Direct activation of a bacterial innate immune system by a viral capsid protein

Bacteria have evolved diverse immunity mechanisms to protect themselves against the constant onslaught of bacteriophages1–3. Similar to how eukaryotic innate immune systems sense foreign invaders through pathogen-associated molecular patterns4 (PAMPs), many bacterial immune systems that respond to bacteriophage infection require phage-specific triggers to be activated. However, the identities of such triggers and the sensing mechanisms remain largely unknown. Here we identify and investigate the anti-phage function of CapRelSJ46, a fused toxin–antitoxin system that protects Escherichia coli against diverse phages. Using genetic, biochemical and structural analyses, we demonstrate that the C-terminal domain of CapRelSJ46 regulates the toxic N-terminal region, serving as both antitoxin and phage infection sensor. Following infection by certain phages, newly synthesized major capsid protein binds directly to the C-terminal domain of CapRelSJ46 to relieve autoinhibition, enabling the toxin domain to pyrophosphorylate tRNAs, which blocks translation to restrict viral infection. Collectively, our results reveal the molecular mechanism by which a bacterial immune system directly senses a conserved, essential component of phages, suggesting a PAMP-like sensing model for toxin–antitoxin-mediated innate immunity in bacteria. We provide evidence that CapRels and their phage-encoded triggers are engaged in a ‘Red Queen conflict’5, revealing a new front in the intense coevolutionary battle between phages and bacteria. Given that capsid proteins of some eukaryotic viruses are known to stimulate innate immune signalling in mammalian hosts6–10, our results reveal a deeply conserved facet of immunity.

In this study, Zhang and co-authors uncover how the CapRelSJ46 TA system that protects E. coli against diverse phages is being activated upon phage infection and leads to abortive infection.
The approach taken and the careful experimental design of this study make the results convincing. This is a very sound contribution to the TA field with key findings: the characterization of the molecular mechanism triggering activation of a TA system upon phage infection; and the discovery of a proteolysis-independent mechanism of activation for a type II TA system.
Authors show that CapRelSJ46 is a fused TA system containing a C-terminal antitoxin domain that senses phage infection through direct binding with the p capsid protein from SEC phage. By crystallization, they show that CapRelSJ46 protein alone adopts a closed conformation where the toxic pyrophosphokinase catalytic site (N-ter) is sterically hindered by the C-terminal antitoxin domain. Their mutation experiment further supports that model given that disrupting the intramolecular interface led to constitutive toxicity of CapRelSJ46 even in the absence of phage infection. To nail down the molecular determinant of phage dependent activation of CapRelSJ46 toxicity, authors used an experimental evolution approach and isolated SEC phages that can infect CapRelSJ46-carrying cells. Authors later demonstrate that SEC mutants are mutated in a hypothetical gene encoding a capsid protein which prevent activation of CapRelSJ46 and therefore, abortive infection. Combination of their in vitro and in vivo data suggests a model where the phage Gp57 capsid protein bind and relieves CapRelSJ46 autoinhibition leading to pyrophosphorylation of the cell's tRNAs by the pyrophosphokinase toxic domain and results in translation arrest leading to abortive infection. I only have minor comments/questions: • Because the Salmonella SJ46 phage contains a fused CapRel, is its infection abortive? How do you explain it had been selected? • About the in vitro transcription-translation assay presented in figure 3j: if I understood correctly, it's a Gp57-encoding template that is being added to the in vitro reaction. I got confused by the text page 6 line 1 that says in the presence of the SEC 7 major capsid protein Gp57 maybe it should state explicitly that it's a Gp57-encoding template that is being added to the reaction. Related to this experiment, it seems that during the last 60min incubation some extra Gp57wt is being translated even in presence of CapRelSJ46, when no DHFR is being made. Could some cis elements make Gp57 translation more robust to translation inhibition (transiently)? • In figure 3, panels h and i: SEC 7 infection and gp57 overexpression lead to different extents of translation shutdown in cells harboring CapRelSJ46, could you comment on this result? Zhang et al. present an elegant series of biochemical, structural, genetic, and in vivo analyses that demonstrate a toxin-antitoxin system in bacteria named CapRel functions through direct recognition of bacteriophage capsid protein. The experiments are thorough, well presented, and the new results are exceptionally exciting for multiple fields related to understanding mechanisms of phage defense and the emerging connections between antiviral immunity in bacteria and animals. Some control experiments are necessary to complete understanding of the proposed mechanism of phage detection. Otherwise, I have only minor comments to help improve the manuscript for a general audience.

1)
A key open question necessary to understand the proposed model of capsid sensing is to define the stoichiometry of CapRel-Gp57 complex formation. The authors' model proposes 1:1 complex formation based solely on structural modeling. This question is particularly important as the authors seek to compare CapRel with Trim5a activation in mammalian cells that relies on recognition of the intact HIV capsid lattice. Does CapRel sense individual Gp57 monomers or perhaps initial stages of capsid lattice formation? Experimental evidence to support stoichiometry (SEC-MALS, electron microscopy analysis etc.) should be readily available as the authors already demonstrate purification of CapRel and Gp57 and co-complex formation in vitro with ITC and HDX. Although determining the structure of the CapRel-Gp57 complex would significantly enhance the paper, only biochemical or lower-resolution stoichiometry information are necessary.
2) Is there is a fitness cost associated with evasion of CapRel sensing? These experiments are particularly interesting as previous results with the defense system Pycsar demonstrated that T5 phages that acquire capsid mutations to escape defense are significantly less fit than wildtype viruses (Tal and Morehouse et al. Cell 2021 PMID 34644530).

Minor Points
3) It is surprising that CapRel crystallized in an apparently "active" conformation. Can the authors further comment on why this may be? Does analysis of packing in the CapRel crystal lattice provide further insight into conformational changes required for toxin activation?
4) The CapRel chimera data demonstrate that the CapRel SJ46-Ebc chimera is significantly less capable of defending against phage T7 compared to the parental CapRel-Ebc construct ( Figure  2b). The text description that the chimeric CapRel "gained protection against T7, manifesting as decreased EOP and smaller plaques" doesn't appear to match the data well. Can the authors comment on why sensing is severely compromised and potentially clarify the text? 5) In Figure 4a soluble Gp57 mutant proteins appear to express to significantly higher levels than wildtype protein. Is this difference meaningful, or perhaps related to capsid lattice stability and the mechanism of CapRel escape? 6) While no further experiments related to cell death are necessary, it would be helpful if the authors discussed how CapRel activation and translation repression leads to abortive infection. This step of the process is not outlined in the model Figure 5h. 7) In the main text figure panels, it is confusing which data are experimentally derived and which models are created with AlphaFold. Especially in Figure 4c presentation of the modeled CapRel-Gp57 complex is potentially misleading to the reader. Although the text legends are clear, it would be helpful in there was an indication in the panels themselves which structures are experimental and which are models. 8) Text Comments: -Line 52: The phrase "which are effectively in constant surveillance mode" as used to selectively describe RM systems is confusing, aren't CapRel, CRISPR, and other pathogen-sensing machineries also always in a constant surveillance mode? -Line 88: To understand use of the CapRel-SJ46 system, it may help to explain that defense systems are often encoded within phage genomes as part of phage competition. -Line 710: Please list the composition of crystal solution LMB C9 and the solutions used for cryoprotection.
I hope the authors will find my comments useful, thank you for the opportunity to read this exciting manuscript.

Philip Kranzusch
Referee #3 (Remarks to the Author): Phage defense systems have received a lot of attention in recent years, but for most, their mechanisms of action remain mysterious. For defensive Abi systems (which encompass many mechanistically distinct systems, including TA systems), phage infection must be specifically sensed for activation of defense. However, known mechanisms underlying the activation of such defenses are few and far between, with many relying on incomplete data to draw definitive conclusions (e.g. escape phages used as the only evidence to identify said activating cue). In this manuscript, Zhang et al convincingly demonstrate the direct activation of a defensive TA system by a phage capsid protein in vivo and in vitro. This manuscript was an absolute pleasure to readoutstanding science described in a very clear manner. I am extremely enthusiastic about this being published in Nature, as it really represents a milestone in our understanding of bacterial immune systems. I have a series of fairly minor suggestions, though I'll label some as 'major' to distinguish from truly minor spelling-type issues, none are truly major in the sense of significantly decreasing my enthusiasm for this outstanding work.

Major:
The authors use strong language about the broad applicability of their findings: 'anticipating that major capsid proteins may emerge as common, direct triggers for a diverse range of anti-phage defense systems.' Yet they haven't shown that major capsid proteins are common triggers for CapRel homologs. I am not suggesting the authors repeat all the in vivo/in vitro studies with different pairs, but given that the identity of the capsid protein is known for several of the phages tested here, it would be worth testing for CFUs during co-expression of the inhibitory CapRel and cognate inhibited phage capsid protein (e.g. for T7 capsid co-expressed with CapRelEBC as was done in Fig 3g for Gp57+CapRelSJ46). One would expect to see toxicity upon co-expression only, as was observed for Gp57+CapRelSJ46. Some of the major capsid proteins of phages tested are also solved, and these could be used in some structural modeling with an inhibitory CapRel to look for evidence of an interaction. This would strengthen the story significantly. If, however, the authors find no evidence of capsid triggering other CapRels, this is a valuable discussion point and not a deal-breaker for the paper.
The bulk of the paper focuses on Secphi27. As such, it would be much more relevant to include the one-step growth curve for this phage +/-CapRelSJ46 in fig 1e (T4 data should be moved to the supplement); time-course data for CapRel and translation inhibition of Secphi27 is presented in Fig  3a/i but we have no baseline for the kinetics of this phage's replication.
Line 124 -the authors construct a chimera in which the C-term of CapRelSJ46 was replaced by the corresponding region of CapRelEbc. They observe that the chimeric CapRel no longer protected against Secphi27 but gained (some) capacity to restrict T7. The description of this data is stronger than what is shown (Fig 2B), which is a very modest reduction in T7 plaquing. Given that the chimera does not phenocopy CapRelEbc, I would like to see the reciprocal chimera (N term swaps), which, according to their model/conclusions, should not alter the specificity. The chimeras also may not be stable/well expressed, which may explain the results -I suggest the authors blot for the WT/chimeras at a minimum. Figure 3a -is there a loading control here? There appears to be a slight decrease at 60 minutesagain, it would valuable to know if, at this point, there is already evidence that CapRel is inhibiting phage production (referring to comment regarding missing one-step growth curve).
In figure 3J&K, it appears the authors switched to testing the double mutant, unclear why the single mutants weren't tested. Some explanation/pointing this out is necessary.
In Figure 5d -F120L/124F don't ablate toxicity and yet allow for apparent complete restoration of EOP -an explanation is warranted, and also, it should be confirmed that these alleles are not suddenly toxic on their own as was done in previous experiments.
My comments regarding the kinetics point to a missing piece in the discussion that I hope can be accommodated, though, of course, I understand there are word limits. The assumption is that Gp57 is produced late -but as far as I can tell, this is not known (or evaluated here). Gp57 should be blotted for (for example, in parallel with the CapRel blot in Fig 3a). Some discussion of why a late expressed phage gene product would be a valuable trigger is necessary -especially for those not immersed in the field. Logically it may be because it is a 'last line of defense' which usually operates in a cell with other defenses. Or does this phage produce capsid early, and thus triggering early makes more intuitive sense to more robustly inhibit phage. For T7, capsid is expressed late, and yet CapRelEBC provides near-complete protection against this phage, this is not intuitive. Perhaps CapRelEBC uses a different trigger (see above). In keeping in mind that the CapRel system is found on resident prophages, perhaps the later induction provides ample time for the prophage to enter the lytic cycle and escape the cell under attack by a different incoming phage.
In the discussion (line 307) I would like to see some additional discussion around the abundance of fused TA systems as proteolysis of the antitoxin for activation in such a context is difficult to imagine. Do the authors anticipate that direct binding of the antitoxin is more pervasive /more realistic for fused TA systems specifically? It is speculative, of course, but one added sentence would be nice. It could be valuable to have added the relative abundance of fused systems vs others in Figure 1a to this point.
Lastly, most presented data are representative images of experiments performed with replicates -I would not accept graphs generated from single data points, so I just want to ensure replicates are available/organized/included in a repository and a link to that be included in the 'Data and Materials Availability Statement.' I do not doubt the rigor of the data, but this standard should be applied to all and should be required for submission.

Minor:
Fig 3B -explain white vs brown wells in legend. Last well to the right I assume is no phage, this is not clear from the cartoon of the dilution series.
Line 184 -A supplementary table of all mutations found should be provided. I am convinced by the follow-up data, even if others were observed, but those data should be made transparent.
Line 324 -perhaps a bit strong to say all domains of life unless known for Archaea "Uptake" is misspelled on line 763 (as upatke). The work by Zhang et al. reported that the C-terminus of a fused TA can bind to the major coat protein of phages and demonstrated that this binding could activate the toxicity of the N-terminus of the fused TA by relieving autoinhibition. The authors used several in vitro evidence to draw the main conclusion that the phage capsid proteins stimulate the innate immune system of bacteria. Although many new anti-phage elements have been reported over the last three years, how bacteria sense phage attack and initiate 'altruistic' cell death via activating toxins in the population remains largely unclear.
Although the finding that the MCP can bind to the antitoxin of the TA systems is intriguing and novel, the robustness of the binding and the outcome of the binding (conformational change of the fused TA) are not verified by in vivo assays, and the resolved structural data by X-ray crystal diffraction and the predicted structures by AlphaFold which they used for the proposed model is contradictory.
The "Direct activation", as the authors stated in the title, is not fully supported by the presented assays. At least two direct pieces of evidence are needed to support the finding that the Cterminus of CapRel serves as the real sensor for phage attack by binding directly to the phage MCP. Firstly, most of the presented assays are based on the overexpression of TA in plasmid. In order to show this CapRel functions as a sensor and player in anti-phage, it is crucial to have a high baseline expression of the fused TA. However, the author only showed that the CapRel protein level is not changed when phage SEC 27 is infected by western blot analysis (Fig. 3a). What is the expression level of CapRel in the host bacteria with a chromosomally-encoded CapRel that has the anti-phage activity? Is there enough active state of CapRels to halt host translation when MCP binds to the C-terminus?
Secondly, in the model proposed here, the toxin activation relies on the conformational change of the fused CapRel. Regarding the activation mechanism, there is an apparent contradiction between the crystal structure of CapRelSJ46 solved in this paper (an open state without the presence of phage major capsid protein) and the mechanism proposed by the authors based on AlphaFold prediction: CapRelSJ46 should be a closed state without the presence of phage major capsid protein (on lines 301-306). If so, the major capsid protein is not the main trigger for the transition between the two conformations. Small-angle scattering or other experiments are needed to prove the conformation of CapRelSJ46 in solution with and without MCP.
There are many overreaching statements regarding the proven implications of the work. For example, another primary concern is the claim of the "innate immune system." As the authors illustrate in the beginning (on lines 88-92), the fused TA is encoded by phage or prophages. Why do the authors name them as the "innate immune system of bacteria"? Many genes in prophages of E. coli are kept at relatively deficient levels or silenced. If CapRels are encoded by phage or prophage, the results presented in this paper is phage-phage interaction instead of host-phage interaction.

Reviewer Reports on the First Revision:
Referees' comments: Referee #1 (Remarks to the Author): Very nice revision of a beautiful work. In my opinion this work is now ready for Nature.
Referee #2 (Remarks to the Author): The authors' revised manuscript is significantly improved. All of my reviewer points have been addressed with new experimental data and analysis of the CapRel crystal structure, and I congratulate the authors on a very exciting story. I have two minor points for the text, but otherwise recommend the manuscript for publication.
1) The authors present new analysis of the CapRel crystal lattice to support that a proposed inactive conformation is incompatible with crystal packing. While the authors' model is reasonable, a minor concern is that relatively little evidence is available to define what is actually the "inactive" CapRel conformation and what is the "active" CapRel conformation. The authors may want to be more cautious in the text to indicate that the current captured structural state by crystallography is consistent with a conformational change leading to activation but that determination of what is the fully active state will require future structural understanding.
2) The authors comment that capsid recognition may be common across different defense systems. Recent, elegant work by Gao, Zhang and colleagues demonstrates that phage portal and terminase proteins are conserved PAMPs recognized by STAND/Avs defense systems (Gao, Wilkinson, and Strecker et al. Science 2022 PMID 35951700). The authors should consider citing this work and perhaps amend their discussion to be less specifically focused on phage capsid recognition. The two best characterized systems CapRel and STAND/Avs together argue that structural proteins in general are likely conserved PAMPs for defense activation.
Referee #3 (Remarks to the Author): I thank the authors for the careful consideration of my comments. I am however not satisfied with the lack of changes to the manuscript in response to my first major comment (regarding the narrative that capsid proteins are common triggers without even acknowledging that even among CapRel homologs there may be a diversity of triggers). I generally agree -capsid seems like a great trigger to evolve a response too (assuming the kinetics do not pose a challenge being 'too late' to establish a protective response). However, I think the authors need to add an explicit caveat to the generalizability of their manuscript given that they do not provide evidence that capsid is a common trigger among CapRel homologs.

[Redacted]
Another small comment: the brief explanation added in now lines 318-322 to explain the difference in toxicity/restoring EOP does not reflect the much better explanation in the rebuttal. I recognize the need to be brief here, but the idea of binding affinity should be included in the main text (not just the accumulation of protein). Lastly, for extended data figure 1 please include a legend defining the corresponding color in the figure rather than relying on a description of colors written in the legend. Congratulations to the authors again on an exciting manuscript.