fig. 7). Likewise, isothermal calorimetry determined that the dissociation constant between pHipTLp and AMP–PNP was about 70 µM (Fig. 4a), which is within the range of ATP-binding affinity expected for other kinases29.
To elucidate how pHipTLp binds ATP, we determined the crystal structure of pHipTLp in complex with AMP–PNP at 1.36 Å resolution (Table 1). The structure of pHipTLp–ATP reveals that AMP–PNP is bound to the P-loop like in other representative kinases (Fig. 4b). The backbone of pHipTLp in the complex is virtually identical to the apo structure (RMSD = 0.35 Å), with the exception of P-loop that bends towards helix α2 to accommodate the AMP–PNP. ATP (AMP–PNP)-interacting residues appear to be conserved among the bacterial HipT toxins, implying a shared mechanism for ATP binding (Fig. 4c). In pHipTLp, the γ-phosphate of AMP–PNP is stabilized by V58, H199, and D219, the β-phosphate forms hydrogen bonds with Q59, K61, and K85, whereas the α-phosphate interacts with K85 and N202. The adenosine moiety interacts with the main chain of K130 and forms π-stacking interactions with the side chain of F132 (Fig. 4b).
A previous study demonstrated that the kinase activity of E. coli HipA is essential for the growth arrest of host cells36, and cell growth was inhibited when HipTO127 was expressed in E. coli BL21 (DE3) cells18. To confirm the role of the residues involved into the ATP binding in the HipT toxins in vivo, HipTO127 TA is used to perform the growth inhibition assays due to the easy manipulation of E. coli compared to L. pneumophila. To investigate whether the above-mentioned residues responsible for ATP binding are essential for the kinase activity of HipT, we performed in vivo toxicity assays with HipTO127 variants in which residues corresponding to the S54 and the highly conserved ATP-binding residues of HipTLp were substituted with alanine. Intriguingly, mutation on the residues corresponding to S54 of HipTLp (S57A and S57D HipTO127) remain toxic to E. coli cells, whereas substitutions of K64 (K61), K86 (K85), H212 (H199), N215 (N202), and D233 (D219) of HipTO127 (corresponding residues in HipTLp are indicated in parentheses) eliminated the toxic phenotype (Fig. 4d). Taken together, these results suggest that unlike in E. coli HipA, HipT retains the ATP-binding ability independent of the autophosphorylation on the conserved S54 in the P-loop and that HipT uses a universal mode for ATP recognition.
Structural basis for the toxin HipTLp recognition by the antitoxin HipSLp
Although the toxic activity of HipT in the HipBST TA system has been demonstrated to be counteracted by the antitoxin HipS17,18, the underlying molecular mechanism remains unknown. We therefore sought to determine the structure of the HipTLp–HipSLp complex. To express the HipTLp–HipSLp complex, a ribosomal-binding site (RBS, AGGAGA)37 was introduced between the stop codon of HipSLp and the start codon of HipTLp. The resultant HipSLp–RBS-HipTLp was cloned into pET21a (+) vector. The crystal structure of the SeMet-labeled HipTLp–HipSLp complex was determined and refined at 1.89 Å resolution (Table 1).
In the structure of HipTLp–HipSLp complex, a copy of HipTLp and HipSLp each were observed per crystal asymmetric unit. Residues belonging to helices α1 and α2 of HipTLp were not visible in the electron density map, whereas the density of the remaining residues was unambiguous (Fig. 5a). All 102 residues of HipSLp were successfully built into the model, showing that HipSLp is a small single-domain protein composed of five β-strands and three α-helices. The overall structure of the HipTLp-HipSLp complex is highly similar to E. coli HipA, with HipSLp and HipTLp aligning with the N- and C-terminal portions of E. coli HipA, respectively (Supplementary Fig. 8a). HipSLp superimposed with the N-terminus of E. coli HipA with an overall RMSD of 0.932 Å across 64 Cα. However, a notable difference can be observed on the β4–α2 loop of HipSLp, which is twisted and rotated by ~45° with respect to its counterpart in the N-terminus of E. coli HipA (Supplementary Fig. 8b).
In the structure of HipTLp–HipSLp complex, three α-helices of HipSLp form a helix bundle that sits above the cleft formed by the β-sheet in the N-terminal lobe of HipTLp, whereas the β-strands form a flank region in HipSLp (Fig. 5b). HipSLp binds the toxin HipTLp via hydrogen bonding in three main interacting regions, which constitute more than 1100 Å2 of total buried surface area (Fig. 5c–e). The intermolecular interactions are mainly formed between helices α1, α2 and α1–α2 loop of HipSLp and helix α3 and strand β5 of HipTLp. In the first interacting region, side chains of HipSLp E63 and HipTLp K157 form a salt bridge, side chain of HipSLp E58 engages in polar interactions with the main chain amide and side chain of HipTLp R154, and hydrogen bonds are additionally formed between main chain of HipSLp G59 and side chain of HipTLp K201 and main chains of HipSLp I65 and HipTLp G57 (Fig. 5c). The second interacting region includes a salt bridge between HipTLp D77 and HipSLp K73 and hydrogen bonds between (i) side chain of HipTLp D133 and HipTLp G94, (ii) side chain of HipTLp R154 and side chain of HipSLp N91 as well as main chain of HipSLp V92, (iii) side chain of HipSLp N91 and main chain of HipTLp Y79, and (iv) side chains of HipTLp Q78 and HipSLp Q90 (Fig. 5d). In the third interacting region, the side chain of HipTLp Q148 hydrogen bonds with the main chain of HipSLp F56, whereas the side chain of HipTLp E144 forms hydrogen bonds with the side chain of HipSLp S38 and the main chain of HipSLp L39 (Fig. 5e).
To verify the importance of these interactions for stable binding of HipSLp to HipTLp, we performed pull-down assays with untagged wild-type HipSLp and wild-type or mutant HipTLp carrying a N-terminal 6×His-tag. The HipTLp mutants D133A, R134A, and E144A completely lost their ability to bind HipSLp and the mutants R154A, K157A, K201A exhibited severely reduced HipSLp binding, suggesting that these residues form key interactions with HipSLp (Fig. 5f).
Molecular mechanism for toxin neutralization in the HipBST TA systems
One of the most striking features of the HipBST TA systems is that the role of antitoxin is taken by HipS which corresponds to the N-terminal portion of HipA toxin from the E. coli HipBA system. To better understand how the toxic activity of HipT is neutralized by HipS, we reinspected and compared the structures of apo pHipTLp, pHipTLp–AMP–PNP complex, and HipTLp–HipSLp complex. Apo pHipTLp and HipTLp from the HipTLp–HipSLp complex superimpose with RMSD of 0.464 Å over 215 Cα atoms. Notably, Ser54 of HipTLp is phosphorylated in the structure of apo pHipTLp but not in the HipTLp–HipSLp complex (please note that HipTLp–HipSLp was co-expressed in E. coli BL21) (Fig. 2a, b and Supplementary Fig. 8c). Since the residue S54 is phosphorylated when HipTLp is expressed alone, we wondered whether the phosphorylation on S54 influences the interaction between HipTLp-HipSLp and HipBLp. Size-exclusion chromatography revealed that the phosphorylation state of S54 does not appear to have a noticeable effect on interactions between HipTLp-HipSLp and HipBLp (Supplementary Fig. 9). Moreover, structural comparison suggests that the P-loop of HipTLp, which encircles ATP and is critical for catalytic activities in typical Ser/Thr kinases, underwent a conformational change from loop to helix upon HipSLp binding (Fig. 6a, b). Such allosteric regulation induced by the antitoxin binding has not been observed in E. coli HipBA TA system29. A conformational change similar to the loop-to-helix change of the HipTLp P-loop in HipTLp-HipSLp can also be observed in the recently released structure of HipBSTO127 trimer19 (Supplementary Fig. 10), suggesting a common mechanism of toxin neutralization.
These observations led us to hypothesize that the loop-to-helix conformational transition induced upon HipSLp binding may obstruct the access of ATP to the kinase active site, resulting in inhibition of the HipTLp kinase activity. Superimposition of the structures of pHipTLp, pHipTLp–AMP–PNP, and the HipTLp–HipSLp further revealed that P-loop in the HipTLp–HipSLp complex overlaps with AMP–PNP in the pHipTLp–AMP–PNP complex (Fig. 6c, d and Supplementary Fig. 11). More specifically, the γ-phosphate and β-phosphate groups of AMP–PNP would clash with the side chains of Q59 and D219, respectively, whereas the α-phosphate would clash with the side chains of K61 and K85 (Fig. 6d). This may account for the unphosphorylated state of the P-loop in the HipTLp–HipSLp complex when they were co-expression (Supplementary Fig. 8c). Such conformational change also occurs in the P-loop of HipBSTO127, suggesting that a similar mechanism is utilized by HipBSTO12719 (Fig. 6e, f). To further verify whether ATP binding is abolished, we measured the binding affinity between the HipTLp–HipSLp complex and AMP–PNP with ITC and found that HipTLp completely lost the AMP–PNP binding affinity when binding with HipSLp (Fig. 6g). Consistent with these results, the thermal stability of the HipTLp-HipSLp complex did not change upon the addition of 4 mM AMP–PNP (Supplementary Fig. 11a). Considering that HipAEc in autophosphorylated form can bind ADP and AMP but not ATP30, we also investigated whether the HipTLp–HipSLp complex binds ADP and AMP. Again, the results of ITC experiments suggested that the HipTLp–HipSLp complex has no detectable affinity for ADP or AMP, even at concentrations of 1 mM (Supplementary Fig. 12). The allosteric regulation of the P-loop induced by the antitoxin binding was also observed in the HipBSTO12719. Together, these findings suggest that HipSLp binding induces conformational changes in the P-loop of HipTLp, which blocks ATP binding and consequently inhibits the HipTLp kinase activity (Fig. 6h).