A Comprehensive Structural and Functional Analysis of Saccharomyces Killer Toxins

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript provides an extensive structural analysis of Saccharomyces killer toxins based on AlphaFold predictions and molecular dynamics simulations. While the computational work is thorough, the text occasionally overstates functional conclusions. In several cases, predicted structural features are discussed as if they directly demonstrate mechanisms such as pore formation or ionophoric activity. The authors should more clearly distinguish between model-based hypotheses and experimentally validated mechanisms, particularly for less well-characterized toxins.

1. In multiple cases, regions with low or moderate AlphaFold confidence (pLDDT <70) are discussed as functionally important. For example, in the structural analysis of K1 and K2, extended loop regions and terminal segments with low pLDDT values (as shown in Fig. 3 and Supplementary Fig. S2) are interpreted as contributing to membrane interaction and pore formation. Similar interpretations are made for the C-terminal region of K74, where structural flexibility is inferred to support ionophoric activity. These conclusions should be restricted to high-confidence regions or explicitly reframed as speculative.
2. The manuscript attributes functional relevance to terminal regions in several toxins despite limited structural confidence. In Klus and K62, both N- and C-terminal segments with poor confidence scores are discussed as potentially mediating membrane insertion or oligomerization (Results, Section 3.3). Given the low model confidence in these regions, such interpretations are not sufficiently supported and should be clearly qualified or removed.
3. The manuscript repeatedly notes that K28 differs structurally and mechanistically from other killer toxins, including the absence of predicted pore-forming motifs and its intracellular mode of action. However, these observations are distributed across multiple sections without synthesis. A dedicated subsection explicitly summarizing K28’s structural features, trafficking pathway, and mode of action would clarify why it should be considered a separate mechanistic class rather than an exception within the ionophoric toxin group.
4. Structural similarity between several killer toxins and bacterial pore-forming toxins or lectin-like domains is reported. However, the manuscript does not explicitly state whether these similarities are interpreted as evidence of shared evolutionary ancestry or as cases of convergent structural adaptation. This distinction should be addressed directly, as it substantially affects the evolutionary interpretation of the data.

5. Molecular dynamics simulations are used to support conclusions regarding structural stability and functional plausibility. However, the rationale for the selected 1 μs simulation length is not stated. In addition, RMSD plots for individual toxins show variable convergence behavior, yet all simulations are treated equivalently in the discussion. The authors should clarify which conclusions rely on MD-derived stability versus static AlphaFold models and justify the simulation parameters more explicitly.

6. The FoldX analysis classifies mutations as destabilizing based on a defined ΔΔG cutoff, but no reference or benchmarking is provided to support this threshold for secreted fungal toxins. Given the sensitivity of ΔΔG-based classification, the authors should justify the chosen cutoff with references or explain its applicability to this protein class.

7. The structural descriptions of K1/K2, K74/K62, and Klus/KHR repeat similar background information and motif descriptions across multiple subsections. These sections could be condensed by focusing on comparative differences rather than reiterating shared features already introduced earlier.
8. While individual structural figures are clear, the manuscript lacks a single figure summarizing toxin families, confidence levels of structural regions, and proposed mechanisms. Such a figure would help integrate the extensive modeling data and understand how the authors’ classification framework is derived.

 

 

Author Response

The manuscript provides an extensive structural analysis of Saccharomyces killer toxins based on AlphaFold predictions and molecular dynamics simulations. While the computational work is thorough, the text occasionally overstates functional conclusions. In several cases, predicted structural features are discussed as if they directly demonstrate mechanisms such as pore formation or ionophoric activity. The authors should more clearly distinguish between model-based hypotheses and experimentally validated mechanisms, particularly for less well-characterized toxins.

Done. We have updated the section titles to better reflect molecular modeling results. We have also added extra titles and ensured that within the results and discussion section, the major headings are limited to the sections that define each killer toxin family.

1. In multiple cases, regions with low or moderate AlphaFold confidence (pLDDT <70) are discussed as functionally important. For example, in the structural analysis of K1 and K2, extended loop regions and terminal segments with low pLDDT values (as shown in Fig. 3 and Supplementary Fig. S2) are interpreted as contributing to membrane interaction and pore formation. Similar interpretations are made for the C-terminal region of K74, where structural flexibility is inferred to support ionophoric activity. These conclusions should be restricted to high-confidence regions or explicitly reframed as speculative.

 

Comment. The regions that we have marked as contributing to membrane interaction and pore formation are actually among the highest confidence in the models (pLDDT >70). We are unsure what low confidence region the reviewer is referring to.

 

 

The model of K74 is of overall low confidence and we were unable to make any valid predictions. For that reason, we dedicated a significant portion of the K74 family section to modeling homologs and characterized the KTS1 toxin from another fungus, which showed a high pLDDT score (> 80) for the predicted pore-forming region. Therefore, we disagree with the reviewer’s analysis of our data and would welcome further clarification if we have misinterpreted their comments.

 

 

1. The manuscript attributes functional relevance to terminal regions in several toxins despite limited structural confidence. In Klus and K62, both N- and C-terminal segments with poor confidence scores are discussed as potentially mediating membrane insertion or oligomerization (Results, Section 3.3). Given the low model confidence in these regions, such interpretations are not sufficiently supported and should be clearly qualified or removed.

 

Comment. The C-terminal domains of K62 is of high confidence – pLDDT >80, as shown in figure S3. There is a low confidence unstructured C-terminal tail of Klus, but the majority of the structured C-terminal domain is of high confidence of ~80. We show that Klus and other members of our defined Klus family have structural homology to empirically-determined structural models of other toxins – most notably SMKT. Again, we disagree with the reviewer’s analysis of the presented data.

 

 

1. The manuscript repeatedly notes that K28 differs structurally and mechanistically from other killer toxins, including the absence of predicted pore-forming motifs and its intracellular mode of action. However, these observations are distributed across multiple sections without synthesis. A dedicated subsection explicitly summarizing K28’s structural features, trafficking pathway, and mode of action would clarify why it should be considered a separate mechanistic class rather than an exception within the ionophoric toxin group.

 

Comment. The K28 toxin has its own section entitled “2.8 Introduction - the K28 killer toxin family”

from pages 31-35. The only other mention of K28 is sporadic in the introduction to the paper, where it is mentioned alongside the most well-studied killer toxins.

 

 

1. Structural similarity between several killer toxins and bacterial pore-forming toxins or lectin-like domains is reported. However, the manuscript does not explicitly state whether these similarities are interpreted as evidence of shared evolutionary ancestry or as cases of convergent structural adaptation. This distinction should be addressed directly, as it substantially affects the evolutionary interpretation of the data.

 

Done. We discussed the likelihood of horizontal gene transfer of toxins in the opening paragraph of the results Line 222: “Finding killer toxin homologs encoded in the genomes of a wide diversity of organisms supports their horizontal transfer between species and ongoing gene diversification and expansion.”

 

We have now speculated further about the evolutionary trajectory of killer toxins with additional text: Line 1736 “Although there is evidence of horizontal gene transfer of these aerolysin-like toxins from fungi to bacteria and plants, whether they share ancient ancestry with canonical aerolysins or arose through a process of convergent evolution is an open question.”

 

1. Molecular dynamics simulations are used to support conclusions regarding structural stability and functional plausibility. However, the rationale for the selected 1 μs simulation length is not stated. In addition, RMSD plots for individual toxins show variable convergence behavior, yet all simulations are treated equivalently in the discussion. The authors should clarify which conclusions rely on MD-derived stability versus static AlphaFold models and justify the simulation parameters more explicitly.

Done. We describe all of the modeling trajectories in the modeling section (2.1 Molecular Modeling of Saccharomyces Killer Toxins) and also clarify that the MD models are the subject of further analysis in the manuscript “The highest average confidence structures generated by AlphaFold2 were subjected to a 1 μs MD simulation to model their behavior in a solvated environment and to improve the quality of the predicted structures; these were used for further functional analysis.”

We do not justify the longer simulation based on the fact that the majority of the simulations easily reached convergence after 1 μs. We describe the simulations in the text: Line :Most models stabilized after 100 to 200 ns, with K1, K2, KHS, and K45 stabilizing quickly and showing minimal RMSD movement for the remainder of the simulation. In contrast, KHR, Klus, and K62 each had a shift in RMSD during simulation, likely due to a conformational change. For K62, this shift occurred within 50 ns and stabilized after 100 ns, which was primarily due to N-terminal flexibility. For Klus and KHR, the shift occurred after 600 ns due to flexibility in the first 15 N-terminal residues and a large flexible loop (amino acids 111-161), respectively. For K62, Klus, and KHR, removal of these flexible regions resulted in more stable structures (Figure S4). Running longer simulations would likely not have benefited the modeling of the majority of toxins, and it was not technically feasible to run them given the cost of time on the supercomputer at our institution.

1. The FoldX analysis classifies mutations as destabilizing based on a defined ΔΔG cutoff, but no reference or benchmarking is provided to support this threshold for secreted fungal toxins. Given the sensitivity of ΔΔG-based classification, the authors should justify the chosen cutoff with references or explain its applicability to this protein class.

Done. We have added the first manuscript that was used to support for the ΔΔG cutoff (see reference: Patel, J. S.; Quates, C. J.; Johnson, E. L.; Ytreberg, F. M. Expanding the Watch List for Potential Ebola Virus Antibody Escape Mutations. Plos One 2019, 14 (3), e0211093. https://doi.org/10.1371/journal.pone.0211093.)

 

1. The structural descriptions of K1/K2, K74/K62, and Klus/KHR repeat similar background information and motif descriptions across multiple subsections. These sections could be condensed by focusing on comparative differences rather than reiterating shared features already introduced earlier.

 

Comment. The K1, K2, K45, and K74 families are compared in the section entitled: 2.6 Mechanistic insights into the K1, K2, K45, and K74 killer toxin families, as we designate them the “K1 family” of toxins. Similar sections on the mechanistic discussion are written for Klus and KHR, as they represent the “Klus family”. However, K62 and K74 are unrelated and do not have similar background information, so it would not make sense to have a section combining these as suggested by the reviewer.

 

 

1. While individual structural figures are clear, the manuscript lacks a single figure summarizing toxin families, confidence levels of structural regions, and proposed mechanisms. Such a figure would help integrate the extensive modeling data and understand how the authors’ classification framework is derived.

 

Comment. Please see Figure 14. This is a comparative figure for all of the toxin models in the manuscript. Confidence levels for all the toxins are found in figure S2 and S3. The mechanisms of the toxins are shown in Table 1.

Reviewer 2 Report

Comments and Suggestions for Authors

This manuscript presents a structural bioinformatics survey of Saccharomyces killer toxins using AlphaFold and molecular dynamics simulations. While the topic is relevant, I cannot recommend publication in its current form due to several critical issues:

First, the study would benefit from a clearer, hypothesis-driven framework. As it stands, the manuscript reads more as a structural catalog or resource than a focused research article addressing a specific biological question.

Inadequate Materials and Methods — The modeling and simulation workflows are not reproducible. Accession numbers of input sequences are missing.

Novelty is overstated — The study’s main contribution is in organizing and modeling existing data using modern tools — valuable, but not novel enough to justify the strong claims made by the authors in the Abstract and elsewhere. Unless new biological insights or validations are added, the paper should frame itself as an integrative annotation or resource, not as a major discovery.

Figures are numerous but often repetitive in presentation, authors could summarise or compare them more effectively.

Results and Discussion are merged. The separation of descriptive results from interpretive discussion would improve focus and readability.

I encourage the authors to consider re-framing this study with a sharper hypothesis and higher standards of computational reporting if they wish to resubmit elsewhere.

Author Response

This manuscript presents a structural bioinformatics survey of Saccharomyces killer toxins using AlphaFold and molecular dynamics simulations. While the topic is relevant, I cannot recommend publication in its current form due to several critical issues:

First, the study would benefit from a clearer, hypothesis-driven framework. As it stands, the manuscript reads more as a structural catalog or resource than a focused research article addressing a specific biological question.

Done: We have added the section title: 1.4 A lack of tertiary structure models of killer toxins.. This section explains the limited structural information available on killer toxins and the opportunities to leverage advances in machine learning to gain functional insights – which is the primary goal of the manuscript.

Inadequate Materials and Methods — The modeling and simulation workflows are not reproducible.

Comment: The modelling workflows are clearly written in the methods sections with details of programs and parameters used for each step of the process. These details are more than adequate for reproduction by any researcher. If the reviewer wishes to provide specifics on why they make the bold claim that our work is not reproducible, we would be happy to accommodate their suggestions.

Accession numbers of input sequences are missing.

            Done. We now include a new supplementary table S14 with all of the accession numbers.

 

Novelty is overstated — The study’s main contribution is in organizing and modeling existing data using modern tools — valuable, but not novel enough to justify the strong claims made by the authors in the Abstract and elsewhere. Unless new biological insights or validations are added, the paper should frame itself as an integrative annotation or resource, not as a major discovery.

Done. We have tempered the language in the abstract and the conclusions sections and have limited the text to what we have discovered through molecular modeling of these killer toxins. We consider the work to provide novel structural insights into killer toxins as we have moved beyond the publically available alphafold models to perform rigorous molecular dynamics simulations. This manuscript is the first to describe models for the majority of the Saccharomyces toxins, as only K2 and K62 have been published to date.

Figures are numerous but often repetitive in presentation, authors could summarise or compare them more effectively.

Comment. The figures are presented in the same format to allow for direct comparison between each of the killer toxins. If the figures were presented in different formats, then comparison would be more difficult. We previously included several supplementary figures (S2, S5, S7) and a summary figure (Figure 14) to facilitate comparisons.

Results and Discussion are merged. The separation of descriptive results from interpretive discussion would improve focus and readability.

Done. We have now added extra subtitles to help the readers navigate the paper to more clearly deliminate the introductions of each toxin family from the molecular modeling results. The paper ends with a conclusion section, which, with Figure 14, brings all of the families together for a final comparison.

Reviewer 3 Report

Comments and Suggestions for Authors

This manuscript presents a comprehensive analysis of all canonical Saccharomyces killer toxins, combining AlphaFold2-generated models and molecular dynamics (MD) simulations with empirical data. Key findings include the identification of a "K1 superfamily", the reclassification of K62 as an aerolysin-family toxin, and the proposal of a rational nomenclature system for killer toxin families. I truly appreciate this work that I find scientifically sound and  timely, and that represents an advance in the structural biology of killer toxins. There are only some concerns regarding the organisation of the manuscript, the handling of low-confidence structural models and the inferential nature of  structural claims. All these points are detailed below.  However, in my opinion that these points can be addressed through minor revision, after which I would certainly recommend the manuscript for publication.

(1)The manuscript is pretty long  and I find a structural challenge in its organization: for each toxin family, extensive  background is presented together with the novel computational results, making it difficult for the reader to clearly distinguish what constitutes the original contribution of this work from what was previously known. In its current form, each section appears in part as a narrative review of the existing literature and in part as a structural analysis, without a clear boundary between the two. This is particularly evident for well-studied toxins such as K1, K2, and K28, where much empirical background precede the modeling results, risking that the computational findings may not be perceived as the central contribution of the study. Indeed, this imbalance is less pronounced for poorly characterised toxins such as K62, Klus, and KHR, where the computational contribution carries greater weight due to the scarcity of empirical data. In my opinion the authors could consider delineating prior empirical knowledge from newly generated computational findings This would substantially improve readability and allow to better appreciate the genuine contribution of the study.

(2) The authors handle the low-confidence K28 model with caution and conclude that the structure and antifungal mechanism of K28 remain enigmatic. This reflects a the appreciation of the limitations of AI-based structural prediction. In my opinion the same level of caution should be applied to K74. In this case the low confidence of the AlphaFold2 model led the authors to base their analysis on a homolog from Cadophora malorum.  In spite of that the toxin is included in the K1 superfamily classification with toxins with considerably more reliable models. I wonder whether the authors can distinguish, particularly in the concluding sections and in Figure 14, between mechanistic predictions based on high-confidence models and those derived from low-confidence predictions, so that readers can calibrate their interpretation of the proposed classification.

(3) Figures S2 and S3, which present the 3D structural models coloured by pLDDT confidence  mix two biologically distinct phenomena: intrinsic protein disorder and model uncertainty. Although the authors address this distinction in the text, the figures alone do not allow readers to differentiate between low-confidence regions and those that indicate a near-complete failure of structural prediction, as in K28. In my opinion this is problematic when multiple toxin models are displayed side by side, as the shared colour scale implicitly invites direct comparison without signalling differences in model reliability. I would suggest to supplement the standard pLDDT colouring with an annotation distinguishing disordered regions from regions that are poorly predicted due to insufficient sequence homology.

(4) The reassignment of the Klus domain order from the canonical delta/alpha/gamma/beta to gamma/delta/alpha/beta is unexpected in the absence of direct experimental validation. This concern is reinforced by the proposed domain organisation of KHR for which a different configuration is suggested. This inconsistency within the same toxin family raises the question whether alternative domain assignments might fit the data equally well. For example, could the N-terminal domain of Klus be interpreted as a structurally unusual delta domain? This would preserve the conventional N-terminal delta assignment with an atypical structure, rather than requiring a complete inversion of the domain order.  

Author Response

This manuscript presents a comprehensive analysis of all canonical Saccharomyces killer toxins, combining AlphaFold2-generated models and molecular dynamics (MD) simulations with empirical data. Key findings include the identification of a "K1 superfamily", the reclassification of K62 as an aerolysin-family toxin, and the proposal of a rational nomenclature system for killer toxin families. I truly appreciate this work that I find scientifically sound and  timely, and that represents an advance in the structural biology of killer toxins. There are only some concerns regarding the organisation of the manuscript, the handling of low-confidence structural models and the inferential nature of  structural claims. All these points are detailed below.  However, in my opinion that these points can be addressed through minor revision, after which I would certainly recommend the manuscript for publication.

1. The manuscript is pretty long  and I find a structural challenge in its organization: for each toxin family, extensive  background is presented together with the novel computational results, making it difficult for the reader to clearly distinguish what constitutes the original contribution of this work from what was previously known. In its current form, each section appears in part as a narrative review of the existing literature and in part as a structural analysis, without a clear boundary between the two. This is particularly evident for well-studied toxins such as K1, K2, and K28, where much empirical background precede the modeling results, risking that the computational findings may not be perceived as the central contribution of the study. Indeed, this imbalance is less pronounced for poorly characterised toxins such as K62, Klus, and KHR, where the computational contribution carries greater weight due to the scarcity of empirical data. In my opinion the authors could consider delineating prior empirical knowledge from newly generated computational findings This would substantially improve readability and allow to better appreciate the genuine contribution of the study.

Done. Based on the reviewers' comments, we have added many more subheadings in the text and clearly labeled them as “introduction” or indicated that they are results, depending on the section. We feel that this better delineates the sections of the manuscript and has increased the readability. Moreover we have also tried to be more succinct in the K1 and K2 sections and slightly reduced the text.

 

2. The authors handle the low-confidence K28 model with caution and conclude that the structure and antifungal mechanism of K28 remain enigmatic. This reflects a the appreciation of the limitations of AI-based structural prediction. In my opinion the same level of caution should be applied to K74. In this case the low confidence of the AlphaFold2 model led the authors to base their analysis on a homolog from Cadophora malorum.  In spite of that the toxin is included in the K1 superfamily classification with toxins with considerably more reliable models. I wonder whether the authors can distinguish, particularly in the concluding sections and in Figure 14, between mechanistic predictions based on high-confidence models and those derived from low-confidence predictions, so that readers can calibrate their interpretation of the proposed classification.

Done. We have now added more text highlighting the low confidence of the K74 model: Line 866 “…so we cannot be confident about the structure of this toxin.”. We have amended Figure 14 to include a disclaimer regarding the low confidence of K74 and have only shown the predicted structure of the confident model of the K74 homolog from C. malorum. In the conclusion section we also add additional text to ensure that the reader understands the caveats regarding our analysis of the K74 family and its inclusion in the K1 superfamily. Line 1657 “Some of the structural models in this group have lower confidence regions, or in the case of K74, had a low overall confidence score that forced the modeling of a sequence homolog that has not been confirmed as a killer toxin (KTS1^Cmal).”

 

3. Figures S2 and S3, which present the 3D structural models coloured by pLDDT confidence mix two biologically distinct phenomena: intrinsic protein disorder and model uncertainty. Although the authors address this distinction in the text, the figures alone do not allow readers to differentiate between low-confidence regions and those that indicate a near-complete failure of structural prediction, as in K28. In my opinion this is problematic when multiple toxin models are displayed side by side, as the shared colour scale implicitly invites direct comparison without signalling differences in model reliability. I would suggest to supplement the standard pLDDT colouring with an annotation distinguishing disordered regions from regions that are poorly predicted due to insufficient sequence homology.

Done. Firstly, we realized that the attached supplementary files are of lower quality than would allow a reader to look closely at Figure S2 to see that for some proteins, the areas of disorder correlate well with regions with low pLDDT. We have now included a high-resolution image. Secondly, we have recolored the secondary structure predictions in Figure S3 to better illustrate the predicted regions of secondary structure and disorder, and how they map to the per-residue pLDDT values. These figures now distinguish between disordered regions and regions that are poorly predicted by AlphaFold.

4. The reassignment of the Klus domain order from the canonical delta/alpha/gamma/beta to gamma/delta/alpha/beta is unexpected in the absence of direct experimental validation. This concern is reinforced by the proposed domain organisation of KHR for which a different configuration is suggested. This inconsistency within the same toxin family raises the question whether alternative domain assignments might fit the data equally well. For example, could the N-terminal domain of Klus be interpreted as a structurally unusual delta domain? This would preserve the conventional N-terminal delta assignment with an atypical structure, rather than requiring a complete inversion of the domain order.  

Comment: We agree that this altered organization was quite an interesting finding! We are confident in our prediction for several reasons, the most straightforward assignment was the lpha domain as in both models it has the hydrophobic helix that is cradles by a beta sheet. The other domains depend on their contribution to the beta sheet that wraps the alpha domain helix. In other killer toxins, these sheets are donated from the gamma and beta domains and not the delta domain. In our models we assign the delta domain in the same way – it does not contribute to the beta sheet. This leaves two other domains, one that we describe as beta because of it size, position at the C-terminus and close association with the predicted alpha domain. That leave the naming of the gamma domain which is positioned different ly in both Klus and KHS.

We appreciate that this was not well described in the text and have included a more detailed description: line 1130 – “The inclusion of 1β into the β-sheet that wraps the hydrophobic α-helix is similar to the gamma-domain interactions observed in other Saccharomyces killer toxins. This justifies assigning the N-terminal domain as gamma and the second domain as delta, since the latter does not contribute structurally to the beta sheet wrapping the alpha domain helix.” and Line 1159 – “As with all other killer toxins of similar structure, the delta domain does not contribute beta strands to the structure of the beta sheet and is assigned as the first domain of KHR.”

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript presents a comprehensive structural and functional analysis of Saccharomyces killer toxins using AlphaFold-based modeling and molecular dynamics simulations integrated with existing empirical knowledge. Overall, the manuscript is suitable for publication after minor revisions.

1. While the authors have addressed concerns regarding confidence levels, the manuscript would benefit from more consistent language distinguishing predicted structural/functional hypotheses vs. experimentally validated mechanisms. In several sections, mechanistic interpretations (e.g., ionophoric activity, membrane insertion, oligomerization) are still presented with strong wording. Even if supported by high-confidence regions, these should be systematically framed as model-informed hypotheses, especially for less-characterized toxins.

2. The authors provide justification that key regions discussed are of moderate-to-high confidence, which addresses the original concern. However, for clarity to readers, it would be helpful to explicitly state confidence thresholds used for interpretation (e.g., pLDDT >70 or >80) and briefly clarify how low-confidence regions were handled (excluded vs. cautiously interpreted). 

3. The explanation of MD simulations has improved. However, the manuscript would benefit from a slightly clearer distinction between conclusions supported by MD-derived stability vs. those based on static AlphaFold structures. Additionally, while convergence behavior is described, a brief explicit rationale for using 1 μs simulations (e.g., consistency across systems, capturing slow conformational shifts) would further strengthen this section.

4. The authors justify the current structure, and the organization by toxin families is reasonable. However, minor tightening such as reducing repetition of general features (e.g., domain organization, ionophoric mechanism) where already introduced and emphasizing comparative differences rather than restating shared features will improve readability. 

5. The authors indicate that Figure 14 and supplementary figures address integration. This is acceptable; however, to improve accessibility, ensure that Figure 14 (or its legend) clearly communicates toxin family classification, structural confidence (directly or via reference), and mechanistic grouping. 

   •  

 

Author Response

The manuscript presents a comprehensive structural and functional analysis of Saccharomyces killer toxins using AlphaFold-based modeling and molecular dynamics simulations integrated with existing empirical knowledge. Overall, the manuscript is suitable for publication after minor revisions.

1. While the authors have addressed concerns regarding confidence levels, the manuscript would benefit from more consistent language distinguishing predicted structural/functional hypotheses vs. experimentally validated mechanisms. In several sections, mechanistic interpretations (e.g., ionophoric activity, membrane insertion, oligomerization) are still presented with strong wording. Even if supported by high-confidence regions, these should be systematically framed as model-informed hypotheses, especially for less-characterized toxins.

 

Done. We have added clearer subheadings to the manuscript that indicate sections that are dedicated to results. We have also made it clear from these subtitles that mechanistic insights into these toxins is based on structural models. We have also amended several sections to more clearly define empirical from modeling data and temper the language. Such as: Line 1038 “The central α-helix of the K1, K1L, K2, K21, KHS, and K45 killer toxins was consistently located within the alpha domain, which has been empirically shown to be responsible for cytotoxicity in K1 and K2 [115,170].” Line 1443 “The similarity of the modeled structures of Klus and KHR to those of other alpha/beta sandwich proteins provides insight into their potential mechanisms of action.” Line 1234 “The modeled 1α helix is amphipathic and, according to PSIPRED, is predicted to be pore-lining, consistent with a possible ionophoric mechanism of action.” Line 1082 “ The close predicted structural similarities of these toxins also support the proposal that their mechanism of intoxication is conserved and that they represent a broader “K1 superfamily””

 

1. The authors provide justification that key regions discussed are of moderate-to-high confidence, which addresses the original concern. However, for clarity to readers, it would be helpful to explicitly state confidence thresholds used for interpretation (e.g., pLDDT >70 or >80) and briefly clarify how low-confidence regions were handled (excluded vs. cautiously interpreted). 

Done: We have stated in line 276: “Values greater than 80.0 are considered of high confidence and most often correlate with regions of secondary structure, while lower scores indicate less confident structural predictions and disordered regions between elements of secondary structure.”  As we draw our functional conclusions from the MD simulations (see our comment to address additional reviewers concerns below), we have provided the pLDDT scores mainly to highlight the low confidence alphafold models of K74 and K28. Based on the concerns of another reviewer we have also added more text highlighting the low confidence of the K74 model: Line 954 “…so we cannot be confident about the structure of this toxin.”. We have amended Figure 14 to include a disclaimer regarding the low confidence of K74 and have only shown the predicted structure of the confident model of the K74 homolog from C. malorum. In the conclusion section we also add additional text to ensure that the reader understands the caveats regarding our analysis of the K74 family and its inclusion in the K1 superfamily. Line 1792 “Some of the structural models in this group have lower confidence regions, or in the case of K74, had a low overall confidence score that forced the modeling of a sequence homolog that has not been confirmed as a killer toxin (KTS1^Cmal).” Discussion of K28 was similarly cautious.  

 

1. The explanation of MD simulations has improved. However, the manuscript would benefit from a slightly clearer distinction between conclusions supported by MD-derived stability vs. those based on static AlphaFold structures.

Done. As suggested by the reviewer, we have clarified in the text in Section 2.2. Molecular Modeling of Saccharomyces Killer Toxins to clearly indicate that all mechanistic predictions were derived from models of Saccharomyces killer toxins: Line 319 – “Structural models generated by MD simulations were primarily used to infer killer toxin structure and function of all Saccharomyces killer toxins.”

 

1. Additionally, while convergence behavior is described, a brief explicit rationale for using 1 μs simulations (e.g., consistency across systems, capturing slow conformational shifts) would further strengthen this section.

Done. We have added the rationale behind the 1 us simulation. Line 290 “The 1.0 μs simulations allow us to capture slow backbone and loop dynamics that are not adequately sampled at shorter timescales and provide a more thorough test of protein structure stability.”

 

1. The authors justify the current structure, and the organization by toxin families is reasonable. However, minor tightening such as reducing repetition of general features (e.g., domain organization, ionophoric mechanism) where already introduced and emphasizing comparative differences rather than restating shared features will improve readability. 

Done.  We have removed repetitive text from the discussion of the K2 family that had three similar toxins with an overall similar description. This includes placing the domain descriptions and order in parentheses so as to improve flow. The same was done for the K1 family, although we left more details as it is the first family discussed and still has the most available information. This provides context for the remaining killer toxins. The KHR/Klus family section was also improved by removing redundancy and tightening the KHR description. All other toxins families have only one toxin that is discussed so there were no edits to reduce redundancy needed.

 

1. The authors indicate that Figure 14 and supplementary figures address integration. This is acceptable; however, to improve accessibility, ensure that Figure 14 (or its legend) clearly communicates toxin family classification, structural confidence (directly or via reference), and mechanistic grouping. 

Done. In addition to the text in the figure, we have added a legend entry describing the grouping of Saccharomyces killer toxins by families and the definition of the K1 superfamily. We also added additional text descriptions in the text to indicate the predicted mechanisms of these ionophoric  toxins with a specific reference to the aerolysin-like beta barrel pores. We also include a reference to figures containing model confidence data and disclaimers about the confidence of the K74 and K28 models. The legend now reads, line 1777 – “Figure 14. Summary of the proposed family and superfamily organization of ionophoric Saccharomyces killer toxins. Confident tertiary structure models of killer toxins are summarized and grouped into the K1, K2, K45, K74, Klus, and K62 families. Based on tertiary structure homology, the K1, K2, K45, and K74 families are also proposed as the K1 superfamily. Tertiary structure models are colored by domains as depicted with delta (black), alpha (yellow), gamma (cyan), beta (dark blue), with the exception of K62, which is colored according to the predicted aerolysin beta barrel pore-forming core domain (yellow) and N-terminal domain (dark blue). **The K74 family is included for comparison despite low-confidence predictions of the K74 tertiary structure. K28 is not included due to poor confidence of its molecular models. Metrics of modeling confidence for all killer toxins can be found in Figures S2, S3, and S4.”