Colicin E1 opens its hinge to plug TolC

Decision letter after peer review:

Thank you for submitting your article "Colicin E1 opens its hinge to plug TolC" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, including David Drew as the Reviewing Editor and Reviewer #1, and the evaluation has been overseen José Faraldo-Gómez as the Senior Editor. The following individuals involved in review of your submission have agreed to reveal their identity: Bert van den Berg (Reviewer #2); Karen Jakes (Reviewer #3).

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Essential revisions:

1. Membrane localization experiments with ColE1-T and ColE1-TR are described in Figure 4. However, more experimental details need to be included in the figure legends and Methods, especially regarding protein concentrations in the experiments shown in Figure 4. For example, what is analysed in intact cells? Is bound toxin washed off into the supernatant? If the toxin has very high affinity will it come off the OMP receptor(s)? The relative concentrations of the colicin constructs used is not given, so there's no way to easily interpret some of the data. Furthermore, Figure 4C shows one cell each, and one could argue that only the best data is shown involved. Could the authors (perhaps as source data) show panels with many cells to show any differences between strains more objectively?

2. The Discussion section needs to better describe the differences between the "pillar" and "total-thread" models, perhaps by also including a schematic of both. Furthermore, the experimental evidence in this paper supports the total-thread model, whereas the more ambiguous experiments (trypsin digestion) are only suggestive of the pillar model. Please make the final proposed model clearer in the discussion and include how the downstream pore-forming cytotoxic domain is translocated.

3. Experiments with GFP attached to colE1-TR suggest efficient uptake in a tolC-dependent process. What would be a very informative further would be to delete the TolA box from the N terminus of the T domain of the T-R construct and determine whether uptake still occurs. If so, that would argue further against a translocation model whereby the C-terminal channel-forming domain of colicin E1 slithers down the outside of TolC, with its C terminus first, as proposed by Cramer et al. (2018). Additional control would be to mutate the hinge-proline.

4. The Housden klebicin paper was published very recently describing detailed studies of the interaction of TolC with another bacteriocin, including cryo-EM structures (Housden et al., Nature Commun 12(1):4625). While the work described here has been carried out independently, some acknowledgement of this work in the discussion is warranted.

Reviewer #1:

Colicin E1 is an E. coli bacteriocin that hijacks the outer membrane proteins TolC and BtuB to enter the cell. Here the authors reveal the cryo EM structure of TolC in complex with colicin E1. They show how Colicin E1 structure rearranges from a helical hairpin into an single, long extended helix that plugs the TolC pore. They further show that E. coli cells that have TolC proteins plugged with a Colicin E1 fragments are more susceptible to antibiotics. Taken together, they have outlined the structural basis for Colicin E1 recognition and shown the applicability of this approach for bacterial pathogenicity.

Overall, I can find little technical issues with the paper and I think the conclusions drawn from the experiments are sound. The paper is well-written and the figures are easy to follow. From a mechanistic viewpoint the major findings was that it was unexpected for colicin E1 to open up from a helical hairpin into a single helix inside TolC. While the author have succeeded to understanding how the toxin engages with its receptor its still unclear how the downstream pore-forming cytotoxic domain is translocated across TolC.

Reviewer #2:

The paper by Slusky and colleagues presents cryo-EM structures of E. coli TolC in the absence and presence of the N-terminal translocation domain of Colicin-E1 (ColE1-T). Via SEC they demonstrate that the toxin domain binds tightly to TolC. Interestingly, the structure shows that the colicin is bound inside the long TolC barrel, with most of the colicin-OMP contacts provided by the β-barrel part of TolC. The cryo-EM data are nice and the map of the complex is of very good quality and supports the ColE1-T model built by the authors. While the involvement of TolC (and another OMP named BtuB) in ColE1 cell entry was known, the structure provides in my opinion clear support for a model in which the entire ColE1 is transported into the cell via the TolC conduit, utilising the proton motive force to power translocation via TolA. This latter part is not well described by the authors and I find the discussion of the two models not very clear and too much in favor of the "pillar" model (see detailed comments; I think also a schematic figure is warranted showing the two models). Another major issue that needs addressing is the fact that the authors have missed (or ignored) a very similar paper that was published very recently and which describes detailed studies of the interaction of TolC with another bacteriocin (Klebicin), including cryo-EM structures (Housden et al., Nature Commun 12(1):4625). Thus, in contrast to what is claimed by the authors, theirs is not the first structure of a bacteriocin bound to TolC. The authors should compare their structure with that of Housden et al. (which is very similar) and rewrite their discussion based on both papers.

To support their structural data the authors also carried out membrane localization experiments with ColE1-T and ColE1-TR (includes the receptor binding domain that interacts with the primary receptor BtuB), described in Figure 4. I didn't find this very clear. In particular I'm not sure what is analysed in Figure 4B? What do you analyse in intact cells? Bound toxin that is washed off and ends up in the sup? If so, what if the toxin has very high affinity and won't come off the OMP receptor(s)? The methods section is not complete and, rather than refer to a previous paper, the procedures should be fully described here. Figure 4C shows one cell each, and one could argue that cherry picking might be involved. Could the authors (perhaps as source data) show panels with many cells to show any differences between strains more objectively?

The final data provided by the authors utilises the fact that TolC is part of the AcrAB-TolC RND efflux pump that moves noxious substances including antibiotics out of the cell. They hypothesise that, when ColE1-T or ColE1-TR is bound to TolC (plugging the efflux channel), the pump should not work or work less efficiently. Indeed, MIC values for several antibiotics are lower in the presence of certain toxin fragments, showing that cells are sensitised due to reduced drug efflux. The required concentration of the colicins are high, however, suggesting that the pump is still partly functional while the colicin is bound, a notion supported by the structure. Together, these data lead to the interesting hypothesis that modified bacteriocins that hijack functionally important OMPs might be used to potentiate antibiotics. Together, this paper provides a nice advance in our thinking on how bacteriocins gain entry into cells (but the discussion should be improved substantially). It is also consistent with the more detailed study by Housden et al., which is satisfying. The potential application towards potentiating antibiotics via colicin-based targeting of OMPs like TolC could be an important future area of research.

Reviewer #3:

Colicins are plasmid-encoded toxins produced by bacteria to kill closely related bacteria that compete for scarce resources. While the individual proteins in the target bacteria used for binding and uptake of colicins are well-known, there remains some controversy about the precise pathway that those toxins ultimately take to reach the target of their ultimate lethal act. In this work by Budiardjo et al., the path for colicin E1, which ultimately kills target E. coli by making an ion-permeable channel in their inner membrane, is clarified.

All of the E colicins use the vitamin B12 receptor, BtuB, as their primary outer membrane receptor to concentrate the colicin at the cell surface via an interaction with the central receptor-binding (R) domain. The enzymatic E colicins, E2-E9, then use the trimeric porin, OmpF, and its interactions with the N-terminal translocation (T) domain of the colicin to thread through one of the three pores of OmpF to ultimately bring the catalytic, carboxy-terminal end of the colicin into the periplasm from which it ultimately reaches and crosses the inner membrane. That process has been very thoroughly examined and documented, principally by the work of Kleanthous, Housden and their co-workers, as cited in references in this manuscript and others cited in this review. Colicin E1, which kills target bacteria by making a channel in their inner membrane and thereby depolarizing the cell, also uses BtuB as its primary outer membrane receptor. But then it differs from the other E colicins in hijacking the TolC channel, whose normal role in the cell is in drug and toxin export, and uses that channel as a means to enter the target cell. The isolated colicin E1 T domain has been found to exist in solution as a helical hairpin (Zakharov et al., 2016), but whether in vivo binding to TolC occurs as that same hairpin, with both ends pointing outside the target bacterium, or whether the hairpin unfolds, possibly around a TolC box previously shown to be essential for cytotoxicity, as proposed (Jakes, 2017), was not known. In this work, it is now clearly shown by cryoEM that the T domain hairpin unfolds completely and is captured inserted inside TolC. The previously identified TolC box acts as a hinge with a kink at a central proline in the 21-residue box sequence (residues 100-120). Residues in the TolC box are involved in specific interactions with residues of TolC within the channel. The interaction with the T domain peptide alone was sufficient to dilate both the periplasmic and extracellular apertures of TolC, without involvement of any other proteins or motive force, a very signficant finding. The structure and the specific residue interactions elucidated in this work help explain earlier results with mutations in TolC that block killing by colicin E1 (ref. #37). The structure reported here negates earlier arguments against models in which colicin E1 threads through TolC to enter the periplasm. The direction of insertion is with the N terminus facing the periplasmic end of TolC. This arrangement would logically be expected, since the known TolA box sequence lies very near the N terminus of the colicin E1 T domain and needs to interact in the periplasm with TolA, a bacterial protein anchored in the inner membrane which extends into the periplasm, is thought to provide the energy necessary to mobilize translocation of the colicin and has been demonstrated genetically to be absolutely required for colicin lethality.

Budiardjo et al. have failed to acknowledge the recent publication of nearly identical results with the closely-related Klebsiella pneumoniae klebicinC uptake via TolC (Housden et al., 2021, Nat. Commun. 12(1): 4625), in which the klebicinC T domain was also captured inside the channel of the K. pneumoniae TolC channel. In those experiments, the stretch of T domain bound inside of TolC was protected from digestion by trypsin, with residues at both ends protruding from TolC subject to protease digestion. Thus, these two different bacteriocins, targeting different bacteria, both insert their N-terminal translocation domains into the TolC protein of their respective targets.

A second very significant result from this manuscript is that binding of colicin E1 T domain inhibits drug efflux and thereby reduces the minimum concentration of some drugs to effectively kill bacteria. A significant effect was only seen in constructs that also included the R domain of the colicin, which may simply be serving to increase the concentration of T domain at the cell surface and thereby making its interaction with TolC more efficient. A shortcoming of these experiments is that the relative concentrations of the colicin constructs used is not given, so there's no way to interpret the lack of much effect from T domain alone. But the basic result suggests that there may be a route to using the E1 T domain as a means to reduce resistance of pathogenic bacteria to standard antibiotics, as has been suggested previously (Jakes, 2017).

In vivo experiments to determine whether the T domain or T-R domain were protected from trypsin digestion by virtue of having been internalized by E. coli are difficult to interpret. The protocol cited in the manuscript is a general protocol, and the relative amounts of protein used are not given. It is perhaps unsurprising that no T domain is seen bound to either intact or lysed cells, while T-R is detected, since the binding of isolated T domain, via TolC, has been shown to be orders of magnitude less efficient than that of R domain to BtuB (Jakes, 2017). The authors point out that the proteolysis results may just be a reflection of a small number of molecules entering the cells. Similarly, the failure to see much transport of the fluorescently-labeled colE1-TR-Cy3 may also be a reflection of a combination of inefficient binding and the bulky Cy3 label impeding passage through TolC. In fact, recent work on transport of colicin E9 through OmpF suggests that relatively small differences in the size of fluorescent labels attached to colicin domains can have a measurable effect on uptake of the labeled proteins (Francis et al. EMBO J. e108610/2021), and the labels used in that work had lower molecular weights than the Cy3 used in the experiments described here. Experiments with GFP attached to colE1-TR suggest more efficient uptake in a tolC-dependent process. What would be a very informative further experiment would be to delete the TolA box from the N terminus of the T domain of the T-R construct and determine whether uptake still occurs. If so, that would argue further against a translocation model whereby the C-terminal channel-forming domain of colicin E1 slithers down the outside of TolC, with its C terminus first, as proposed by Cramer et al. (2018).

[Editors' note: further revisions were suggested prior to acceptance, as described below.]

Thank you for resubmitting your work entitled "Colicin E1 opens its hinge to plug TolC" for further consideration by eLife. Your revised article has been evaluated by José Faraldo-Gómez (Senior Editor) and a Reviewing Editor.

The manuscript has been improved but there are some remaining issues that need to be addressed, as outlined below:

1. While we realise the PDBs were released earlier than those of Housden et al. it would be better to remove the instances of "first" in the text as, after all, they are not really needed.

2. The modified text for points 23 and 24 in the previous submission was still unclear to Referee 2. We would ask you to try to re-address these points (re-pasted below for clarity).

23. line 253: "need to be in close proximity for colE1 binding to occur" needs a reference.

24. line 254: "when BtuB clusters together in groups of 12 or more, it may exclude TolC" needs a reference. Also, please explain "These cluster geometries would therefore lower the number of full binding sites available for the T and R domains of colicin E1."

3. Throughout the manuscript, reference is made to the TolA box of colicins E3 and E9. However, neither of those colicins actually has a TolA box. Beginning with the early work of Bouveret et al. (Mol. Micro 1998 27, 143-157) and subsequently by others, it has been shown that what had been called the TolA box in earlier work is actually a TolB box. Colicins E2-E9 and A all interact with TolA indirectly, by binding to TolB in the periplasm, and TolB then engages TolA. The fullest picture of those interactions is in the recent work of Francis et al. (2021), which is cited in this manuscript. The only E colicin known to have a TolA box is E1. A reference (probably Pilsl and Braun, 1995) is needed when referring to the E1 TolA box. All references to a TolA box in colicins E3 or E9 need to be fixed.

4. While the experimental details and source data have been expanded and clarified, the implications of both the trypsin digestion experiments and fluorescence labeling are less clear than the cryo-EM and drug efflux experiments, and some caveats should be added to the discussions of those experiments.

5. The argument between the two competing models may require some further clarification. Referee 3 has made some suggestions below. Please look at these and amend appropriately as you see fit.

Specific comments and corrections:

Line 62: …targets and killing mechanisms. Most colicins…

Lines 69-71 should read: …of ColE3 or ColE9 initiates translocation using the secondary OMP receptor/translocator, OmpF, to access the Tol-Pal system in the periplasm. For most colicins, the T domain requires an OMP distinct from the R domain target for colicin translocation.

Line 72-73: …structurally characterized bound to their OMP primary receptors. (Kurisu…, Sharma…, Buchanan et al., 2010). Those structures all reveal that the plugged β-barrels remain plugged upon colicin binding, so another route through the outer membrane is required.

Line 75-76 ff: …that T domain can pass fully through one pore of its membrane translocator, OmpF, and loops around to insertv in a second pore of the trimer from the periplasmic side, while the R domain initially interacts with…its receptor, BtuB. Those same studies and earlier ones clearly demonstrated that full engagement of the E9 T domain with its TolB translocation partner in the periplasm results in disengagement of the R domain from BtuB (Francis et al. 2021 and references therein). Experiments such as these and those that preceded this most recent work are the basis for the "total thread" model for colicin translocation, in which the colicin is pulled through the same pore of OmpF through which its T domain initially entered. The process is energized by the TolABQR system.

Line 82: death. Unlike any of the other E colicins, the T domain of colicin E1…

Line 83: …and the R domain binds BtuB, the primary receptor shared by all E colicins.

Line 98: …and disrupt TolC channel conductance in vitro, in planar lipid bilayer membranes…

Line 101: …in solution similar to another colicin T domain from colicin Ia, one that does not interact with TolC (Wiener et al., 1997).

Line 223: …these residues are contact sites…

Line 290: Should read: Given the two-dimensional surface…

Line 370: Add Francis et al., 2021 reference here.

Line 376: Why was the hinge opening unanticipated?

Line 378: …as the total thread model hypothesized and consistent with experimental data that support this model for colicin E9 (Housden et al., 2005; Housden et al., 2013, Francis et al., 2021).

Line 380: …clustered and does not diffuse…

Reviewer #1:

I am satisfied with the revised submission and have no further concerns.

Reviewer #2:

The revised manuscript by Slusky has substantially improved based on the reviewer comments and in particular it is now much clearer. I have just a few remaining minor issues for consideration:

point 2: I understand the reasoning of the authors. However, I'm not sure if it is entirely appropriate to allow priority claims based on non peer-reviewed preprints. While I do realise the PDBs were released earlier than those of Housden et al., in my opinion it would be better to remove the instances of "first" (as indeed is the norm in some other journals) in the text as, after all, they are not really needed. I don't know what the policy is of eLife in this regard, and leave it up to the editor to decide.

points 23 and 24: unfortunately the modified text is still unclear to me.

Reviewer #3:

The authors have very thoroughly addressed all of the reviewers' comments, provided additional experiments and methodological descriptions, and fixed the reference list to remove duplications and make it a more user-friendly format.

From that perspective, I should say that the paper is acceptable now. But in the process of expanding parts of the Introduction and Discussion, I feel they have done their work and the field a disservice by giving "first billing" to a model for which there is scant experimental support and giving short shrift to the model for which there is now extensive experimental support and which common sense should support.

In their expanded Introduction, the authors once again lead with a lengthy elaboration of the "pillar model" for the translocation of the colicin E1 pore-forming domain across the outer membrane of target E. coli. That model was introduced simply because of the assumption by Zakharov et al. (2016) that the helical hairpin structure of the E1 T domain observed in solution from CD spectra is maintained when the T domain binds in vitro to TolC, as it was shown to do by SEC, occlusion of TolC channels in planar lipid bilayer membranes (Zakharov et al., 2016), and in vivo by protecting sensitive E. coli from killing by active colicin E1 (Jakes, 2017). On the contrary, the current work has clearly demonstrated that the hairpin unfolds and inserts, amino-terminus first, into TolC., so the whole basis for a model with both N- and C-termini pointing to the outside of the target cell is gone. (In addition, recent work on the Klebicin C bacteriocin has shown that locking that protein's T domain helix in its closed state, with an inserted disulfide bond, kills its activity in the oxidized state, but not when it's reduced (Housden et al., 2021)). Another argument made by Zakharov et al. against insertion of the T domain into TolC was that the periplasmic aperture of the TolC channel is too narrow for a polypeptide to exit, so therefore the colicin couldn't possibly thread into the periplasm via that route. The structure shown in Figure 1B demonstrates that the binding of T domain within the channel opens that aperture; the colicin domain is easily accommodated within TolC. Thus, the colicin polypeptide or a part of it could transit through TolC.

The sections on potentiation of antibiotics by the E1 T domain are additional evidence that the colicin domain is occupying TolC and inhibiting its normal role in drug efflux. It's likely that the failure to see much of an effect with T domain, as compared to R-T, is simply the result of the much tighter binding of R to BtuB, which then concentrates T domain at the cell surface and makes it more likely to find and enter a TolC channel.

While the experimental details and source data have been expanded and clarified, the implications of both the trypsin digestion experiments and fluorescence labeling are less clear than the cryo-EM and drug efflux experiments, and some caveats should be added to the discussions of those experiments. It's not clear what not finding protected T domain tells us. Recently published experiments with the related KlebC and the K. quasipneumoniae TolC (Housden et al. 2021) clearly show that a significant segment of the KlebC T domain is protected from trypsin digestion when bound and digested in vitro. Mass spec analysis of the protease digestion products showed that the KlebC T domain extends out of the TolC channel at both the periplasmic and extracellular ends, and those parts of the peptide were degraded by trypsin, but there remained a protected band of >40 residues that coincides with residues shown inside the TolC channel in the cryoEM structure.

The observation in most cases of just a single punctum from fluorescently-labeled colicin peptides is puzzling, although the presence of a fluorescent signal is clearly BtuB and TolC-dependent. Frankly, I'm not sure what we learn from these experiments. Somehow the videos referred to did not come through with the copy of the manuscript I received.

The new experiments with P110A and T∆1-40 are interesting and somewhat surprising. There is nothing in the Methods about how either of those mutants was generated.

Throughout the manuscript, reference is made to the TolA box of colicins E3 and E9. However, neither of those colicins actually has a TolA box! Beginning with the early work of Bouveret et al. (Mol. Micro 1998 27, 143-157) and subsequently by others, it has been shown that what had been called the TolA box in earlier work is actually a TolB box. Colicins E2-E9 and A all interact with TolA indirectly, by binding to TolB in the periplasm, and TolB then engages TolA. The fullest picture of those interactions is in the recent work of Francis et al. (2021), which is cited in this manuscript. The only E colicin known to have a TolA box is E1. A reference (probably Pilsl and Braun, 1995) is needed when referring to the E1 TolA box. All references to a TolA box in colicins E3 or E9 need to be fixed.

I have taken the liberty of making a number of additions and corrections, in italics, to add some balance to the descriptions of the two models. Obviously, these changes are just suggestions, but I hope the authors feel they improve the paper, which is a very important addition to this field. (Note: unfortunately I see that in copying my Word document, formatting was lost so the italics cannot be distinguished from the original text.)