eLife Assessment
Mechanical transduction channels of sensory hair cells possess lipid scramblase activity. Membrane lipid disruption resulting from mechanical transduction is thought to be restored by flippase activities. This fundamental study provides compelling evidence that ATP8B1, a P4-ATP flippase and its subunit TMEM30B, are key in mediating this restorative function in outer hair cells of the mammalian cochlea.
Reviewer #1 (Public review):
Sensory hair cells of the inner ear convert mechanical sound vibrations into electrical signals through mechano-electrical transduction (MET). While the protein components of the MET machinery have been studied extensively, much less is known about how the surrounding membrane lipid environment contributes to hair cell function. The recent discovery that TMC1 and TMC2 also function as lipid scramblases has brought renewed attention to the importance of membrane lipid asymmetry and the mechanisms that maintain it in sensory hair cells.
In this study, the authors identify the P4-ATPase ATP8B1 and its partner TMEM30B as key regulators of membrane lipid asymmetry in outer hair cells. Using complementary genetic models, HA-tagged knock-in mice, localization analyses, and functional experiments, they show that ATP8B1-TMEM30B is enriched in stereocilia and the apical membrane of outer hair cells and is required to maintain phosphatidylserine asymmetry, support hair cell survival, and preserve normal hearing. The parallels between the ATP8B1/TMEM30B loss-of-function phenotypes and TMC1 deafness-associated mutants with constitutive scrambling support a model in which ATP8B1-TMEM30B flippase activity maintains membrane lipid asymmetry and homeostasis, whereas constitutive TMC1-mediated phospholipid scrambling disrupts this balance and contributes to membrane instability.
The authors have addressed the points raised during the initial review thoroughly. The revised manuscript includes clearer methodological details, additional physiological characterization, improved presentation and quantification of several datasets, and a more balanced interpretation of the localization and mechanistic findings. These changes improve both the clarity and rigor of the study while leaving its main conclusions unchanged.
As with any study that opens a new area of investigation, important mechanistic questions remain. In particular, it will be interesting to determine how disruption of membrane lipid asymmetry ultimately impairs MET function and triggers hair cell degeneration, how flippase and scramblase activities are coordinated in vivo, and how these pathways are integrated with the broader molecular machinery underlying mechanotransduction. These questions highlight the exciting directions that this study opens for the field.
Overall, this work provides evidence that ATP8B1-TMEM30B is a critical regulator of stereocilia membrane lipid asymmetry and represents an important contribution to our understanding of membrane homeostasis in auditory hair cells. I have no further major concerns and support publication.
Comments on second revision:
I appreciate the authors' efforts to address the issue and perform the additional experiment, and I agree that removing these data is reasonable and does not affect the main conclusions of the manuscript.
Reviewer #2 (Public review):
Summary:
Prior work identified TMEM30B (knockout mice) as well as ATP8B1 (human genetics and mouse model), ATP8A2 (knockout mice), and ATP811A (human genetics) as relevant for hearing. The authors also reasoned that given the recent discovery of TMC1 and TMC2's dual function as mechanotransduction channels of the inner ear and as lipid scramblases, a counterpart flippase should be in the sensory hair-cell stereocilia bundle where mechanotransduction happens. They use CRISPR/CAS to modify the endogenous mouse genes and add an HA tag at the N-terminus of the ATP8B1, ATP8A1, ATP8A2, and ATP11A proteins. Their experiments with these mice unambiguously localized ATP8B1 at the base of outer hair cell stereocilia bundles. Knockout of ATP8B1 results in loss of outer hair cells, deficient auditory function (ABR), and degeneration of outer hair cell stereocilia bundles. Similarly, hair cells from genetically modified mice with endogenous HA-tagged TMEM30B proteins show localization of this protein to outer hair cell stereocilia bundles. TMEM30B knock out mice phenocopy the ATP8B1 knock out model. Interestingly, the authors show that annexing V staining precedes hair cell loss in ATP8B1 and TMEM30B knockout mice and that proper localization of these proteins is lost in mice that lack CIB2, a protein essential for hair cell mechanotransduction.
Strengths:
(1) Use of knock-in HA-tagged proteins to unambiguously localize ATP8B1 and TMEM30B
(2) Systematic characterization of auditory function (ABR), hair cell loss, and hair-cell stereocilia bundle morphology.
(3) Advances our understanding of the role played by lipid homeostasis in auditory function.
(4) Reports on mouse models that will be helpful to further understand the mechanistic role played by ATP8B1 and TMEM30B in normal hearing and hereditary deafness.
Comments on second revision:
It is interesting that in the images selected as representative of the new experiments the effect seems to be opposite to what was reported before - more ATP8B1-HA in the bundle with transduction blockage. However, this does not seem to be supported by the quantification, and further exploration at different ages, blocker concentrations, effective transduction blockage, etc. might be needed to really have a solid result. Thus, I agree that leaving out these results, and the previous sections linking mechanotransduction to localization of ATP8B1, is the best option.
Author response:
The following is the authors’ response to the previous reviews.
Public Reviews:
Reviewer #1 (Public review):
Sensory hair cells of the inner ear convert mechanical sound vibrations into electrical signals through mechano-electrical transduction (MET). While the protein components of the MET machinery have been studied extensively, much less is known about how the surrounding membrane lipid environment contributes to hair cell function. The recent discovery that TMC1 and TMC2 also function as lipid scramblases has brought renewed attention to the importance of membrane lipid asymmetry and the mechanisms that maintain it in sensory hair cells.
In this study, the authors identify the P4-ATPase ATP8B1 and its partner TMEM30B as key regulators of membrane lipid asymmetry in outer hair cells. Using complementary genetic models, HA-tagged knock-in mice, localization analyses, and functional experiments, they show that ATP8B1-TMEM30B is enriched in stereocilia and the apical membrane of outer hair cells and is required to maintain phosphatidylserine asymmetry, support hair cell survival, and preserve normal hearing. The parallels between the ATP8B1/TMEM30B loss-of-function phenotypes and TMC1 deafness-associated mutants with constitutive scrambling support a model in which ATP8B1-TMEM30B flippase activity maintains membrane lipid asymmetry and homeostasis, whereas constitutive TMC1-mediated phospholipid scrambling disrupts this balance and contributes to membrane instability.
The authors have addressed the points raised during the initial review thoroughly. The revised manuscript includes clearer methodological details, additional physiological characterization, improved presentation and quantification of several datasets, and a more balanced interpretation of the localization and mechanistic findings. These changes improve both the clarity and rigor of the study while leaving its main conclusions unchanged.
As with any study that opens a new area of investigation, important mechanistic questions remain. In particular, it will be interesting to determine how disruption of membrane lipid asymmetry ultimately impairs MET function and triggers hair cell degeneration, how flippase and scramblase activities are coordinated in vivo, and how these pathways are integrated with the broader molecular machinery underlying mechanotransduction. These questions highlight the exciting directions that this study opens for the field.
Overall, this work provides evidence that ATP8B1-TMEM30B is a critical regulator of stereocilia membrane lipid asymmetry and represents an important contribution to our understanding of membrane homeostasis in auditory hair cells. I have no further major concerns and support publication.
Reviewer #2 (Public review):
Summary:
Prior work identified TMEM30B (knockout mice) as well as ATP8B1 (human genetics and mouse model), ATP8A2 (knockout mice), and ATP811A (human genetics) as relevant for hearing. The authors also reasoned that given the recent discovery of TMC1 and TMC2's dual function as mechanotransduction channels of the inner ear and as lipid scramblases, a counterpart flippase should be in the sensory hair-cell stereocilia bundle where mechanotransduction happens. They use CRISPR/CAS to modify the endogenous mouse genes and add an HA tag at the N-terminus of the ATP8B1, ATP8A1, ATP8A2, and ATP11A proteins. Their experiments with these mice unambiguously localized ATP8B1 at the base of outer hair cell stereocilia bundles. Knockout of ATP8B1 results in loss of outer hair cells, deficient auditory function (ABR), and degeneration of outer hair cell stereocilia bundles. Similarly, hair cells from genetically modified mice with endogenous HA-tagged TMEM30B proteins show localization of this protein to outer hair cell stereocilia bundles. TMEM30B knock out mice phenocopy the ATP8B1 knock out model. Interestingly, the authors show that annexing V staining precedes hair cell loss in ATP8B1 and TMEM30B knockout mice and that proper localization of these proteins is lost in mice that lack CIB2, a protein essential for hair cell mechanotransduction.
Strengths:
(1) Use of knock-in HA-tagged proteins to unambiguously localize ATP8B1 and TMEM30B
(2) Systematic characterization of auditory function (ABR), hair cell loss, and hair-cell stereocilia bundle morphology.
(3) Advances our understanding of the role played by lipid homeostasis in auditory function.
(4) Reports on mouse models that will be helpful to further understand the mechanistic role played by ATP8B1 and TMEM30B in normal hearing and hereditary deafness.
Weaknesses:
(1) Are the HA tags causing any functional issues? Function and localization of tagged proteins can sometimes be compromised. This is checked for TMEM30B and ATP8B1, but not for ATP8A1, ATP8A2, and ATP11A.
(2) Following on the point above, is it possible that ATP8B1-HA is well localized, but localization for the other three flippases (ATP8A1-HA, ATP8A2-HA, and ATP11A-HA) is compromised by the tag? Is this potential miss-localization causing any functional phenotypes? I find surprising that there are flippases only in outer hair cells and only formed by ATP8B1. A possible explanation is that the tag is interfering with trafficking. If so, there should be a phenotype (ABRs), although this might be masked by redundancy among these flippases or caused by systemic issues (admittedly difficult to sort out).
We would like to apologize for the oversight on our end that caused the delay and repeat consultation with you, and would like to thank you for your patience.
As a reminder, after our manuscript underwent the full review, we discovered an issue with the final figure of the paper. This experiment is ancillary to the main story, and its removal does not alter the paper’s central findings or conclusions. This Figure 7 asked whether genetic loss of mechanotransduction (MET) alters ATP8B1/TMEM30B distribution in hair cells. After we discovered a problem with the MET-negative mouse model, we agreed with the reviewers to instead test whether pharmacological MET block has the same effect. We found that a 30-minute incubation with benzamil (MET blocker) did not alter ATP8B1 distribution (see uploaded PDF named “ATP8B1-HA_Benzamil”).
Because this experiment was limited in scope and is difficult to compare directly with genetic deletion of MET, we think it is best to proceed with the manuscript version that removes this line of inquiry.
We have made the following changes (compared to the most recent submission):
- Removed Figure 7
eLife Assessment
Franziscus et al. present an important study introducing SPEx, an elegant, broadly accessible approach to subcellular spatial proteomics that combines cell expansion, laser microdissection, and mass spectrometry to enable protein profiling within defined subcellular regions. The authors provide convincing evidence that this flexible integration of established techniques offers a robust, practical, and cost-effective strategy for compartment-resolved proteomic analysis. The method is appropriately validated across multiple subcellular compartments and complementary workflows, demonstrating its ability to recover known protein markers while identifying previously uncharacterised components. The study also provides a characterisation of the method's capabilities and limitations. Overall, SPEx is likely to be of broad interest to cell biologists seeking to map organelle composition.
Reviewer #1 (Public review):
Summary:
The authors present a novel approach to subcellular spatial proteomics by combining laser microdissection with expansion microscopy and LC-MS/MS analysis (SPEx). They implement two different workflows for LMD and LC-MS/MS quantification:
(1)The standard approach in which an area of interest is cut out by LMD, subjected to proteomics analysis and compared to the rest of the cell without the dissected ROI.
(2) The subtraction approach in which ROIs are removed and the remaining cellular material is compared to samples containing both the surrounding material and the ROI.
The authors assess the technique by applying it to subcellular targets of various size, volume and protein compositions such as the nucleus, nucleoli and Golgi. They demonstrate that SPEx can identify proteins enriched or reduced in ROIs.
Strengths:
The general, relatively easy and inexpensive nature of this approach, as well as its potential applicability to many cell types and subcellular areas of interest provides an exciting alternative to subcellular fractionation, native immunoprecipitation or genetically encoded proximity labeling constructs. Moreover, by visually selecting ROIs for subsequent analysis, subcellular context or organelle morphology can be taken into account as discussed by the authors in the discussion section.
Weaknesses:
The authors addressed our concerns and technical questions in revision.
A main limitation of the current method is that it relies on isolation of structures of interest in 2D. The authors discuss some alternative approaches that might allow future access to more 3D target isolation.
There remain several unexplored minor extensions of the technical development of the method, including comparing the proteomes of organelles that are visually distinct, and applying different anchoring procedures to try to increase protein retention.
Reviewer #2 (Public review):
Summary:
This study introduces SPEx, an approach that combines physical expansion, imaging-guided microdissection, and mass spectrometry for compartment-resolved proteomic analysis. Using the nucleus, nucleolus, and Golgi apparatus as examples, the authors demonstrate that SPEx can identify proteins enriched or depleted in defined subcellular regions.
Strengths:
A major strength of the method is its accessibility and flexibility. It combines established technologies without requiring genetic manipulation or specialized proximity-labeling strategies, while direct imaging-based selection allows regions of interest to be defined according to their spatial and morphological context. The complementary comparative strategies used in the study provide convincing support for compartment-specific protein enrichment and depletion.
An important limitation is the lack of depth resolution during microdissection, which can introduce contributions from material above or below the selected region, particularly for small structures. Appropriate comparative controls are therefore important for interpreting SPEx measurements. The current study is also primarily a proof of principle in a single cultured cell type, and its ability to capture changes across biological states or perturbations remains to be established.
Overall, the study provides convincing evidence that SPEx is a useful and flexible addition to the spatial proteomics toolbox, with particular potential for applications requiring direct image-based selection of subcellular regions.
Reviewer #3 (Public review):
Franziscus et al. describe an elegant approach for spatially specific proteome analysis. To achieve this, they expand fixed cells and subsequently use a laser to micro-dissect a region of interest, which is then analyzed by MS.
They demonstrate the effectiveness of their approach by analyzing nucleus, nucleolus and the Golgi and benchmark their hits against previous dataset for these organelles.
The manuscript is very well written and nicely guides the reader through the applied methods. The presented data is convincing and I do not see the need for additional experimental verification of the protocol. The only minor concern is the novelty of the method and the presentation. A combination of expansion, laser microdissection and proteomics has been applied in the past (PMID: 36450705, PMID: 39477916). In the manuscript one of these studies is cited, though it does not become clear that this approach is already described. However, Franziscus et al. describe the approach better and makes it more accessible to the reader, especially since the other studies described this methodology in combination with tissue expansion and not in combination with single cell expansion as it is done here. I would ask the authors to be clearer in the introduction with what others have already done and what is there contribution here. In general, I am convinced that the community will benefit from the presented protocol to analyze organelle proteomics in detail.
Comments on revised version.
I had only minor issues with the manuscript, which have now been addressed.
Author response:
The following is the authors’ response to the original reviews.
Reviewer #1 (Public review):
Summary:
The authors present a novel approach to subcellular spatial proteomics by combining laser microdissection with expansion microscopy and LC-MS/MS analysis (SPEx). They implement two different workflows for LMD and LC-MS/MS quantification:
(1)The standard approach, where an area of interest is cut out by LMD, subjected to proteomics analysis, and compared to the rest of the cell without the dissected ROI.
(2) The subtraction approach, where ROIs are removed, and the remaining cellular material is compared to samples containing both the surrounding material and the ROI.
The authors assess the technique by applying it to subcellular targets of various sizes, volumes, and protein compositions such as the nucleus, nucleoli, and Golgi. They demonstrate that SPEx can identify proteins enriched or reduced in ROIs.
Strengths:
The broad, relatively easy, and inexpensive applicability of this approach to potentially many cell types and subcellular areas of interest provides an exciting alternative to subcellular fractionation, native immunoprecipitation, or genetically encoded proximity labeling constructs. Moreover, by visually selecting ROIs for subsequent analysis, subcellular context or organelle morphology can be taken into account, as discussed by the authors in the discussion section.
Weaknesses:
While strongly supporting the sharing of this approach, we have a number of comments and questions that will improve the impact of the manuscript:
We thank the reviewer for the careful evaluation of our manuscript and the generally positive assessment.
(1) General:
(a) The manuscript would benefit from restructuring and language revision. In its current form, the writing is sometimes dense and verbose (in particular, the Results section). This makes it difficult to follow the authors' arguments.
We improved readability and clarity in the revised manuscript. We added subheadings to facilitate reading.
(b) The authors mention the possibility of selecting organelles based on morphology. This is left for the discussion, but it seems like a missed opportunity - the authors could compare individual organelles in different morphological states, e.g., connected vs. fragmented mitochondria.
The authors agree with the reviewers’ assessment that investigating proteome of organelles based on morphology or cellular state is an exciting application of SPEx. While we plan experiments along this line in the future, we think that these experiments are beyond the scope of this manuscript, which is meant to describe the method and its general usefulness.
(2) Technical:
(a) Why do the authors strive and optimize for a 10x expansion factor? Is SPEx compatible with a more standard 4x expansion, as e.g., used in the classic U-ExM approach (https://www.nature.com/articles/s41592-018-0238-1)? This could be added to the discussion.
We aimed for 10x expansion solely because our ultimate goal is to cut out very small structures. 10x expansion results in a 100x increase in area, while in U-ExM the area increase would only by 16x. Thus, this makes a huge difference for small structures such as nucleoli. We did not assess the compatibility with U-ExM. Still, we performed pilots with just 4x expansion (using TREx) in the early stages of optimization. We were able to isolate single cells and obtain similar protein coverage as with 10x expansion. We further clarified our motivation to use 10x expansion in the discussion.
(b) The U-ExM approach shows improved ultrastructural preservation when using 3%FA with 0.1% glutaraldehyde fixation (GA). Is SPEx compatible with the use of low amounts of GA for fixation?
We tried different fixation methods in the early stages of this study (where expansion was not yet close to 10x). We saw a mild negative effect of GA on the expansion factor, so we avoided it in the later experiments since it also did not seem necessary to preserve the structure of our organelles of interest. However, the use of GA would generally be compatible with SPEx, potentially at the cost of a mild negative effect on expansion factor (see Author response image 1) and proteome coverage. We discussed the use of GA in the discussion section.
Author response image 1.
Fixation methods mini-screen. Cells were fixed with the indicated reagents for 10 minutes at 37°C. After TREx expansion, the diameter of the nucleus was measured (A) and the resulting expansion factor compared to the non-expanded control was determined (B).
Related to the above, was the anchoring efficiency reduced only to achieve a 10x expansion factor or does this additionally affect the proteome coverage?
We solely lowered the anchoring to increase expansion factors. In earlier pilots we performed proteomic analysis on samples that were just expanded 4x using standard TREx expansion (also using the original anchoring strategy from the TREx publication, consisting of 0.2 mg/ml AcX for overnight at RT). We presented the results of this pilot in Fig S1A. We still detected over 2,000 proteins from 10 cells, a coverage, which is highly similar to what we found in the final experiments (Figure 2F), in which the anchoring was lower for 10x expansion. Based on these data, we think that anchoring (and expansion factor!) has a negligible impact on protein coverage. We clarified this point in the manuscript.
(d) Have the authors considered using alternative anchoring approaches, such as GMA (https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0291506#pone.0291506.s001), which potentially increase the amount of sample retained in the hydrogel, thus allowing for better proteome coverage? This could be added to the discussion.
We did not use alternative anchoring approaches. We modified the TREx protocol to fit our purposes and since this was sufficient, we did not explore alternatives. However, using anchoring approaches, in which higher amounts of sample could be retained in the gel might be beneficial for the proteomics coverage. We will keep this suggestion in mind for future experiments. Thank you for the suggestion!
(e) The limitation of the approach to near-2D samples should be mentioned, and alternative approaches for more 3D samples could be discussed.
The authors agree that SPEx is limited to near-2D samples at this point. We suggest that SPEx is applicable for 3D samples (e.g. in tissues) by performing cryosectioning. TREx has been shown to be compatible with sectioned tissue (Damstra et al., 2022). We discuss this point in the discussion.
(f) How are peptides that are directly anchored to the hydrogel dealt with during LC-MS/MS analysis? Are they excluded, or can they be identified during the spectral search? The latter would allow us to get a deeper structural understanding of how proteins are actually anchored into hydrogels, which so far has not been assessed.
The reviewer raises an interesting point. In general, peptides carrying the anchoring modification are analysed by LC-MS, but we did not include these specific modifications in the database search. Overall, we assumed that the labeling would be low and stochastic and hence should, if at all, only minimally affect the detection of peptides. Nevertheless, in response to the reviewers’ comment, we searched the MS data again for the crosslinking reagent linked to lysine residues. However, we could not get any confident hit for any peptide containing this modification. Since we cannot exclude that the modification precludes the identification of the corresponding peptides, we compared the number peptides generated by trypsin cleavage after arginine and lysine. As the human genome contains similar proportions of both amino acids, one would expect similar numbers for both peptide types being identified. Any modifications of lysine by the anchoring reagent, would prevent tryptic cleavage and thus reduce the number of lysine peptides. As shown in Figure S1C, the number of lysine-terminating is only slightly lower compared to arginine-terminating peptides. Notably, the proteomics results of a different fixed human tissue sample directly extracted by laser capture micro dissection without expansion showed a very similar lysine-to-arginine peptide ratio. This indicates that the large majority of lysine residues is not modified and affected by the hydrogel anchoring. We included these data into the manuscript.
An alternative approach to address this question would be to investigate if the peptide coverage of proteins detected by SPEx is enriched for peptides representing the folded core of proteins as opposed to the surface-exposed regions, which likely get more anchored into the hydrogel.
Because of the negligible amounts of modified peptides, we did not investigate this potential bias of surface-exposed versus folded-core peptides.
(g) Same question regarding peptides with NHS labeling. Can they be identified, or do they just compete for ionization and thus negatively affect coverage and dynamic range of the LC-MS/MS approach?
The reviewer raises a similar point as above for another lysine labeling used during the SPEx protocol. Again, we specifically looked for this modification by re-searching the raw MS data, but still could not identify any peptides, carrying this modification on a lysine residue. Even though we cannot exclude that this rather large modification prevents detection, considering the high number of lysine-terminating peptides in our dataset (see Figure S1 C), we would expect that also this labeling step is stochastic and affects only a minor proportion of the proteins. We discuss the potential NHS labeling in the manuscript.
(h) How are the primary and secondary antibodies affecting the proteomics analysis identified as contaminants?
We thank the reviewer for this comment. Since antibodies bind to proteins in a non-covalent manner, they will be released during the denaturing steps of the protocol. Of course, the antibodies will stay in the sample, be digested and analyzed and could, if very abundant, affect the analysis of the proteins from the samples. To check this possibility, we re-searched the MS data including the sequences of the antibodies used. To our surprise, we could not detect any peptides of these antibodies. This suggests that the concentrations of the antibodies used are much lower than those of the sample proteins and thus should not have any impact on the proteomics results. We interpret this result also as a benefit of our method compared to organellar-IP.
(i) Have the authors observed differences in proteomics coverage of only antibody vs NHS-labeling? Depending on the questions above, could pure antibody-based labeling increase proteomic coverage?
We did not perform this comparative analysis, since we always used NHS dyes. In the experiments presented in this manuscript, NHS dyes allowed easy visualization of the whole cell without the use of antibodies. This NHS staining was essential for this particular setup for sample acquisition. We cut out entire cells, cells lacking the nucleus and cells lacking the Golgi apparatus, which served as critical controls. However, other ways of detecting cell boundaries could be used to avoid NHS staining. As shown above, both, the anchor and NHS labeling are likewise sparse and stochastic. Moreover, we could not detect any impact of the antibody labeling to our results. Thus, we assume that both labeling procedures could be used.
Reviewer #1 (Recommendations for the authors):
(1) Introduce the aberration PPS-membrane.
We apologize for the omission and now explain the PPS abbreviation.
(2) Please state which cells are used already in the first section "Development of the SPEx method" as opposed to only further down in the manuscript.
We apologize for the omission, which has been fixed.
(3) In Figures 2D and E, the authors should also show the hydrated expansion factors for better comparison.
We agree that the inclusion of the hydrated expanded gels allows for better comparison. These data are now included in the manuscript.
Reviewer #2 (Public review):
Summary:
This study introduces a method that combines physical expansion of cells, imaging-guided isolation of defined regions, and protein identification to enable compartment-resolved analysis of protein composition at the subcellular scale. The authors aim to address a central limitation in existing approaches, namely the loss of spatial information during sample preparation or the indirect nature of proximity-based labeling methods. Using several cellular compartments as examples, they demonstrate that their approach can recover compartment-enriched protein sets and identify candidate proteins with previously unassigned localization.
Strengths:
A major strength of this work is the conceptual simplicity and accessibility of the approach. By combining established techniques in a modular way, the method avoids the need for genetic manipulation or specialized labeling strategies, making it broadly adaptable across experimental systems. The ability to directly select regions of interest based on imaging represents a clear advantage over indirect enrichment strategies and allows flexible targeting of both membrane-bound and non-membrane-bound compartments.
The experimental design is also a strong aspect of the study. The use of complementary comparison strategies-analyzing isolated compartments alongside matched "subtracted" controls-provides an internal framework for assessing enrichment and depletion, increasing confidence in spatial assignment. The application of the method across multiple organelles of different sizes and properties demonstrates versatility, and the reported specificity for several compartments is encouraging. In particular, the ability to profile small and biochemically challenging structures highlights a potentially important niche for the approach.
Weaknesses:
Despite these strengths, several methodological limitations constrain the interpretation of the results. The most important relates to spatial accuracy in three dimensions. While lateral resolution is improved through physical expansion, the lack of depth resolution introduces uncertainty regarding contributions from structures above and below the selected region. Although the authors argue that this does not substantially affect specificity, the current evidence is largely indirect, and a more rigorous quantification of potential contamination would strengthen this conclusion.
Quantitative interpretation also remains challenging. Because the measurements reflect total protein abundance rather than local concentration, differences in compartment size and protein density can influence enrichment values, particularly for small structures embedded within larger volumes. This issue is evident in the analysis of smaller compartments and complicates direct comparison across conditions. Additional normalization or modeling would help clarify how to interpret these measurements.
Another limitation concerns variability in the expansion process and its downstream consequences. Differences in expansion factor across samples may affect the definition of regions of interest and introduce variability in sampling, yet the impact of this variability is not fully explored. Similarly, the use of a modified chemical treatment to preserve proteins for downstream analysis is central to the workflow but is not extensively validated with respect to preservation of spatial organization.
While the identification of previously unannotated proteins is an appealing aspect of the study, validation is limited to a small number of examples, and broader support from independent datasets or literature context is lacking. In addition, the study primarily focuses on steady-state measurements in a single cell type, and therefore does not yet demonstrate the ability of the method to capture dynamic or condition-dependent changes in protein localization.
Finally, the positioning of the method relative to existing approaches could be more clearly articulated. Although qualitative comparisons are provided, a more systematic and quantitative benchmarking against alternative strategies would help readers better understand the specific advantages and trade-offs.
We thank the reviewer for the careful evaluation of the manuscript and for the constructive feedback. The reviewer raises valid points, which we addressed in the revised manuscript.
Reviewer #2 (Recommendations for the authors):
Additional experiments and analyses
(1) To strengthen confidence in the spatial specificity of the method, the authors should include experiments or analyses that directly estimate potential cross-compartment contamination. For example, using markers that are known to localize to mutually exclusive cellular regions would allow an empirical assessment of signal carryover. Alternatively, a computational or geometric model could be used to estimate how much material from adjacent regions is expected to contribute to the sampled proteome, given the loss of depth information.
We agree with the reviewer’s concern regarding spatial resolution and potential contaminations resulting from it. We had the same concerns. We have obviously no resolution in z and we detected about 700 proteins (Fig. 5F) in our all samples, likely representing proteins residing above and below of (and potentially next to) the structures we cut out. To overcome this issue, we always compare the proteome of the structure of interest with an appropriate control. We also determined that mitochondria proteins are specifically omitted from our nuclear samples. Moreover, we provide evidence that our analysis methods can be used also to determine proteins that are specifically lost from a structure. Omitting the comparative approach would clearly result in very low specificity and a high number of false positives. However, we argue that comparison to a control largely removes these unspecific contaminations and we think that this is illustrated by our excellent specificity. Nevertheless, to address these concerns further, we have determined the masses of all quantified proteins using iBAQ values. We then summed the masses of all proteins being annotated to a specific organelle and calculated the percentage to the summed masses of all quantified proteins. As expected, we found that in our excised nuclei samples, 75.5% of the protein molecules originate from the nucleus and only 24.5% of protein mass was derived from other regions. Likewise, we found 84.7% of the protein mass in the excised Golgi samples annotated with Golgi or extracellular by GO. This analysis revealed that we detect an expected amount of contaminant proteins derived from other compartments due to the lack of resolution in z, but most of protein mass excised is already organelle-specific. The additional quantification step further enhances assignment accuracy, boosting the overall specificity of our approach. In the case of the nucleoli the mass contribution from known nucleoli proteins is significantly lower (~4%), which would be expected for a structure that only covers a fraction of the cell in z-direction. In the case of such small targets comparison to a fitting control is therefore essential. We added this analysis to the revised version of the manuscript and discuss it in detail (see “SPEx is best used as a comparative approach”). We thank the reviewer for bringing this up!
(2) Given that drying results in compression along the vertical axis while largely preserving the lateral dimensions, it would be valuable to provide a quantitative analysis of how this affects spatial accuracy. This could include comparing marker distributions before and after drying or examining signal distribution across depth.
Even though we were unable to image a particular cell before and after drying (due to the physical properties of the gels and the change in size during drying), we did so for the bulk of cells. We did not observe discernible morphological alterations after drying. The investigated structures (Golgi, nucleus and nucleoli) still looked morphologically similar to expanded cells (before drying) and non-expanded cells. We cannot exclude that the drying process introduced minor distortions of the sample. We would like to point out, however, that our focus is mainly on the spatial proteomics and to a lesser extent on the perfect conservation of the ultrastructure during drying. As long as we can maintain very good spatial separation and are able to cut out individual structures, our method should work just fine.
In addition, trimming or sectioning the expanded samples along the z-axis prior to microdissection could help minimize contributions from structures above and below the region of interest, thereby improving spatial specificity.
We tried to cut expanded samples (monolayer of cells) with a microtome prior to laser dissection. This approach did unfortunately not work. Mounting a monolayer of cells for microtome cutting is non-trivial! The very thin monolayer of cells would need to be perfectly aligned with the plane where the cryo-sectioning takes place. Moreover, the different densities of the hydrogel without cells and the expanded cells in the same hydrogel appeared to be particular challenging during sectioning. Therefore, we gave up on this approach for monolayers. For a monolayer of cells, the contamination in the lateral dimension seems to be small enough so it can be eliminated by comparing the sample to appropriate controls, preventing the need for cryo-sectioning. We do think, however, that cryo-sectioning should be feasible for tissues since the combination of cryo-sectioning and TREx expansion in tissues has been described previously (Damstra et al., 2022).
(3) Because the method relies on imaging-guided spatial dissection of a three-dimensional structure, further validation of spatial precision would strengthen the manuscript. Applying z-stack trimming to reduce depth-related ambiguity and comparing results obtained with more conservative and spatially restricted cutting boundaries would help determine how sensitive the proteomic output is to segmentation strategy and provide a clearer assessment of robustness.
As pointed out above, we were unable to perform trimming in z. Therefore, the proposed analysis cannot be performed.
(4) The use of SDS-based denaturation instead of protease digestion represents a key methodological deviation from conventional expansion workflows and is central to enabling downstream proteomic analysis. The authors should provide additional validation demonstrating that this approach preserves spatial fidelity without introducing distortion or bias in protein localization.
We respectfully disagree with the reviewer. SDS denaturation was used previously in both the TREx and the U-ExM protocols (Damstra et al., 2022; Gambarotto et al., 2021). The preservation of ultrastructure has been established in both studies. Still, as pointed out by the reviewer, protease digestion is frequently used in ExM, likely because it minimizes distortion of the sample by eliminating any mechanical resistance to expansion. We omitted protease treatment to be able to perform LC-MS/MS. Protease treatment would likely reduce the amounts of peptides we would be able to detect. However, with the extended SDS denaturation, we were able to maintain excellent expansion even in the absence of protease treatment.
(5) The manuscript reports variability in expansion factors across conditions, and it would be important to more clearly establish reproducibility. The authors should quantify variability across biological replicates and discuss how differences in expansion factor influence region-of-interest selection, spatial resolution, and downstream quantitative analysis.
We were surprised that this reviewer thinks that we have variability in the expansion factors across conditions. In fact, the expansion is rather reproducible, and we only observe very minor variability in expansion factors between samples. The minor differences in expansion factor are unlikely to affect sampling because sampling happens manually and differences in the size of an organelle are accounted for by just adjusting the size of the isolated ROI. Even if we would have some discernible difference in the expansion factor between samples, this would not matter. The organelle of interest is cut out from the same cell, which will be subsequently used as control. Each sample subjected to LC-MS/MS contained at least 10 individually cut-out objects, further eliminating variability potentially introduced by minor changes in expansion levels.
(6) The authors should also address how differences in compartment size and protein density influence quantitative interpretation. Because proteomic measurements reflect total protein abundance rather than local concentration, smaller compartments may be underrepresented. Incorporating normalization strategies or simulations would help clarify how enrichment values should be interpreted across compartments.
We agree with the reviewer that comparison of structures of very different sizes might be technically difficult and that in such cases, controls should be collected in a similar size range to the actually studied compartments. We have implemented this change already in our workflow and mention this consideration in the discussion.
Regarding normalization, we employed median normalization to our proteomics data abundances which controls for different sample amounts and eventually results in a comparison of protein concentration rather than pure protein abundances. This approach eliminates any knowledge of cell or organelle size and allows a more uniform comparison. We do agree, however, with the reviewer that for differential proteomics analysis, regions of interest should be preferable of the similar size and protein amount to avoid normalization biases. We discussed this point in the revised manuscript.
(7) Validation of newly identified compartment-associated proteins should be expanded. In particular, the interpretation of proteins such as C7orf50 and SPOUT1 would benefit from deeper contextual discussion, including whether prior studies have reported alternative localizations under different cellular conditions and whether their localization may be cell-type-specific or context-dependent.
We agree that more contextual discussion of the newly localized proteins will benefit the manuscript. We added more information about the newly identified proteins to the results section.
(8) To support broader claims about the utility of the method, the authors should consider demonstrating that the approach can capture biologically meaningful, context-dependent changes in protein localization. For example, comparing different cell types or applying defined perturbations such as stress or apoptosis would help establish the ability of the method to resolve dynamic intracellular remodeling.
The authors agree with the reviewers’ assessment that studying context-dependent changes at a subcellular level is an exciting idea to explore. While we are currently planning such experiments, we think that these experiments are beyond the scope of this manuscript, which is meant to describe the method and its general usefulness.
(9) The manuscript would benefit from a more direct and quantitative comparison with existing spatial proteomics approaches. In particular, benchmarking against proximity labeling methods would help clarify relative performance in terms of specificity and false positive rates, and comparison with published datasets could provide an objective frame of reference.
We are a bit puzzled by the reviewer’s request. We already compared SPEx to a number of established spatial proteomics methods in the paper (Figure S1I-J, Figure S2F-G, Figure 5F-G) and discuss coverage and specificity. However, we expanded this comparison further to methods using Exm combined with LMD in the new manuscript (see below).
(10) In addition, the positioning of the method relative to other expansion-based spatial proteomics approaches remains insufficiently defined. The authors are encouraged to include a structured comparison, ideally in the form of a summary table, outlining key parameters such as sampling strategy, spatial resolution, proteome depth, spatial specificity, throughput, and limitations. This would help clearly delineate the unique contributions of the approach within the broader methodological landscape.
Peer reviewer’s request, we added a table (S15) on spatial proteomics methods combining expansion microscopy with manual or laser dissection and also provide information about lateral resolution, sampling strategy, proteome coverage, advantages and limitations.
Writing and presentation
(1) The presentation of expansion-related measurements would benefit from clearer organization. Currently, expansion factors from non-dried and dried samples are presented in separate figures, which may create redundancy and confusion. Reorganizing these data to directly compare pre- and post-drying conditions, or moving non-essential measurements to supplementary material, would improve clarity.
We now also show the hydrated expansion factors in Figure 2C-D for comparison.
(2) The method used for quantifying expansion should be described more explicitly. It should be clarified whether the same structures were tracked before and after expansion or whether independent samples were used, particularly for nuclei, which vary in size.
We use the nucleus because this also used by other groups in the field (Liffner et al., 2024). The tracked structures are from independent samples. Thus, we average over a number of structures and calculate the expansion factor based on the mean diameter. We made this clearer in the manuscript.
(3) Additional clarification is needed regarding the fluorescent labeling strategy. The rationale for using different dyes across experiments should be explained to distinguish whether these choices are driven by technical constraints, experimental design, or visualization preferences.
We agree with the reviewer that the manuscript will profit from a more in-depth discussion of the labeling strategy. The structure of interest was always labeled in red because in our hands more red-shifted dyes are more resistant to drying than e.g. green dyes. We added this information to the manuscript.
(4) Several figure legends and result descriptions would benefit from revision for clarity and completeness. In particular, complex panels such as intersection analyses should be more fully explained to ensure accessibility to readers.
We agree with the reviewer’s assessment. The intersection analysis and the PCA, which represent the most complex panels in the manuscript, are now more clearly described in the legends.
(5) The authors should also improve clarity in describing structural changes associated with drying. Including representative images of gels before and after drying would help illustrate the preservation of lateral dimensions and compression along the vertical axis.
We agree that this addition would improve the manuscript. We included the images as Figure S1B. The gel post-drying is outlined by black dots because the gels become very thin and near-invisible post-drying.
Minor corrections
(1) Minor issues should be addressed to improve consistency and readability. These include correcting figure cross-references, standardizing statistical notation throughout the manuscript, and resolving typographical inconsistencies.
Done.
(2) Specific figure-related issues should also be addressed, including clarifying mislabeled figure references and improving color schemes in plots where group distinctions are currently difficult to interpret.
Done.
Data and reproducibility
(1) The authors should ensure that all datasets and analysis workflows are fully accessible and sufficiently documented to enable reproducibility. This includes providing raw and processed proteomics data, as well as detailed descriptions of analysis pipelines and parameter settings.
Raw proteomics data for all experiments will be publicly available upon publication of the final version and workflows are described in a detailed manner. To improve reproducibility, we also included a step-by-step explanation of the assembly of the gelation chamber, which represents the handling-wise most challenging step of the protocol, in Figure S3D.
Reviewer #3 (Public review):
Franziscus et al. describe an elegant approach for spatially specific proteome analysis. To achieve this, they expand fixed cells and subsequently use a laser to micro-dissect a region of interest, which is then analyzed by mass spectrometry.
They demonstrate the effectiveness of their approach by analyzing the nucleus, nucleolus, and the Golgi, and benchmark their hits against previous datasets for these organelles.
The manuscript is very well written and nicely guides the reader through the applied methods. The presented data is convincing, and I do not see the need for additional experimental verification of the protocol. The only minor concern is the novelty of the method and the presentation. A combination of expansion, laser microdissection, and proteomics has been applied in the past (PMID: 36450705, PMID: 39477916). In the manuscript, one of these studies is cited, though it does not become clear that this approach is already described. However, Franziscus et al. describe the approach better and make it more accessible to the reader, especially since the other studies described this methodology in combination with tissue expansion and not in combination with single cell expansion as it is done here. I would ask the authors to be clearer in the introduction about what others have already done and what their contribution is here. In general, I am convinced that the community will benefit from the presented protocol to analyze organelle proteomics in detail.
We thank the reviewer for the careful evaluation of our manuscript and overwhelmingly positive assessment. We apologize for the omission of the mentioned citations, which are now included in the manuscript together with a direct comparison with the mentioned methods (Table S15).
References
Damstra HG, Mohar B, Eddison M, Akhmanova A, Kapitein LC, Tillberg PW. 2022. Visualizing cellular and tissue ultrastructure using Ten-fold Robust Expansion Microscopy (TREx). eLife 11:e73775. DOI: https://doi.org/10.7554/eLife.73775
Gambarotto D, Hamel V, Guichard P. 2021. Ultrastructure expansion microscopy (U-ExM). Methods in Cell Biology 161:57–81. DOI: https://doi.org/10.1016/bs.mcb.2020.05.006, PMID: 33478697
Liffner B, Silva TLA e., Vega-Rodriguez J, Absalon S. 2024. Mosquito Tissue Ultrastructure-Expansion Microscopy (MoTissU-ExM) enables ultrastructural and anatomical analysis of malaria parasites and their mosquito. BMC Methods 1:13. DOI: https://doi.org/10.1186/s44330-024-00013-4
eLife Assessment
This important study identifies and characterizes a set of amino acid states that can rescue protein function in the presence of substantially deleterious mutations. Some of these super-compensatory substitutions also confer substantial mutational robustness, with broader implications for understanding epistasis, protein evolution, and protein engineering. The evidence is compelling, supported by a creative reanalysis of a large deep-mutational-scanning dataset, experimental validation, an in-depth investigation of the evolutionary implications, and structural mechanistic analysis.
Reviewer #1 (Public review):
Summary:
The study identifies and characterizes a set of amino acid states that make the protein robust to other mutations, to the point of being able to compensate mutations that render wildtype proteins entirely non-functional. The study uses a previously published dataset and uses it to find and study such super-compensators. It then analyzes the biophysics and fitness landscape structure of what may be behind the compensation, identifying stability as an important parameter that, nevertheless, is not sufficient to explain all of the compensatory effect. These findings have important implications for our understanding of protein evolution, with these super-compensators possibly acting in a role of "permissive mutations" and open up evolutionary trajectories that may be closed without them. Perhaps the identification of such super-compensator substitutions can be incorporated into various protein design approaches.
Strengths:
The paper presents a compelling case with a rigorous analysis of the expected error rates of observation. While not unique, the current state-of-the-art in the field typically does include experimental error rate estimation like this work. The paper also does a good job in exploring the issue, including looking at plausible biophysical basis of super-compensators.
Comments on revised version.
The authors have addressed all of my concerns in the revision.
Reviewer #2 (Public review):
Summary:
This manuscript presents an interesting and conceptually valuable analysis of compensatory evolution using a large combinatorial deep-mutational-scanning dataset for yeast His3p.
Strengths:
I particularly like the identification of "super compensatory" substitutions that improve fitness across diverse genetic backgrounds and apparently reduce the sensitivity of the local fitness landscape to subsequent mutations. The work connects epistasis, protein stability, mutational robustness, and evolvability in a clear and potentially broadly relevant manner.
The authors provide several complementary lines of evidence in support of this central conclusion. In particular, the new experimental validation of S189A is an important strength because it directly demonstrates that a predicted super compensator can buffer the effects of diverse deleterious substitutions, while analyses of additional DMS datasets from other proteins and assay systems suggest that the phenomenon is not restricted to the original His3p landscape.
Comments on revised version.
The authors have satisfactorily addressed all of my previous concerns.
Reviewer #3 (Public review):
Summary:
The manuscript by Jiang and co-authors presents an analysis of experimental measurements (about 400k variants) from a deep mutational scan of HIS3 enzyme. The authors assess the ability of a genotype to be "rescued" and show that this depends on mutation sites (in particular their solvent accessibility) and mutations effects (should be mild on folding stability or binding affinity). They further identify a set of super-compensatory mutations and their results suggest that these mutations flatten the fitness landscape.
Strengths:
This finding is interesting and likely of interest to a broad community. The analysis seems sound.
Comments on revised version.
The authors have satisfactorily addressed my concern. Additional robustness experiments and case studies from ProteinGym are valuable.
Author response:
The following is the authors’ response to the original reviews.
eLife Assessment:
This important study identifies and characterizes a set of amino acid states that can rescue protein function in the presence of substantially deleterious mutations. Some of these super-compensatory substitutions also confer substantial mutational robustness, with broader implications for understanding epistasis, protein evolution, and protein engineering. The evidence is convincing, supported by a creative reanalysis of a large deep-mutational-scanning dataset, statistically rigorous treatment of anticipated error rates, experimental validation, and analyses of additional proteins, although clearer presentation and deeper investigation of the evolutionary implications and structural mechanisms would further strengthen the study.
We sincerely thank the Reviewing Editor, the Senior Editor, and the three reviewers for this assessment. The two directions identified in it are addressed directly in the present revision. The structural mechanism of super compensation is now examined with an inverse-folding model in addition to the Rosetta calculations (Reviewer #2, Comment 1); the evolutionary implications are tested on a phylogeny rather than by co-occurrence among extant species, with the remaining limitations stated explicitly (Reviewer #2, Comment 3); and the presentation is revised throughout, including the title, the terminology used for fitness landscapes, the two section titles, the numbering of equations, and the figures (Reviewer #1). We hope that these revisions strengthen the clarity and the evolutionary interpretation of the study.
Reviewer #1 (Public review):
Summary:
The study identifies and characterizes a set of amino acid states that make the protein robust to other mutations, to the point of being able to compensate mutations that render wildtype proteins entirely non-functional. The study uses a previously published dataset and uses it to find and study such super-compensators. It then analyzes the biophysics and fitness landscape structure of what may be behind the compensation, identifying stability as an important parameter that, nevertheless, is not sufficient to explain all of the compensatory effect. These findings have important implications for our understanding of protein evolution, with these super-compensators possibly acting in a role of "permissive mutations" and opening up evolutionary trajectories that may be closed without them. Perhaps the identification of such super-compensator substitutions can be incorporated into various protein design approaches.
Strengths:
The paper presents a compelling case with a rigorous analysis of the expected error rates of observation. While not unique, the current state-of-the-art in the field typically does include experimental error rate estimation like this work. The paper also does a good job in exploring the issue, including looking at plausible biophysical basis of super-compensators.
We are deeply grateful to Reviewer #1 for the constructive feedback and the careful reading of our manuscript. We are particularly encouraged by the assessment that the paper “presents a compelling case with a rigorous analysis” of the expected error rates of observation, and by the recognition that the identification of super compensatory substitutions may inform protein design approaches. The comments on the rigor of the evolutionary framing, on the presentation of the fitness-potential analysis, and on the figures were highly valuable in improving the manuscript. We have carefully considered each point and itemize the corresponding revisions below.
Weaknesses:
(1) The paper lacks rigor in talking about evolutionary-related issues of the state of the fitness landscape. As an example, the paper mentions that these super-compensators flatten the landscape. While I understand where this is coming from, I think that the fitness landscape in this context is a static entity and cannot be flattened or otherwise altered. A much more accurate description is that a sequence with a super-compensator is located in a flatter-than-expected segment of the fitness landscape, or on a flat fitness ridge. These issues are more semantic in nature, and while the manuscript would benefit from it being shown to an expert in molecular evolution or fitness landscapes, this issue does not take away from the importance of the results.
We thank the reviewer for the suggestions. The proposed title is “Super-compensatory substitutions restore protein function and confer local mutational robustness”. The replacement text distinguishes the fixed genotype-to-fitness map from comparisons among local neighborhoods within that map. “Local mutational robustness” is defined operationally as a smaller absolute measured effect of the same second-site substitution in matched backgrounds carrying the focal state. The prepared prose removes statements that a substitution “flattens,” “smooths,” or changes the landscape and instead says that a genotype occupies a locally flatter region or fitness ridge.
Reviewer #1 (Recommendations for the authors):
I think that there should be more rigor when talking about or painting a mental picture of the fitness landscape (see above). More minor comments below:
Main text
(1) Line 106: "pooled competitive growth assays" -> is it correct to call it "competitive"? By design, there should not be competition in the sense that the presence or absence of cells with genotype A should not influence the growth rate of cells with genotype B.
We thank the reviewer for the careful terminology check. We fully agree that “competitive” is misleading here, as the assay does not measure direct competition between genotypes. We have revised the term to “pooled growth assays” throughout the manuscript to more accurately reflect the experimental design.
(2) Line 125: 3,750 rescuable genotypes are identified among 31,701 lethal ones (Line 111), but in the abstract the claim is one-third of non-functional missense variants are rescuable (Line 26-27). I am not sure I understand.
We thank the reviewer for catching this inconsistency. Upon re‑checking, we realize that 3,750/31,701 is approximately 11.8% (roughly one‑tenth), not one‑third. The “one‑third” figure in the Abstract was a misstatement and has been corrected to reflect the correct proportion. All other numbers in the main text and abstract have been cross‑checked and are consistent.
(3) Line 149: add citation to His3p being a homo-24-mer.
We thank the reviewer for the suggestion. The citation has been added accordingly (Rawson et al., 2018).
(4) Lines 178-187: I understood the paragraph about the fitness potential of super compensators, but I think that it should be shown to someone who does not think of fitness potentials as much as I do and see if it makes sense to them.
Please refer to the question 2 of reviewer 3
(5) Line 188: The title (and the paragraph) could focus on buffering beneficial ones; for deleterious, it is given by definition (although strictly only for lethals).
We thank the reviewer for the suggestion. We have revised the title to “Super Compensators Buffer the Effects of Beneficial Substitutions” to clearly emphasize that our study focuses on the buffering of beneficial mutations
(6) Line 238-239: Even more strongly than previously, the title is describing a property given by definition of (super) compensators, is it not?
We thank the reviewer for this critical and insightful point. We fully agree that the original title, “Super Compensators Enhance Organisms’ Tolerance to Deleterious Mutations,” risked stating a tautological property inherent to the definition of compensators. To address this, we revised it to “Validating Super Compensators as Robust Buffers of Deleterious Substitutions”, which now emphasizes the experimental validation aspect rather than a definitional property.
(7) Line 260: Regarding the robustness of the findings in other datasets - first, I did not see any results based on these datasets (maybe that is for another publication), and it is not clear which results are being referred to. Is it the existence of super-compensatory mutations, or their prevalence or their global stabilizing effect, or all of that?
We thank the reviewer for this important query. We have now clarified in the revised manuscript that our analysis of the other datasets was intended to test whether supercompensators can be detected in other proteins, not to estimate their global stabilizing effects. The results (Fig. S7A–B) show that such high‑compensation amino acid states do exist in other systems.
(8) Figure 1A: why does it jump in the diagram from WT 2 mutations away rather than 1?
We thank the reviewer for pointing out the inconsistency in Figure 1A. We have revised the figure as suggested, and the updated version now properly shows the stepwise progression from WT through the single‑mutation intermediate to the double‑mutation state, removing the original jump.
(9) Lines 655-656: Figure 1C dark blue vs black is hard to read.
We thank the reviewer for this suggestion. We have changed the dark blue to light blue in Figure 1C to enhance readability ‘
(10) Figure 2D: given axes, it would read better if it is a square.
We thank the reviewer for raising this issue. We have revised Figure 2D to a square (1:1) aspect ratio, as suggested, to improve readability. The updated figure is now included in the revised manuscript.
(11) Lines 681-682: when the definition of compensatory ability is the sum of incoming weights - although mathematically easiest, I am not sure what the sum represents. Because these weights are products, wouldn't a product be more natural? Not sure about this myself.
We thank the reviewer for raising this important issue. The graph direction to represent transitions from the existing background state to added substitutions. Each edge records the fraction of pairs showing a fitness increase ≥ 0.5. A product of incoming weights was not adopted because zero-valued weights would force the product to zero, obscuring differences in compensatory ability across substitutions.
Supplementary
(1) Figure S8 - the figure itself is confusing; I cannot connect it with the conclusion.
We apologize for the confusion caused by Figure S8. To address this, we have revised the figure layout and significantly expanded the legend to explicitly describe: (1) the origin and processing of the data, (2) the definition of the row/column of the heatmap, and (3) a direct interpretation to the corresponding conclusion. We believe the updated version now clearly conveys the intended message. The revised figure and legend are included in the updated supplementary materials.
(2) The font is inconsistent with the rest of the document.
We thank the reviewer for this careful observation. We have revised the font in the indicated section to match the formatting used throughout the manuscript. We have also reviewed the rest of the document to ensure overall formatting consistency.
We appreciate Reviewer #1’s careful comments on the rigor of the evolutionary framing and on the presentation of the manuscript. The revisions replace the landscape-altering language with statements about local neighborhoods, correct the inconsistent proportion reported in the Abstract, revise the two section titles so that they no longer restate a definitional property, number all equations, and improve the readability of Figures 1A, 1C, 2D, and S8.
Reviewer #2 (Public review):
Summary:
This manuscript presents an interesting and conceptually valuable analysis of compensatory evolution using a large combinatorial deep-mutational-scanning dataset for yeast His3p.
I particularly like the identification of "super compensatory" substitutions that improve fitness across diverse genetic backgrounds and apparently reduce the sensitivity of the local fitness landscape to subsequent mutations. The work connects epistasis, protein stability, mutational robustness, and evolvability in a clear and potentially broadly relevant manner.
The authors provide several complementary lines of evidence in support of this central conclusion. In particular, the new experimental validation of S189A is an important strength because it directly demonstrates that a predicted super compensator can buffer the effects of diverse deleterious substitutions, while analyses of additional DMS datasets from other proteins and assay systems suggest that the phenomenon is not restricted to the original His3p landscape.
We first appreciate Reviewer #2’s positive assessment of the conceptual value of the study, and in particular the recognition that the work connects epistasis, protein stability, mutational robustness, and evolvability, and that the new experimental validation of S189A is an important strength. We are also grateful for the three specific and constructive suggestions on structure-aware modelling, on the scope of the ProteinGym analysis, and on tree-aware testing of ortholog co-occurrence. These suggestions led to the largest additions in this revision. We have carefully considered each point, and our responses and the corresponding revisions are detailed below.
Weaknesses:
(1) The structural analysis currently relies primarily on correlations with RSA, weighted contact number, conservation, and Rosetta-predicted changes in folding or binding energy. For super compensators, the mechanistic evidence is largely limited to predicted stabilization and individual examples, such as the proposed salt bridge between 110D and R112. I believe that the newly developed structure-aware deep-learning approaches could provide useful information on the mechanism of super compensators. For example, an inverse-folding model such as ESM-IF1 could score complete multi-mutant sequences conditioned on the His3p backbone and test whether adding a super compensator restores sequence-structure compatibility across backgrounds. More recent multimodal mutation-effect or stability models could similarly be used to cross-check the Rosetta results, including models that explicitly support combinatorial mutations. I would not recommend simply comparing AlphaFold confidence scores between mutants, because current structure predictors are not necessarily sensitive to subtle mutation-induced energetic or conformational changes.
Following the reviewer's suggestion, we scored complete multi-mutant His3p sequences with ESM-IF1 (esm_if1_gvp4_t16_142M_UR50) conditioned on the 6EZM backbone, comparing exact measured genotype pairs that differ only by the state of interest (up to 1,000 backgrounds per state; 7,000 pairs). Five of the seven reported super compensatory substitutions in the Table S1 increased sequence–backbone compatibility: 144N (+0.0160), 169R (+0.0115), 110D (+0.0086), 189A (+0.0021) and 93V (+0.0016), each with a bootstrap interval excluding zero. Two decreased (73A, −0.0153; 137G, −0.0080). Against score-blind matched controls the contrast favoured the reported state for those same five (exact two-sided binomial P = 0.45). All seven directions were preserved in a contacting-chain context (chains A, R and L; Cα within 8 Å). We also used these scores to cross-check the archived Rosetta energies over the 1,217 exact pairs carrying archived entries on both sides (Zenodo 17769278). Higher archived energy accompanies lower measured fitness, so concordance would appear as a negative correlation; the observed Spearman correlations are positive (ρ = +0.085 to +0.094 pooled, +0.200 to +0.211 within focal state). The two predictors therefore do not track one another. We did not compare AlphaFold confidence scores between mutants.
In a word, ESM-IF1 supports the idea that most supercompensatory substitutions can promote the folding of His3p.
(2) The manuscript states that the pipeline was applied to 217 ProteinGym datasets and concludes that super compensators are broadly distributed across proteins and assays. However, this central generalization is described in only a few sentences and is largely relegated to Figure S7. The Methods do not explain which datasets contained sufficient combinatorial mutants to calculate compensatory ability or buffering, how many genotype pairs or quadruplets were available per substitution, or how differences in assay scale and library design were handled. This point requires clarification because super compensation is inherently a background-dependent property and cannot be established from single-mutant measurements alone. ProteinGym is widely used as a substitution-effect benchmark, and many of its constituent assays primarily contain single substitutions; for example, an analysis of an earlier ProteinGym collection reported that 76 of 87 assays contained only single substitutions. It is therefore unclear how the same compensatory-interaction pipeline could be applied uniformly to all 217 datasets.
We thank the reviewer for the suggestions. We have re-performed this analysis based on the pinned PG_v1.3 manifest. The 217 DMS assays cover 186 unique proteins. To identify super-competitors, we filtered for DMS datasets containing multiple substitutions per genotype—specifically requiring at least 10 sets of quadruplet genotypes to evaluate whether a specific amino acid state can mitigate mutational effects. This strict filtering yielded four qualified DMS datasets: three assayed for in vitro protein folding stability using cDNA display proteolysis (E3 ubiquitin-protein ligase AMFR, iron-regulated surface determinant protein H, and E3 SUMO-protein ligase CBX4; Tsuboyama et al., 2023), and one assayed for fluorescence intensity via FACS (green fluorescent protein 2; Gonzalez Somermeyer et al., 2022). We have updated the Methods section to detail this selection process.
(3) The analysis of 335 His3p orthologs in Discussion is potentially very interesting, but co-occurrence between super compensators and putatively deleterious amino-acid states does not by itself demonstrate evolutionary compensation. Closely related species share substitutions through common ancestry, and both states could be associated with a particular lineage or ecological context. A tree-aware analysis would considerably strengthen this result. The authors could reconstruct ancestral states and ask whether acquisition of a super compensator tends to precede or accompany otherwise deleterious substitutions. Alternatively, they could use phylogenetically informed permutations that preserve substitution frequencies and shared ancestry.
We agree. Co-occurrence among living species is not by itself evidence of compensation, because related species inherit substitutions from a common ancestor. We have therefore redone the analysis on a phylogeny, using the phylogenetically informed permutation the reviewer suggests.
We realigned the archived orthologs with MAFFT L-INS-i and built a maximum-likelihood tree with IQ-TREE: 329 species, 225 alignment columns, the Q. YEAST model selected by BIC, with 1,000 ultrafast-bootstrap and 1,000 SH-aLRT replicates. Six rows carrying three ambiguous duplicate species labels were removed, which also corrects the earlier figure caption that stated 355 species. We then tested 28 substitution pairs against a null in which both substitutions are simulated on the tree and thresholded to their observed frequencies, so the null preserves both how common each substitution is and how the species are related.
Two pairs stand out, and both involve 189A. 189A occurs together with 204I in 112 species, where the tree-based null expects 78 — an excess of 34 species. 189A occurs with 211E in 66 species against an expectation of 43 — an excess of 23. Both excesses are significant on their own under the tree-based simulation (p = 0.021 and 0.036, from 100,000 simulations) and under a separate Pagel test of correlated evolution (p = 0.046 and 0.023). No other pair reaches significance under both tests. 189A is also the substitution we validated experimentally, so the strongest evolutionary signal and the experimental result point to the same substitution.
We note two limits. Neither pair survives correction for having tested 28 pairs. And both pairs involve common substitutions (189A is present in 177 of the 329 species), so part of their high rank may reflect the greater statistical power available for common substitutions. We therefore present these as the strongest associations in the data rather than as proof of historical compensation.
We sincerely thank Reviewer #2 for recommending a broader mechanistic and evolutionary evaluation. The added ESM-IF1 analysis, the re-performed ProteinGym dataset selection, and the tree-aware analysis of ortholog co-occurrence provide more direct evidence for the properties of super compensatory substitutions, and we have stated the remaining limitations of each analysis explicitly.
Reviewer #3 (Public review):
The manuscript by Jiang and co-authors presents an analysis of experimental measurements (about 400k variants) from a deep mutational scan of the HIS3 enzyme. The authors assess the ability of a genotype to be "rescued" and show that this depends on mutation sites (in particular their solvent accessibility) and mutation effects (should be mild on folding stability or binding affinity). They further identify a set of super-compensatory mutations, and their results suggest that these mutations flatten the fitness landscape.
This finding is interesting and likely of interest to a broad community. The analysis seems sound. However, I have a number of major concerns regarding the presentation and positioning of the work.
We sincerely thank Reviewer #3 for the careful reading of our manuscript and for the positive comments that the finding is “interesting and likely of interest to a broad community” and that the analysis “seems sound.” The concerns regarding the positioning of the work relative to Pokusaeva et al. (2019), the description of the model used to estimate fitness potentials, the structural and citation details, and the sensitivity of rescuability to the threshold choices were valuable in improving the clarity and the rigor of the manuscript. We address each point below.
(1) It would improve the manuscript to clarify the present contribution with respect to a previous study by the same authors, namely Pokusaeva et al. 2019. Did the authors apply the same protocol to generate a new library of mutants, or did they re-analyse an already published library? If the library is not new, ambiguous sentences like "Nevertheless, to our knowledge, the His3p library remains one of the largest and most comprehensive resources that contains multi-site mutants" should be reformulated.
We thank the reviewer for noting this ambiguity; we agree and will revise accordingly. The large-scale His3p landscape was reported in Pokusaeva et al. (2019), and no new version was constructed here. Instead, we provide a computational re-analysis of those measurements to quantify rescuability, broad positive effects of super compensatory substitutions, and local buffering of fitness effects, plus a newly generated targeted DMS validation library (randomizing positions 181 and 202 in WT S189 vs. 189A) that experimentally validates compensation. To clarify provenance, we will change the sentence “Nevertheless, to our knowledge, the His3p library remains one of the largest...” to “While the large-scale library was measured previously (Pokusaeva et al., 2019), the current work combines a newly generated targeted dataset with a fresh re-analysis of this public His3p library to reveal previously unappreciated rules of compensation.” This revision clearly separates prior data from our novel contributions.
(2) Pokusaeva et al. 2019 is cited for the library and also for the deep neural network. It would be beneficial to briefly describe the architecture, the inputs and outputs, and the training procedure. Was the network trained on the current library? What is the purpose of this network? It looks more like an additive linear model (except for the global sigmoid) than a deep neural network. How does it relate to global epistasis models? The sigmoid function is designed to capture plateauing effects; doesn't that introduce some circularity issue in the reasoning?
Pokusaeva et al. (2019) describes a three-layer feed-forward model that is best understood as a global-epistasis model rather than a conventional deep neural network. The input is a one-hot encoding of amino-acid states. These inputs are first mapped to a one-dimensional additive latent variable, in which the state-specific coefficients are summed to produce a genotype-level latent “fitness potential.” That latent value is then passed through a learned nonlinear output layer with 22 units to generate the predicted fitness. The model was trained on the current library.
Its purpose is to separate the intrinsic contribution of each amino-acid state from the background-dependent saturation of the His3p assay. Raw fitness values are not directly comparable across genetic backgrounds: in already low-fitness backgrounds, additional substitutions may remain phenotypically invisible, whereas in high-fitness backgrounds, mildly deleterious effects may be masked. A global-epistasis formulation addresses this by placing states on an additive latent scale and capturing assay saturation through a monotonic nonlinear link function. In this framework, each state has an estimated coefficient that is interpreted as its background-independent fitness contribution.
This coefficient is not hypothetical; it is explicitly provided in File/DeepLearningWeights/DLWeightsFI.tsv as FI_mean, one row per state. These coefficients are used downstream to quantify each state’s fitness impact, which is essential for showing that supercompensatory substitutions are not defined in a circular way by construction.
(3) Are the super-compensatory mutations observed (conserved) across evolution? Beyond the fact that they are accompanied by mildly deleterious mutations in natural sequences. Can we predict them with variant effect predictors?
We thank the reviewer for these questions. Regarding evolutionary conservation, we have not performed a systematic phylogenetic analysis, as our primary focus here was the experimental characterization of compensation within the His3p system. Nevertheless, our data show that super‑compensatory mutations are consistently stabilizing across backgrounds (Fig. 4L)—precisely the property that makes them robustly beneficial regardless of context. This suggests that these mutations may be identifiable through stability‑predicting tools (e.g., FoldX, DeepDDG), which would flag them as stabilizing. Whether this stabilization is reflected in evolutionary conservation patterns is a separate question; while we observe their co‑occurrence with mildly deleterious variants in natural sequences (as the reviewer notes), we have not systematically tested whether conservation‑based predictors (e.g., EVE, DeepSequence) can predict them.
(4) The AAindex mention should be accompanied by a citation.
We thank the reviewer for the reminder. We will add the appropriate citation for the AAindex database in the revised manuscript
(5) Equations should be numbered. WCN formula seems to contain misformatting issues.
We thank the reviewer for the reminder. We numbered all equations sequentially and correct the formatting errors in the WCN formula.
(6) A more explicit description of the structural data analysed (which PDB entry?) should be provided.
We thank the reviewer for the reminder. The PDB entry used is 6EZM.
(7) I believe the citation Van Cleve and Weissman 2015 for the ProteinGym benchmark is incorrect. Additionally, is the Rosetta citation adequate?
We thank the reviewer for pointing out these citation issues. We have double-checked the references and agree with these concerns.
Regarding the Rosetta citation, we acknowledge that there is no single “Rosetta citation”. In our work, we used the ddg_monomer function for ΔΔG_fold calculations and the flex ddG function for ΔΔG of interface binding calculations. Therefore, we cited the corresponding methodological papers: Kellogg et al. (2011) for ddg_monomer and Barlow et al. (2018) for flex ddG.
(8) How is the definition of rescueability sensitive to the threshold choice?
We thank the reviewer for raising this important issue. To systematically evaluate the robustness of our rescuability measurements, we tested three key parameters: (i) the fitness cutoff used to define unfit genotypes, (ii) the minimum number of genotypes in the superset required for an unfit genotype to be considered and (iii) the fitness cutoff used to define fit genotypes. The first two parameters collectively determine the size of the unfit genotype set, while the third controls how rescuability was calculation.
As shown in Fig. S1D of the manuscript, varying the fitness cutoffs for defining both unfit and fit genotypes has only a minor effect on the calculated rescuability, as reflected by the highly correlated rescuability values across different cutoff combinations.
Besides, the primary goal was to examine how rescuability correlates with three biochemical features of each position in His3p: relative solvent accessibility (RSA), weighted contact number (WCN), and evolutionary conservation. When we applied different fitness cutoffs to define unfit genotypes while with at least 10 genotype in the superset, the correlations between rescuability and these biochemical features remained highly consistent (see Author response image 1A-D). This demonstrates that the observed associations between rescuability and RSA, WCN, and conservation are robust to the choice of fitness cutoff used to filter the unfit genotype sets.
We next assessed the sensitivity of our results to the superset (the whole genotypes with additional substitutions) threshold. Using a fixed fitness cutoff of zero to classify unfit genotypes, we found that, although this shifted the fraction of genotypes classified as unfit, the observed correlations between rescuability and the three biochemical features were largely unchanged, even when the superset was defined by as few as a single genotype (see Author response image 1E-F).
In summary, despite variations in the parameter settings used to filter unfit genotypes for rescuability calculation, the overarching conclusion—that rescuability is consistently correlated with RSA, WCN, and conservation—remains robust and reproducible across all tested conditions.
Author response image 1.
Different rescuability calculation methods and filtering thresholds produce similar results. (A-F) Different filtering thresholds used to calculate the average rescuability for each position of the monomer of His3p give similar results. (A-B) The average rescuability for each position along His3p for unfit genotype with fitness < 0.5. The total count of genotypes containing additional substitutions is at least 10. (C-D) The average rescuability for each position along His3p for unfit genotype with fitness < 1. The total count of genotypes containing additional substitutions is at least 10. (E-F) The average rescuability for each position along His3p for unfit genotype with fitness = 0 and the total count of genotypes containing additional substitutions is at least 1.
We are grateful to Reviewer #3 for the questions on data provenance, model description, structural data, citations, and threshold sensitivity. The revisions separate the previously published library from the newly generated targeted dataset, describe the global-epistasis model and its role explicitly, correct the citations, and document the robustness of rescuability across all tested filtering thresholds.
eLife Assessment
This valuable study links NAD⁺-dependent HSD3B6 steroidogenic activity, androgen receptor signaling, and CD38-associated inflammation to meibomian gland atrophy, suggesting a potentially targetable feedback cycle relevant to dry eye disease. The revised work provides solid evidence for this relationship, with additional measurements showing no significant differences in circulating testosterone or aldosterone and persistence of elevated Cd38 expression and reduced glandular NAD⁺ levels after castration, strengthening support for a local steroidogenic mechanism. However, the global knockout does not definitively establish a gland-specific intracrine mechanism, and the absence of tear-film and ocular-surface functional measurements limits conclusions about the consequences for dry eye disease.
Reviewer #1 (Public review):
Summary:
The authors provide a compelling story linking NAD dependent HSD3b6 gene expression in the eyelid to a viscous cycle involving decreased steroidogenesis and AR signaling, pro-inflammatory cytokine release, inflammation, CD38 activation and further NAD+ decline that induces meibomian gland atrophy leading to dry eye disease. The work follows on studies published by the authors in Nature Aging and uses a previously reported HSD3b6 knockout mouse to try and show meibomian gland atrophy, inflammation and CD38+ cell recruitment. Overall the data presented does not appear to support the authors conclusions, and the described interactions seem to be overstated, in that if HSD3b6 was critical to meibomian gland function, one might expect complete loss of the meibomian gland in knockout mice, which is not the case.
Comments on revised version:
It should be noted that while the authors indicate the CD38 is significantly up-regulated in the HSD3b6 KO mouse, the increase was not sufficient to show a significant adjusted P-value after revision of the manuscript. Bulk RNA sequencing also shows no significant change in meibum lipid gene expression for aged mice that are treated with 78c, an inhibitor of CD38, that the authors indicate increases NAD levels leading to increased meibomian gland size compared to vehicle treated mice. Unfortunately, there was no increase in meibum lipid gene expression with 78c or with 78c+NMN treatment as identified by adjusted P-value.
Reviewer #2 (Public review):
Summary:
In the manuscript entitled, "NAD boosting mediated by CD38 inhibition drives reversal of a pathological vicious cycle of intracrine activity and inflammation in eyelid meibomian gland dysfunction", the authors demonstrate strong correlations between a pro-inflammatory state, the activity of an intracrine hormone (3 beta-hydroxysteroid dehydrogenase, 3B-HSD) and the NAD co-factor. Specifically, in a 3B-HSD knockout mouse, there was an upregulation in pro-inflammatory cytokines and increased CD38+ cells (CD38 is an enzyme that depletes NAD, a necessary cofactor for 3B-HSD activity). Conversely, induction of inflammation in the eyelids resulted in reductions in 3B-HSD activity. Supplementation with 5 alpha-dihydrotestosterone (DHT) or the NAD precursor NMN, and inhibition of CD38 activity (78c), corrected the pathologies observed in both the 3B-HSD knockout mouse and the pro-inflammatory model (LPS injection into eyelids).
Strengths:
The experiments were performed with good rigor, assessing the impact of inflammation and 3B-HSD activity using multiple model systems. The endpoints represented a combination of transcriptional changes, protein quantification, enzymatic activity, and immunofluorescent microscopy. The authors use human tissue from both younger and older individuals to justify their hypotheses that increased CD38 + cells and reduced 3B-HSD quantity exist in older individuals. The data provide the foundation for assessing more global changes to the tear film and ocular surface.
Weaknesses:
The main weaknesses of the study include:
(1) An absence of information on meibomian gland health, tear film and ocular surface.
(2) Too few human subjects to validate the hypotheses.
Overall, this study demonstrates an important relationship that exists between intracrine signaling, inflammation and cofactor signaling, and represents a novel approach in therapeutic design for patients with meibomian gland dysfunction.
Comments on the revised version:
The title of the manuscript is a bit deceiving, particularly because it mentions "meibomian gland dysfunction". The authors would need to demonstrate abnormalities in lipid synthesis (quantity or quality) and ocular surface changes associated with loss of this intracrine signaling. The data presented are strong for demonstrating the role of NAD synthesis and utilization in meibomian gland homeostasis, but do not explain the role in ophthalmic diseases, such as MGD. This could possibly be applicable to forms of blepharitis or conjunctivitis as well given their findings of CD38+ cells present in the eyelids of the KO mice. Therefore, the authors should remove the claims of meibomian gland dysfunction in the title and comments in the abstract. Instead, they could use meibomian gland homeostasis for both sections.
Author response:
The following is the authors’ response to the original reviews.
Public Reviews:
Reviewer #1 (Public review):
Summary:
While the results show some loss in the eyelid meibomian glands, there is significant gland retention in HSD3b6 KO mice, as shown in Figure 2. This is supported by the lack of DEG patterns showing downregulation of Meibum lipid genes (AWAT2, Far2, Soat1, Plin2, SCD, etc.), and no decrease in Pparg expression, known to be critical for meibomian gland lipid gene expression.
Weaknesses:
It should be noted that while the authors indicate that CD38 is significantly up-regulated in the HSD3b6 KO mouse, the increase was not sufficient to show a significant adjusted P-value. Bulk RNA sequencing also shows no significant change in meibum lipid gene expression for aged mice that are treated with 78c, an inhibitor of CD38, which the authors indicate increases NAD levels, leading to increased meibomian gland size compared to vehicle-treated mice.
Unfortunately, there was no increase in meibum lipid gene expression with 78c, as identified by adjusted P-value. However, it should be noted that the supplemental file covering DEG expression was labeled as a Microarray analysis. This did not include the 78c+NMN treated mice, which the authors contend show a more impactful effect on the meibomian gland.
We thank the reviewer for the careful evaluation of our work and for the constructive comments regarding the interpretation of the transcriptomic data and the molecular basis of meibomian gland dysfunction.
First, regarding the statistical significance of Cd38 expression in the microarray analysis, we performed additional validation using gene-specific quantitative PCR in independent eyelid tarsal plate samples. These new data are now presented in Fig. 2C, and the Results section has been revised as follows:
“… To validate the upregulation of Cd38, we performed real-time PCR analysis using gene-specific primers in independent eyelid tarsal plate samples, which confirmed a significant increase in Cd38 expression in Hsd3b6<sup>−/−</sup> mice compared to Hsd3b6<sup>+/+</sup> controls. This difference was also observed in castrated mice; thus, the effect of Hsd3b6 deletion on Cd38 expression is independent of testicular testosterone (Fig. 2C).”
Second, we apologize for the error in the original supplementary DEG file, which caused confusion during the review process. We have now replaced it with a complete RNA-seq dataset that includes all expression values for aged mice treated with 78c+NMN (please see Supplementary Data 2). As shown in Fig. 4E, F, treatment with 78c+NMN significantly increased the expression of Elovl4 and Elovl7, elongases responsible for the synthesis of very long-chain fatty acids in the meibomian gland. We also re-examined representative meibum lipid-related genes (Awat2, Far2, Soat1, Plin2, and Scd1). Although none reached statistical significance individually, collective analysis of these representative lipid-related genes by two-way ANOVA with Bonferroni's post hoc test demonstrated a significant overall increase following 78c+NMN treatment in aged mice, compared with controls, supporting coordinated upregulation of the meibomian gland lipid synthesis program.
These analyses indicate that, although the individual lipid-related genes showed only modest changes, they exhibit a coordinated increase following 78c+NMN treatment in aged mice. By contrast, the same gene set showed no significant overall change in young (3-month-old) mice, consistent with the original RNA-seq results presented in Fig. 4E. To better illustrate these findings, we have summarized the expression of these representative genes in a new Supplementary Fig. S5 and revised the Results accordingly, as follows:
“…Furthermore, representative meibum lipid-related genes, including Awat2, Far2, Soat1, Plin2, and Scd1, exhibited a coordinated increase following 78c-NMN treatment (Fig. S5). While the changes in individual genes were modest and did not reach statistical significance (Fig. 4E), analysis of the gene set as a whole demonstrated a significant overall increase compared with vehicle-treated controls.”
“In contrast, the same treatment in young mice did not significantly alter gene expression profiles (Fig. 4E, young), and no coordinated increase was observed for Awat2, Far2, Soat1, Plin2, or Scd1 (Fig. S5). This result is consistent with previous studies [44,47,48] showing that 78c (or NMN) exerts minimal effects in young mice compared with aged mice across multiple physiological parameters, including glucose homeostasis, exercise tolerance, and cardiac function.”
We also would like to note that the complete RNA-seq datasets provided in Supplementary Data 2 allow readers to examine the expression profiles of additional genes of interest.
Finally, regarding the observation that meibomian glands remain partially preserved despite the presence of gland atrophy in Hsd3b6<sup>−/−</sup> mice, together with the limited significant changes in classical lipid-related genes, we agree that this is an important point for interpreting the mechanism(s) of meibomian gland dysfunction. Our previous transcriptomic analysis of FACS-purified meibomian acinar cells in Hsd3b6<sup>−/−</sup> mice (PMID: 37117756) demonstrated reduced expression of genes involved in cell proliferation/differentiation, despite relatively preserved expression of canonical Pparg-dependent lipid genes. Our findings therefore extend the current framework by suggesting that meibomian gland atrophy can occur, at least in part, independently of detectable changes in Pparg expression, supporting the idea that intracrine steroidogenesis contributes to maintaining the proliferative activity of basal acinar cells alongside the Pparg-dependent pathway. Accordingly, we have revised the Discussion to clarify the relationship between our findings and the established Pparg-centred model of meibomian gland maintenance, as follows:
“Although 78c+NMN treatment effectively restored meibomian gland size, the molecular mechanisms underlying this recovery remain incompletely understood. Treatment with 78c+NMN significantly increased the expression of Elovl4, Elovl7, and Hsd3b6. However, representative meibum lipid-related genes, including Awat2, Far2, Soat1, Plin2, and Scd1, exhibited only a modest coordinated increase following treatment. Interestingly, the expression of Pparg, a key transcriptional regulator of meibocyte differentiation and lipid synthesis [88-90], remained unchanged. These findings suggest that restoration of meibomian gland size by 78c+NMN does not require detectable changes in Pparg mRNA expression, although Pparg activity may instead be regulated at the level of protein abundance, subcellular localization, and/or post-translational modifications such as phosphorylation or SUMOylation [88,91].
Our previous transcriptomic analysis of FACS-purified meibomian gland basal acinar cells (meibomian precursor cells expressing androgen receptors) from Hsd3b6<sup>−/−</sup> mice demonstrated impaired expression of genes involved in cell proliferation and differentiation despite relatively preserved expression of lipid-related genes [6]. In contrast, a recent single-cell study showed that Pparg expression progressively increases during meibocyte differentiation, although it is already detectable in basal acinar cells [92]. These observations suggest that Hsd3b6-dependent intracrine activity primarily supports the maintenance and proliferation of basal acinar cells, whereas the canonical Pparg pathway predominantly regulates subsequent meibocyte differentiation and lipid synthesis. Together, these findings suggest that Hsd3b6 dependent intracrine activity contributes to meibomian gland homeostasis through mechanisms that complement the Pparg-associated pathway. Thus, our findings extend the current framework by indicating that meibomian gland homeostasis is maintained not only through canonical PPARγ-mediated regulation but also through intracrine steroidogenesis.
Newly added references:
(88) Jester, J.V., Potma, E., and Brown, D.J. (2016). PPARγ Regulates Mouse Meibocyte Differentiation and Lipid Synthesis. Ocul Surf 14, 484-494. 10.1016/j.jtos.2016.08.001.
(89) Kim, S.W., Xie, Y., Nguyen, P.Q., Bui, V.T., Huynh, K., Kang, J.S., Brown, D.J., and Jester, J.V. (2018). PPARγ regulates meibocyte differentiation and lipid synthesis of cultured human meibomian gland epithelial cells (hMGEC). Ocul Surf 16, 463-469. 10.1016/j.jtos.2018.07.004.
(90) Hwang, H.S., Parfitt, G.J., Brown, D.J., and Jester, J.V. (2017). Meibocyte differentiation and renewal: Insights into novel mechanisms of meibomian gland dysfunction (MGD). Exp Eye Res 163, 37-45. 10.1016/j.exer.2017.02.008.
(91) Ahmadian, M., Suh, J.M., Hah, N., Liddle, C., Atkins, A.R., Downes, M., and Evans, R.M. (2013). PPARγ signaling and metabolism: the good, the bad and the future. Nat Med 19, 557-566. 10.1038/nm.3159.
(92) Zhu, X., Xu, M., Portal, C., Lin, Y., Ferdinand, A., Peng, T., Morrisey, E.E., Dlugosz, A.A., Castellano, J.M., Lee, V., et al. (2025). Identification of Meibomian gland stem cell populations and mechanisms of aging. Nat Commun 16, 1663. 10.1038/s41467-02556907-6.
Reviewer #2 (Public review):
Summary:
In this manuscript, the authors demonstrate strong correlations between a pro-inflammatory state, the activity of an intracrine hormone (3 beta-hydroxysteroid dehydrogenase, 3B-HSD), and the NAD co-factor. Specifically, in a 3B-HSD knockout mouse, there was an upregulation in proinflammatory cytokines and increased CD38+ cells (CD38 is an enzyme that depletes NAD, a necessary cofactor for 3B-HSD activity). Conversely, induction of inflammation in the eyelids resulted in reductions in 3B-HSD activity. Supplementation with 5 alpha-dihydrotestosterone (DHT) or the NAD precursor NMN, and inhibition of CD38 activity (78c), corrected the pathologies observed in both the 3B-HSD knockout mouse and the pro-inflammatory model (LPS injection into eyelids).
Strengths:
The experiments were performed with good rigor, assessing the impact of inflammation and 3BHSD activity using multiple model systems. The endpoints represented a combination of transcriptional changes, protein quantification, enzymatic activity, and immunofluorescent microscopy. The authors use human tissue from both younger and older individuals to justify their hypotheses that increased CD38 + cells and reduced 3B-HSD quantity exist in older individuals. The data provide the foundation for assessing more global changes to the tear film and ocular surface.
Weaknesses:
The main weaknesses of the study include the following:
(1) An absence of information on meibomian gland health, tear film, and ocular surface.
(2) Too few human subjects to validate the hypotheses.
Conclusion:
Overall, this study demonstrates an important relationship that exists between intracrine signaling, inflammation, and cofactor signaling. It represents a novel approach in therapeutic design for patients with meibomian gland dysfunction.
We thank the reviewer for the careful evaluation of our work and for recognizing the novelty and potential therapeutic implications of the relationship among intracrine signaling, inflammation, and NAD metabolism in meibomian gland dysfunction. We also appreciate the reviewer’s constructive comments on the limitations of the present study and areas for improvement in the manuscript.
Regarding the lack of information on meibomian gland health, tear film, and ocular surface, we have added a limitations section to explicitly acknowledge that these functional outcomes were not directly assessed in the present study. As the reviewer kindly noted, we only focused on a key pathological hallmark of MGD, namely age-associated tissue atrophy, and based on this we explored the NAD/3β-HSD/inflammation-mediated mechanism. Thus, a limitation of the present study is the lack of assessment of the downstream consequences of MGD. This point is now explicitly discussed in the manuscript; please note that additional details are provided in our responses to Reviewer #2’s Recommendations for the authors.
Regarding the human data, we added two additional human eyelid samples, increasing the sample size of the older group from 2 to 4. Nevertheless, we agree that the sample size remains limited and have now explicitly acknowledged this limitation in Discussion as follows:
“A caveat of this study is that the number of human eyelid samples analyzed was limited. Larger human cohorts will be necessary to establish the translational relevance of the age-associated CD38 accumulation observed in human tissues.”
Taken together, we believe that these revisions have clarified the scope and limitations of our present study.
Reviewer #3 (Public review):
Summary:
The authors aimed to investigate whether disruption of intracrine steroid hormone metabolism contributes to meibomian gland dysfunction and proposed a "vicious cycle" of gland dysfunction and inflammation, using a global Had3b6 knockout mouse model. The work addresses an important aspect of MGD, but its impact may be limited unless the intracrine mechanism can be more clearly distinguished from systemic hormonal effects.
Strengths:
This study addressed an important question. The hormonal regulation of the meibomian gland has long been recognized. If clarified, the concept of local steroid metabolism influencing gland homeostasis could have implications for understanding disease mechanisms and identifying therapeutic targets.
Weaknesses:
The use of a global knockout makes it difficult to separate local intracrine effects from systemic hormonal changes, and key controls and hormone measurements are lacking. LPS-induced inflammation may not reflect the chronic nature of MGD.
We thank the reviewer for the thoughtful evaluation of our work and for highlighting the importance of distinguishing local intracrine mechanisms from systemic hormonal effects. In response, we performed several additional experiments and substantially revised the manuscript. Specifically, we measured plasma testosterone and aldosterone levels and found no significant differences between Hsd3b6<sup>−/−</sup> and Hsd3b6<sup>+/+</sup> mice. In addition, we repeated key experiments in castrated mice and observed that the effects of Hsd3b6 deletion on tarsal plate Cd38 expression and meibomian NAD levels were preserved despite the absence of testicular testosterone. These new findings provide additional support for a local intracrine mechanism in the eyelid. We also revised the Introduction and Discussion to better explain the proposed intracrine–inflammation vicious cycle and clarified the rationale and limitations of the LPS-based inflammation model, including its inability to fully recapitulate the chronic low-grade inflammation characteristic of MGD. Further details are provided in our responses to Reviewer #3’s Recommendations for the authors. We hope that these additions and revisions have strengthened the manuscript and addressed the reviewer’s concerns.
Recommendations for the authors:
Reviewer #1 (Recommendations for the authors):
Overall, the authors need to provide bulk RNA sequencing data for all treatment groups and provide some explanation of the measurements in individual mice for gland growth.
We thank the reviewer for this helpful comment. We have now replaced the original supplementary file with a complete RNA-seq dataset that includes all treatment groups (Supplementary Data 2). In addition, as described in our response to the Public Review comment above, we extensively revised our manuscript to provide explanation of gene expression profiles (particularly, representative meibum lipid-related genes) in the dataset.
Accordingly, we have expanded the Results and reconstructed the Discussion by incorporating two new discussion paragraphs. The relevant revisions are provided below for your perusal:
“… Furthermore, representative meibum lipid-related genes, including Awat2, Far2, Soat1, Plin2, and Scd1, exhibited a coordinated increase following 78c-NMN treatment (Fig. S5). While the changes in individual genes were modest and did not reach statistical significance (Fig. 4E), analysis of the gene set as a whole demonstrated a significant overall increase compared with vehicle-treated controls.”
“In contrast, the same treatment in young mice did not significantly alter gene expression profiles (Fig. 4E, young), and no coordinated increase was observed for Awat2, Far2, Soat1, Plin2, or Scd1 (Fig. S5). This result is consistent with previous studies [44,47,48] showing that 78c (or NMN) exerts minimal effects in young mice compared with aged mice across multiple physiological parameters, including glucose homeostasis, exercise tolerance, and cardiac function.”
“Although 78c+NMN treatment effectively restored meibomian gland size, the molecular mechanisms underlying this recovery remain incompletely understood. Treatment with 78c+NMN significantly increased the expression of Elovl4, Elovl7, and Hsd3b6. However, representative meibum lipid-related genes, including Awat2, Far2, Soat1, Plin2, and Scd1, exhibited only a modest coordinated increase following treatment. Interestingly, the expression of Pparg, a key transcriptional regulator of meibocyte differentiation and lipid synthesis [88-90], remained unchanged. These findings suggest that restoration of meibomian gland size by 78c+NMN does not require detectable changes in Pparg mRNA expression, although Pparg activity may instead be regulated at the level of protein abundance, subcellular localization, and/or post-translational modifications such as phosphorylation or SUMOylation [88,91].
Our previous transcriptomic analysis of FACS-purified meibomian gland basal acinar cells (meibomian precursor cells expressing androgen receptors) from Hsd3b6<sup>−/−</sup> mice demonstrated impaired expression of genes involved in cell proliferation and differentiation despite relatively preserved expression of lipid-related genes [6]. In contrast, a recent single-cell study showed that Pparg expression progressively increases during meibocyte differentiation, although it is already detectable in basal acinar cells [92]. These observations suggest that Hsd3b6-dependent intracrine activity primarily supports the maintenance and proliferation of basal acinar cells, whereas the canonical Pparg pathway predominantly regulates subsequent meibocyte differentiation and lipid synthesis. Together, these findings suggest that Hsd3b6-dependent intracrine activity contributes to meibomian gland homeostasis through mechanisms that complement the Pparg-associated pathway. Thus, our findings extend the current framework by indicating that meibomian gland homeostasis is maintained not only through canonical PPARγ-mediated regulation but also through intracrine steroidogenesis.”
Reviewer #2 (Recommendations for the authors):
The main suggestion for the study is the correlation with meibomian gland function. While the study correlates changes with 3B-HSD activity, pro-inflammatory cytokines, and NAD cofactor levels, there is little support describing the pathological changes to the meibomian glands and the impact on the tear film and ocular surface. The changes in the lipid profiles, tear film stability, aqueous tear production, ocular surface staining properties, corneal epithelial transcriptional abnormalities, and leukocyte infiltration into the corneal stroma could all be assessed. While the authors have done a thorough job of demonstrating the correlation of the three major factors listed above, the correlation with meibomian gland function is not completely assessed in the study.
We thank the reviewer for this insightful comment. We agree that functional assessments such as tear film stability, ocular surface pathology, lipid composition, and other downstream manifestations of MGD would complement the mechanistic findings reported here and provide additional insight into their physiological consequences.
However, the primary objective of the present study was to define the molecular mechanism underlying MGD, specifically the pathological interaction between meibomian gland intracrine activity and inflammation, and the role of the CD38–NAD axis in this process. Accordingly, our experiments were designed to focus on mechanistic endpoints directly related to this hypothesis, including 3β-HSD activity, NAD availability, inflammatory gene expression, and CD38 accumulation. In addition, we assessed a key pathological hallmark of MGD, namely age-associated meibomian gland atrophy and its recovery following 78c/NMN treatment.
While direct evaluation of tear film and ocular surface parameters was beyond the scope of the present mechanistic study, we agree that such analyses represent important future directions. To acknowledge this, we have added the following statement to the Discussion:
“… Firstly, while this study focused on defining the molecular and structural mechanisms underlying meibomian gland dysfunction, functional outcomes such as tear film stability, ocular surface integrity, meibum lipid composition, and other clinically relevant manifestations of disease were not directly assessed. Given the well-established link between meibomian gland dysfunction and tear film instability [57-59], future studies incorporating these functional assessments will be important to determine how the molecular and structural changes identified here contribute to ocular surface pathology and disease progression.”
The description of the vicious cycle seemed a bit out of place in the introduction. This would have been most appropriate in the conclusion/discussion in the manuscript. The general nature of the description lacked focus on the eyelids and meibomian glands, and the narrative took time to get to the interactions that would be assessed in the study.
We thank the reviewer for this insightful suggestion. We agree that the original Introduction placed excessive emphasis on the general concept of vicious cycles and that the discussion of disease-wide examples diverted attention from the specific focus of the present study on the meibomian gland.
In response, we substantially shortened and refocused the Introduction. Specifically, we removed the broad descriptions of vicious cycles in neurodegenerative, metabolic, and cardiovascular diseases. The revised Introduction now introduces the concept of a vicious cycle only briefly and moves more directly to the biological context of meibomian gland dysfunction and local intracrine steroidogenesis.
At the same time, we agree that the broader implications of this concept are better suited for the Discussion. We therefore added a brief concluding paragraph to discuss how our findings may inform the understanding of age-associated vicious-cycle processes in other diseases and may also provide a framework for considering intracrine-related pathology beyond the meibomian gland. The corresponding text has been added to the Discussion, as follows.
“Finally, although the present study focused on the eyelid meibomian gland, the concept of an intracrine–inflammation vicious cycle may have broader relevance. Self-reinforcing pathological processes are implicated in several age-associated diseases, and altered intracrine activity has been described in multiple extra gonadal and cancerous tissues [5,7,8,93,94]. Our findings may therefore provide a useful framework for understanding how local hormone metabolism contributes to the initiation and maintenance of tissue-specific pathological cycles during ageing.”
Reviewer #3 (Recommendations for the authors):
(1) Introduction:
(a) The first paragraph of the introduction defines the term "vicious cycle," but its relevance to the current study is low. It is not until line 54 that the specific tissue of interest is introduced. It is suggested that the beginning of the introduction be shortened to emphasize information that is directly relevant to the study.
We thank the reviewer for this helpful suggestion. We agree that the original opening of the Introduction was too general and that the specific focus of the study on the meibomian gland should be introduced earlier.
In response, we substantially shortened the first part of the Introduction. Specifically, we removed the broad discussion of vicious cycles across multiple disease contexts and revised the opening so that the concept of a vicious cycle is introduced only briefly. The revised Introduction now moves directly to the relevance of local intracrine steroidogenesis in the eyelid meibomian gland, thereby better emphasizing the biological context and focus of the present study.
At the same time, we recognized that some of the broader implications of the study remain important. Therefore, we incorporated these points into the Discussion, where we briefly discuss the potential relevance of intracrine–inflammation vicious cycles to other age-associated diseases and intracrine-related pathologies beyond the meibomian gland, as follows:
“Finally, although the present study focused on the eyelid meibomian gland, the concept of an intracrine–inflammation vicious cycle may have broader relevance. Self-reinforcing pathological processes are implicated in several age-associated diseases, and altered intracrine activity has been described in multiple extragonadal and cancerous tissues [5,7,8,93,94]. Our findings may therefore provide a useful framework for understanding how local hormone metabolism contributes to the initiation and maintenance of tissue-specific pathological cycles during ageing.”
(b) In the introduction, the authors should also include common types of intracrine hormones, such as DHT.
We thank the reviewer again for this valuable suggestion. In response, we revised the Introduction to include representative examples of intracrine steroid hormones, including 5α-dihydrotestosterone (DHT) and estrogens, together with a brief description of their local production and action within peripheral tissues. We believe that this addition helps clarify the concept of intracrinology for readers less familiar with the field.
“Common examples of intracrine steroid hormones include 5α-dihydrotestosterone (DHT) produced in the skin, prostate, and eyelid meibomian glands, as well as estrogens synthesized within the breast and endometrium [3-8]. These locally produced hormones act within the tissue in which they are synthesized.”
(c) The hypothesis of a "vicious cycle" in MGD is not fully explained. Specifically, how intracrine steroids drive the meibomian gland into a vicious cycle is not described. The current information only discusses a possible association between steroid hormones and meibomian gland health.
According to this important suggestion, we performed additional experiments and rewrote our introduction. Please note that plasma testosterone levels were not impaired in Hsd3b6<sup>−/−</sup> mice and their eyelid proinflammatory phenotypes (compared to wt mouse eyelid) remained preserved in castrated mice. These data are now provided, by performing additional experiments (please see our replies to your comments, below). Taking these observations into account, we have tempered our introduction.
(2) Results:
(a) One major concern is that sex hormones have long been recognized to be associated with the meibomian gland and have been extensively studied (see the TFOS guidelines). There have also been clinical trials using androgens to treat MGD. With the global knockout of Had3b6, systemic sex hormone levels could also be altered, which may result in MGD (estrogens are synthesized from androgens and would therefore also be affected).
This thoughtful concern is well received, as we were not attentive to describe systemic testosterone in Hsd3b6<sup>+/+</sup> and Hsd3b6<sup>−/−</sup> mice in our original manuscript (for which we should apologize). The corresponding plasma testosterone data are now incorporated as Fig. S1A and presented below for your perusal.
Importantly, plasma testosterone levels in Hsd3b6<sup>−/−</sup> mice were comparable with those in wt Hsd3b6<sup>+/+</sup> mice, consistent with the known compensatory role of the paralogous enzyme Hsd3b1 in the mouse testis [21], which was confirmed to be expressed in Hsd3b6<sup>−/−</sup> mice (see new Fig. S1B). This type of compensation by paralogous steroidogenic enzymes has also been reported in other knockout model, such as Hsd17b3<sup>−/−</sup> mice, in which circulating testosterone levels were preserved due to a compensatory role of Hsd17b12 in the testis (new refs [83,84]). Based on these results, we rewrote our Results and Discussion, as follows.
“plasma testosterone levels were indistinguishable between Hsd3b6<sup>−/−</sup> and Hsd3b6<sup>+/+</sup> mice (Fig. S1A), consistent with the known compensatory role of the paralogous enzyme Hsd3b1 in the mouse testis [21], which was confirmed to be expressed in Hsd3b6<sup>−/−</sup> mice (Fig. S1B). These results indicate that local steroidogenic activity in the eyelid exerts anti-inflammatory effects independent of circulating testosterone levels.”
“For the interpretation of our meibomian intracrine activity model, it is important to note that in Hsd3b6<sup>−/−</sup> mice, plasma testosterone levels were maintained at the levels comparable with those in Hsd3b6<sup>+/+</sup> mice (Fig. S1A). This appears due to a redundant expression of Hsd3b1 in the testis [6], although in the meibomian gland, Hsd3b6 is the sole enzyme (Fig. S1B) and accordingly its deletion causes the complete abolition of meibomian 3β-HSD activity [6]. The resilience of plasma testosterone levels has also been reported in a knockout mouse model deficient in Hsd17b3, a member of 17-ketosteroid reductase enzyme family required for the formation of testosterone [83], likely due to a potential compensatory role of the paralogous enzyme Hsd17b12 in the testis [84]. The Hsd3b6<sup>−/−</sup> vs. Hsd3b6<sup>+/+</sup> phenotypes in the tarsal plate Cd38 expression and meibomian NAD levels were preserved in castrated mice (Fig. 2C, F), which provide additional evidence to support that non-systemic local steroidogenic activity in the eyelid, rather than circulating testicular testosterone, underlies the phenotypes observed in Hsd3b6<sup>−/−</sup> mice.”
Newly added references:
(83) Lawrence, B.M., O'Donnell, L., Gannon, A.L., Smith, S., Curley, M.K., Darbey, A., Mackay, R., O'Shaughnessy, P.J., Smith, L.B., and Rebourcet, D. (2024). Compensatory mechanisms that maintain androgen production in mice lacking key androgen biosynthetic enzymes. FASEB J 38, e70177. 10.1096/fj.202402093R.
(84) Rebourcet, D., Mackay, R., Darbey, A., Curley, M.K., Jørgensen, A., Frederiksen, H., Mitchell, R.T., O'Shaughnessy, P.J., Nef, S., and Smith, L.B. (2020). Ablation of the canonical testosterone production pathway via knockout of the steroidogenic enzyme HSD17B3, reveals a novel mechanism of testicular testosterone production. FASEB J 34, 1037310386. 10.1096/fj.202000361R.
- For context, please note that plasma testosterone levels showed substantial inter-individual variability in both genotypes. Such variability is a well-established physiological feature of male mice, reflecting episodic testosterone secretion. Indeed, classical studies showed that variation in testosterone levels within the same mouse sampled on different occasions is comparable to that observed between individual mice, and a recent longitudinal LC–MS/MS study confirmed 10- to 100-fold fluctuations within individual adult male mice (Bartke and Dalterio, 1975, PMID: 1216260; Kettering et al., 2026, PMID: 41880754). Thus, the variability observed in our dataset is consistent with the known physiological dynamics of circulating testosterone in male mice.
(b) To support the hypothesis of intracrine regulation in MGD, the following would be essential:
- a local or targeted knockout model, and
- a global knockout with appropriate controls, such as Had3b6+/+ mice combined with gonadectomy.
Otherwise, the current study reflects the role of general steroid hormones in the meibomian gland rather than an intracrine-specific mechanism.
We repeated our experiments using castrated Hsd3b6<sup>−/−</sup> mice. As shown in new Fig. 2C, we observed a significant increase in Cd38 expression in the tarsal plate of Hsd3b6<sup>−/−</sup> mice (n = 6, castrated) as compared to Hsd3b6<sup>+/+</sup> control mice (n = 4, castrated). In addition, in Fig. 2F, we observed a significant decrease in NAD levels in the meibomian gland of Hsd3b6<sup>−/−</sup> mice (n = 3, castrated) as compared to Hsd3b6<sup>+/+</sup> control mice (n = 5, castrated). These results provide additional evidence to support that local steroidogenic activity in the eyelid, rather than circulating testicular testosterone, underlies the phenotypes observed in Hsd3b6<sup>−/−</sup> mice. Based on these results and consideration, we have modified our manuscript as follows.
“… This difference was also observed in castrated mice; thus, the effect of Hsd3b6 deletion on Cd38 expression is independent of testicular testosterone (Fig. 2C).”
“… Note that regardless of castration, Hsd3b6 deletion reduced meibomian gland NAD levels.”
“…The Hsd3b6<sup>−/−</sup> vs. Hsd3b6<sup>+/+</sup> phenotypes in the tarsal plate Cd38 expression and meibomian NAD levels were preserved in castrated mice (Fig. 2C, F), which provide additional evidence to support that non-systemic local steroidogenic activity in the eyelid, rather than circulating testicular testosterone, underlies the phenotypes observed in Hsd3b6<sup>−/−</sup> mice.”
(c) With the knockout of Had3b6, other steroid hormones may also be altered systemically, such as increased synthesis of the mineralocorticoid aldosterone. These hormonal changes could cause systemic effects such as hypertension, with unknown consequences for the meibomian gland. Please provide data on systemic steroid hormone levels to better interpret the current findings.
We thank the reviewer for raising this important point. To address the possibility that systemic mineralocorticoid alterations contribute to the observed meibomian gland phenotypes, we measured plasma aldosterone levels in Hsd3b6<sup>−/−</sup> and Hsd3b6<sup>+/+</sup> mice. These new data are now included as Fig. S6.
Importantly, plasma aldosterone levels were not significantly different between Hsd3b6<sup>−/−</sup> and Hsd3b6<sup>+/+</sup> mice (Fig. S6). This finding is consistent with the known expression pattern of steroidogenic enzymes in the adrenal zona glomerulosa, where both Hsd3b1 and Hsd3b6 are redundantly expressed and contribute to aldosterone biosynthesis (refs. 85-87). Thus, similar to the compensatory role of Hsd3b1 in maintaining systemic testosterone production in the testis, the redundant expression of Hsd3b1 in aldosterone-producing cells likely preserves systemic aldosterone levels in Hsd3b6<sup>−/−</sup> mice.
Based on these new data, we have added the plasma aldosterone measurements as Fig. S6 and revised the Discussion accordingly. These results further support our conclusion that the phenotypes observed in Hsd3b6<sup>−/−</sup> mice are unlikely to be secondary to systemic endocrine alterations and instead reflect the loss of local steroidogenic activity within the eyelid.
“…Plasma aldosterone levels were likewise preserved in Hsd3b6<sup>−/−</sup> mice (Fig. S6), consistent with the redundant expression of Hsd3b6 and Hsd3b1 in the adrenal gland zona glomerulosa cells [85-87].”
(d) What is the age of the mice that underwent microarray analysis? The Methods section only specifies "young" and "old" without providing the actual ages. Although the ages can be inferred from the figures, they should also be clearly stated in the Methods section.
We thank the reviewer for this comment. We have now clarified the age of the mice in the Methods section. All mice used for the microarray analysis were age-matched and were 9 weeks old at the time of tissue collection (please see page 35, lines 818-819).
(e) Line 148: CD11b+ cells in the eye are usually referred to as dendritic cells derived from monocytes that function as antigen-presenting cells (APCs), rather than as phagocytes.
(f) Line 149: How was it determined that the CD38+ cells are macrophages? CD38 can also be expressed by T cells, plasma cells, NK cells, and other immune cell types.
Heeding these important instructions and based on their co-labeling with the myeloid marker CD11b, we have rewritten our Results on the cell identity of CD38 cells, as follows.
“… We also found that the myeloid cell marker CD11b stained the same cells (Fig. S2), indicating an increased number of myeloid immune cells that express CD38 in the tarsal plate of Hsd3b6<sup>−/−</sup> mice (Fig. 2E, vs. Hsd3b6<sup>+/+</sup>).”
In addition, to clarify that multiple immune cell types can contribute to CD38 expression in the tarsal plate and ocular tissues, we have extensively revised our manuscript by adding a new paragraph to the Discussion as follows.
“Secondly, the cellular identity of the CD38-expressing cells observed in the tarsal plate warrants careful interpretation. In the present study, we observed co-localization of CD38 with the myeloid marker CD11b, indicating that these cells are of myeloid origin. However, CD11b<sup>+</sup> cells in ocular tissues such as in conjunctiva are often described as monocyte-derived dendritic cells functioning as antigen-presenting cells, rather than as classical phagocytic macrophages [95,96]. Therefore, our data do not allow us to further resolve the specific myeloid subpopulation (e.g., macrophages, dendritic cells, or other subsets) responsible for CD38 expression. In addition, CD38 is known to be expressed across multiple immune cell types, including T cells, plasma cells, and NK cells [97-99], suggesting that non-myeloid immune cells may also contribute to the overall CD38 signal in the tarsal plate. While previous studies in other tissues, such as in adipose tissue and liver [39,40], have attributed age-associated CD38 upregulation primarily to macrophages, our findings highlight the need for more precise cell-type–resolved analyses in ocular tissues. Future studies using lineage-specific markers or single-cell approaches will be required to define the exact cellular sources of CD38 and their respective contributions to NAD metabolism and inflammation in the eyelid.” Added references:
(94) Khandelwal, P., Blanco-Mezquita, T., Emami, P., Lee, H.S., Reyes, N.J., Mathew, R., Huang, R., and Saban, D.R. (2013). Ocular mucosal CD11b+ and CD103+ mouse dendritic cells under normal conditions and in allergic immune responses. PLoS One 8, e64193. 10.1371/journal.pone.0064193.
(95) Ahadome, S.D., Mathew, R., Reyes, N.J., Mettu, P.S., Cousins, S.W., Calder, V.L., and Saban, D.R. (2016). Classical dendritic cells mediate fibrosis directly via the retinoic acid pathway in severe eye allergy. JCI Insight 1. 10.1172/jci.insight.87012.
(96) Malavasi, F., Deaglio, S., Funaro, A., Ferrero, E., Horenstein, A.L., Ortolan, E., Vaisitti, T., and Aydin, S. (2008). Evolution and function of the ADP ribosyl cyclase/CD38 gene family in physiology and pathology. Physiol Rev 88, 841-886. 10.1152/physrev.00035.2007.
(97) Katsuyama, E., Humbel, M., Suarez-Fueyo, A., Satyam, A., Yoshida, N., Kyttaris, V.C., Tsokos, M.G., and Tsokos, G.C. (2024). CD38 in SLE CD4 T cells promotes Ca2+ flux and suppresses interleukin-2 production by enhancing the expression of GM2 on the surface membrane. Nat Commun 15, 8304. 10.1038/s41467-024-52617-7.
(98) Peralta Ramos, J.M., Castellani, G., Kviatcovsky, D., Croese, T., Tsitsou-Kampeli, A., Burgaletto, C., Abellanas, M.A., Cahalon, L., Phoebeluc Colaiuta, S., Salame, T.M., et al. (2025). Targeting CD38 immunometabolic checkpoint improves metabolic fitness and cognition in a mouse model of Alzheimer's disease. Nat Commun 16, 3736. 10.1038/s41467-025-58494-y.
(g) LPS-induced inflammation is acute and severe and may not accurately model MGD, which is typically characterized by chronic, low-grade inflammation. It is suggested to use a more relevant model, such as benzalkonium chloride exposure or a desiccating stress chamber.
We thank the reviewer for this important suggestion. We agree that benzalkonium chloride exposure and desiccating stress are widely used and well-established models for ocular surface inflammation and dry eye [30-35], and we have also tested these approaches in our experimental system. However, while these treatments reproducibly induced inflammatory responses in the cornea, conjunctiva, and meibomian orifice, we did not observe detectable inflammatory changes within the tarsal plate (i.e., the internal eyelid tissue containing the meibomian glands). To our knowledge, induction of inflammation in the tarsal plate by these approaches has not been reported in the literature. As such, in our current study, we adopted subcutaneous LPS injection into the eyelid. However, as noted by the reviewer, this approach represents an acute and relatively strong inflammatory stimulus and does not fully recapitulate the chronic low-grade inflammation characteristic of MGD. Please note that, by describing both the rationale and limitations of this approach, we have improved our manuscript as follows:
“Eye-drop instillation of inflammatory stimuli such as those by lipopolysaccharide (LPS) or benzalkonium chloride has been shown to be effective to cause inflammation in the cornea and conjunctiva [30-35]. However, we found this approach ineffective to induce inflammation in the tarsal plate (inside the eyelid tissue). Therefore, we performed subcutaneous injection of LPS to the eyelids (see Methods).”
Methods [Page 30, lines 694-701] “LPS (Sigma) was reconstituted in saline and injected subcutaneously (5 µg/eyelid) 8 h before tissue collection for qRT-PCR analysis. LPS was administered every other day for a total of three or six doses before the measurement of meibomian NAD⁺ concentration or meibomian 3β-HSD enzymatic activity, respectively. Subcutaneous microinjection was performed using an insulin syringe (Becton Dickinson, 30G). This LPS-based approach represents an acute inflammatory stimulus and does not fully recapitulate the chronic low-grade inflammation characteristic of MGD — while it provides a practical means to induce inflammation in the tarsal plate.” Added references:
(30) Lin, Z., Liu, X., Zhou, T., Wang, Y., Bai, L., He, H., and Liu, Z. (2011). A mouse dry eye model induced by topical administration of benzalkonium chloride. Mol Vis 17, 257-264.
(31) Lin, Z., He, H., Zhou, T., Liu, X., Wang, Y., Wu, H., and Liu, Z. (2013). A mouse model of limbal stem cell deficiency induced by topical medication with the preservative benzalkonium chloride. Invest Ophthalmol Vis Sci 54, 6314-6325. 10.1167/iovs.12-10725.
(32) Wu, M., Downie, L.E., Hill, L.J., and Chinnery, H.R. (2023). Topical Decorin Reduces Corneal Inflammation and Imparts Neuroprotection in a Mouse Model of Benzalkonium Chloride-induced Corneal Neuropathy. Invest Ophthalmol Vis Sci 64, 20. 10.1167/iovs.64.2.20.
(33) Wu, C.Y., Wu, M., Huang, X., Gu, B.J., Maldonado-Codina, C., Morgan, P.B., Downie, L.E., and Chinnery, H.R. (2022). Topographical Distribution and Phenotype of Resident Meibomian Gland Orifice Immune Cells (MOICs) in Mice and the Effects of Topical Benzalkonium Chloride (BAK). Int J Mol Sci 23. 10.3390/ijms23179589.
(34) Simmons, K.T., Xiao, Y., Pflugfelder, S.C., and de Paiva, C.S. (2016). Inflammatory Response to Lipopolysaccharide on the Ocular Surface in a Murine Dry Eye Model. Invest Ophthalmol Vis Sci 57, 2443-2451. 10.1167/iovs.15-18396.
(35) Corrales, R.M., Villarreal, A., Farley, W., Stern, M.E., Li, D.Q., and Pflugfelder, S.C. (2007). Strain-related cytokine profiles on the murine ocular surface in response to desiccating stress. Cornea 26, 579-584. 10.1097/ICO.0b013e318033a729.
eLife Assessment
This fundamental study identifies GVIN as a previously unrecognized interferon-induced antimicrobial GTPase and establishes its role in cell-autonomous immunity against intracellular Burkholderia. The compelling data demonstrate that GVIN assembles into coat-like structures on the bacterial surface and restricts BimA-dependent actin-based motility through a mechanism independent of GBP1. These findings expand our understanding of interferon-induced host defense mechanisms and reveal a distinct pathway by which human cells restrict the intracellular spread of bacterial pathogens.
Reviewer #1 (Public review):
This is a very elegant and experimentally convincing study. Using systematic screening of actin tail formation in two bacterial strains and employing a panel of CRISPR-CAS ko cell lines, the authors identify the dynamin-related GTPases GVIN as antibacterial agent, which forms an oligomeric coat around an intracellular Burkholderia strain. The bacterial O-antigen LPS layer is required for the formation of the GVIN coat, which disturbs the polar localization of the bacterial actin-polymerizing BimA protein.
The experiments appear of high quality, figures are well prepared, and clean and statistically significant results are provided. I have no criticism on the presented approaches.
The identification of a novel GBP1-independent pathway targeting intracellular bacteria is not only of fundamental importance for the immunity field, but also of high interest to researchers in other areas, for example, evolutionary or structural biologists.
Comments on revised version.
This is a beautiful study.
Reviewer #2 (Public review):
Public summary
Summary:
The authors wanted to investigate how cells defend against intracellular pathogens, such as Shigella and Burkholderia species, that coopt the host actin machinery to spread from cell-to-cell. Previous work has identified IFNg-inducible GTPase of the Guanylate Binding Protein (GBP) family in cytosolic defence against Gram-negative bacteria. By forming a coat around Shigella, human GBP1 suppresses actin-based motility by displacing IcsA, which is the actin-polymerising virulence factor present at bacterial poles. In addition, GBP1 recruits the cytosolic LPS-sensor, caspase-4, to the bacterial surface, which results in the removal of bacterial replicative niches via pyroptotic cell death. Here they followed up their finding that GBP1 can reduce actin-based motility of Shigella in HeLa cells and surprisingly fails to do so during Burkholderia infection. In contrast, in T24 bladder epithelial cells, GBP1 is competent in blocking Burkholderia actin-tails. They therefore wanted to identify the GBP1-independent factor that blocks actin-based motility in IFNg-treated cells that is absent in HeLa cells.
Major strengths and weaknesses of the methods and results:
The authors report a second IFNg-dependent pathway involving the protein product of the gene GVIN1, which was previously thought to be a pseudogene. GVIN1 (GTPase, very large interferon inducible 1) is thus the first human member of this family of ~250 kDa putative GTPases to be demonstrated to be functional and have potential antimicrobial roles. The discovery that GVIN1 is indeed functional, forms coats on Burkholderia in an LPS O-antigen-dependent manner, and limits actin-dependent motility are the main strengths of this paper. The authors use CRISPR/Cas9-based knockouts in HeLa and T24 cells, and complement them, to demonstrate GBP1 and GVIN1 are both required to inhibit actin-based motility.
An appraisal of whether the authors achieved their aims, and whether the results support their conclusions.
The authors achieved their main goals through well-planned experiments and unbiased screens. They succeeded in finding the factor that blocks actin-based motility independently of GBP1. This is driven by GVIN1, which coats bacteria and limiting actin-tail formation by reducing the expression of BimA through currently unknown mechanisms. Further, they found that an O-antigen mutant can escape coating by GVIN1, indicating the requirement for these polysaccharides in GVIN1-dependent bacterial sensing. However, the authors have not investigated whether GVIN1, which has two GTPase-domains, does indeed have GTPase activity and whether GVIN1 and GBP1 together completely block cell-to-cell spread by Burkholderia and thereby restrict bacterial numbers over the infection time course. They also do not show whether GBP1 and GVIN1 target the same bacterial cell or different populations of bacteria.
A discussion of the likely impact of the work on the field, and the utility of the methods and data to the community.
This work uncovers the antimicrobial actions of a member of yet another family of IFNg-induced GTPases, which potentially acts against other intracellular pathogens. GVIN1 appears to operate independently and in parallel to GBP1, pointing to the breadth and complexity of the IFNg-inducible GTPase families.
Comments on revised version.
They have addressed my comments.
Reviewer #3 (Public review):
Summary:
Here, Guo et al. (2025) propose that the IFN-induced GTPase GVIN1 forms a coat on cytosolic Burkholderia thailandensis, blocking actin tail formation through a mechanism analogous to GBP1-mediated restriction of Shigella motility.
Their study was prompted by the intriguing observation that IFNγ priming and GBP1 coat formation fail to inhibit B. thailandensis actin-based motility in HeLa cells, yet IFNγ restricts the motility of Burkholderia in T24 cells. Further investigation revealed that IFNγ restricts B. thailandensis motility in T24 cells via both GBP1-dependent and -independent mechanisms, suggesting that HeLa cells lack a critical GBP1 co-factor required to inhibit actin tail formation.
To identify the GBP1-independent mechanism, the authors performed an siRNA screen of interferon-stimulated genes (ISGs) and identified GVIN1, a large IFN-induced GTPase, as essential for restricting B. thailandensis motility. To identify the GBP1-independent mechanism, perform a knock-down screen for ISGs and find that the loss of GVIN, a very large IFN-induced GTPase, results in higher actin tail positive B. thailandensis in T24 cells. They further demonstrate that GVIN forms coats on the surface of B. thailandensis which prevent the polar localization of BimA and thus actin tail formation. In summary, the data reveal two independent IFNγ-induced pathways that restrict bacterial motility: one GBP1-dependent and the other GVIN1-dependent, each relying on distinct host co-factors.
Global assessment:
This is a well-executed study that convincingly demonstrates how GVIN1 restricts actin-based motility of B. thailandensis through the assembly of coatomers. The results are clearly described, the manuscript is easy to follow, and the data are overall compelling and well presented. I have only a few suggestions on how the manuscript could be further improved.
Comments on revised version.
The authors have answered all my points.
Author response:
The following is the authors’ response to the original reviews.
Public Reviews:
Reviewer #1 (Public review):
This is a very elegant and convincing study. Using systematic screening of actin tail formation in two bacterial strains and employing a panel of CRISPR-CAS ko cell lines, the authors identify a novel dynamin-related GTPase GVIN, which forms an oligomeric coat around an intracellular Burkholderia strain. The bacterial O-antigen LPS layer is required for the formation of the GVIN coat, which disturbs the polar localization of the bacterial actin-polymerizing BimA protein.
I am not an expert in infection studies, but the experiments appear to be of high quality, the figures are well prepared, and clean and statistically significant results are provided. I have no criticism of the presented approaches.
The identification of a novel GBP1-independent pathway targeting intracellular bacteria is not only of fundamental importance for the immunity field but also of high interest to researchers in other areas, for example, evolutionary or structural biologists.
We sincerely thank the reviewer for their generous appraisal of our work. We appreciate their encouraging feedback.
Reviewer #2 (Public review):
Summary:
The authors wanted to investigate how cells defend against intracellular pathogens, such as Shigella and Burkholderia species, that co-opt the host actin machinery to spread from cell-to-cell. Previous work has identified IFNg-inducible GTPase of the Guanylate Binding Protein (GBP) family in cytosolic defence against Gram-negative bacteria. By forming a coat around Shigella, human GBP1 suppresses actin-based motility by displacing IcsA, which is the actin-polymerising virulence factor present at bacterial poles. In addition, GBP1 recruits the cytosolic LPS-sensor, caspase-4, to the bacterial surface, which results in the removal of bacterial replicative niches via pyroptotic cell death. Here, they followed up their finding that GBP1 can reduce actin-based motility of Shigella in HeLa cells and, surprisingly, fails to do so during Burkholderia infection. In contrast, in T24 bladder epithelial cells, GBP1 is competent in blocking Burkholderia actin-tails. They therefore wanted to identify the GBP1-independent factor that blocks actin-based motility in IFNγ-treated cells that is absent in HeLa cells.
Major strengths and weaknesses of the methods and results:
The authors report a second IFNγ-dependent pathway involving the protein product of the gene GVIN1, which was previously thought to be a pseudogene. GVIN1 (GTPase, very large interferon inducible 1) is thus the first human member of this family of ~250 kDa putative GTPases to be demonstrated to be functional and have potential antimicrobial roles. The discovery that GVIN1 is indeed functional, forms coats on Burkholderia in an LPS O-antigen-dependent manner, and limits actin-dependent motility are the main strengths of this paper. The authors use CRISPR/Cas9-based knockouts in HeLa and T24 cells, and complement them to demonstrate that GBP1 and GVIN1 are both required to inhibit actin-based motility.
An appraisal of whether the authors achieved their aims and whether the results support their conclusions:
The authors achieved their main goals through well-planned experiments and unbiased screens. They succeeded in finding the factor that blocks actin-based motility independently of GBP1. This is driven by GVIN1, which coats bacteria and limits actin-tail formation by reducing the expression of BimA through currently unknown mechanisms. Further, they found that an O-antigen mutant can escape coating by GVIN1, indicating the requirement for these polysaccharides in GVIN1-dependent bacterial sensing. However, the authors have not investigated whether GVIN1, which has two GTPase domains, does indeed have GTPase activity and whether GVIN1 and GBP1 together completely block cell-to-cell spread by Burkholderia and thereby restrict bacterial numbers over the infection time course. They also do not show whether GBP1 and GVIN1 target the same bacterial cell or different populations of bacteria.
A discussion of the likely impact of the work on the field, and the utility of the methods and data to the community:
This work uncovers the antimicrobial actions of a member of yet another family of IFNg-induced GTPases, which potentially acts against other intracellular pathogens. GVIN1 appears to operate independently and in parallel to GBP1, pointing to the breadth and complexity of the IFNγ-inducible GTPase families.
We thank the reviewer for their overall positive assessment of our work and their suggestions for additional studies. We addressed all the comments and in response we have added additional experimental data to the revised manuscript. However, while we agree that a detailed biochemical characterization of GVIN1 protein is warranted, we consider such experiments outside of the scope of the current work.
Reviewer #3 (Public review):
Summary:
Here, Guo et al. (2025) propose that the IFN-induced GTPase GVIN1 forms a coat on cytosolic Burkholderia thailandensis, blocking actin tail formation through a mechanism analogous to GBP1-mediated restriction of Shigella motility.
Their study was prompted by the intriguing observation that IFNγ priming and GBP1 coat formation fail to inhibit B. thailandensis actin-based motility in HeLa cells, yet IFNγ restricts the motility of Burkholderia in T24 cells. Further investigation revealed that IFNγ restricts B. thailandensis motility in T24 cells via both GBP1-dependent and -independent mechanisms, suggesting that HeLa cells lack a critical GBP1 co-factor required to inhibit actin tail formation.
To identify the GBP1-independent mechanism, the authors performed an siRNA screen of interferon-stimulated genes (ISGs) and identified GVIN1, a large IFN-induced GTPase, as essential for restricting B. thailandensis motility. To identify the GBP1-independent mechanism, perform a knockdown screen for ISGs and find that the loss of GVIN, a very large IFN-induced GTPase, results in higher actin tail-positive B. thailandensis in T24 cells. They further demonstrate that GVIN forms coats on the surface of B. thailandensis, which prevent the polar localization of BimA and thus actin tail formation. In summary, the data reveal two independent IFNγ-induced pathways that restrict bacterial motility: one GBP1-dependent and the other GVIN1-dependent, each relying on distinct host co-factors.
Global assessment:
This is a well-executed study that convincingly demonstrates how GVIN1 restricts the actin-based motility of B. thailandensis through the assembly of coatomers. The results are clearly described, the manuscript is easy to follow, and the data are overall compelling and well presented. I have only a few suggestions on how the manuscript could be further improved.
We thank the reviewer for a clear summary of our findings, and we appreciate the reviewer’s thoughtful suggestions on our manuscript, which we addressed in our revised manuscript.
Recommendations for the authors:
Reviewer #1 (Recommendations for the authors):
GVIN/VLIG was previously described in the mouse, mostly. Why was it considered a pseudogene in humans? Some more background should be provided.
Major genome annotation projects (Ensembl, GENCODE, and RefSeq) rely on automated pipelines that flag loci as pseudogenes when they contain features incompatible with a single canonical open reading frame. These pipelines score in-frame premature stop codons, frameshifting indels, or the absence of a full-length matching transcript or protein in reference databases. The human GVINP1 locus displays several such features when compared against a naive single-ORF model of the mouse Vlig1 mRNA, and on that basis GVINP1 was likely categorized as a pseudogene by all three consortia, an assignment that has propagated through UniProt, HGNC, GeneCards, and OMIM ever since (UniProt Q7Z2Y8; GeneCards GVINP1). Importantly, these pipelines are not designed to detect loci that are expressed from more than one ORF or that produce a full-length protein through non-canonical translation, and they do not incorporate mass-spectrometry peptide evidence, ribosome profiling, or interferon-inducible expression data. Therefore, these automated pipelines can draw incorrect conclusions, as is the case here. We have added to the Introduction to provide a bit more background, as requested.
Is GVIN restricting the proliferation of the Burkholderia strain in the tested human cell line(s)? Is it restricting the intra-host mobility of the bacteria?
GVIN1 limits intracellular bacterial replication, albeit only modestly (see new data in Fig. S5B). We found that although GBP1 and GVIN1 block actin tail formation through independent, non-overlapping mechanisms, control of intracellular B. thailandensis burden required both proteins: IFN-γ failed to restrict bacterial replication in either GBP1KO or GVIN1KO T24 cells, and the loss-of-restriction phenotypes were quantitatively similar between the two knockouts (Fig. S5B). Together, these data identify GBP1 and GVIN1 as the effectors of two cooperating IFN-γ–induced defense pathways that jointly control B. thailandensis.
P 10: Remarkably, GVIN1 coatomers formed only infrequently on the surface of the O-antigen-deficient delta wbpm B. thailandensis mutant (Figure 5F). This information is not included in Figure 5F (only the quantification in E) and could be added.
We have added representative images of the corresponding microscopy data for both WT and DwbpM in a new Figure S8B.
Reviewer #2 (Recommendations for the authors):
Major points:
(1) The finding of GVIN1 as a protein is important and deserves further characterisation. For instance, it would be useful to investigate the transcripts they find in RNAseq, start sites, length of transcript, etc. There are two GTPase domains in this protein, but is either of them active as a GTPase if key GTP-binding and/or hydrolysis residues are mutated?
We agree that biochemical and enzymatic characterization of the GVIN1 protein is an important future direction. Similarly, careful investigations of GVIN1 expression, ribosome profiling, etc. ought to be conducted. However, these represent substantial, self-contained lines of investigation, and we will address them in a dedicated follow-up studies.
(2) Throughout the paper, the authors have only investigated the proximal steps of actin-tail formation. They do not go on to measure the impact of bacterial targeting by GBP1 or GVIN1, specifically in terms of bacterial spread cell-to-cell and replication (e.g., CFU measurements over time) and the impact on pyroptosis. Does GBP1 and/or GVIN1 targeting reduce bacterial CFUs over time?
re: 1) bacterial burden (CFU data); 2) host cell death (pyroptosis), and 3) cell-to-cell spread: 1) Our revised manuscript now provides CFU data (Fig. S5B) that identifies a modest impact on bacterial burden by GVIN1 expression – please, see also our response to Reviewer #1. 2) We already provided data in our original submission (Fig. S2D) demonstrating that IFNγ priming of either T24 WT or GBP1<sup>KO</sup> cells did not appear to change the degree of host cell death in infected cells, as measured by Lactate Dehydrogenase (LDH) release assays. Accordingly, no further studies into host cell death pathways seem warranted at this time. 3) B. thailandensis is unusual among cytosolic bacterial pathogens in that it can spread between host cells not only through actin tail–driven propulsion but also via flagellar motility (PMID: 21730143). Therefore actin-based motility is not essential for cell-to-cell spread for this specific organism (PMID: 21730143). Consistent with this dual-mode dissemination, IFN-γ priming of WT T24 cells had no measurable impact on plaque-based cell-to-cell spread (new data in Fig. S5A), despite sharply reduced actin tail formation inside IFN-γ primed WT T24.
(3) The enumeration of microscopy could be improved by providing the % of GBP1+ve or GVIN1+ve bacteria that are actin-ve.
We added those data presentations as new panels Fig. S2G for GBP1 and Fig. S7B. As expected, these data show that under IFN-γ primed conditions GBP1-coated or GVIN1-coated bacteria fail to form actin tails.
(4) The authors show a schematic of GBP1 and GVIN1 targeting the same bacterial cell, but they have not demonstrated this experimentally. For many of their claims, it is important to show GVIN1 and GBP1 co-staining during infection with WT and the wbpM mutant Burkholderia.
The schematic depicted GBP1 and GVIN1 on opposite sides of the same bacterium purely for illustrative clarity and was not intended to convey that GBP1 and GVIN1 necessarily coat the same bacterial cell. We agree, however, that whether GBP1 and GVIN1 coatomers can co-decorate a single bacterium is a mechanistically interesting question, and we have now addressed it experimentally. Dual immunostaining in IFNγ-primed cells showed that most targeted bacteria were encased by either a GBP1 or a GVIN1 coatomer, but not both (Fig. S7C). A small subpopulation of GBP1/GVIN1 double-positive bacteria was nonetheless detectable (Fig. S7C–D), indicating that the two coat types are largely, but not strictly, mutually exclusive. The rarity of dual-coated bacteria reinforces our conclusion that the two coatomers operate as predominantly independent effector systems, consistent with their ability to restrict actin tail formation independently of one another.
(5) The authors make points about GBP1 and GVIN1 as being independent pathways and the need for a cofactor. To be certain about these conclusions, have they co-expressed both GBP1 and GVIN1 together in double knockout HeLa & T24 and assessed outcomes? They started out to look for a GBP1 cofactor that is required in HeLa cells, but they found GVIN1, which seems to operate independently of GBP1 (which is not inherently an issue, but I feel this needs to be clarified better). In T24 cells, expression of (mCh-)GBP1 alone without IFNg priming does reduce Actin+ve bacteria (Figure 2C, green bars), so GBP1 can use a factor that is expressed in naïve cells? But the authors have not shown that this factor is GVIN1. The same is true for GVIN1 expressed without IFNγ in T24-GVIN1-/- (Fig 5C). What happens in a similar experiment with an O-antigen mutant?
We believe the individual experiments the reviewer proposes have, in fact, already been performed in some fashion and are presented in the manuscript. Because these experiments are distributed across several figures, we have taken this opportunity to summarize the logic here and to point the reviewer to the relevant panels.
First, we would like to clarify the origin of GVIN1 in our study, as the reviewer's framing suggests that we set out to identify a GBP1 co-factor and instead landed on GVIN1. Our siRNA screen (Fig. 4A–B) was in fact designed to identify ISGs required for the GBP1-independent anti-actin-tail pathway and was therefore conducted in a GBP1<sup>KO</sup> T24 background. GVIN1 emerged as a factor whose depletion restored actin tail formation in the absence of GBP1, and it is therefore, by design of the screen, an effector of a GBP1-independent pathway rather than a candidate GBP1 co-factor.
Second, we show that single deletion of either GBP1 or GVIN1 produces a partial and quantitatively similar de-restriction phenotype, whereas only the DKO fully abolishes IFNγ-mediated restriction of actin tail formation (Fig. 4C). Thus, we are seeing the classical additive signature of two non-redundant pathways acting on the same phenotype. In the gain-of-function direction, ectopic re-expression of GVIN1 in the DKO background restores IFNγ-dependent restriction to a level indistinguishable from the GBP1<sup>KO</sup> single knockout (Fig. 5D). A reciprocal experiment co-expressing both GBP1 and GVIN1 in the DKO would only test for additivity of two effects we have already shown to occur independently and would not provide an orthogonal test of independence beyond what Fig. 4C and Fig. 5D already establish.
Third, we do not think GVIN1 is a likely candidate for the basal GBP1 co-factor that the reviewer infers from Fig. 2C. GVIN1 is IFNγ-inducible and effectively absent from unprimed T24 cells (the putative co-factor, while IFNγ-inducible, would be expected to be expressed at appreciable levels in naïve cells). Indeed, GVIN1 entered our screen precisely because of that inducibility (Fig. 4A and RNA-seq data in Fig. S3A–B). More decisively, we show that IFNγ-primed HeLa cells allow for ectopically expressed GVIN1 but not for ectopically expressed GBP1 to restrict actin tail formation. In other words, the ISG co-factor for the GVIN1 pathway must be distinct from the ISG co-factor for the GBP1 pathway. Lastly, we now show that GBP1 and GVIN1 coatomers are largely mutually exclusive, further arguing against GVIN1 acting as the GBP1 co-factor or vice versa in blocking actin tail formation.
Fourth, the O-antigen mutant experiment the reviewer asked about has also been performed and provides another point of mechanistic separation between the two pathways. Ectopically expressed GVIN1 assembles a coatomer on WT B. thailandensis but only rarely on the ΔwbpM O-antigen-deficient mutant (Fig. 5F). This observation parallels (and thus mechanistically explains) why ΔwbpM is resistant specifically to the GBP1-independent (i.e., GVIN1-dependent) branch of IFNγ defense (Fig. 3E). GBP1, by contrast, engages both B. thailandensis WT and the O-antigen ΔwbpM mutant efficiently (new Fig. S8C).
Taken together, four convergent lines of evidence (additive genetics in the DKO (Fig. 4C), cross-complementation of GVIN1 in the GBP1-null background (Fig. 5D), non-overlapping co-factor requirements in HeLa versus T24 cells (Fig. 2, Fig. 5), and distinct O-antigen dependencies for the two coatomers (Fig. 5E, Fig. 8C)) establish that GBP1 and GVIN1 constitute two parallel, independently operating antimicrobial modules, each likely with its own set of IFN-inducible co-factors.
(6) The discussion does not currently discuss GBP1/Caspase-4-driven pyroptosis as a host-defence mechanism, which should be included. The authors should also discuss why Shigella and Burkholderia are differentially targeted, given the differences in the mechanism of actin polymerisation by IcsA and BimA.
We previously showed that a GBP1 polybasic motif mutant that is unable to form coatomers on bacteria remains able to activate Caspase-4-driven pyroptosis and thus demonstrated that coatomers are not strictly required for GBP1-dependent Caspase-4 activation (PMID: 37023136). We now provide this information in the Discussion of the revised manuscript.
We provide new data showing that GVIN1 forms coatomers on BimA-deficient and WT B. thailandensis with similar efficiency (new Fig. S8A) and we previously demonstrated that icsA plays no role in GBP1 coatomer formation (PMID: 33885766). Therefore, actin polymerization appears to play no role in either GVIN1 or GBP1 coatomer formation.
Minor points:
(1) Have the authors used anti-actin antibodies or phalloidin staining? Please clarify.
We used phalloidin staining and this information is now also provided in the Results section (lines 118-119) in addition to the Materials and Methods section of the paper.
(2) Provide the Gene ID, Ensembl ID, etc., for GVIN1 at first mention.
We inserted the text: “Because automated annotation pipelines flag predicted in-frame stop codons, GVIN1 has long been catalogued as the pseudogene GVINP1 (GTPase, Very large IFN-Inducible Pseudogene-1) by Ensembl, GENCODE, and RefSeq (HGNC 25813; NCBI Gene 387751; OMIM 616121).” in the Introduction (lines 104 – 107)
(3) The bond linking the core sugars with the O-antigen is missing in the schematic of LPS in Figure 3A. Please mention which sugars these are, if it is known for Burkholderia.
We have corrected the schematic (Fig.3A) to include the bond linking the core sugars with the O-antigen.
(4) RNA-seq datasets should be deposited to GEO or similar public databases, and project IDs provided.
The RNA-seq datasets was deposited to the NCBI Gene Expression Omnibus (GEO), and the project accession IDs are is cited in the Data Availability statement.
(5) The description of the transcriptomics analyses was not clear - are the heatmaps in Figure S3 of genes that were induced by IFNg in these cell lines, but not HeLa or A594, but the description states fold-change over naïve cells?
The reviewer is correct and we apologize for the error in the figure legends, which we edited to clarify that the heatmap compares gene expression across IFNγ-primed cell lines.
Reviewer #3 (Recommendations for the authors):
Major comments:
(1) It would be nice to include some additional analysis on GVIN: GBP1 and GVIN1 seem to recognize different structures and need different cofactors. Are they structurally similar (alignment)? Can you map functionally important GBP1 regions on GVIN1, such as a polybasic motif, similar to motifs found in GBP1? Or does GVIN have a lipidic membrane anchor or a CAAX box? In other words, is GVIN binding directly to the outer membrane or through the unknown cofactor? Is GVIN an effector of another protein that binds to the OM?
Detailed structural studies including the mapping and biochemical characterization of GVIN1's membrane-targeting domain(s) represent substantial, self-contained lines of investigation that are beyond the scope of this study and that we hope to present in a future follow-up study.
(2) In T24 cells, GVIN1 and GBP1 (and their cofactors) get induced upon IFNg priming. Upon B. thailandensis infection, is it possible to see mixed or double coats? Does GBP1 have a higher affinity for Shigella and GVIN for B. thailandensis? Is GVIN1 B. thailandensis-specific? It would be a nice addition to discuss other virulent Burkholderia species that make actin tails (such as B. pseudomallei) to further enhance the possible relevance of GVIN1.
Please see our response to reviewer #2: dual immunostaining in IFNγ-primed cells showed that most targeted bacteria were encased by either a GBP1 or a GVIN1 coatomer, but not both (Fig. S7C). A small subpopulation of GBP1/GVIN1 double-positive bacteria was nonetheless detectable (Fig. S7C–D), indicating that the two coat types are largely, but not strictly, mutually exclusive. Re: the measurement of GBP1/ GVIN1 affinities for different bacteria – we agree that this is an area of future exploration but not within the scope of this study. Extending research on GVIN1 to other bacterial organisms including other Burkholderia species is indeed among our future research goals. We more broadly speculate on the potential importance of GVIN1-mediated host defense targeting other Burkholderia species in the Discussion (see lines 363 ff)
(3) In panel F of Figure 6, It appears that GVIN forms full coats in the absence of IFNγ priming, but relocalizes to the bacterial poles upon IFNγ treatment. Is this shift in localization a representative and consistent observation, or could it be an artifact of the specific image selected? Additionally, could this relocalization be driven by the presence of IFNγ-induced cofactors? For example, might GVIN interact directly with BimA, or could an IFNγ-dependent cofactor mediate its recruitment to the poles?
We believe the author is commenting on Figure 5F instead of 6 based on the question. The perceived shift in GVIN1 localization to the bacterial pole under IFNγ-primed conditions was not something we routinely observe and the original image shown in Fig. 5F is likely an outlier. We therefore replaced the original image with a more representative image.
(4) The authors conclude that BimA is eliminated from the outer membrane, but it is unclear what evidence this conclusion is based on. Could it be that BimA is not eliminated, but rather becomes diffusely distributed in the membrane in the presence of GVIN1, and is therefore more difficult to detect by microscopy? One way to test this hypothesis would be to delete BimC. If BimA localization is indeed dependent on BimC, its signal should become diffuse. If a BimA signal can be detected in BimC-deficient bacteria, but not in GVIN-positive bacteria, it would indeed strengthen the claim that BimA is removed from the membrane.
Prior literature demonstrates that on a bimC mutant, BimA remains detectable on the bacterial surface in a dispersed manner and that this dispersed BimA localization correlates with an "actin cloud" surrounding the bacteria. In our GVIN1-coated bacteria, we do not detect an actin cloud or residually localized BimA signals. We have now mentioned this in the Results section line 305 - 308 referencing this published phenotype to strengthen our argument that BimA is likely cleared from the membrane rather than merely redistributed.
(5) The siRNA mini-screen would benefit from appropriate controls demonstrating that the targeted genes were effectively knocked down. Without this, it's difficult to interpret negative results with confidence. It remains possible that the co-factor of GVIN1 is among the ISGs tested, but that the knockdown was insufficient to produce a detectable phenotype.
We now acknowledge this limitation. We have updated the text on lines 221-223 to explicitly state that our mini-screen does not rule out potential false negatives due to varying knockdown efficiencies.
(6) The authors convincingly demonstrate that GVIN1 coats prevent actin tail formation by B. thailandensis, but they do not explore the broader biological consequences of this observation. An important next step would be to assess whether GVIN1 coating restricts B. thailandensis cell-to-cell spread. By 8 hours post-infection, a clear difference in spread should be observable if actin-based motility is impaired. Additionally, it would be valuable to determine whether GVIN1-coated bacteria are still capable of intracellular replication. These functional assays would provide important insight into the biological relevance of GVIN1-mediated coating.
We have now assessed the effect of GVIN1 on intracellular bacterial replication (see new CFU data in Fig. S5B) and cell-cell motility (see new data in Fig. S5A). Please, also refer to our comments in response to Reviewer 1 for more details.
(7) As far as I am aware, the exact function of the GBP1 coat remains unclear. The authors refer to their earlier work (Kutsch et al., 2020) and claim to have demonstrated an "emulsifying activity." However, this overinterpretes the findings of that publication, since Kutsch et al. (2020) showed that the GBP1 coat disrupts the barrier function of the O-antigen, rather than compromising the integrity of the outer membrane itself. This disruption increases bacterial susceptibility to polymyxin B, but does not show an 'emulsifying effect'. Furthermore, the same group later reported that GBP1 exhibits an "aggregating" activity on LPS micelles (Dickinson et al., 2023), which seems to contradict the notion of an emulsifying function. I would recommend tempering the current claims about GBP1's activity. The precise impact of the GBP1 coat on the bacterial outer membrane remains unresolved and should be presented as an open question.
Our previous work demonstrated GBP1 coatomers on the bacterial surface disrupt the membrane-stiffening and membrane-shielding activities of O-antigen. We stand by the interpretation of our results and think it is likely that GBP1 coatomers act like surface-active agents or emulsifiers. However, we agree with the reviewer that more biophysical studies are needed to definitively test this model. Accordingly, we have revised some of the language and clearly articulated that GBP1 coatomers acting as a surfactant is a model that we are proposing.
As the reviewer correctly stated we have also shown that GBP1 can aggregate “free” LPS, i.e. to act more like a flocculant. Substances that are both flocculants and surfactants exist (e.g. long-chain fatty acid soaps). However, these substances typically do not act as flocculants and surfactants simultaneously under the exact same conditions. Similarly, we propose that GBP1’s role depends on context: we propose that GBP1 coatomers on the bacterial surface act as surfactants, while polymerizing GBPs bound to LPS acts a flocculant.
Minor comments:
(1) Figure 3: Can a GBP1 coat assemble on the delta wbpM mutant of B. thailandensis? The authors suggest GBP binds to the O-antigen, but other publications suggest that the O-antigen is dispensable in certain bacteria.
We conducted these experiments – see also our response to reviewer #2. Briefly, ectopically expressed GVIN1 assembles a coatomer on WT B. thailandensis but only rarely on the ΔwbpM O-antigen-deficient mutant (Fig. 5F). GBP1, by contrast, engages both B. thailandensis WT and the O-antigen ΔwbpM mutant efficiently (new Fig. S8C). We do not argue that GBP1 requires O-antigen binding for coatomer formation on Burkholderia. However, direct or indirect interactions between GVIN1 and O-antigen may be required for GVIN1 coatomer formation.
(2) Figure 4: GBP6 appears as a hit in the T24 screening but turned out to be a false positive. How does that look in the other cell lines that were part of the screen (U937, THP1)?
the characterization of T24 GBP6<sup>KO</sup> cells clearly demonstrated that GBP6 is dispensable for either the GBP1 or the GVIN1 pathway in these cells. While we cannot formally rule out that GBP6 would be required for the GBP1/GVIN1 pathways in other cell lines, there are no compelling arguments in favor of this model – as far as we can tell - and we therefore did not investigate this further.
(3) Figure 5F: Please show the blots confirming exogenous GVIN expression.
We have provided an immunoblot showing the expression of exogenous GVIN1in HEK293T cells in new Fig. S7A
(4) Figure 6: Can you induce GVIN expression after actin tails have been formed and stop disassembling them?
This is an interesting idea and will require a technically challenging video microscopy approach that we will consider for future investigations.
(5) Figure 6: Are the bacteria showing no BimA signal GVIN1-coated? Eventually, stain for LPS to see if bacteria are lacking BimA also lack LPS.
Our preliminary studies indeed support this and thus indicate that the elimination of BimA from the bacterial surface requires the concerted effort of the GVIN1 coatomer and its putative co-factor (or the GBP1 coatomer and its putative co-factor, respectively). A detailed characterization of the mechanism by which BimA is removed from the bacterial surface will be part of future studies.
eLife Assessment
This study represents an important contribution to the study of decision-making under risk, bringing an interdisciplinary approach spanning economic theory, behavioral neuroscience, and computational modeling to test how choice preference is influenced by rare and extreme events. Using a novel experimental design, the authors present evidence that rats are indeed sensitive to these rare and extreme events despite their infrequent occurrence, driven primarily by an almost complete avoidance of "Black Swans" - rare and extreme losses. Motivated by economic theory, computational models of the choice behavior provide solid evidence of a distinct distributional effect, in which rare outcomes are treated as a differently-weighed contributor to the overall expectation associated with each option. Noting the inevitable difficulty in empirically testing the impact of rare events in animal tasks, these results bring a novel interdisciplinary perspective to a question of broad interest for understanding decision-making in complex probabilistic environments through the lens of economic choice.
Reviewer #1 (Public Review):
[Editor's note: this version has been assessed by the Reviewing Editor without further input from the original reviewers, given the substantial revisions completed since the original submission: these have improved the clarity and transparency of the results in a relatively complex study. These changes largely address those concerns raised in review that do not relate to the underlying experimental design and the control of the distinct task conditions.]
Summary:
In this manuscript, the authors investigate the impact of rare and extreme events on rodents' decision-making under risk, in gain and loss contexts. They describe the behavior of 20 rats performing a four-armed bandit task, where probabilistic gains (sugar pellets) and losses (time-out punishments) can - in some arms - incorporate extremely large - but rare - outcomes. They report that most rats are sensitive to rare and extreme outcomes despite their infrequent occurrence, and that this sensitivity is primarily driven by extreme loss events which they try to avoid, rather than extreme gains that they seek to obtain.
They finally propose a modification of standard reinforcement-learning, which features a specific sensitivity to rare and extreme outcomes and can account for the observed behavior.
Strengths:
The manuscript really taps into a surprisingly neglected but very relevant aspect of decision-making: the effect of rare and extreme events (REE). The authors have developed an experimental setup that seemingly allows investigation of this aspect, which is not trivial given the idiosyncratic properties of rare and extreme events.
The parameters of the experimental setup seem also to be well thought off: basically, in the absence of REE, some options are objectively better than others (because, in expectation, they overall deliver more food, or minimize time-out punishments), but this ordering reverses if REE are taken into account. This allows for a clean test of the integration of REE in the rodent's decision-making model.
The data is presented and analyzed in a very descriptive but exhaustive and transparent way, down to the description of individual rodent's behavior.
Reviewer #2 (Public review):
Summary:
This paper attempts to examine how rare, extreme events impact decision-making in rats. The paper used an extensive behavioural study with rats to evaluate how the probability and magnitude of outcomes impact preference. The paper, however, provides limited evidence for the conclusions because the design did not allow for the isolation of the rare, extreme events in choice. There are many confounding factors, including the outcome variance and presence of less-rare, and less-extreme outcome in the same conditions.
Strengths.
(1) The major strength of the paper is the significant volume of behavioural data with a reasonable sample size of 20 rats.
(2) The paper attempts to examine losses with rats (a notoriously tricky problem with non-human animals) by substituting time-outs as a proxy for losses. This allows for mixed gambles that have both gain and loss possible outcomes.
(3) The paper integrates both a behavioural and a modelling approach to get at the factors that drive decision-making.
(4) The paper takes seriously the question of what it means for an event to be rare, pushing to less frequent outcomes than usually used with non-human animals.
Author response:
The following is the authors’ response to the previous reviews.
Public Reviews:
Reviewer #2 (Public review):
Summary:
This paper attempts to examine how rare, extreme events impact decision-making in rats. The paper used an extensive behavioural study with rats to evaluate how the probability and magnitude of outcomes impact preference. The paper, however, provides limited evidence for the conclusions because the design did not allow for the isolation of the rare, extreme events in choice. There are many confounding factors, including the outcome variance and presence of less-rare, and less-extreme outcome in the same conditions.
Strengths.
(1) The major strength of the paper is the significant volume of behavioural data with a reasonable sample size of 20 rats.
(2) The paper attempts to examine losses with rats (a notoriously tricky problem with non-human animals) by substituting time-outs as a proxy for losses. This allows for mixed gambles that have both gain and loss possible outcomes.
(3) The paper integrates both a behavioural and a modelling approach to get at the factors that drive decision-making.
(4) The paper takes seriously the question of what it means for an event to be rare, pushing to less frequent outcomes than usually used with non-human animals.
Weaknesses:
(1) The primary issue with this work is that the primary experimental manipulation fails to isolate the rare, extreme events in choice. As I understand the task, in all the conditions with a rare extreme event (e.g., 80 pellets with probability epsilon), there is also a less-rare, less-extreme event (e.g., 12 pellets with probability 5). In addition, the variance differs between the two conditions. So, any impact attributable to the rare, extreme event could be due to the less rare event or due difference in the variance (or other statistical moments, like skew or kurtosis). That the distributions can be shown to be different under specific assumption to value maximizing agents (e.g., with Jensen Gaps and Table 2) is not really relevant to what rats are sensitive and what drive their behaviour. The design here does not support the conclusions. Finally, by deliberately confounding rarity and extremity, the design does not allow for assessing the impact of either aspect on rat behaviour.
This is a repetition almost word for word of point (1) appearing in your previous report, which we have addressed at length in our first-round reply. Therefore, we do not reiterate here to save space.
(2) The RL modelling work also fails to show a specific impact of the rare extreme event. As best as I can understand Eq 2, the model provides a free parameter that adds a bonus to the value of either the two options with high-variance gains (A and V in the paper) or to the two options with high-variance losses (F and V in the paper). Or equivalently to the ones with "Jackpots" vs the ones with "Black Swans" (see Point 1 above as to how these different aspects are all confounded in this design). This parameter seems to only depends on whether this option could have possibly yielded the rare, extreme outcome (i.e., based on the generative probability) and was not connected to its actual appearance. [This point is unclear as the text says this, but the rebuttal states otherwise; plus some options never received the REE, see Table S11]. That makes it a free parameter that just bumps up (or down) the probability of selecting a pair of options. That may be due to presence of the REE or the other rare event or just the variance difference. Moreover, in the case of the "black swan" or high-variance loss conditions, this seems very much like a loss aversion parameter, but an additive one instead of a multiplicative one. Is there a theoretical claim here that "extreme losses" need an additive loss-aversion parameter?
This is a repetition almost word for word of point (2) appearing in your previous report, which we have addressed at length in our first-round reply. Therefore, we do not reiterate here to save space.
(3) The paper presented the methods and results with lots of neologisms and fairly obscure jargon (e.g., fragility, total REE sensitivity). That might it very hard to decipher exactly what was done and what was found. For example, on p. 4, the use of concave and convex was very hard to decipher; the text even has to repeat itself 3 times (i.e., "to repeat" and "in other words") and is still not clear. It would be much clearer (and probably accurate) to say that the options varied along the variance dimension, separately for gains and losses. Option A was low-variance gains and losses. Option B was low-variance losses and high-variance gains. Option C was high-variance losses and low-variance gains, and Option D was high-variance losses gains. That tells much more clearly what the animals experienced without the reader having to master a set of new terminologies around fragility and robustness, which brings a set of theoretical assumption unnecessarily into the description of the experimental design. Alternatively, if the authors are wary of using the term "variance" because other moments of the distribution also differ, they could use "high-value gains" or "high-value losses" or something else which does not obscure the experimental design with jargon. Again, this goes back to point 1 above, whereby the different options differ on so many dimensions (as is made even more apparent in the rebuttal) that the design cannot isolate the impact of the variables of interest.
This is a repetition almost word for word of point (3) appearing in your previous report, which we have addressed at length in our first-round reply. Therefore, we do not reiterate here to save space. The new words that we use are an attempt at reflecting the novelty of the approach and design.
(4) Were the probabilities shuffled or truly random (seem to be fixed sequences, so neither)? What were the experienced probabilities? Given the fixed sequences, these experienced ("ex-post") probabilities, could differ tremendously from the scheduled ("ex ante") probabilities. It's quite possible than an animal never experienced the rare, extreme event for a specific option. From Table S11, that is guaranteed to have happened in that 4 animals only ever experienced the "black swan" outcome once. It's even possible (if they only picked a specific option on the 10th/60th choices by chance), that they only ever experienced that rare extreme event. This point still cannot be known given the information provided, which does not break down outcomes by options. The Supplemental in Table S11 only gives overall numbers but does not indicate what the rats experienced for each choice/option-which is what matters here. A simple table that indicates for each of the 4 options, how often they were selected, and how often the animals experienced each of the 6-8 possible outcome would make it much clearer how closely the experience matched the planned outcomes. In addition, by restricting the rare outcome to either the 10th or 60th activations in a session, these are not random. Did the animals learn this association? The text states that they did not, but no evidence is provided.
This is a repetition almost word for word of point (4) appearing in your previous report, which we have addressed at length in our first-round reply. Therefore, we do not reiterate here to save space.
(5) The choice data are generally presented in an over-processed fashion with a sum and a difference (in both figures and tables). The basic datum (probability/frequency of selecting each of the 4 options) is not provided directly in the main text, even if it can theoretically be inferred from the sum and the difference. New right side of Table S4 is probably the most valuable piece in terms of explaining what rats did and should be highlighted a lot more. Inspection of that table reveals some interesting (and potentially worrying) results. Most notably, the vast majority of responding happens on the "anti-fragile" and "robust" option, often totalling around 90% of all selections, especially amongst the most common blue rats. Alas, those were all those the two options that were deliberately assigned to the two most preferred holes in the training phase (see p. 26). Does this reflect genuine preference for reward distributions or does this reflect a spatial hole bias? The assignment strategy makes this impossible to tell apart.
This is a repetition almost word for word of point (5) appearing in your previous report, which we have addressed at length in our first-round reply. Therefore, we do not reiterate here to save space. In addition, in Figure 8 of Material and Methods, we report the changes in behavior between the training and final sessions. The related discussion stresses that half of the rats changed more than 30% of their choices in the final experimental sessions compared to the training sessions. See also our previous reply to point (8).
(6) There is insufficient detail provided on the inferential statistical tests (e.g., no degrees of freedom or effect sizes), and only limited information on exactly what tests were run and how (bootstrapping, but little detail). Without code or data (only summary information is provided in the supplement), this is difficult to evaluate. In addition, the studies seem not to pre-registered in any way, leaving many research degrees of freedom. Not all studies need to be pre-registered and sometimes discovery of new things requires exploratory work, but preregistration does provide additional safeguards against overemphasizing post-hoc detected patterns-a serious issue in behavioural science. Moreover, this promotes transparency in reporting results and analyses, allowing for a better assessment of the strength of evidence for a claim. For example, here, were any alternative analysis pipelines attempted? Also, there were many sub-groupings of the animals and subsequent comparisons between them which all seemed post-hoc. On what grounds were these divisions made-were other divisions examined as well?
This is a repetition almost word for word of point (6) appearing in your previous report, which we have addressed at length in our first-round reply. Therefore, we do not reiterate here to save space.
(7) On p. 12 (Fig 4), there is an attempt to look at the impact of a rare, extreme event by plotting a measure of preference for the 10 trials before/after the rare, extreme event. In the human literature, the main impact of experiencing a rare, extreme event is what is known as the wavy recency effect (See Plonsky et al. 2015 in Psych Review for example, now cited). What this means is that there tends to there tends to be some immediate negative recency (e.g., avoiding a rare gain) followed by positive recency (e.g., chasing the rare gain). Typically, this refers to the specific option that yielded that outcome. First, as the other analyses do, the current analysis combines choice of the option that yielded the rare outcome with choice of other options, so that cannot directly assess the impact of the rare, extreme event on choice. Also, using a 10-trial window would thus obscure any impact of this rare, extreme event. There is mention of the very next trial, but an analysis that looks at the 10-trial time course trial-by-trial could reveal any impact that might be predicted from the human literature.
This is a repetition almost word for word of point (7) appearing in your previous report, which we have addressed at length in our first-round reply. Therefore, we do not reiterate here to save space.
(8) As I understood the method (p. 31), the assignment of options to physical locations was not random or counterbalanced, but deliberately biased to have one of the options in the preferred location. This would seem to create a bias towards a particular option and a bias away from the other options, which confounds the preference data in subsequent analyses. Table S4 reinforces this concern where the vast majority of response are clustered in the two most preferred options from training.
This is a repetition almost word for word of point (8) appearing in your previous report, which we have addressed at length in our first-round reply. Therefore, we do not reiterate here to save space.
(9) Are delays really losses? This is a big assumption. Magnitude and delay are different aspects of experience, which are not necessarily commensurable and can be manipulated independently. And, for the model, how were these delays transformed into outcomes for the model. Eq 1 skips over that. Is there an assumption of linearity? In addition, I was not wholly clear if the delays meant fewer trials in a session or if the delays merely extended the session and meant longer delays until the next choice period.
This is an exact repetition of point (9) appearing in your previous report, which we have addressed at length in our first-round reply. Therefore, we do not reiterate here to save space.
Other points:
(1) I think the authors still misunderstand the concept of "hot-stove effects". The idea is that the experience of a very bad outcome can lead to avoiding the situation again (i.e., not sampling that option) and can provide the appearance of oversensitivity to that bad outcome. Here, that might be more thought as "black-swan avoidance". Imagine if, to the rat, all options are equal in value, then some initial bad luck in encountering the black swan might make the animal avoid that option, even though with enough experience, then it would have been equal in value.
This point is reminiscent of recommendation (5) appearing in your previous report, which we have addressed at length in our first-round reply. Therefore, we do not reiterate here to save space.
(2) I am still not convinced that the Jensen inequalities add to this paper in terms of understanding the rat behaviour. That may be more suited for a different paper about the statistical and mathematical properties of certain generative distributions, but not here given what rats actually choose and experience.
This point is reminiscent of recommendation (8) appearing in your previous report, which we have addressed at length in our first-round reply. Therefore, we do not reiterate here to save space.
(3) Providing the data open access is very good. The code, however, should be equally available and not just upon request. Code needs to be available for assessment during peer review and for reproducibility checks. There are substantial enough problems with reproducibility in the field that code availability should be a minimum criterion for publication (see Miske et al., 2026 in Nature for the most recent large-scale evaluation of this problem).
This point is reminiscent of recommendation (9) appearing in your previous report, which we have addressed at length in our first-round reply. Therefore, we do not reiterate here to save space.
(4) The paper still somewhat mischaracterizes the literature on rare events, posing it as a series of "exceptions", rather than recognizing that a huge chunk of the literature uses rare events rarer than 10%. Also, there is even existing terminology in that literature for exactly the situation that is being created here-rare treasures (aka jackpots here) and rare disasters (aka Black Swans here).
This is reminiscent of point (10) appearing in your previous report, which we have addressed at length in our first-round reply. Therefore, we do not reiterate here to save space.
(5) Defining the observed behaviour in terms convexity, instead of stating choices more plainly obscures what is done/found. This is especially the case here because convex and concave mean different things when applied to gains/losses in terms of whether or not that option can lead to the REE. The use of the terms obscures rather than clarifies and probably is best left for the discussion (and maybe the intro) when mapping from theoretical distributions to the experiment at hand. In the paper, even the bottom of p.5 seems to incorrectly define "Total Sensitivity" as the combined proportion of selecting convex options in either domain, which does not map how convex is defined in Fig 1B or elsewhere in the text.
This is reminiscent of point (3) appearing in your previous report, which as stated above we have addressed at length in our first-round reply. Therefore, we do not reiterate here to save space.
(6) Fig 1C is baffling. Why are probabilities drawn moving away from the origin? The standard scientific plotting convention is for numbers to grow when moving away from the origin. That would be vastly clearer. Also, the color coding is confusing. Green-red maps onto convex-concave, but that would naturally seem to indicate gains vs losses, not convex vs concave. And why are probabilities growing larger in both directions from the origin? Much more sensible to communicate the procedure would likely be a standard plot of magnitude vs probability.
As explained in the main text of the article, panel (c) in Figure 1 reveals an important and novel feature of the design: gains and losses are either convex or concave in decreasing probabilities, meaning that the gain/loss gets more rare, it becomes larger (if convex) or smaller (if concave) in a nonlinear fashion.
(7) Discussion: I think the main difference between the human situations discussed and this experiment is that humans have not experienced those rare "black swan" outcomes. Rather, they hear about the disasters that are possible and do not incorporate that information, as discussed in the description-experience literature already cited in this paper (though not in that context).
We thank the referee for pointing this out. However, it is not clear to us that the various trauma experienced in real life by humans do not share properties with Black Swans in our experimental design.
Recommendations for the authors:
Reviewer #2 (Recommendations for the authors):
Small points:
- p. 4. Options are not strategies. Picking two out of the 4 options more often, is not picking two strategies. That's picking two options (also on p. 7, but really throughout).
This has been changed in the text.
- The sums don't add up in probability of outcomes. How were they 0.9, 0.1, 0.01? What were they exactly? What exactly is epsilon?
Exact probabilities, which of course add up to one, are given in Material and Methods (page 29).
- The term "one-sided sensitivity" to refer to the selection of options include rare-extreme outcomes is also confusing. There are many other words that might work better here, including "encounter", "choice", "selection", or even "preference".
One-sided Sensitivity if a proposed measure of the choice asymmetry, as explained in the text (pages 5-6, including caption of Figure 1).
- I don't think that Figure 3 is more intuitive. It's still a summed measure, with rats sub-divided into post-hoc groups. Also, aren't the groups divided precisely in part by the selections that are being shown in this plot?
There is no claim that Figure 3 is more intuitive. As explained in Material and Methods (pages 33-34), it provides an alternative way to represent behaviors (BSA/JPS vs TSREE/OSREE).
- Fig 3 has no black or dark and light grey. There is no one asterisk in the figure. The light grey is irrelevant because all bars must sum to 1. No error bars or equivalent on the plot.
Dark and grey bars sum to one but have a different interpretation, as explained in the text (pages 10-11). We have dropped the one asterisk reference.
- p. 14. Technicality, but alpha need not be bound at 1 for a convergent equation. It only needs to be bound at 2 (real maximum value). Above 1 (but below 2) would be a type of overcompensation that would also converge at same Q value.
The assumption that alpha is positive and bounded by one is common in the literature and is important for stochastic convergence. Assuming that alpha is between one and two has undesirable implications, beyond possible non-convergence, in particular that Q values may fluctuate.
- p. 24. What was the generative frequency (not experienced frequency). Not adequately explained. Less than half a percent of all outcomes? Of all times that the relevant choice was made. This is not explained.
The sequences used to generate gains and losse are detailed in the Supplementary Material.
- p. 4. Says 41 sessions, but methods say 40 (e.g., Fig 7 and surrounding text)
This has been corrected in the text.
- My understanding of Tables S5 and S6 is that the best models did not include REE in Q sub-values. That seems to be opposite to the textual conclusion.
As stated in Table 1 (page 15), which summarizes results in Tables S5-S6, the selected models for rats 3, 11, and 12 do not integrate REE while the selected models for rats 4, 16 and 20 integrate REE though Q values and REE decision weights. For most rats, REE are integrated in the selected models through the latter only.
- What are the parameters in Table S7? The full 8-parameter model? But how are these the ones selected by BIC as per caption? Also, there are a number of anomalous values-lots of 1s and 0s, plus a +40 on the value of the losses for Rat 20. Columns are always gain then loss except first column.
We report in Table S7 all estimated parameters for each rat, but only for the model selected by the BIC criterion (see Table S5). Admissible values for alphas include 0 and 1. All estimates have been checked for identification. The first and second columns have been switched to have gain first.
- Table S11 does not subdivide by option selected. So, we cannot tell if the rats actually experienced the outcome distribution associated with some options.
Given the large number of nosepokes generated by each rat, reporting all actual gains and losses would consume too much space. On the other hand, given that each option has only three frequent and rare values, it is expected that each will be experiences by each rat, in view of the choice frequencies reported in Table S2 (page 52).
eLife Assessment
The study provides an important advance towards understanding how spatial and temporal transcriptional programs are integrated to regulate lineage-specific chromatin and enhancer activation. The functional evidence is currently based on limited evidence, but the current data provide a solid correlative and conceptual foundation. Functional experiments directly linking Gsb occupancy to chromatin state and regulation of some lineage-specific targets would further strengthen the causal interpretation of the model. Clarifying the scope of conclusions and explicitly acknowledging the technical limitations of current chromatin assays would provide a more balanced interpretation of the manuscript.
Reviewer #1 (Public review):
The authors did very little to address our concerns outside of written clarifications. While they altered wording to describe Gsb from "pioneering" to "chromatin priming" our stance remains that there is no evidence shown to confirm that Gsb plays a role in chromatin remodelling. (IE while ectopic expression of Gsb results in a different chromatin conformation, there is no evidence that the ectopic expression of Gsb alters an already established chromatin state.) Additionally, the phenotypic experiments show loss of NB5-6 reporter and neuronal markers upon loss of Gsb which confirmed the developmental requirement for Gsb within the endogenous NB5-6 domain, but another striking feature that was described was that loss of Gsb also results in loss of NB3-5 identity. The abrupt change to a new NB lineage made this portion of the manuscript feel a bit disjoined and while this is described as an additional reciprocal phenotype, there are still no sequencing data or other evidence as to how this loss has occurred.
We disagree that pioneer or chromatin-priming activity should be inferred strictly from single timepoint binding profile. We also disagree that any "remodeling" was shown. To sufficiently show remodeling, one must express a factor at/after a time where the chromatin conformation has been established to be already open/closed and then show the expression of that factor induces conformation changes to the opposite conformation. Otherwise, it is hard to infer whether these chromatin changes are due to the presence of the factor blocking the normal conformational changes that should occur.
The authors mention they walked back the terminology from "pioneering" to "chromatin priming" and make it clear that Gsb does not act alone to establish the chromatin landscape. We are not sure that this is proper terminology either.
We mention that the phenotypic data do not align with the sequencing data. The authors added text to clarify the rationale behind each experiment - some being to induce outside of the endogenous domain with clean genetic access while others to obtain information within the endogenous spatial domain.
Reviewer #2 (Public review):
Summary:
This article by Bhattacharya et al. investigates how neural stem cells (NSCs, NBs) in Drosophila integrate spatial and temporal cues to activate neuron-specific terminal selector (TS) genes. Prior to this work, it was understood that NSCs utilize spatial transcription factors (STFs) and temporal transcription factors (TTFs) to determine lineage identity and birth order, but the mechanisms of integration were not fully elucidated. The authors employed chromatin profiling techniques to analyze the binding of STFs and TTFs in two specific neuroblast lineages, NB5-6 and NB7-4. They found that Gsb (an STF) binds both accessible and less-accessible chromatin in NB5-6, while En (another STF) binds only to pre-accessible chromatin in NB7-4. The findings support an "STF code" where the combination of pioneer and non-pioneer spatial factors, along with temporal factors, triggers neuroblast-specific enhancer activation and determines lineage identity.
Comments on revised version.
The authors have made extensive revisions and addressed the concerns raised in my previous review. Overall, I find the revised manuscript substantially improved, both experimentally and conceptually. In particular, the additional locus-level and genome-wide analyses, the motif analysis of En occupancy, the attempted endogenous Gsb depletion, and the revised interpretations and distributed STF-code model have strengthened the manuscript considerably.
(1) Adding Gsb occupancy profiles at specific gene loci such as lbe, ap, and eya, along with chromatin accessibility, Hb binding, and expression data at ems, enhances the link between TaDa results and lineage phenotypes. Although these data do not prove direct regulation of each locus, they strongly support a correlation between Gsb binding, chromatin state, and neural stem cell identity. Therefore, this aspect is mostly addressed, as long as the manuscript does not present these correlations as conclusive evidence of direct regulation.
(2) We appreciate the additional genome-wide analysis distinguishing genomic loci that gain versus lose accessibility following Gsb misexpression, as well as the comparison with two independent Gsb occupancy datasets and the corresponding Hb occupancy analysis. These data strongly support the conclusion that Gsb has chromatin-remodelling capacity and that accessibility loss occurs at loci associated with Gsb occupancy.<br /> We also appreciate the authors attempt to examine endogenous Gsb function using early CRISPR depletion and their explanation of why this approach could not subsequently be coupled to lineage-specific TaDa. The inability to directly assess chromatin consequences of Gsb loss in native NB5-6 remains a limitation, but the authors have now acknowledged this appropriately. We therefore consider this comment substantially addressed.<br /> We would, however, encourage the authors to distinguish clearly between demonstrated chromatin-remodelling capacity and the stronger, physiologically unresolved claim that Gsb establishes chromatin priming in NB5-6.
(3) The response to the question about "En not being a pioneer factor" is particularly satisfactory. The new motif analysis directly tackles the concern that the lack of En occupancy at NB5-6-specific sites might just be due to the absence of En binding motifs. The similar presence of En motifs at NB5-6- and NB7-4-specific sites strengthens the conclusion that differing occupancy is unlikely caused by motif absence. Additionally, comparing the accessibility dependence of En and Gsb occupancy helps argue against the idea that TaDa cannot detect binding at less accessible loci. Importantly, the authors have removed the categorical claim that "En is not a pioneer factor" and now limit their conclusion to the specific developmental and cellular context studied. This response adequately addresses the comment.
(4) The revised distributed STF-code model is clearer and provides a useful framework for explaining the complementary roles of Gsb, En, and other spatial factors. I would encourage the authors to clarify that the proposed stepwise model is a conceptual framework rather than an experimentally established temporal sequence. In particular, the data do not directly demonstrate that Gsb-dependent chromatin priming is causally followed by Hb recruitment at the same loci. Similarly, the division of labour among STF components should not be interpreted as mutually exclusive, since individual STFs could contribute to both chromatin remodelling and TTF-dependent enhancer regulation. The schematic and accompanying text should therefore distinguish the demonstrated differences in Gsb and En chromatin engagement from the broader proposed model.
(5) The authors correctly recognise that the factors determining the NB7-4 accessibility landscape are still unknown. Pinpointing Msh and other motif-enriched candidates clarifies this issue and offers valuable leads for future research.
Overall, the authors have responded constructively to the major concerns raised in the review. The revised manuscript presents a substantially more coherent model in which STF-code components have distinct but complementary roles in shaping the chromatin landscape and enabling lineage-specific Hb engagement.
Author response:
The following is the authors’ response to the original reviews.
eLife Assessment:
The study provides an important advance towards understanding how spatial and temporal transcriptional programs are integrated to regulate lineage-specific chromatin and enhancer activation. The functional evidence is currently incomplete, but the current data provide a solid correlative and conceptual foundation. Functional experiments directly linking Gsb occupancy to chromatin state and regulation of some lineage-specific targets would further strengthen the causal interpretation of the model. Clarifying the scope of conclusions and explicitly acknowledging the technical limitations of current chromatin assays would provide a more balanced interpretation of the manuscript.
We thank the reviewers and editors for their comments on our manuscript. We address here the concerns raised by them.
Public Reviews:
Reviewer #1 (Public review):
Summary:
It has long been known that Drosophila embryonic ventral nerve cord neuroblasts incorporate both spatial and temporal transcription factor expression to generate 30 distinct neuroblasts and lineages per hemisegment. This manuscript aims to elucidate the mechanism by which this integration of spatial and temporal transcription factors occurs through "direct regulation" or "epigenetic regulation". Direct regulation is defined as both spatial and temporal factors binding to open chromatin and working together to dictate specific lineages. Epigenetic regulation is defined as a spatial factor priming the chromatin in a neuroblast-specific manner to allow for the integration of temporal factors to generate specific lineages. The authors conclude that there is a two-step model in which a spatial transcription factor code "primes" the chromatin in terms of accessibility and then recruits temporal factors to ensure lineage-specific enhancer activation.
We thank the reviewer for this clear and succinct summary and for accurately capturing the central idea of the model we propose. In particular, we appreciate that the reviewer highlights the distinction between the previously proposed “direct regulation” and “epigenetic regulation” models, which our work suggests may operate together within neuroblast lineages through a combinatorial spatial transcription factor code — a model we refer to as the “distributed STF code model”.
Strengths:
The authors tested two models, "direct regulation" vs "epigenetic regulation" in a well-defined pool of neural stem cells during normal development.
We thank the reviewer for recognizing this aspect of the study.
Weaknesses:
The data in this study cannot clearly substantiate these two models.
Overall, there are a number of issues that are inconsistent and not supportive of the model proposed in this manuscript. Firstly, there is no evidence of pioneer factor activity in any of the NB lineages described - i.e., any changes in chromatin accessibility being shown over time. The authors must show chromatin conformation changes during the window of spatial transcription factor expression in order to convince the readers of this phenomenon.
Thank you for raising this point. We agree that a developmental time course of chromatin accessibility would be very informative. The difficulty is that, in this system, the relevant sequence unfolds extremely rapidly and across two cellular contexts. Spatial transcription factors such as Gsb are expressed in the neuroectoderm; NBs then delaminate, and Hb expression begins almost immediately — on the order of minutes to tens of minutes. Before delamination there is no NB that can be targeted with lineage-specific drivers, whereas after NB formation the temporal program is already under way. This transition therefore cannot currently be resolved using lineage-restricted Dam profiling.
In most studies, however, pioneer or chromatin-priming activity is inferred from a transcription factor’s ability to bind regions of relatively low accessibility and alter accessibility in a heterologous cellular context, rather than from direct developmental timecourse measurements. Our study provides both types of evidence: Gsb occupies accessible and relatively less-accessible regions in NB5-6 (Figure 4), while ectopic Gsb expression in NB7-4 causes bidirectional chromatin remodelling, with corresponding changes in Hb occupancy (Figure 6 and Figure 6—figure supplement 2).
In light of this technical constraint, and in response to the reviewer’s broader concern, we have revisited our original interpretation. We now emphasise Gsb’s ability to alter chromatin accessibility bidirectionally and use the term “chromatin priming” rather than “pioneering.” We also make clear that Gsb does not act alone to establish the NB5-6 chromatin landscape, but functions within a broader, distributed STF code.
Some examples of relevant manuscript changes are:
- “We found, however, that Gsb’s role in chromatin remodelling was more complex: ectopic Gsb remodelled chromatin bidirectionally with regions both losing and gaining accessibility upon Gsb misexpression.”
- “Thus, Gsb’s chromatin-remodelling capacity is not equivalent to autonomous opening of lineage-appropriate enhancers. Rather, these data support a model in which Gsb contributes to chromatin priming as part of a broader NB-specific STF code…”
- “Thus, Gsb can engage loci and directly restrict chromatin accessibility at bound sites, but its ability to promote accessibility and Hb recruitment likely requires additional NB5-6-specific inputs.”
Secondly, the phenotypic data do not align with the sequencing data - the story would be more cohesive if the sequencing data and phenotypic data were in the same NB subtypes. On one hand, we are shown that Gsb misexpression induces loss of chromatin accessibility in NB 7-4, however in the widespread loss model, we are not shown a phenotype in these NB7-4 - which suggest that the chromatin accessibility at these sites (sites that have already been distinguished as SoIs for that NB subtype) does not play an important role in distinguishing NB 7-4 identity. However, the authors report loss of NB3-5 identity but have no evidence as to how the chromatin has changed (or if it has at all) in that subtype, leaving the readers to wonder how the loss of identity occurred
Thank you for raising this point regarding the alignment between the chromatin and phenotypic analyses. The reviewer’s comment made us realise that the rationale for these experiments may not have been sufficiently clear in the original manuscript and could therefore be perceived as misaligned. We therefore explain the logic of the experimental design here and have edited the manuscript to clarify this point for readers.
The chromatin experiments were designed to test whether Gsb is capable of remodelling chromatin when introduced outside its endogenous domain. NB7-4 provided a convenient heterologous context with clean genetic access for TaDa/CATaDa profiling.
Importantly, although NB7-4 was selected primarily for this technical reason, Gsb misexpression did affect NB7-4’s identity readout: at higher induction levels, the NB7-4-specific Gal4 reporter was lost. Because this same driver was required for lineage-restricted chromatin profiling, we titrated Gsb misexpression to a lower level that preserved reporter activity while still allowing us to assay chromatin accessibility and Hb occupancy. This calibration is documented in the new Figure 6—figure supplement 1. Thus, the calibrated profiling condition should not be interpreted as evidence that Gsb has no effect on NB7-4 identity; rather, it was necessary to retain genetic access to the lineage.
The phenotypic experiments addressed a different question: the developmental requirement for Gsb within its endogenous spatial domain. Our analysis was therefore centred on NB5-6. Gsb depletion caused loss of the NB5-6-specific Lbe reporter and the NB5-6 neuronal markers Ap and Eya, whereas Gsb misexpression produced a partial gain of these markers (Figures 7 and 8). The accompanying changes in row-3/NB3-5 markers emerged as an additional reciprocal phenotype, consistent with Gsb established role in specifying row-5/6 identity while repressing row-3 fates, rather than as an independent endpoint that we had selected.
Thus, the chromatin and phenotypic experiments are complementary: the NB7-4 experiment tests Gsb’s capacity to remodel chromatin in a convenient heterologous context, whereas the NB5-6 experiments establish its developmental function within its endogenous domain. We have revised the manuscript to make this distinction explicit.
Relevant edits made to manuscript:
“Having shown that ectopic Gsb remodels chromatin and alters Hb occupancy in the NB7-4 lineage, we next asked whether altering Gsb function changes NB lineage identity. The misexpression experiments described above had already shown that, when induced at high levels, Gsb misexpression turned off the NB7-4-lineage Gal4 reporter, consistent with Gsb suppressing NB7-4 identity at sufficient levels. In the chromatin-profiling experiments above, this was addressed by inducing lower Gsb misexpression levels that preserved Gal4 expression. Together, these observations suggested that Gsb misexpression can influence lineage identity as well as chromatin state. However, since NB7-4 is not an endogenous Gsb-expressing context, we turned to NB5-6 to test the developmental consequences of altering Gsb function.”
Reviewer #2 (Public review):
Summary:
This article by Bhattacharya et al. investigates how neural stem cells (NSCs, NBs) in Drosophila integrate spatial and temporal cues to activate neuron-specific terminal selector (TS) genes. Prior to this work, it was understood that NSCs utilize spatial transcription factors (STFs) and temporal transcription factors (TTFs) to determine lineage identity and birth order, but the mechanisms of integration were not fully elucidated. The authors employed chromatin profiling techniques to analyze the binding of STFs and TTFs in two specific neuroblast lineages, NB5-6 and NB7-4. They found that Gsb (an STF) binds both accessible and less-accessible chromatin in NB5-6, while En (another STF) binds only to pre-accessible chromatin in NB7-4. The findings support an "STF code" where the combination of pioneer and non-pioneer spatial factors, along with temporal factors, triggers neuroblast-specific enhancer activation and determines lineage identity.
We appreciate the reviewer’s careful summary of our findings and their clear articulation of the STF-code framework that emerges from the work.
Strengths:
The experiments are well-executed, the interpretations are generally sound, and the figures are clear and elegant. However, some conclusions are drawn too broadly without essential functional data. Therefore, additional work is needed to more effectively convey the central message.
We thank the reviewer for their positive assessment of the experiments, interpretation, and figures, and we respond to their specific concerns below.
Weaknesses:
(1) Integration of TaDa and functional data on Gsb for the STF model
The authors demonstrate that TaDa profiling maps Gsb binding across the genome and identifies candidate chromatin-priming sites in NB5-6. Gsb LOF/GOF experiments reveal effects on NB identity. Combining TaDa data with LOF and GOF analyses indicates that Gsb influences NB5-6 specification by binding to both open and relatively closed chromatin, helping maintain NB5-6 identity while limiting NB3-5 fate.
However, the study does not establish a direct link between specific LOF/GOF phenotypes and particular genomic targets. For instance, analyzing Gsb occupancy at lineage-specific identity factors or terminal selector genes (such as Lbe, Ap, or Eya for NB5-6; and Ems, etc., for NB3-5) in wild-type and manipulated conditions (Gsb misexpression) would directly connect chromatin binding to the regulation of fate determinants. These investigations would strengthen the mechanistic connection between the correlative TaDa profiles and the observed identity changes, supporting the idea that Gsb functions as a context-dependent chromatin-priming factor within the STF code, rather than as a generic transcription factor.
We thank the reviewer for this suggestion. We agree that examining Gsb occupancy at known lineage determinants helps connect the genome-wide TaDa profiles to the developmental phenotypes. We have therefore added browser profiles showing endogenous Gsb occupancy at several loci associated with NB5-6 identity and its neuronal output, including lbe, ap, and eya (new Figure 2E). These are the same markers that are lost following Gsb depletion and partially gained following Gsb misexpression (Figures 7 and 8), providing a locus-level connection between Gsb binding and the observed changes in NB5-6 identity.
The precise comparison suggested by the reviewer is difficult because the binding and misexpression experiments were necessarily performed in different contexts. Gsb occupancy was measured in NB5-6, where Gsb is normally expressed, whereas the effect of Gsb misexpression on chromatin was measured in the heterologous NB7-4 context where Gsb is not normally expressed. We therefore did not measure Gsb occupancy following its misexpression in NB7-4 and so cannot compare Gsb binding before and after misexpression at individual loci in NB7-4.
We have nevertheless added genome-browser examples showing how Gsb misexpression affects accessibility and Hb occupancy at lineage-associated loci (new Figure 6M). The ems locus provides one such example: ectopic Gsb reduces both accessibility and Hb occupancy at ems in NB7-4, while broad Gsb misexpression causes loss of Ems expression and of the independent NB3-5 reporter R59E09>GFP in the phenotypic experiments (Figure 8).
(2) Gsb misexpression reveals bidirectional chromatin remodelling
Experiments with ectopic Gsb expression demonstrate bidirectional chromatin remodelling in NB7-4, showing decreases in accessibility at some binding sites and increases at others. While the authors show that Gsb can disrupt chromatin upon misexpression, interpreting its "pioneer-like" or chromatin-priming activity is complex due to several factors: the misexpression occurs in a non-native lineage, the direct versus indirect effects rely on whole-embryo Dam-Gsb peaks instead of NB7-4-specific binding, and heat-shock-induced chromatin changes are not fully accounted for. These issues make it challenging to definitively determine Gsb's role in chromatin priming.
A complementary approach would be to perform Gsb knockdown/loss-of-function in its native NB5-6 lineage and profile chromatin accessibility (TaDa or CATaDa). This would allow a cleaner, more physiologically relevant assessment of Gsb's contribution to priming, SoI establishment, and Hb recruitment. Such an experiment would strengthen the causal link between Gsb occupancy and chromatin state and clarify whether Gsb truly acts as a context-dependent pioneer in vivo, rather than producing indirect effects due to ectopic misexpression.
We thank the reviewer for this thoughtful comment. The reviewer raises four related but distinct issues, which we address separately below.
(1) Gsb was misexpressed in a non-native lineage
As discussed in our response to Reviewer 1 regarding pioneer activity, expression in a heterologous cellular context is a standard experimental strategy for testing whether a transcription factor has the capacity to engage and remodel chromatin outside its endogenous setting. The use of NB7-4 was therefore deliberate: this experiment tested the sufficiency of Gsb to alter chromatin, rather than attempting to reproduce its physiological activity in NB5-6. Nevertheless, because Gsb produced bidirectional changes rather than autonomous chromatin opening, we have revised our interpretation. We now describe this experiment as demonstrating chromatin remodelling capacity and use “chromatin priming” rather than claiming a classical pioneer for its activity.
Relevant edits made to the manuscript: Results, Figure 7 and Figure 7—figure supplements 1 and 2.
(2) Direct and indirect effects were inferred using whole-embryo Gsb occupancy
We have not profiled Gsb occupancy directly in NB7-4 following its misexpression. However, we have now performed an unbiased genome-wide analysis using two independent Gsb-binding datasets: stage-matched Gsb occupancy from NB5-6 and whole-embryo Gsb occupancy. Since Dam methylation is cumulative across the window of Gal4 activity, the whole-embryo dataset serves as a proxy for embryowide sites that are compatible with Gsb occupancy. Across both datasets, the result was strikingly consistent. The set of 1,340 loci that lost accessibility upon Gsb misexpression showed strong Gsb occupancy in both datasets, whereas the 606 loci that gained accessibility showed little or no Gsb occupancy in either dataset (Figure 6E,H). Thus, two independently generated Gsb-binding maps systematically distinguish the loci that close from those that open.
This reproducibility — across many sites and two datasets — provides strong evidence that accessibility loss occurs at loci that Gsb can bind and is therefore likely to reflect a direct consequence of Gsb activity. Conversely, lower Gsb occupancy at sites that gain accessibility suggests that these gains are more likely to be indirect. The corresponding changes in Hb occupancy further strengthen this interpretation (Figure 6C–G).
In summary, we find it particularly compelling that two independently generated Gsb-binding maps, obtained at different developmental stages and in different cellular contexts, identify the same systematic distinction: loci that close upon Gsb misexpression are Gsb-bound, whereas loci that open are not. We therefore consider this a strong and unbiased result supporting a direct role for Gsb in chromatin closing.
Relevant edits made to the manuscript: (Results; Figure 6E, H)
“Although we did not directly profile Gsb occupancy in NB7-4 upon misexpression, we asked whether the sites that lose or gain accessibility differ in their propensity to be bound by Gsb. To do this, we mapped independent Gsb occupancy datasets onto these two classes of remodelled sites: stage-matched Dam: Gsb from NB5-6, where Gsb is normally expressed, and whole-embryo Dam: Gsb, which provides a broader map of Gsb-compatible loci. In both datasets, sites that lost accessibility upon Gsb misexpression in NB7-4 showed strong Gsb occupancy, whereas sites that gained accessibility showed little or no Gsb signal. Thus, the most direct consequence of ectopic Gsb is likely accessibility loss at loci that Gsb can bind, and this loss is accompanied by reduced Hb occupancy. By contrast, accessibility gains occur mainly at loci with little evidence of Gsb binding in independent datasets, suggesting that these gains are more likely indirect consequences of Gsb misexpression.”
(3) Heat-shock-induced chromatin changes were not fully accounted for
We agree that, because heat shock is required to induce Gsb, a contribution from the induction regime cannot be completely excluded. However, several observations argue that the chromatin changes are not dominated by a non-specific heat-shock response. First, the accessibility changes were reproducible across biological replicates (Figure 6—figure supplement 2). Second, as described above, sites that lost accessibility were consistently occupied by Gsb in two independent datasets and showed a corresponding reduction in Hb occupancy, whereas sites that gained accessibility showed little or no Gsb occupancy (Figure 6C–H). This structured relationship would not be readily explained by a general heat-shock response.
As an additional test, we performed a Gene Ontology analysis of genes associated with sites that gained accessibility. The enriched terms were predominantly related to development and cell-fate specification, including neuroblast differentiation and segment specification, while heat-shock and stress-response terms were not enriched (Author response image 1). Although none of these observations alone formally excludes an effect of heat shock, together they argue that the observed pattern reflects structured, Gsb-associated chromatin remodelling rather than a general stress response. We nevertheless interpret the accessibility gains cautiously as likely indirect effects of Gsb misexpression.
Author response image 1.
Gene ontology analysis of sites showing gain in accessibility upon Gsb misexpression in NB7-4. Genes overlapping the 606 sites showing at least 2-fold increase in accessibility (DamOnly signal) were annotated using the UCSC Table Browser (dm6). The resulting gene list was analysed for associated gene ontology (GO) terms using Metascape (Zhou et al. 2019). The top 20 enriched GO terms are represented as a bar chart, coloured by p-values.
(4) Chromatin accessibility following Gsb loss in NB5-6 would provide a more physiological test
We agree that this would be an informative experiment, and we explored several ways of perturbing Gsb early enough to address its endogenous function. A NB5-6-specific RNAi approach could not be used because the available NB5-6 Gal4 driver becomes active only after the period in which Gsb specifies the lineage.
The difficulty with conventional gsb mutants is different but equally limiting. Homozygous embryos would have to be identified during the brief developmental window in which Hb is expressed, while each TaDa/CATaDa replicate requires at least 30 mg of precisely staged embryos. The available fluorescent balancer markers are either not detectable at these early stages or become reliably detectable only towards the end of the window. It is therefore not feasible to identify and collect sufficient homozygous embryos — either manually or by fluorescence-assisted whole-embryo sorting — within the required time frame.
We therefore generated a gsb guide-RNA line and used Vasa-Cas9 to deplete Gsb in the early embryo. This strategy produced a high rate of Gsb editing, embryonic lethality and the expected axonal defects (Figure 7—figure supplement 1). When we examined NB5-6 in these embryos, we found that the Lbe-K reporter and the NB5-6 neuronal markers Ap and Eya were lost, confirming the requirement for Gsb in specifying this lineage (Figure 7).
Unfortunately, this also meant that we could not take the experiment forward for TaDa/CATaDa. Lbe-K-Gal4 is the driver we use to target NB5-6, and its expression was lost following early Gsb depletion. Without the driver, we can no longer restrict Dam expression to NB5-6 and therefore cannot obtain a lineage-specific chromatin profile. Thus, although the early CRISPR strategy allowed us to examine the developmental consequences of Gsb loss, the same experiment could not be used to assay chromatin accessibility in NB5-6. We have made this limitation explicit in the revised Discussion.
Relevant edits made to the manuscript: (Result, Figure 7 and Figure 7— figure supplement 1). We also now state explicitly in the “Scope and future directions” section that the current tools did not allow us to profile the chromatin consequences of Gsb loss directly in NB5-6 (Discussion).
(3) En is not a pioneer factor
The authors conclude that Engrailed (En) is not a pioneer factor, based on the observation that En binding correlates with accessible chromatin and that En is not enriched at NB5-6-specific SOIs. However, this conclusion is not sufficiently supported by the functional data.
First, the absence of En binding at NB5-6-specific SOIs does not necessarily indicate an inability to engage closed chromatin. These regions were not selected for the presence of En consensus motifs, so their lack of occupancy may simply reflect the absence of En binding motifs rather than a lack of pioneering capacity. A systematic motif analysis at NB5-6-specific SOIs is needed to determine whether En binding sites are present but unoccupied.
Second, the claim that En lacks pioneer activity relies solely on a single steady-state TaDa/DamID occupancy assay at one developmental stage. Because pioneer factor interactions can be transient, low-affinity, and stage-specific, such binding may not be detected by TaDa, which also depends on local GATC density and methylation kinetics and may yield false negatives. Given these technical limitations, the absence of En binding at less accessible regions does not definitively rule out a priming role.
In the absence of direct functional assays (En LOF/GOF), the authors should explicitly acknowledge these technical and conceptual limitations and tone down the claim that "En lacks pioneer activity".
We thank the reviewer for raising this point. We agree that the statement “En is not a pioneer factor” could be read as a general conclusion about En activity across developmental contexts. Our conclusion is more specific: in NB7-4 during the Hb-competence window, En occupancy is strongly biased towards pre-accessible chromatin and does not display the chromatin-engagement pattern expected of a pioneer factor. We have revised the manuscript throughout to make this context-specific interpretation clear. Below, we also address each of the reviewer’s points.
(1) The absence of En at NB5-6-specific SoIs could reflect an absence of En-binding motifs
We agree that the absence of En occupancy at NB5-6 SoIs would not be informative if these loci lacked En-binding motifs. We therefore performed the suggested motif analysis. En motifs were present at comparable frequencies in NB5-6- and NB7-4-specific SoIs (88.0% and 86.5%, respectively), with both enriched relative to matched shuffled controls (new Figure 3—figure supplement 1L). Thus, NB5-6 SoIs are motif-compatible but remain unoccupied by En in NB7-4. Their lack of En occupancy cannot therefore be explained simply by motif absence and instead points to additional determinants such as chromatin accessibility or cofactor context.
Relevant edits made to the manuscript: Result, Figure 3B–C and new Figure 3—figure supplement 1L)
“As expected, En was enriched at NB7-4-specific SoIs, where Hb uniquely binds in this lineage. However, there was no detectable En signal in NB7-4 at sites known to be NB5-6-specific SoIs, which remain relatively less accessible and Hb-free in NB7-4. This difference is unlikely to be explained simply by En motif availability, since En motifs are similarly represented at NB7-4- and NB5-6-specific SoIs. Thus, at SoIs, En occupancy is associated with the cognate, accessible NB7-4 SoIs, with little detectable occupancy at the non-cognate, relatively less-accessible NB5-6 SoIs.”
(2) TaDa may fail to detect transient, low-affinity or stage-specific En binding
We agree that no occupancy assay can exclude interactions below its detection threshold. However, TaDa is cumulative rather than a steady-state snapshot: Dam methylation accumulates during the expression window, allowing transient interactions to leave a detectable signal. Genome-wide, En occupancy was strongly weighted towards accessible chromatin, and accessibility predicted En binding with an AUROC of 0.74 (Figure 3E–I, Figure 3—figure supplement 2 and new Figure 3— figure supplement 3E).
Gsb provides an internal comparison using the same experimental framework. Gsb binding was detected at both accessible and relatively less-accessible chromatin, and accessibility did not predict Gsb occupancy better than chance (AUROC = 0.41; Figure 4E–I and new Figure 3—figure supplement 3E). This argues that the En pattern is not caused by a general inability of TaDa to detect binding at less-accessible loci. We nevertheless acknowledge that rare interactions, interactions below the detection threshold or binding outside the assayed window cannot be excluded. The resolution and cumulative nature of TaDa are now also described explicitly in the revised manuscript.
Relevant edits made to the manuscript: Result section for the genome-wide analysis and Discussion section for the TaDa limitations; Figure 3E–I, Figure 4E–I, Figure 3—figure supplement 2 and new Figure 3—figure supplement 3E).
(3) Direct functional assays of En were not performed
We agree that determining whether En actively establishes or maintains chromatin accessibility would require En loss- or gain-of-function experiments combined with chromatin profiling. These experiments were not performed in this study, and we have therefore removed the categorical statement that “En is not a pioneer factor.”
The revised manuscript now states the more specific conclusion supported by our data: En occupancy in NB7-4 during the Hb-competence window is strongly biased towards pre-accessible chromatin. We also acknowledge that the current experiments cannot exclude rare or transient interactions below the detection threshold, or chromatin-remodelling activity by En at other loci, developmental stages or cellular contexts. The factors that establish the NB7-4 accessibility landscape therefore remain unresolved.
Relevant edits made to the manuscript: (Result; Figure 3)
“Taken together, these results show that En occupancy is strongly biased toward preaccessible chromatin in NB7-4, including at NB7-4-specific SoIs. Thus, En is well placed to integrate spatial and temporal information through direct regulation at preaccessible enhancers. However, because NB7-4 has a distinct accessibility landscape, the factors that establish this landscape remain unresolved. We therefore propose that NB7-4 may use a combinatorial STF-code logic, in which other lineage-specific factors help establish NB7-4-specific SoIs, while En occupies this pre-accessible landscape and likely collaborates with Hb to activate lineage-specific enhancers.”
(4) Clarity of STF-code Model and Central Message
The manuscript begins by presenting two models, direct and epigenetic, but the central takeaway of the paper is not clear. Specifically, the nuanced roles of the spatial factors Gsb and En as chromatin-priming versus stabilizing/effector factors within an STF code, and the resulting division of labor, are not clearly illustrated. The distinction between Gsb as a chromatin-priming factor and En as a cofactor-dependent activator/stabilizer should be explicitly presented in a stepwise model for better clarity. The authors could strengthen this by providing a schematic with two sequential stages illustrating how neuroblast identity factors (STF code) change chromatin states to drive lineage-specific enhancer activation. The schematic can be shown from the neuroectoderm to individual NB lineages to make it more panoramic.
We thank the reviewer for this suggestion. This comment, together with related points raised below, made us realise that the manuscript did not bring out strongly enough that our data lead us to propose a third model: the distributed STF-code model. Rather than treating direct and epigenetic regulation as mutually exclusive alternatives, this model distributes these activities across different members of a NB-specific STF code. We have now revised the manuscript to reflect this transition. For example, we now state this explicitly in the Abstract, at the end of the Introduction and at the beginning of the Discussion.
Relevant edits made to the manuscript: Abstract, Introduction, Discussion, and have added a new summary schematic (new Figure 9).
(5) Identification of Priming Factors in NB7-4
While the authors suggest that an unknown priming factor might be responsible for establishing sites of integration in NB7-4, they do not identify or explore potential candidates for this role. Further investigation into what factors might be involved in chromatin priming in NB7-4 could provide a more complete understanding of the mechanisms at play.
We thank the reviewer for this suggestion. We agree that identifying the factors that establish the NB7-4 accessibility landscape is an important next step. Our current data do not identify such a factor, and doing so would require systematic perturbation of candidate STF-code members together with lineage-specific chromatin profiling.
We have nevertheless made this question more concrete in the revised Discussion. We identify Msh as a candidate shared input in NB5-6 and NB7-4, highlight additional candidates suggested by motif enrichment at NB7-4 SoIs, and propose NB6-4 — which coexpresses Gsb and En — as a useful context in which to test the division of labour between STF-code members. We present these as testable candidates rather than established priming factors.
(6) Functional Validation of STF Code Components
The study proposes an STF code for each neuroblast lineage, but the specific components of these codes, beyond Gsb and En, are not fully explored. Identifying and validating additional factors that contribute to the STF code in each lineage could strengthen the conclusions.
We thank the reviewer for this point. We agree that the complete composition of the NB5-6 and NB7-4 STF codes remains to be established, and we now make this explicit. The present study proposes a mechanistic framework based on the different chromatin-engagement properties of Gsb and En, rather than claiming to have defined or validated every component of either code.
As noted in our response above, the revised Discussion identifies Msh, Iroquois-complex members and TF families enriched at NB7-4 SoIs as candidate additional components. Testing these factors individually and in combination, with chromatin, Hb-occupancy and lineage-identity readouts, will be required to define the minimal functional code for each lineage. We present this as an important direction for future work (lines 759–770).
Recommendations for the authors:
Reviewer #1 (Recommendations for the authors):
(1) The reviewers appreciate the content shown in Figure 1, which is helpful for people who are unfamiliar with the system as a whole. One key issue is that later in the manuscript, the authors look at Gsb and describe it as a factor that works to close chromatin, making the model depicted in panel E misleading when reading the first half of the manuscript and leading to confusion/whiplash when getting to Figure 6, where a spatial transcription factor is thought to play a role in closing chromatin.
Thank you for pointing this out. Figure 1 was intended to present the two models currently proposed in the field and as the starting hypotheses for the study. It was not intended to present the conclusions from our study. We realise, however that this distinction was not clear and since we did not present the updated model, it made the later finding appear contradictory.
We have now made substantial edits to the text and have made the transition from the original models to our revised interpretation explicit in the Introduction and Discussion. We have also added a new summary model (new Figure 9), which shows that chromatin opening, chromatin closing and STF–Hb engagement can be distributed across different members of an NB-specific STF code.
(2) It is unclear what "less accessible" actually means. The reviewers would like to see quantitative standardization to define the degree of chromatin accessibility when using the Dam expression alone.
We thank the reviewer for this point. The comment made us revisit how we defined accessibility, which we had previously treated too much as a binary distinction between accessible and inaccessible chromatin. We have now revised our analyses to represent chromatin accessibility as a continuum of states.
We defined a common set of 27,119 DamOnly-marked regions across both NBs, ranked them by mean DamOnly signal and divided them into deciles from least to most accessible. The construction and validation of this continuum are shown in new Figure 2G and new Figure 2—figure supplement 2A–D. We also use this framework to examine STF occupancy and Gsb-dependent chromatin remodelling across the accessibility spectrum (new Figure 3— figure supplement 2A and new Figure 6—figure supplement 2E–F).
Where a discrete lineage-specific comparison was required, we applied a minimum two-fold difference in DamOnly signal between the NBs. This identified 845 NB5-6-preferential and 1,067 NB7-4-preferential regions. A region preferentially accessible in one NB is therefore described as relatively less accessible in the other, rather than as completely inaccessible.
Relevant edits made to the manuscript: Results, new Figure 2G and new Figure 2—figure supplement 2A–D.
(3) A key takeaway in Figure 3 is that En is binding already accessible chromatin. Another key point that this manuscript makes is that spatial-temporal factors (and codes) are responsible for this opening. We have not yet been shown that chromatin accessibility of this neuroblast identity actually changes upon spatial-temporal factor expression. I suggest including data of the dam-only control as a time course. Showing chromatin accessibility just before temporal factor expression and then after to validate that pioneer factor activity appears to be likely.
We appreciate this suggestion and agree that, conceptually, the model implies a temporal sequence in which spatial factors first act during neuroectoderm patterning and temporal factors such as Hb subsequently engage the chromatin after neuroblast formation. In practice, however, these events occur in very rapid succession during development. Spatial transcription factors are expressed in the neuroectoderm, neuroblasts are specified and delaminate, and Hb expression begins almost immediately after neuroblast formation — all within a short developmental window on the order of minutes to tens of minutes.
This makes it difficult to directly test chromatin accessibility changes across these stages using lineage-restricted Dam-based approaches. Before delamination there is no neuroblast to target with the available drivers, and once the neuroblast forms the temporal program is already underway. As a result, the sequential changes in accessibility that the reviewer is suggesting cannot currently be resolved with the available tools.
We agree that directly visualizing how chromatin accessibility changes during this brief developmental transition would be very interesting, but doing so would likely require approaches with much higher temporal resolution, such as live imaging-based methods for monitoring chromatin accessibility.
(4) Similar to the comment for Figure 3, showing chromatin accessibility changes in the NB subtype via a time course that includes both before and after spatial transcription factor expression begins.
This point is closely related to the reviewer’s previous comment regarding a temporal analysis of chromatin accessibility. Please refer to our response above describing the developmental timing and technical limitations that make such an experiment difficult in this system.
(5) It is difficult to reconcile the conclusion in Figure 5 that regions bound by both En/Hb or Gsb/Hb are considered the highest accessibility. Can the authors touch on the logic for this, given that in the next figure, they are going to suggest that Gsb works as a factor that is primarily responsible for closing chromatin? It is unclear how Gsb is closing chromatin, but it is also binding Hb, and in those regions, chromatin is most accessible.
We thank the reviewer for raising this point. The findings in Figures 5 and 6 may initially appear difficult to reconcile, but we think that they both actually point to the same conclusion: that Gsb’s effect on chromatin is context-dependent.
An important point is that Gsb binding is not restricted to a single chromatin state. In its endogenous NB5-6 context, it occupies both accessible and relatively less-accessible regions (Figure 4B–C, E–F). Its presence at accessible chromatin does not establish that Gsb opened that chromatin nor that it closed it because these occupancy data do not reveal the direction of chromatin change. From these data we can only conclude that sites co-bound by Gsb and Hb are more accessible than sites bound by Gsb alone, suggesting that the outcome of Gsb binding depends on the surrounding regulatory context. In the presence of Hb — and likely other members of the NB5-6 STF code — Gsb can participate in a highly accessible, productive regulatory state (Figure 5C–E).
The misexpression experiment helps us interpret these relationships in the endogenous context. When Gsb is introduced into NB7-4, sites that lose accessibility are strongly Gsb-bound in two independent datasets, and this loss is accompanied by reduced Hb occupancy (Figure 6C–H). Some Gsb-bound sites also gain accessibility, consistent with bidirectional remodelling, but Gsb occupancy is negligible at these gain sites. Thus, outside its normal NB5-6 regulatory environment, the more direct effect of Gsb binding is likely chromatin closing.
This provides a possible explanation for the endogenous (NB5-6) data in Figure 5. The lower accessibility of sites bound by Gsb alone may reflect this closure-biased activity, whereas the highly accessible state at Gsb–Hb co-bound sites is likely to arise from the broader NB5-6 regulatory context. We do not conclude that every Gsb-only site is actively closed, or that Gsb itself opens the co-bound sites. Rather, our data suggest that Gsb can participate in different chromatin states, with the outcome of its binding shaped by Hb and other members of the STF code.
We have revised the Results and Discussion to make this interpretation clearer.
(6) The most striking piece of data from Figure 6 is that it appears the physiological role of Gsb is to reduce chromatin accessibility and that opening chromatin is an indirect effect. I am not sure that the field widely recognizes closing as chromatin as "pioneer-like" activity. Moreover, this finding seemed very out of place given the way the story has been sold thus far - discussing "priming of chromatin" and referring to this as opening in figure 1 and discussion of binding of these factors at open and less accessible.
Thank you for raising this issue, which overlaps with the reviewer’s previous point. We were equally struck by the result from Figure 6: the sites that lose accessibility are more consistent with a direct effect of Gsb, whereas the sites that gain accessibility are more likely to represent indirect effects. This suggests that Gsb can remodel chromatin bidirectionally.
We agree that the field does not generally describe chromatin closing as pioneer activity. We therefore no longer refer to Gsb as a pioneer factor. Instead, we describe Gsb as having bidirectional chromatin-remodelling capacity and refer to it as “chromatin priming” activity.
Finally, we agree that the original presentation made this progression difficult to follow. Figure 1 presents the two existing models as the starting point, whereas our data lead us to propose a third, distributed STF-code model. We now make this transition explicit throughout the revised manuscript and have added a new summary schematic (new Figure 9; See Results and Discussion).
Reviewer #2 (Recommendations for the authors):
Minor comment:
(1) Have the authors validated the specificity of the Lbe-GAL4 and 19B03AD;18F07DBD drivers? It is important to clarify whether these drivers label only NB5-6 and NB7-4 neuroblasts or also their progeny. Since the interpretation of chromatin dynamics is in the context of neuroblasts, reads from differentiated neurons could affect the conclusions.
We thank the reviewer for raising this point. These driver lines were extensively characterised in our previous study and are highly specific to the NB5-6 and NB7-4 lineages, respectively. Importantly, as the reviewer points out, they are not neuroblast-exclusive and also label progeny. Because our profiling is restricted to the early Hb competence window, this contribution should be limited to a small number of GMCs and the earliest-born neurons.
On rereading the manuscript, we realised that there were several instances where we used “NB-specific” too loosely when referring to these datasets. We have therefore revised the manuscript throughout to make clear that the TaDa/CATaDa profiles represent the NB5-6 and NB7-4 lineages during the Hb window and are not exclusively derived from the neuroblast itself. We have also added this explicitly as a limitation in the Scope and Future Directions section. Importantly, this does not change the essence of our conclusions, which are based on comparisons between these two defined lineages and on perturbations examined within the same lineage-restricted framework.
Relevant edits made to the manuscript:
- “The driver lines were previously validated as being specific to these lineages”.
- “First, the lineage-specific Gal4 drivers also label the immediate progeny of each NB; thus, although profiling was restricted to the early Hb competence window, when only a small number of GMCs and early-born neurons are present, the resulting profiles are not exclusively derived from the NB itself.”
(2) The lineage NB6-4 co-expresses both Gsb and En and represents an informative intermediate case. This context supports a cooperative STF-code model in which Gsb provides chromatin priming while En acts as an effector or stabilizer, without requiring autonomous pioneer activity. The authors should briefly discuss how NB6-4 fits into their model and whether Gsb priming in this lineage enables En-Hb co-occupancy and enhancer activation.
We thank the reviewer for this comment. We agree that NB6-4 is an especially interesting case in the context of our model, because it co-expresses both Gsb and En and therefore may provide a lineage in which both aspects of the proposed two-step mechanism can be examined together: Gsb could contribute to chromatin priming and En could function at already primed enhancers together with Hb. We now include this explicit example in the Discussion section.
(3) The authors' data support a role for Gsb in lineage-specific chromatin priming, but the functional experiments reveal important limitations. Gsb alone does not efficiently open closed chromatin, install NB5-6-specific SOIs, or recruit Hb, and ectopic Gsb often antagonizes native chromatin states. Thus, the results do not support Gsb as a classical standalone pioneer factor, as implied in the discussion. The authors should replace the term "pioneer-like potential or activity" with "context-dependent pioneer activity" in the results, as they have rightly mentioned in the discussion.
We agree with the reviewer. As discussed in our responses to the related comments above, our data do not support describing Gsb as a classical pioneer factor. Gsb remodels chromatin bidirectionally, with both gains and losses of accessibility, whereas pioneer activity is traditionally associated with chromatin opening. We have therefore gone further than the reviewer’s suggested change: throughout the manuscript, we have replaced “pioneer-like potential/activity” and “context-dependent pioneer activity” with “context-dependent chromatin remodelling” or “chromatin priming.” The Results and Discussion now make it clear that Gsb’s activity depends on the surrounding STF and cofactor context rather than representing autonomous pioneer activity.
(4) The use of ectopic expression systems may introduce artifacts or off-target effects that could affect the interpretation of the results. These potential limitations are not extensively discussed.
We agree that ectopic expression can produce indirect or non-physiological effects. The Gsb misexpression experiment was designed to test whether Gsb is sufficient to remodel chromatin in the heterologous NB7-4 context, rather than to reproduce its endogenous function in NB5-6. We therefore do not interpret every change in accessibility in the misexpression experiment as a direct or physiological role for Gsb in NB5-6. However, the changes are not random. Sites that lose accessibility are strongly associated with Gsb occupancy in two independent datasets and they also lose Hb occupancy, whereas sites that gain accessibility show little Gsb occupancy and are therefore more likely to reflect indirect effects (Figure 6C–H). The changes were reproducible between replicates, and genes associated with accessibility gains were not enriched for heat-shock or stress-response terms (Author response image 1). These analyses do not remove the limitations of ectopic expression, but they help distinguish the more likely direct effects from indirect ones. We now make both the purpose and the limitations of this experiment explicit in the Results and Discussion.
(5) The moderate correlation (0.7-0.82) between Dam-En and Dam-only reflects the expected overlap of En binding with accessible chromatin, but it also underscores a limitation: TFs that prefer open regions may show high baseline similarity to Dam-only, potentially masking low-affinity or transient interactions. The authors should acknowledge this when concluding that En lacks pioneer activity.
We thank the reviewer for bringing this up. We want to clarify that our conclusion that En preferentially occupies accessible chromatin is not based only on the moderate genome-wide correlation between Dam: En and DamOnly. We also examined En binding at defined SoIs and at differentially accessible loci across the genome. In both analyses, En is strongly enriched at accessible NB7-4 sites but shows little detectable enrichment at the corresponding less-accessible sites (Figure 3B–F). We have now extended this analysis across 27,119 regions ranked along the full accessibility spectrum. Here too, En signal remains low across the less-accessible bins and increases principally in the most accessible bins (new Figure 3— figure supplement 2E).
A useful comparison here is Gsb. Using the same experimental and analytical framework, we readily detect Gsb binding at both accessible and relatively less-accessible chromatin in NB56 (Figure 4B–F; new Figure 3—figure supplement 2E). Thus, the method is able to detect binding at less-accessible chromatin when it occurs. Consistent with this, accessibility predicts En occupancy but not Gsb occupancy in the corresponding ROC analysis (new Figure 3—figure supplement 3E). We also refer the reviewer to our response to the related comment above, where we discuss this comparison in more detail.
Finally, we want to point out that TaDa is not a snapshot assay. Dam-mediated methylation accumulates over the expression window and across many cells, so repeated weak or transient interactions should, in principle, contribute to the signal. We agree, however, that very rare interactions below the detection threshold cannot be excluded. We have therefore tempered our conclusion throughout the manuscript: rather than stating that En lacks pioneer activity, we now conclude that En occupancy during the NB7-4 Hb window is strongly biased towards pre-accessible chromatin.
(6) The schematic illustrating the NB 5-6 and NB 7-4 SOIs (Figures 3A, 4A, and S2A) should be made clearer. When Hb is bound, sites that are accessible serve as SOIs for a specific lineage, whereas Hb-unbound, inaccessible sites do not. The authors can visually represent this by showing that, in the case of Hb binding, the accessible region functions as a resident SOI for the NB7-4 lineage, while the same region in NB5-6, being Hb-free, is inaccessible. This distinction could be indicated with a shaded or shadowed area from top to bottom in Figures 3A and 4A.
We thank the reviewer for pointing this out. We have revised these panels to make the visual distinction between lineage-specific SoIs and the corresponding Hb-free, less-accessible loci in the other lineage more explicit.
(7) At 472, change "Fig.S6C-F" to "Fig.S6C-D".
Thank you for spotting that!
eLife Assessment
This useful manuscript presents evidence that the major contribution of the lysine methyltransferase KMT2A to ribosomal gene transcription regulation is through promoting histone acetylation rather than H3K4 methylation, in part through interactions of KMT2A with PCAF. Some of the claims in the paper are unsupported by the evidence and therefore viewed as incomplete. Additional experiments demonstrating a more direct and causal link between PCAF and KMT2A would elevate the key conceptual advance that this new axis influences epigenetic pathways through an unconventional histone acetylation pathway. This work will be of interest to those studying transcriptional regulation and chromatin remodeling.
Reviewer #1 (Public review):
Summary:
In this manuscript the authors study the distribution of the KMT2A methytransferase across the human ribosomal gene loci (rDNA) and the effects of its depletion on rDNA chromatin modification using a KMT2A conditional cell line. They used ChIP mapping data and claim to show that KMT2A depletion does not affect H3K4me3 levels but does reduce acetylation of H3K9 and reduces PCAF (KAT2B) binding. They further claim that KMT2A depletion causes stalling of RNA Polymerase I at the upstream "Spacer" promoter duplication on the rDNA.
Weaknesses:
I found the manuscript very confusing. In particular, I found it short on short on experimental detail and I found the data in general to be unconvincing. It is my opinion that the authors overinterpret their data, leading to unrealistic and unsupported claims. The manuscript also avoided discussing relevant recent studies on the regulation of the rDNA and of rDNA chromatin, most citations being to data 15 to 20 or more years old. This is most apparent when discussing the factors such as UBTF and SL1 that determine preinitiation complex assembly and the status of active versus inactive rDNA chromatin. Both have seen significant advances in recent years, and these should be thoroughly discussed.
Specific comments:
The authors do not describe the generation of the KMT2A conditional HEK293T cells or their treatment. Since these cells form the basis for subsequent experiments, they must be explained fully. I found that the conditional cell line had already been used in a previous unrelated publication "Chinchole et al. 2022". This manuscript explained that the KMT2A knockout is induced via Dox controlled Cas9 in a HEK cell clone constitutively expressing guide RNAs to KMT2A. The KMT2A knockout requires a 7 day treatment with Dox, leaving wide open the question of direct versus indirect effects of KMT2A loss, and perhaps explaining some of the discrepancies with published data.
In mapping the interactions of KMT2A across the rDNA by ChIP they use 18 PCR amplicons, some of which the authors inaccurately state as targeting specific rDNA elements. For example, they state Primers (that is, Amplicons) #1-3 target the spacer and core 47S promoters. In fact, amplicon #2 overlaps neither of these promoters and lies centrally within the 1kbp of enhancer repeats between them. Exactly why emphasis is put on KMT2A chipping at these two promoters is anyhow questionable since KMT2A levels (Figure 1) appear equally high at multiple IGS amplicons.
Discrepancies with published data are particularly apparent in their mapping of H3K4me2, H3K4me3 and H3K9ac, all directly relevant to their conclusions. These modifications have been shown in several publications to map to a single major site a few hundred bp upstream of the spacer promoter in multiple human and mouse cell types. Since the authors fail to monitor H3K4 methylation and H3K9 acetylation at this dominant site they also fail to provide significant evidence that KMT2A in fact affects H3 methylation or H3 acetylation on the rDNA.
The KMT2A deletion drastically reduces PolI and UBF (UBTF) engagement across the rDNA, and consistent with this, SL1 and RRN3 engagement is reduced as are 47S levels. Thus, whatever the 7-day KMT2A deletion is doing to the cells it is certainly inducing a very significant shut-down of active rDNA. How such an extreme experimental model can tell anything of KMT2A functions is unclear.
In the last results section, the authors further claim "Loss of KMT2A results in a stalled RNA Pol I complex at the spacer promoter" (Figure 4 and S4). To support this the authors try to map spacer promoter transcripts but mistakenly claim that the amplicon #1 can detect these transcripts. In fact, this amplicon overlaps the Spacer Promoter initiation site and so could not detect transcripts coming from this promoter. Published data, not cited, have demonstrated that PolI stalling at the spacer promoter occurs as a normal phenomenon on the active rDNA. Its loss is associated with suppression, not the opposite as inferred in this manuscript.
Reviewer #2 (Public review):
Summary:
This important manuscript presents solid evidence that the major contribution of KMT2A to ribosomal gene transcription is through promoting histone acetylation rather than H3K4 methylation. It is suggested this occurs in part through interactions of KMT2A with PCAF.
Strengths:
This manuscript advances an important concept that chromatin modifying enzymes can also function to promote transcription through interactions with other transcriptional regulators independently of their catalytic functions. The data are overall supportive of this conclusion for KDM2A.
Weaknesses:
(1) Care needs to be taken regarding interpreting the lack of effect of depleting KMTs or HATs on histone acetylation or methylation levels. There are several overlapping HATS or KMTs that generate these modifications. The most that can be said from these negative results is that the depleted HAT or KMT is not the sole enzyme generating these modifications at these loci.
(2) The biochemistry supporting an interaction between KDM2A and PCAF is under-developed and not strong enough to support this conclusion at this point.
(3) It is concluded that acetylation causes increased rDNA transcription. However, acetylation may be a consequence of transcription instead, especially over transcribed regions.
(4) Some conclusions, for example implicating PCAF, are based on overexpression experiments that are very artificial and difficult to draw conclusions from.
(5) Several figures are generated using publicly available data sets that were reanalyze using updated reference genomes, sometimes from different cell lines/types. This seems to be a useful exercise but the methods should include information regarding the robustness of these data sets.
Reviewer #3 (Public review):
Summary:
This manuscript examines a key mechanism by which KMT2A regulates RNA polymerase I transcription at ribosomal DNA. The authors suggest that KMT2A facilitates the recruitment of PCAF to rDNA. This process enhances histone acetylation and supports assembly of the RNA polymerase I transcription machinery. The study addresses an intriguing question by proposing a catalytic-independent role for KMT2A in rDNA transcription, potentially linking the KMT2A-dependent pathway to histone acetylation. However, several aspects of the proposed mechanism require further experimental validation. Notably, while the current evidence demonstrates associations among KMT2A, PCAF occupancy, histone acetylation, and rRNA transcription, it does not fully support the proposed objectives.
Strengths:
This could be the first report of a lysine methyltransferase regulating an acetyltransferase in Pol I-mediated gene expression.
Weaknesses:
(1) The causal relationship between KMT2A and the recruitment of PCAF needs to be demonstrated. Simply showing reduced PCAF occupancy after depleting KMT2A does not, on its own, prove direct recruitment. The authors should investigate whether KMT2A and PCAF physically interact, and whether this interaction depends on DNA or RNA.
(2) The main prediction of the model is that restoring PCAF activity should reverse the molecular effects caused by the depletion of KMT2A. The authors should test whether PCAF restoration can mitigate the consequences of KMT2A loss. They should also assess PCAF occupancy, H3K9 acetylation levels, Pol I occupancy, and nascent rRNA synthesis.
(3) The catalytic-independent function of KMT2A should be directly investigated. The absence of a significant change in H3K4me3 levels following KMT2A depletion does not confirm that KMT2A's catalytic activity is not essential. Rescue experiments comparing wild-type KMT2A with a catalytically inactive variant would provide a much stronger assessment of the proposed non-catalytic mechanism.
(4) The manuscript highlights H3K9ac; however, various acetyltransferases and histone acetylation marks are linked to active rDNA. The authors should investigate whether KMT2A specifically regulates H3K9ac or leads to broader changes in histone acetylation at rDNA.
(5) Due to the highly repetitive nature of rDNA, the manuscript must include detailed information about the reference assembly, mapping strategy, management of multimapping reads, normalisation processes, and biological replicates. The authors should demonstrate that any observed changes in occupancy are not due to technical differences in rDNA mapping or variations in copy number.
(6) KMT2A regulates a wide range of transcriptional programs, and its depletion can affect proliferation, cell-cycle distribution, or nucleolar function. The authors should include appropriate controls to demonstrate that the rDNA phenotype is not a result of changes in cellular proliferation or a generalised response to transcriptional or nucleolar stress.
(7) The CSB and B-WICH complex have previously been linked to the recruitment of PCAF and the acetylation of H3K9 at active rDNA. The authors should investigate whether KMT2A operates independently of these factors or if KMT2A functions within an established pathway for remodeling rDNA chromatin.
(8) The distinction between active and silent rDNA should be taken into account. The authors need to determine whether KMT2A/PCAF occupancy is preferentially associated with active rDNA copies. Sequential ChIP analysis or the use of established markers for active, poised, and silent rDNA would help clarify the chromatin context in which the proposed mechanism operates.
Author response:
Public Reviews:
Reviewer #1 (Public review):
Summary:
In this manuscript the authors study the distribution of the KMT2A methyltransferase across the human ribosomal gene loci (rDNA) and the effects of its depletion on rDNA chromatin modification using a KMT2A conditional cell line. They used ChIP mapping data and claim to show that KMT2A depletion does not affect H3K4me3 levels but does reduce acetylation of H3K9 and reduces PCAF (KAT2B) binding. They further claim that KMT2A depletion causes stalling of RNA Polymerase I at the upstream "Spacer" promoter duplication on the rDNA.
Weaknesses:
I found the manuscript very confusing. In particular, I found it short on short on experimental detail and I found the data in general to be unconvincing. It is my opinion that the authors overinterpret their data, leading to unrealistic and unsupported claims. The manuscript also avoided discussing relevant recent studies on the regulation of the rDNA and of rDNA chromatin, most citations being to data 15 to 20 or more years old. This is most apparent when discussing the factors such as UBTF and SL1 that determine preinitiation complex assembly and the status of active versus inactive rDNA chromatin. Both have seen significant advances in recent years, and these should be thoroughly discussed.
Specific comments:
The authors do not describe the generation of the KMT2A conditional HEK293T cells or their treatment. Since these cells form the basis for subsequent experiments, they must be explained fully. I found that the conditional cell line had already been used in a previous unrelated publication "Chinchole et al. 2022". This manuscript explained that the KMT2A knockout is induced via Dox-controlled Cas9 in a HEK cell clone constitutively expressing guide RNAs to KMT2A. The KMT2A knockout requires a 7-day treatment with Dox, leaving wide open the question of direct versus indirect effects of KMT2A loss, and perhaps explaining some of the discrepancies with published data.
We thank the Reviewer for their comment. As the Reviewer correctly states, these KMT2A knockouts have been used in previous studies: Chinchole et al., 2022; PMID: 36111497 and Malik et al., 2023; PMID: 37379335. In both cases, the generation of KMT2A inducible knockout was described in detail in the Methods section; hence, in this manuscript we referred to these publications. However, if the Reviewer feels that details should still be provided, we will do so again here.
As the Reviewer is aware, KMT2A is an essential chromatin-bound protein, which is quite stable in its heterodimer form. As we described in our publication, Malik et al., 2023, the inducible knockout clones were obtained after extensive screening and thoroughly characterized. While the protein diminishes in the WB after 3 days, we thought it necessary to confirm that it does not persist on the chromatin in ChIP assays.
We are pasting the excerpt from the manuscript describing the relevant section below.
“Our initial attempts to generate MLL knock-outs in several different cell lines were not successful. Therefore, keeping in mind that MLL is essential for cell viability and growth [26,27], we generated and successfully obtained knock-outs in HEK-293 cell lines using doxycycline-inducible Cas9 expression vectors ([52]; Fig 3A). MLL levels were drastically reduced in the 2 independent MLL-knock-out clones shown here (MLL iKO#11 and iKO#20) and both KOs showed cell cycle profile similar to Control cells (S3A Fig). In the case of chromatin-binding proteins, often cellular levels show reduction but not the chromatin-bound fraction. We, therefore, confirmed that MLL chromatin binding was indeed reduced in our inducible knock-out cell lines by ChIP assays (Figs 3B and S3B), both on centromere and HOXA9 promoter”.
Inducible Cas9-mediated knockout provides an opportunity to generate irreversible knockouts, allowing sufficient time to deplete stable proteins and reveal delayed phenotypes or compensatory responses [PMID: 29036671]. However, this advantage has to be balanced with the potential secondary effects arising from prolonged depletion. Indeed, inducible Cas9 knockouts have been generated after similar induction periods by other laboratories to deplete chromatin-bound proteins [PMID: 28216383; PMID: 37702151], especially for members of the KMT2 family [PMID: 40846851, PMID: 27535106; PMID: 37535603 ]. Further, we have not observed any discrepancy with published data for KMT2A.
In mapping the interactions of KMT2A across the rDNA by ChIP they use 18 PCR amplicons, some of which the authors inaccurately state as targeting specific rDNA elements. For example, they state Primers (that is, Amplicons) #1-3 target the spacer and core 47S promoters. In fact, amplicon #2 overlaps neither of these promoters and lies centrally within the 1kbp of enhancer repeats between them. Exactly why emphasis is put on KMT2A chipping at these two promoters is anyhow questionable since KMT2A levels (Figure 1) appear equally high at multiple IGS amplicons.
As the enhancer repeats (ER) are clearly shown in the schematic of Fig. 1A and SF1B, and the same is stated in the legend of SF1B, rather than an inaccuracy, it is a matter of oversight. We thank the Reviewer for noticing this and bringing it to our attention. We will correct it in the text.
We thank the Reviewer for raising this point regarding our focus on the promoters in the present study and the binding of KMT2A to the IGS region. KMT2A’s role in regulating RNA polymerase II transcription by regulating gene promoters is well established. However, its role in regulating RNA polymerase I transcription at the rDNA promoter has not been characterized. In this study, we provide evidence for a previously unrecognized role of KMT2A in regulating rDNA transcription and hence extend its function to regulation of RNA polymerase I.
We agree that our ChIP data show KMT2A occupancy at multiple regions within the IGS, in addition to the promoter regions. For this reason, we felt it was important to present these data and perform ChIP-seq analyses of the rDNA locus to reveal the complete picture.
Our emphasis on the promoter regions is aimed at understanding the potential role of KMT2A in regulating RNA polymerase I transcription initiation. However, we recognize that the enrichment of KMT2A at multiple IGS regions raises the possibility that it may have additional functions beyond promoter regulation. We are currently investigating the role of KMT2A at the IGS, and in our future studies we hope to address the functional significance of this occupancy in greater detail.
Discrepancies with published data are particularly apparent in their mapping of H3K4me2, H3K4me3 and H3K9ac, all directly relevant to their conclusions. These modifications have been shown in several publications to map to a single major site a few hundred bp upstream of the spacer promoter in multiple human and mouse cell types. Since the authors fail to monitor H3K4 methylation and H3K9 acetylation at this dominant site they also fail to provide significant evidence that KMT2A in fact affects H3 methylation or H3 acetylation on the rDNA.
We thank the reviewer for raising this important point about the distribution of histone modifications at the rDNA locus and how they depend on KMT2A. Previous studies have established that active histone modifications are enriched at specific regions of rDNA. In particular, landmark studies from two groups have pioneered the use of genome assemblies to map these histone modifications. The Scacheri group [Zentner et al. 2011, PMID: 21355038] reported enrichment of H3K4me2, H3K4me3, and H3K9ac at the human rDNA promoter, along with additional active-chromatin-associated enrichment within the intergenic spacer. Subsequently, the same group demonstrated that the distribution of histone modifications in mouse embryonic stem cells differs from that in human embryonic stem cells and other differentiated mouse cell types [Zentner et al. 2014, PMID: 24347625]. These findings underscore the importance of examining the relevant regulatory regions in the appropriate biological system.
Similarly, studies from the Moss group radically redefine our understanding of how histone modifications are distributed across rDNA in mammals. They demonstrated that active histone modifications (including H3K4me2/3, H3K9ac, H3K27ac, and H3K36me3) are not present at the core 47S promoter or along the transcribed gene body of active rDNA repeats, but are instead strictly sequestered to a highly localized "Enhancer Boundary Complex" located a few hundred base pairs upstream of the spacer promoter. Consequently, this precise upstream hub represents the sole site of active histone modifications on active rDNA genes [Herdman et al., 2017, PMID: 28715449 ; Mars et al., 2018, PMID: 29158335]. These findings challenge the classical models of histone modification distribution at rDNA.
However, an extensive body of classical literature presents a complementary view where active histone modifications directly decorate the core 47S promoter and proximal transcribed regions to dynamically coordinate RNA Polymerase I transcription. Work from the Grummt and Santoro laboratories established that unmethylated, active rDNA repeats possess prominent H3K4me3 and hyperacetylated histones directly at the core 47S promoter, serving as a functional "poised" platform for pre-initiation complex (PIC) assembly [Santoro and Grummt, 2002, PMID: 12368916; Santoro and Grummt, 2005, PMID: 15767661; Mayer et al., 2006, PMID: 16678107; Schmitz et al., 2010, PMID: 20952535]. Furthermore, tracking chromatin remodelling complexes—such as the CSB-PCAF complex investigated by the different groups—has shown that active depletion of histone acetyltransferases leads to a direct loss of H3K9ac specifically at the primary 47S initiation site, causing a failure in Pol I loading [Shen et al., 2013, PMID: 23667505; Vintermist et al., 2011, PMID: 21559432]. Besides these reports, a wealth of independent functional data highlights that key chromatin regulators dynamically target and modulate the core 47S promoter region to dictate ribosomal transcription initiation. Some examples are referred to here [Xei et al., 2012, PMID: 22570494; Wang et al., 2011, PMID: 21960006; Zhu et al., 2010, PMID: 20531378; Frescas et al., 2007, PMID: 17994099].
Our investigation of KMT2A directly builds upon this functional promoter-centric model. While we appreciate that mapping variations exist between custom rDNA reference models in the literature, our data show clear and reproducible levels of H3 methylation and acetylation dynamics across the primary regulatory sites (including the 47S promoter and adjacent transcribed unit) upon KMT2A depletion. These changes correlate directly with functional changes in 47S pre-rRNA output, supporting a model where KMT2A-mediated histone modifications operate directly at the core transcription initiation machinery to regulate overall nucleolar output.
We acknowledge that the distribution of histone modifications across the rDNA repeat and their functional dependence on KMT2A represent distinct questions. Accordingly, we interpret the ChIP-seq profiles in conjunction with our site-specific ChIP-qPCR and functional experiments rather than relying exclusively on the genome-wide mapping data. We therefore conclude that our results support a role for KMT2A in regulating the acetylation-associated chromatin environment and transcriptional activity of the rDNA promoter, without claiming that KMT2A depletion reduces all active histone modifications across the locus.
We will clarify the relevant analysis and the scope of our conclusions in the revised manuscript.
The KMT2A deletion drastically reduces PolI and UBF (UBTF) engagement across the rDNA, and consistent with this, SL1 and RRN3 engagement is reduced as are 47S levels. Thus, whatever the 7-day KMT2A deletion is doing to the cells it is certainly inducing a very significant shut-down of active rDNA. How such an extreme experimental model can tell anything of KMT2A functions is unclear.
We feel that this concern is on similar lines as that stated in comment 1. We have already explained our inducible knockout model above. The cell-cycle profile of KMT2A-inducible knockout (iKO) cells reported in Malik et al. (2023), PMID: 37379335. provides relevant context for interpreting these effects.
We further wish to highlight that contrary to expectation, the H3K4me2/3 levels are largely unaffected on the rDNA in KMT2A iKO cells, which is surprising but not unexpected given the role of KMT2A in selectively trimethylating specific gene loci (PMID: 19703992). However, the loss of acetylation on H3 is very specific to the study at hand.
Unchanged H3K4me2/3 levels, along with selectively decreased acetylation, have not been reported for a general shutdown of active rDNA under extreme conditions like serum starvation etc. In case of a non-specific effect, both active methylation and acetylation marks are expected to go down. The presence of the H3K4me2/3 mark in our iKO cells acts as an internal control, confirming that the loss of H3 acetylation is a direct consequence of KMT2A depletion.
Furthermore, in the current manuscript, the effects of KMT2A deletion on the Pol I transcriptional machinery can be rescued by the KMT2A catalytic point mutant carrying the N3906A mutation but not by the mutant with TAD deletion. This rescue provides important evidence that the phenotype is linked to a specific function of KMT2A rather than being a nonspecific consequence of KMT2A deletion.
Our findings, considered alongside Lone et al. (2026) PMID: 42096476, support distinct roles for KMT2A and KMT2F in regulating rDNA transcription. In particular, the effects of KMT2A loss occur without a substantial reduction in rDNA H3K4me3, distinguishing its role from the H3K4me3-dependent function of KMT2F in promoting Pol I pre-initiation complex formation.
In the last results section, the authors further claim "Loss of KMT2A results in a stalled RNA Pol I complex at the spacer promoter" (Figure 4 and S4). To support this the authors try to map spacer promoter transcripts but mistakenly claim that the amplicon #1 can detect these transcripts. In fact, this amplicon overlaps the Spacer Promoter initiation site and so could not detect transcripts coming from this promoter. Published data, not cited, have demonstrated that PolI stalling at the spacer promoter occurs as a normal phenomenon on the active rDNA. Its loss is associated with suppression, not the opposite as inferred in this manuscript.
We thank the Reviewer for raising this point. We have double-checked that amplicon #1 is downstream of the spacer promoter start site. However, since the reviewer has raised this point, we will look into it in greater detail and remap our amplicons. We will also revisit the previous literature and compare our results with the published findings.
Reviewer #2 (Public review):
Summary:
This important manuscript presents solid evidence that the major contribution of KMT2A to ribosomal gene transcription is through promoting histone acetylation rather than H3K4 methylation. It is suggested this occurs in part through interactions of KMT2A with PCAF.
Strengths:
This manuscript advances an important concept that chromatin-modifying enzymes can also function to promote transcription through interactions with other transcriptional regulators independently of their catalytic functions. The data are overall supportive of this conclusion for KDM2A.
Weaknesses:
(1) Care needs to be taken regarding interpreting the lack of effect of depleting KMTs or HATs on histone acetylation or methylation levels. There are several overlapping HATS or KMTs that generate these modifications. The most that can be said from these negative results is that the depleted HAT or KMT is not the sole enzyme generating these modifications at these loci.
We thank the Reviewer for this important comment. We agree that histone modifications are regulated by multiple enzymes with overlapping or partially redundant activities, and that the absence of a detectable change following depletion of an individual KMT or KAT does not exclude a contribution from that enzyme to the regulation of the modification.
Specifically, the lack of a detectable reduction in H3K4me3 following KMT2A depletion indicates that KMT2A depletion alone is insufficient to produce a detectable decrease in H3K4me3 under the conditions examined. The persistence of this mark may reflect the activity of other KMTs and hence functional redundancy. In contrast, our observation that KMT2A depletion reduces H3K9 acetylation, together with our findings implicating PCAF in KMT2A-dependent rDNA transcription, supports a role for KMT2A in regulating an acetylation-associated mechanism at this locus.
Nevertheless, we acknowledge that these findings do not, by themselves, establish that PCAF is the sole enzyme responsible for H3K9 acetylation or that it accounts for all KMT2A-dependent effects on rDNA transcription. We will revise the interpretation of our results to avoid attributing the maintenance or loss of a histone modification exclusively to a single enzyme.
(2) The biochemistry supporting an interaction between KDM2A and PCAF is under-developed and not strong enough to support this conclusion at this point.
We thank the reviewer for their comment. Based on these comments and those from Reviewer 3, we plan to add more experiments to strengthen the KMT2A-PCAF interaction, including a reverse pull-down.
(3) It is concluded that acetylation causes increased rDNA transcription. However, acetylation may be a consequence of transcription instead, especially over transcribed regions.
We thank the reviewer for raising this important point regarding the relationship between histone acetylation and rDNA transcription. We agree that increased histone acetylation over transcriptionally active regions may either contribute to transcriptional activation or result from ongoing transcription. These possibilities are not mutually exclusive.
Our study provides several lines of evidence supporting a functional relationship between KMT2A and PCAF-associated histone acetylation, which indicate that acetylation is the cause and not a consequence of rDNA transcription. KMT2A depletion reduces rDNA transcription and is accompanied by a reduction in H3K9 acetylation. In addition, we present a model in which KMT2A facilitates the transfer of RNA Pol I PIC from the spacer promoter to the 47S promoter following its assembly at the spacer promoter by recruiting PCAF to rDNA through its TA domain. Together, these observations support that acetylation supports efficient rDNA transcription
Nevertheless, we acknowledge that these findings do not completely exclude the possibility that some of the observed reduction in histone acetylation occurs downstream of reduced transcription and will modify the discussion accordingly.
(4) Some conclusions, for example implicating PCAF, are based on overexpression experiments that are very artificial and difficult to draw conclusions from.
We thank the reviewer for their comment. We agree that overexpression experiments can produce non-physiological protein levels and potentially influence experimental readouts. However, our interpretation of PCAF involvement is based on a broader set of observations presented in this study, including the effects of KMT2A depletion on rDNA transcription and histone acetylation, together with our experiments examining the contribution of individual histone acetyltransferases to rRNA synthesis. These findings identify PCAF as a key co-activator of KMT2A-dependent transcriptional activation.
Therefore, the overexpression experiments have been used to provide mechanistic support rather than definitive evidence that endogenous PCAF is the sole or essential mediator of this process.
(5) Several figures are generated using publicly available data sets that were reanalyze using updated reference genomes, sometimes from different cell lines/types. This seems to be a useful exercise but the methods should include information regarding the robustness of these data sets.
We thank the Reviewer for their concern; we will update this information in the Methods.
Reviewer #3 (Public review):
Summary:
This manuscript examines a key mechanism by which KMT2A regulates RNA polymerase I transcription at ribosomal DNA. The authors suggest that KMT2A facilitates the recruitment of PCAF to rDNA. This process enhances histone acetylation and supports assembly of the RNA polymerase I transcription machinery. The study addresses an intriguing question by proposing a catalytic-independent role for KMT2A in rDNA transcription, potentially linking the KMT2A-dependent pathway to histone acetylation. However, several aspects of the proposed mechanism require further experimental validation. Notably, while the current evidence demonstrates associations among KMT2A, PCAF occupancy, histone acetylation, and rRNA transcription, it does not fully support the proposed objectives.
Strengths:
This could be the first report of a lysine methyltransferase regulating an acetyltransferase in Pol I-mediated gene expression.
Weaknesses:
(1) The causal relationship between KMT2A and the recruitment of PCAF needs to be demonstrated. Simply showing reduced PCAF occupancy after depleting KMT2A does not, on its own, prove direct recruitment. The authors should investigate whether KMT2A and PCAF physically interact, and whether this interaction depends on DNA or RNA.
As we have shown that KMT2A and PCAF interact by endogenous IP (Fig. S4D) and pull-down (Fig. S4E) experiments with the KMT2A C-terminal fragment containing the TAD domain, does the Reviewer mean in vitro interaction? Both proteins are large and may be difficult to produce in bacteria.
We do plan to perform reverse IP as suggested by Reviewer 2, and during these experiments, we can plan to test if this interaction is promoted by DNA or RNA by testing it in the presence or absence of DNase or RNase.
(2) The main prediction of the model is that restoring PCAF activity should reverse the molecular effects caused by the depletion of KMT2A. The authors should test whether PCAF restoration can mitigate the consequences of KMT2A loss. They should also assess PCAF occupancy, H3K9 acetylation levels, Pol I occupancy, and nascent rRNA synthesis.
We agree with the Reviewer that this is the central hypothesis of our study. Ideally, a simple global restoration or overexpression of PCAF should be able to rescue the effects caused by the depletion of KMT2A. However, based on our model and existing literature, we expect the following issues:
(1) We propose that KMT2A serves as a recruiter, a physical scaffold, to target PCAF to the rDNA. Therefore, mere overexpression of PCAF, in the absence of KMT2A, will not be able to ensure its localisation to the rDNA locus.
(2) Transcription activation by KMT2A involves tight synergy between its TA domain and specific KATs (PMID: 24253305; PMID: 31515273). The KMT2A TAD contains intrinsically disordered regions that must directly engage co-activator docking sites—specifically interacting with domains like the KIX domain of p300/CREB-binding protein acetyltransferase complexes—to transition into a stable secondary structure and drive downstream chromatin remodelling. Because this functional synergy requires the physical presence of both the KMT2A TAD and the KAT complex at the promoter, in our opinion, restoring PCAF activity in a KMT2A-depleted background cannot rescue localized H3K9 acetylation at the rDNA promoter. While we are not sure how PCAF engages with the KMT2A TAD domain, based on the above literature, the physical presence of KMT2A may be required.
We will test this hypothesis by overexpressing PCAF in the absence (or presence) of KMT2A, monitoring rRNA expression, and then plan to undertake future experiments accordingly.
(3) The catalytic-independent function of KMT2A should be directly investigated. The absence of a significant change in H3K4me3 levels following KMT2A depletion does not confirm that KMT2A's catalytic activity is not essential. Rescue experiments comparing wild-type KMT2A with a catalytically inactive variant would provide a much stronger assessment of the proposed non-catalytic mechanism.
We thank the reviewer for their comment. We completely agree that it is important to rule out whether catalytic activity of KMT2A is essential for rRNA transcription. We have shown in Fig. 4G that the catalytic point mutant carrying the N3906A mutation is able to recruit PCAF and RNA Pol I to the rDNA but not the KMT2A∆TAD mutant. This rescue experiment supports our hypothesis that the catalytic activity of KMT2A is not involved in mediating transcription at rDNA, but TA domain-dependent recruitment of PCAF is critical.
(4) The manuscript highlights H3K9ac; however, various acetyltransferases and histone acetylation marks are linked to active rDNA. The authors should investigate whether KMT2A specifically regulates H3K9ac or leads to broader changes in histone acetylation at rDNA.
We thank the reviewer for their comment. This is indeed an insightful suggestion. As shown in Fig 3. Multiple histone acetylation marks decorate the rDNA, and it is not possible for us to test them all. We propose to test some of these marks during revision to gain insight into the range of KMT2A-mediated regulation.
(5) Due to the highly repetitive nature of rDNA, the manuscript must include detailed information about the reference assembly, mapping strategy, management of multimapping reads, normalisation processes, and biological replicates. The authors should demonstrate that any observed changes in occupancy are not due to technical differences in rDNA mapping or variations in copy number.
We thank the reviewer for their comment. As indicated in the manuscript, we have used the pipeline from George et al. al, 2023, PMID: 37121547, which describes all these parameters in detail. Hence, we have omitted a repetition here but added a brief sentence in the Methods now.
(6) KMT2A regulates a wide range of transcriptional programs, and its depletion can affect proliferation, cell-cycle distribution, or nucleolar function. The authors should include appropriate controls to demonstrate that the rDNA phenotype is not a result of changes in cellular proliferation or a generalised response to transcriptional or nucleolar stress.
We thank the reviewer for highlighting this important point. The transcriptional profiling recently published by our lab [Lone et al. (2026), PMID: 42096476; Malik et al., 2023, PMID: 37379335] provides clear evidence that helps distinguish direct regulatory effects from generalized stress responses. Our global transcript analysis indicates that KMT2A depletion specifically downregulates transcription at the rDNA repeats without inducing hallmark markers of nucleolar stress or broad cell-cycle arrest.
(7) The CSB and B-WICH complex have previously been linked to the recruitment of PCAF and the acetylation of H3K9 at active rDNA. The authors should investigate whether KMT2A operates independently of these factors or if KMT2A functions within an established pathway for remodeling rDNA chromatin.
We agree with the Reviewer that CSB and the B-WICH complex, PCAF recruitment, and H3K9ac deposition are indeed a well-documented axis at active rDNA. In the present study, our focus is on establishing the functional role of KMT2A at the rDNA locus. Given that KMT2A traditionally acts as a histone methyltransferase that often precedes or cooperates with acetyltransferases like PCAF, it is highly plausible that KMT2A operates sequentially or in concert with the B-WICH pathway to remodel chromatin.
However, mapping these biochemical interactions is beyond our current aims. Nonetheless, we agree this is a crucial conceptual link, and we will expand this link in our Discussion.
(8) The distinction between active and silent rDNA should be taken into account. The authors need to determine whether KMT2A/PCAF occupancy is preferentially associated with active rDNA copies. Sequential ChIP analysis or the use of established markers for active, poised, and silent rDNA would help clarify the chromatin context in which the proposed mechanism operates.
We appreciate the Reviewer’s suggestion to determine if KMT2A/PCAF occupancy is restricted to active rDNA copies. While this is an important factor to consider, making a clear distinction is currently not possible due to several constraints. First, the highly repetitive nature of rDNA repeats makes it difficult to differentiate active from silent fractions both experimentally and bioinformatically. Second, performing sequential ChIP to definitively map this co-occupancy is technically challenging and establishing and optimising this protocol within the revision timeframe will be extremely difficult.
Furthermore, the exact histone landscapes defining active, poised, and silent rDNA are not yet fully characterized; much of the existing consensus in epigenetic literature relies heavily on observations drawn from RNA Polymerase II systems rather than Pol I repeats. As demonstrated by Xie et al., 2012 [PMID: 22570494], rDNA promoters frequently feature bivalent marks—bearing both active H3K4me3 and repressive H3K27me3 modifications—that characterize a poised, non-transcribing state. This biological ambiguity makes absolute active/silent categorization extremely difficult in our case.
References:
(1) Chinchole A, Lone KA, Tyagi S. MLL regulates the actin cytoskeleton and cell migration by stabilising Rho GTPases via the expression of RhoGDI1. Journal of Cell Science. 2022;135(20):jcs260042. doi:10.1242/jcs.260042 PMID: 36111497
(2) Frescas D, Guardavaccaro D, Bassermann F, Koyama-Nasu R, Pagano M. JHDM1B/FBXL10 is a nucleolar protein that represses transcription of ribosomal RNA genes. Nature. 2007;450(7167):309-313. doi:10.1038/nature06255 PMID: 17994099
(3) Garg A, Orru R, Ye W, et al. Structural and mechanistic insights into the interaction of the circadian transcription factor BMAL1 with the KIX domain of the CREB-binding protein. J Biol Chem. 2019;294(45):16604-16619. doi:10.1074/jbc.RA119.009845 PMID: 31515273
(4) George SS, Pimkin M, Paralkar VR. Construction and validation of customized genomes for human and mouse ribosomal DNA mapping. J Biol Chem. 2023;299(6):104766. doi:10.1016/j.jbc.2023.104766 PMID: 37121547
(5) Herdman C, Mars JC, Stefanovsky VY, et al. A unique enhancer boundary complex on the mouse ribosomal RNA genes persists after loss of Rrn3 or UBF and the inactivation of RNA polymerase I transcription. PLoS Genet. 2017;13(7):e1006899. doi:10.1371/journal.pgen.1006899 PMID: 28715449
(6) Kühn MWM, Song E, Feng Z, et al. Targeting Chromatin Regulators Inhibits Leukemogenic Gene Expression in NPM1 Mutant Leukemia. Cancer Discov. 2016;6(10):1166-1181. doi:10.1158/2159-8290.CD-16-0237 PMID: 27535106
(7) Lone KA, Karole AM, Ravindran G, Tyagi S. The KMT2F histone methyltransferase interacts with the RNA polymerase I machinery to promote ribosomal RNA transcription. PLOS Biology. 2026;24(5):e3003785. doi:10.1371/journal.pbio.3003785 PMID: 42096476
(8) Malik KK, Sridhara SC, Lone KA, Katariya PD, Pulimamidi D, Tyagi S. MLL methyltransferases regulate H3K4 methylation to ensure CENP-A assembly at human centromeres. PLOS Biology. 2023;21(6):e3002161. doi:10.1371/journal.pbio.3002161 PMID: 37379335
(9) Mars JC, Sabourin-Felix M, Tremblay MG, Moss T. A Deconvolution Protocol for ChIP-Seq Reveals Analogous Enhancer Structures on the Mouse and Human Ribosomal RNA Genes. G3 Genes|Genomes|Genetics. 2018;8(1):303-314. doi:10.1534/g3.117.300225 PMID: 29158335
(10) Mayer C, Schmitz KM, Li J, Grummt I, Santoro R. Intergenic Transcripts Regulate the Epigenetic State of rRNA Genes. Molecular Cell. 2006;22(3):351-361. doi:10.1016/j.molcel.2006.03.028 PMID: 16678107
(11) McKinley KL, Cheeseman IM. Large-Scale Analysis of CRISPR/Cas9 Cell-Cycle Knockouts Reveals the Diversity of p53-Dependent Responses to Cell-Cycle Defects. Developmental Cell. 2017;40(4):405-420.e2. doi:10.1016/j.devcel.2017.01.012 PMID: 28216383
(12) Ning M, Hoshii T, Nakagawa T, et al. Non-catalytic role of SETD1A promotes gastric cancer cell proliferation through the E2F4-TAF6 axis in the cell cycle. Cell Death Dis. 2025;16(1):639. doi:10.1038/s41419-025-07976-4 PMID: 40846851
(13) Perlee S, Kikuchi S, Nakadai T, et al. SETD1A function in leukemia is mediated through interaction with mitotic regulators BuGZ/BUB3. EMBO Rep. 2023;24(10):e57108. doi:10.15252/embr.202357108 PMID: 37535603
(14) Santoro R, Li J, Grummt I. The nucleolar remodeling complex NoRC mediates heterochromatin formation and silencing of ribosomal gene transcription. Nat Genet. 2002;32(3):393-396. doi:10.1038/ng1010 PMID: 12368916
(15) Santoro R, Grummt I. Epigenetic mechanism of rRNA gene silencing: temporal order of NoRC-mediated histone modification, chromatin remodeling, and DNA methylation. Mol Cell Biol. 2005;25(7):2539-2546. doi:10.1128/MCB.25.7.2539-2546.2005 PMID: 15767661
(16) Schmitz KM, Mayer C, Postepska A, Grummt I. Interaction of noncoding RNA with the rDNA promoter mediates recruitment of DNMT3b and silencing of rRNA genes. Genes Dev. 2010;24(20):2264-2269. doi:10.1101/gad.590910 PMID: 20952535
(17) Shen M, Zhou T, Xie W, et al. The chromatin remodeling factor CSB recruits histone acetyltransferase PCAF to rRNA gene promoters in active state for transcription initiation. PLoS One. 2013;8(5):e62668. doi:10.1371/journal.pone.0062668 PMID: 23667505
(18) Thakur JK, Yadav A, Yadav G. Molecular recognition by the KIX domain and its role in gene regulation. Nucleic Acids Res. 2014;42(4):2112-2125. doi:10.1093/nar/gkt1147 PMID: 24253305
(19) van Schie JJM, de Lint K, Molenaar TM, et al. CRISPR screens in sister chromatid cohesion defective cells reveal PAXIP1-PAGR1 as regulator of chromatin association of cohesin. Nucleic Acids Res. 2023;51(18):9594-9609. doi:10.1093/nar/gkad756 PMID: 37702151
(20) Vintermist A, Böhm S, Sadeghifar F, et al. The chromatin remodelling complex B-WICH changes the chromatin structure and recruits histone acetyl-transferases to active rRNA genes. PLoS One. 2011;6(4):e19184. doi:10.1371/journal.pone.0019184 PMID: 21559432
(21) Wang P, Lin C, Smith ER, et al. Global Analysis of H3K4 Methylation Defines MLL Family Member Targets and Points to a Role for MLL1-Mediated H3K4 Methylation in the Regulation of Transcriptional Initiation by RNA Polymerase II. Molecular and Cellular Biology. 2009;29(22):6074-6085. doi:10.1128/MCB.00924-09 PMID: 19703992
(22) Wang W, Chen Z, Mao Z, et al. Nucleolar protein Spindlin1 recognizes H3K4 methylation and stimulates the expression of rRNA genes. EMBO Rep. 2011;12(11):1160-1166. doi:10.1038/embor.2011.184 PMID: 21960006
(23) Xie W, Ling T, Zhou Y, et al. The chromatin remodeling complex NuRD establishes the poised state of rRNA genes characterized by bivalent histone modifications and altered nucleosome positions. Proc Natl Acad Sci U S A. 2012;109(21):8161-8166. doi:10.1073/pnas.1201262109 PMID: 22570494
(24) Yuen G, Khan FJ, Gao S, et al. CRISPR/Cas9-mediated gene knockout is insensitive to target copy number but is dependent on guide RNA potency and Cas9/sgRNA threshold expression level. Nucleic Acids Research. 2017;45(20):12039-12053. doi:10.1093/nar/gkx843 PMID: 29036671
(25) Zentner GE, Saiakhova A, Manaenkov P, Adams MD, Scacheri PC. Integrative genomic analysis of human ribosomal DNA. Nucleic Acids Res. 2011;39(12):4949-4960. doi:10.1093/nar/gkq1326 , PMID: 21355038
(26) Zentner GE, Balow SA, Scacheri PC. Genomic Characterization of the Mouse Ribosomal DNA Locus. G3 Genes|Genomes|Genetics. 2014;4(2):243-254. doi:10.1534/g3.113.009290 PMID: 24347625
(27) Zhu Z, Wang Y, Li X, et al. PHF8 is a histone H3K9me2 demethylase regulating rRNA synthesis. Cell Res. 2010;20(7):794-801. doi:10.1038/cr.2010.75 PMID: 20531378
eLife Assessment
This study provides invaluable insights into the field of genomic prediction of traits by demonstrating how machine learning algorithms can integrate genomic and transcriptomic data and handle nonlinearity, providing important comparative prediction results across environments. The rationale and analyses conducted to integrate the two types of ~omics datasets in two environments are solid. The work will be of interest to plant geneticists and breeders, especially those interested in the genetics of environmental adaptation.
Reviewer #2 (Public review):
The authors examine the relative contributions of SNP and transcriptomic information to predicting trait variation and phenotypic plasticity in switchgrass. Using a diversity panel of 426 genotypes evaluated in Texas and Michigan, USA., this study demonstrates that transcriptomic predictors could outperform genomic markers in predicting environmental differences for traits like biomass yield and tiller count, highlighting the trait-dependent advantage of transcriptomic information over SNPs. However, combining SNPs and transcriptomic predictors generally yielded limited improvement in prediction accuracy.
A principal limitation of this study concerns the experimental design and the robustness of transcriptomic predictors. With only one transcriptomic sample per genotype in each environment, within-genotype variation in transcript abundance cannot be estimated, making it difficult to distinguish genotype-specific expression responses from individual-plant variation. Unlike SNP genotypic information, which provides a stable representation of a genotype's inherited genetic variation underlying phenotypic differences, transcript abundance is highly sensitive to environmental conditions, developmental stages, and physiological state at the time of sampling. As a result, the stability of the transcriptomic predictors and their predictive performance across independent environments, years, or sampling conditions remain untested. Furthermore, the stability of feature selection across different training populations has not been established, raising additional questions about the reproducibility of the identified candidate genes. Together, these limitations constrain their generalizability and the biological interpretation of transcriptomic associations with phenotypic plasticity.
Overall, this study provides valuable comparative evidence regarding the predictive performance of genomic and transcriptomic information. Its principal contribution lies in demonstrating that the relative predictive advantages of these information sources are trait-dependent and their integration does not necessarily improve prediction accuracy. While the predictive findings are supported, the biological interpretations extend beyond the evidence presented, as the identified transcriptomic predictors and their interactions do not establish causal genetic effects, epistasis, or direct regulatory relationships. Therefore, the proposed genetic mechanisms and environment-specific interactions should be regarded as hypotheses requiring independent biological validation, rather than established mechanistic findings.
Author response:
The following is the authors’ response to the original reviews.
eLife Assessment:
The study by Izquierdo and colleagues provides important insights into the field of genomic and transcriptomic prediction of traits across multiple environments. The rationale and analyses conducted to integrate the two types of ~omics datasets across two environments are solid. However, some clarification would be appreciated in the presentation of the results, and adding some statistical control to clarify how the predictors were selected, or assessing their importance using the SHAP framework, would further consolidate the findings.
We appreciate the positive feedback and the constructive evaluation of our work. In response, we have added further clarification to the Results section and provided additional detail on the feature selection procedure to facilitate understanding of how predictors were identified within the modeling framework.
Public Reviews:
Reviewer #1 (Public review):
Summary:
P. Izquierdo et al. investigated the genetic determinism of various traits of interest in switchgrass using large-scale genomic and transcriptomic data. More specifically, they worked on a diversity panel comprising 426 genotypes evaluated in common-garden experiments at two locations (Michigan and Texas). The phenotypic and genomic data were already published. In this work, they produced transcriptomic data for each of the 426 genotypes at each site, and they carried out phenotype predictions using genomic and transcriptomic data separately or together. While they were moderately correlated at each location, both omic information appeared to be complementary for the prediction of phenotype. To further exploit the fact that they have data across two locations, they computed differences for phenotypes and transcripts between locations as indicators of trait and transcript plasticity, respectively. They built predictive models of trait plasticity using genomic information and transcript plasticity, which proved to be quite accurate for traits affected by GxE. Finally, they made use of SHAP values from predictive models of flowering time and biomass at each location, as well as for their plasticity, to gain insight into their genetic determinism. These SHAP values provide the importance of the predictive features (SNP and/or transcripts) for trait prediction. This allowed them to confirm some candidate genes and to propose new candidates for both traits.
We appreciate the reviewer's thorough and constructive summary of our work. We would like to clarify that while the genomic data were previously published, the transcriptomic and phenotypic data were generated as part of the present study. We have revised the text to make this distinction explicit.
Strengths:
I found this study interesting and rich. I think the sample size (426 genotypes) is large enough to support the findings. The use of a modern machine-learning approach (XGBoost) together with SHAP indices to find interesting features and get insights into the biological mechanisms underlying flowering time and biomass production is quite original. The methodology employed is globally sound. I also like the fact that the authors accounted implicitly for the population structure by providing a baseline prediction using the first 5 PCs.
Weaknesses:
While the methodology is globally sound, I sometimes had difficulties following exactly what was done. This is partly due to the fact that the authors used 2 omics (SNPs and transcripts) to predict phenotypes, and sometimes, in the results, it is not clear which of the 2 is the focus. This was especially the case for the importance of the features and the interpretability of the models, where I found it sometimes hard to tell whether the analysis was done on SNPs or transcripts.
We have revised the text throughout to clarify when models using SNP, transcript (T), or both data types (S+T) were the focus of the analysis.
Also, regarding the methodology, I did not understand why the authors needed to perform a feature selection approach. Maybe it was required to perform the interaction analysis, which could not be deployed on all the features? But regarding the importance of the features, I do not get the added value of the selection over the direct use of SHAP indices when using all features. Maybe this is because I am not a specialist in this kind of approach, but maybe the authors could add more details to explain the rationale behind the feature selection.
Feature selection was performed to identify the subset of features that optimized predictive accuracy while reducing model complexity. By retaining only the most informative features, the resulting models achieve higher predictive accuracy than models trained on all features, and model interpretation is simplified. The rationale and procedure are described.
Regarding SHAP interactions specifically, the number of features used to estimate SHAP interaction scores was limited to 5,000 due to the high computational demands of this analysis. Interaction scores are estimated for each individual in the training population, producing a 5,000 × 5,000 interaction matrix per individual. With 306 individuals in the training population and 10 independent model runs, this corresponds to 3,060 matrices of this size.
Reviewer #2 (Public review):
Summary:
The authors aimed to evaluate whether integrating genomic (SNP) and transcriptomic information with machine learning can improve phenotypic prediction of polygenic traits across environments. The manuscript explored not only the predictability across models and predictor feature sets, but also attempted to identify meaningful genes and interactions underlying trait variation.
Strengths:
The main strength of the manuscript is its integration of SNP, transcriptomic, and phenotype datasets for 426 sorghum genotypes between Texas and Michigan. It provides a systematic comparison of predictor types (SNP versus transcriptomic abundance) and model strategies to integrate them.
Weaknesses:
(1) Experimental Design
The experimental design raises several concerns that should be clarified before strong biological conclusions are drawn from the transcriptomic analyses.
First, the transcriptomic sampling is not well aligned with the developmental stages most relevant to the phenotypes being modeled. Leaf tissue was collected at a single time point in each environment, whereas traits such as flowering time, biomass, tiller count, and panicle height arise from developmental processes occurring over extended and potentially distinct temporal windows. Consequently, the measured expression profiles are likely to reflect physiological states specific to the sampling dates (May 5-6 in Texas and June 22-24 in Michigan) rather than the regulatory processes underlying the target phenotypes.
We selected the pre-flowering stage because it represents a critical developmental transition during which plants remain largely vegetative while initiating reproductive development. Gene expression at this stage is therefore expected to capture regulatory processes associated with flowering initiation while also reflecting pathways involved in vegetative growth and biomass accumulation. In our previous work in maize (Azodi et al., 2020), we demonstrated that transcriptomic profiles obtained at a single developmental stage can contain information predictive of complex traits. Although the expression profiles analyzed here do not capture the full temporal dynamics underlying traits such as flowering time and biomass, they provide a biologically meaningful snapshot of molecular variation relevant to these traits. We have revised the manuscript to acknowledge this limitation and to clarify that our findings should be interpreted as associations between gene expression measured at the pre-flowering stage and the phenotypes of interest, rather than as evidence that all identified genes directly regulate these traits throughout development.
Second, the phrase "haphazardly randomized" is questionable for a field experiment. It is unclear whether the design included formal randomization, blocking, row/column structure, or spatial correction. Without explicit accounting for spatial field heterogeneity, environmental variation within sites may confound genotype and transcriptomic effects.
We thank the reviewer for this comment and agree that the original wording was unclear. The text has been revised to more precisely describe the experimental design. The establishment and management of the common gardens are described in detail in Lovell et al. (2021). Briefly, the diversity panel was established as a single-block planting at each location, with genotypes distributed throughout the field and arranged in a honeycomb design in which each individual had four nearest neighbors located 1.5 m apart. We have removed the phrase "haphazardly randomized," which was imprecise.
Because only a single replicate of each genotype was established at each site, formal spatial correction was not performed, as genotype and within-site spatial effects cannot be estimated independently in this design. We acknowledge that additional replication would have improved our ability to account for fine-scale environmental heterogeneity.
However, several aspects of the experimental design reduce the likelihood that spatial variation substantially confounded the results. The diversity panel is part of a coordinated common garden network spanning ten field sites across the United States, designed primarily to evaluate genotype-by-environment interactions across contrasting environments rather than within-site spatial variation. In addition, the AP13 reference genotype was distributed throughout the experimental plots and did not show evidence of substantial spatial heterogeneity at the Michigan site, suggesting that within-site environmental variation was limited. We have added text acknowledging this limitation and its implications for interpreting genotype and genotype-by-environment effects.
Third, the Methods do not clearly describe biological replication for RNA-seq. If each genotype-by-environment combination were represented by a single transcriptomic sample, then within-genotype expression variance cannot be estimated. This is important because transcript abundance is highly sensitive to microenvironment, sampling time, tissue status, developmental stage, and technical variation. The absence of replication significantly weakens confidence in gene-level feature importance and gene-gene interaction claims.
We appreciate this thoughtful comment and fully acknowledge that the absence of biological replication for RNA-seq within each genotype-by-environment combination is an important limitation of this study. Because the diversity panel consisted of a single plant per genotype at each field site, transcriptomic data were generated from one biological sample per genotype-by-environment combination, and within-genotype expression variance could not be directly estimated. We recognize that transcript abundance can be influenced by microenvironmental conditions, developmental stage, tissue status, sampling time, and technical variation, all of which may affect confidence in gene-level feature importance and inferred interaction effects.
The scale of the experiment, involving 426 genotypes evaluated across contrasting common gardens, made biological replication for RNA-seq financially and logistically impractical. This experimental design was established to maximize representation of genetic diversity and environmental variation across a large multi-environment common garden network rather than to obtain replicated transcriptomic measurements within individual sites. To minimize uncontrolled variation, tissue was collected at both locations at approximately the same pre-flowering developmental stage, and standardized protocols for tissue collection, library preparation, sequencing, and data processing were applied across all samples.
Although individual genotype-by-environment combinations were not biologically replicated, the analyses leverage substantial replication at the level of alleles and population structure, as individual alleles and transcripts are represented across many genetically distinct accessions spanning multiple ecotypes. This is analogous to many human population-scale transcriptomic studies, where each genotype is typically represented by a single biological sample and inference is made across genetically diverse individuals rather than replicated measurements of the same genotype.
Importantly, the goal of this study was to identify transcriptomic patterns that are predictive of phenotypic variation across genotypes and environments, rather than to establish causal regulatory relationships. Accordingly, the genes and interaction patterns identified through model interpretation should be viewed as candidate predictive associations and hypothesis-generating signals that warrant validation in future studies with replicated sampling and controlled experiments. We have revised the manuscript to more clearly acknowledge these limitations and to align the interpretation of gene importance and interaction analyses with the predictive scope of the study.
Four, the analysis of expression differences across environments is based on a simple subtraction (TX - MI) followed by correlation with genetic similarity. This approach is not standard in transcriptomic analysis and does not account for variability, replication, or statistical uncertainty. Conventional methods for assessing differential expression and genotype-by-environment interactions rely on model-based frameworks that explicitly estimate variance components and test for interaction effects. Without such modeling, the observed expression differences may reflect noise or confounding factors rather than genotype-driven responses.
We appreciate this comment. Our approach to estimating expression differences across environments was intended to quantify transcriptomic plasticity and evaluate whether variation in gene expression responses across environments could help explain and predict phenotypic plasticity. While this is not a conventional differential expression framework, our objective was not to identify differentially expressed genes or formally test genotype-by-environment interaction effects at the transcript level.
Expression differences were calculated as TX − MI using log-transformed expression values. On the logarithmic scale, this difference is equivalent to the log expression ratio between environments and therefore provides a straightforward measure of the magnitude and direction of transcriptomic responses across locations. We used this metric as a descriptive measure of transcriptomic plasticity that could be related to genetic similarity and incorporated into predictive models of phenotypic plasticity. Consistent with this rationale, models based on expression differences across environments were able to predict differences in phenotypic performance between locations.
(2) SHAP contribution values
Although SHAP is a well-established framework for decomposing model predictions into feature-level contributions, its use in this manuscript raises several concerns regarding interpretation, statistical validity, and biological inference.
First, SHAP values quantify the contribution of features within the fitted model, conditional on the joint distribution of inputs and the model structure. They do not represent causal effects or direct biological importance.
We agree that SHAP values do not represent causal effects or direct measures of biological importance. Rather, SHAP quantifies the contribution of features to model predictions within the context of a fitted model and the observed input distribution. Our rationale for using SHAP was not to infer causality, but to identify features that are consistently informative for predicting phenotypic variation. The underlying hypothesis is that features contributing strongly to accurate predictions may be associated with biological processes influencing the trait and can therefore serve as candidate genes or genomic regions for further investigation. This strategy has been widely adopted in genomics and plant biology, where interpretable machine learning methods are used to extract biological insights from predictive models. Accordingly, the genes and interactions highlighted by SHAP should be interpreted as predictive associations and candidate mechanisms requiring validation through independent experiments and functional studies. We have revised the manuscript to more clearly distinguish predictive importance from causal biological effects.
There is a difference where SHAP values are often in log-odds and the regression model uses absolute units.
We thank the reviewer for this comment. We believe there may be a misunderstanding regarding the scale of the SHAP values reported in this study. The models analyzed here are regression models predicting continuous traits, not classification models. Consequently, SHAP values are expressed in the same units as the predicted phenotype and represent the additive contribution of each feature to individual predictions. The interpretation of SHAP values in log-odds units is specific to certain classification models where predictions are generated on the log-odds scale. We have revised the manuscript to clarify this point and avoid potential confusion.
Without a fair evaluation of model fit, the interpretation of SHAP values needs to take a cautious step because a model could fit poorly when a feature shows very high SHAP values.
We agree that SHAP values should be interpreted in the context of model performance. For this reason, model performance was evaluated prior to feature interpretation. Figure 1 presents the predictive performance of all models, and we focused our biological interpretation on flowering time and biomass because these traits are agronomically important and showed relatively stronger predictive performance compared to other traits, providing a more reliable basis for interpreting feature importance. We have revised the manuscript to emphasize that model interpretation should be considered in conjunction with model accuracy.
In genomic data, where features are highly correlated due to linkage disequilibrium and co-expression, SHAP values can distribute contribution values across correlated variables in ways that are not uniquely identifiable. As a result, features highlighted as "important" may reflect correlation structure rather than true functional relevance.
We agree that multicollinearity can complicate the interpretation of feature importance, as correlated features may share predictive information, resulting in importance being distributed among them or assigned preferentially to one feature over another. This issue is particularly relevant in genomic datasets due to linkage disequilibrium among SNPs and co-expression among genes. To reduce redundancy among genetic markers, SNPs were filtered using a linkage disequilibrium threshold of r<sup>2</sup> < 0.1 prior to model training, substantially reducing the number of highly correlated markers. We acknowledge, however, that correlation among features, particularly among co-expressed genes, remains present and may influence SHAP-based feature attribution.
To improve the robustness of feature interpretation, models were trained across 10 independent runs and SHAP values were averaged across runs, reducing sensitivity to individual model fits and providing more stable estimates of feature contributions. Nevertheless, SHAP values should not be interpreted as uniquely identifying the most functionally relevant feature within a correlated group of predictors. The genes and genetic variants highlighted by SHAP should therefore be viewed as candidate predictive features associated with trait variation rather than definitive evidence of causal biological relevance.
This correlative structure can be exacerbated in this manuscript because of the use of TPM-normalized transcript abundances as predictor variables without biological replicates. Assume the estimates of transcript abundances are robust, TPM values are compositional, with a constant-sum constraint that creates dependencies among all genes that induce negative correlations. This issue is particularly relevant for the interpretation of gene importance and interaction effects, where correlated predictors can lead to unstable and non-unique attributions. This biological interpretation of transcript-based features remains uncertain.
We agree that TPM-normalized expression values are compositional and may introduce dependencies among genes due to the constant-sum constraint. TPM normalization was selected because it facilitates comparisons of transcript abundance across samples by accounting for differences in sequencing depth and transcript length. Expression values were further transformed using log<sub>10</sub>(TPM + 1), which reduces skewness and stabilizes variance, making the data more suitable for machine learning analyses. We acknowledge that compositional effects and gene co-expression may influence feature attribution and the interpretation of gene importance. However, the primary objective of this study was predictive rather than inferential. In this context, log-transformed TPM values provide a standardized and comparable representation of transcript abundance across genotypes and environments, serving as consistent input for predictive models. Consequently, genes identified through SHAP analyses should be interpreted as candidate predictive features associated with trait variation, rather than as definitive evidence of direct causal biological importance.
(3) Result interpretation
For example, in page 11, "plasticity SNP- and transcriptomic-based models generally outperformed single-environment models for traits with low cross-environment correlation, such as green-up (Fig. 2c, r = -0.13, p < 8.3 × 10<sup>-3</sup>) and tiller count (Fig. 2f, r = -0.08, p = 0.1) (Supplementary Fig. S1)." is too broad. For green-up, the Diff model appears much better than MI, but not clearly better than TX.
We thank the reviewer for this observation and agree that the original wording was too broad. We have revised the text to more accurately reflect the observed pattern, noting that difference-based models showed relatively stronger predictive performance for traits with low cross-environment correlation, while generally performing worse than single-environment models for traits with high cross-environment correlation. We believe this pattern is biologically meaningful, as transcriptomic differences across environments are expected to be most informative when phenotypic responses differ substantially between locations. We have revised to more accurately characterize these results.
And, same page 11, "...Diffexp was more predictive than SNPs for trait plasticity in biomass, flowering time, and tiller count..." only holds true for biomass, not flowering time, or tiller count.
We thank the reviewer for this observation and agree that the original statement was too strong. Diffexp models showed higher predictive performance for biomass and tiller count, whereas performance for flowering time was comparable between Diffexp and SNP models. We have revised to more accurately characterize these results.
The aspect of "complementary information" between SNP and transcriptomic models in page 12 is stronger than what is supported by Figure 2. Figure 2 shows different predictive performance, but it does not by itself demonstrate complementarity. Establishing complementarity requires evidence that combining SNP+T improves prediction consistently or captures distinct, non-overlapping signals. Yet the preceding section says SNP+T outperformed either single data type in only 15% of cases, with modest gains. This is confusing. Also, there was not G+T in Figure 2; it is SNP+T.
We agree that the term "complementary" was stronger than what the results in Figure 2 directly support. We have revised the text to describe these findings more cautiously, emphasizing that SNP and transcriptomic models may capture partially distinct aspects of biological variation. In the following section, we further address this point by showing that the two model types identify largely distinct sets of predictive features, suggesting that they reflect different sources of information relevant to trait prediction. We also thank the reviewer for noting the SNP+T vs. G+T inconsistency, which has been corrected throughout the manuscript.
Recommendations for the authors:
Reviewer #1 (Recommendations for the authors):
(1) I found that the title does not properly reflect the content of the paper. The term "artificial intelligence" is a bit misleading: I was expecting deep learning-based predictions. GxE and plasticity are clear strengths of the paper that are not present at all in the title. The same holds for multi-omic integration. Maybe you could consider revisiting the title.
We thank the reviewer for this suggestion and have revised the title to: "Multiomic integration using interpretable machine learning reveals genetic mechanisms of trait variation and phenotypic plasticity in switchgrass." This revision highlights the integration of genomic and transcriptomic data, explicitly mentions phenotypic plasticity, and replaces "artificial intelligence" with "interpretable machine learning," which more accurately describes the modeling framework employed.
(2) In general, the quality of the figures is very poor, and it did not allow me to perform a deep assessment of their content, which is sometimes hardly visible.
We agree that figure quality in the submitted PDF was suboptimal. The reduction in image resolution occurred during conversion from the original Word document to PDF. We have uploaded a revised manuscript with higher-resolution figures to ensure all elements are clearly visible.
(3) To estimate plasticity in phenotypes and transcripts, you use differences between locations. Other options have been proposed in the literature, like ratios, for example. I think the choice of the difference is probably a good one, but maybe you could justify it in the introduction.
We chose to estimate phenotypic and transcriptomic plasticity using differences between locations because this metric preserves both the magnitude and direction of environmental responses while avoiding the instability associated with ratio-based measures, particularly when values are small. Furthermore, because transcript abundance values were modeled using log<sub>10</sub>(TPM + 1)-transformed data, differences between environments are conceptually equivalent to log expression ratios, as log(A) − log(B) = log(A/B). This justification has been added to the manuscript.
(4) Figure 1. I wonder why you computed correlation matrices between genotypes for transcriptomic data instead of simply covariance matrices, which would be closer to the genomic relationship matrix.
Our intention in Figure 1 was to visualize overall similarity among samples based on transcriptomic expression profiles across genes. We used Pearson correlation because it is widely applied in transcriptomic analyses to capture relative expression patterns while remaining independent of absolute expression magnitude and scale differences among samples. Covariance matrices, by contrast, are strongly influenced by feature variance and scale, which can complicate interpretation of overall transcriptomic similarity in high-dimensional expression datasets.
(5) Figure 1. You calculate correlations between phenotypic distance and genomic distances for the different traits, but I think this is somehow related to the genomic heritability estimation. Could you please elaborate on this or provide the rationale under your analysis?
We agree that the correlation between genomic relationships and phenotypic similarity is conceptually related to genomic heritability, as both analyses evaluate the extent to which genetic similarity is associated with phenotypic variation. However, their objectives differ. The correlation analysis between SNP, eCor, and phenotypic distance was intended as an exploratory assessment of whether genetically or transcriptomically similar genotypes tend to exhibit similar phenotypic responses across traits and environments. Genomic heritability estimation, in contrast, provides a formal variance component framework to quantify the proportion of phenotypic variance explained by genome-wide marker information. While related conceptually, these analyses address complementary but distinct aspects of the relationship between genomic, transcriptomic, and phenotypic variation.
(6) L163-164: "Because genotypes with similar genetic information are expected to exhibit similar trait value": this is the definition of the heritability, so I would say that this is true to the extent of the heritability of the trait... Maybe you could consider modifying this sentence.
Thank you for this suggestion. Our intention was not to invoke heritability directly, but rather to assess the associations among kinship, eCor, and phenotypic similarity within each environment and to compare the patterns captured by these different data types. We agree that the original wording could be interpreted as implying a deterministic relationship between genetic similarity and phenotypic resemblance. We have revised the text to focus on the association between genetic relatedness and similarity in expression and phenotypic profiles.
(7) Figure 2. For the prediction with transcriptomic data, I understood that you matched data by site for the prediction, which is OK. But I wonder what would happen if you had used transcriptomic data from the other site and vice versa for the prediction, how much predictability you would have lost? I guess that this would somehow be related to the extent of GxE for each trait. Maybe it would be something interesting to look at?
We thank the reviewer for this thoughtful suggestion. While cross-environment prediction is an interesting avenue for future research, our primary objective was not to maximize prediction across environments, as predictive ability would be expected to decrease substantially under strong G×E effects. Instead, we focused on single-environment models to: (1) identify transcriptomic features associated with trait prediction within each environment, and (2) evaluate which features were consistently important across environments. Furthermore, because G×E effects were observed, we developed TX − MI difference models to identify transcriptomic features associated with phenotypic plasticity. We therefore believe that the within-environment and plasticity-focused analyses were more closely aligned with the objectives of this study than cross-environment prediction would have been.
(8) The calculation of local SHAP values for each genotype mentioned L331 and the clustering of genotypes based on such values, as presented in Figure 4, are not presented in the method section. Could you please provide more details?
The estimation of SHAP values is described in the Methods section. Briefly, local SHAP values were calculated for each genotype, and genotypes were subsequently clustered based on these values using hierarchical clustering with Euclidean distance.
(9) Gene-by-gene interactions paragraph: I am a bit confused here regarding which omic was used for looking at the interaction. Naively, I would have gone with the SNPs, but maybe you would rather use the transcripts instead? This would explain why in Figure 5 c-h, the x axes seemed to correspond to transcripts rather than SNPs. Could you please clarify and explain why you focused here on transcripts rather than SNPs?
We thank the reviewer for this comment. To clarify, gene–gene interactions were assessed using transcriptomic models. This choice was motivated by three observations: (1) transcriptomic features were enriched among the most important predictors in the multi-omic models; (2) transcriptomic models identified a greater number of benchmark genes than SNP-based models; and (3) gene expression data provide a more biologically interpretable framework for assessing gene–gene interactions, as transcripts more directly reflect gene activity and regulatory processes than individual genetic variants. We have revised the manuscript to make this rationale explicit.
(10) Materials and Methods:
- Were there replicates of the genotypes within each site?
This diversity panel is part of a multi-institutional project maintaining 10 common gardens across the United States. Given the scale of the experiment and the large number of genotypes evaluated, the design included a single plant per genotype at each site rather than within-site replication. We acknowledge this as a limitation and have discussed it explicitly in the revised manuscript.
- Have the data (phenotypic/transcriptomic) been corrected for spatial effects?
Because only a single replicate per genotype was evaluated at each site, genotype and local spatial effects could not be estimated independently. Applying a spatial correction under these conditions could inadvertently remove true biological variation. The randomized honeycomb design was intended to minimize systematic spatial confounding; however, we acknowledge that residual environmental heterogeneity may still contribute to phenotypic and transcriptomic variation.
- L502-503: Could you please write the model formally instead of R code?
The model has been written formally.
- Genomic heritability estimates: I wonder how useful they really are? I suppose you could have used the model L502-503 to estimate genomic heritability in general across the 2 sites.
We agree that the mixed model described in the revised manuscript could be used to estimate genomic heritability jointly across both sites. In this study, however, genomic heritability was estimated within individual environments to evaluate trait predictability within each location and to assess the heritable component of phenotypic plasticity. Genomic heritability also provides an important benchmark for interpreting prediction results, as the theoretical upper limit of prediction accuracy is approximately proportional to the square root of heritability under quantitative genetic theory.
- L523: I am not sure what LD> 0.1 means. Could you please give more details, e.g., which measure of LD, which software, and which parameters?
We have clarified this point in the Methods. LD pruning was performed using the +prune plugin in BCFtools, with pairwise LD measured as the squared Pearson correlation coefficient (r<sup>2</sup>). SNPs with r<sup>2</sup> > 0.1 within a 5 Mb sliding window were removed to retain a set of more independent markers.
- L526: How many leaves were sampled per plant? Or maybe a particular leaf was chosen?
For each plant, we collected the fully expanded leaves at the highest point on each tiller. The proximal ~4 cm (length of microcentrifuge tube) of that harvested leaf was then flash-frozen in liquid nitrogen for subsequent RNA extractions.
- L592-597: I understand that training and test sets were stratified according to biomass (log-transformed) in TX. Could you please explain why this trait/location was chosen for the stratification?
Training/test stratification: We chose log-transformed biomass in TX for stratification because biomass is the primary agronomic trait of interest in switchgrass and exhibits greater phenotypic variation in TX compared to MI, allowing stratification to better preserve the full range of values in both the training and testing sets. The same training/testing split was subsequently applied to all analyses to ensure consistent and fair comparisons across traits, environments, and data types.
(11) Data availability: Maybe the read counts could also be provided on top of the raw RNA sequences. This would ensure, to some extent, reproducibility of the analyses, especially because the code for processing RNA-seq reads does not seem to be released in the GitHub repository.
The RNA-seq pipeline used in this study is described in the revised manuscript. The pipeline code is maintained in an internal GitLab repository at JGI and is therefore only accessible within JGI systems. To facilitate reproducibility, we have deposited the TPM count matrix used in our analyses in the following Zenodo repository: https://zenodo.org/records/21781035
(12) Code availability: The code to run Bayesian regression models does not seem to be released.
The code for the Bayesian regression models has been added to the manuscript repository.
Reviewer #2 (Recommendations for the authors):
The manuscript will benefit the most from additional biological replicates for RNAseq, as well as the phenology alignment with data collection. Further recommendations for improving the clarity in experimental design and alignment between claims and evidence.
We agree that additional biological replication for RNA-seq and transcriptomic sampling across multiple developmental stages would further strengthen the study. However, given the scale of the experiment, 426 genotypes evaluated across contrasting field environments, generating replicated transcriptomic datasets at multiple developmental stages was not logistically or financially feasible. Despite these constraints, the experimental design enabled characterization of transcriptomic and phenotypic variation across a large natural diversity panel and the identification of predictive patterns associated with trait variation across environments. We have revised the manuscript to more clearly describe the experimental design, sampling strategy, and study limitations, and to ensure that conclusions are appropriately aligned with the evidence presented.
Minor corrections:
(1) MAF threshold in Line 522, page 27. It should be > 0.05?
Corrected to > 0.05.
(2) SNP+T versus G+T.
Revised to SNP+T consistently throughout the manuscript.
(3) Include FDR-adjusted p-value thresholds.
FDR-adjusted p-value thresholds: For the Pearson correlations reported in this study, each represents a single pre-specified comparison rather than a family of simultaneous tests, and raw p-values are therefore appropriate. For the benchmark gene enrichment and Wilcoxon rank-sum tests, where multiple comparisons were performed across models and traits, FDR correction has been applied using the Benjamini– Hochberg procedure and the reported p-values have been updated accordingly.
(4) Avoid ambiguous phrasing like "generally consistent" and "generally outperformed".
Revised throughout the manuscript.
(5) The mention of artificial intelligence in the Abstract is largely conceptual and reads more like discussion material rather than the main finding of the manuscript, per eLife's Abstract requirements.
The mention of artificial intelligence has been revised in the Abstract in accordance with eLife's guidelines.
eLife Assessment
This useful study shows that loss of Samp1, a protein at the boundary of the cell nucleus, disrupts muscle cell development and is accompanied by changes in gene activity and genome organization. Some of the evidence is solid, but data to support the overall claims is judged to be incomplete. Controlled experiments and mutants to establish whether the changes in organization contribute to impaired development or follow from it would elevate the work. This work will be of interest to those in the genome organization and developmental fields.
Reviewer #1 (Public review):
Summary:
The authors ask whether Samp1 contributes to the reorganization of the genome and the regulation of gene expression during muscle-cell differentiation. Using C2C12 cells, they find that Samp1 depletion impairs differentiation and is accompanied by altered transcriptional responses and changes in peripheral chromatin and chromosome 8 distribution. These findings support Samp1's involvement in the differentiation program, although the causal relationship between the spatial and transcriptional phenotypes remains unresolved.
Strengths:
The study addresses a biologically significant question: how proteins at the nuclear envelope contribute to the coordinated changes in genome organization and gene expression that accompany cell differentiation. Its main contribution is to extend the established requirement for Samp1 in myogenesis to measurable spatial and transcriptional phenotypes, providing a more specific basis for investigating its function.
The complementary imaging approaches examine chromatin organization at different scales. Fluorescence ratiometric imaging assesses the relative distribution of chromatin reporters across the nucleus, whereas chromosome 8 imaging provides a chromosome-specific measurement. Agreement between these approaches strengthens the observation that Samp1-depleted cells undergo an altered spatial response to differentiation conditions, while leaving the underlying mechanism open.
The four-time-point transcriptional analysis adds temporal information that an endpoint comparison would miss. It allows readers to examine baseline differences and subsequent responses to differentiation conditions, and provides a potentially useful resource for identifying genes and pathways associated with the impaired differentiation program.
The work also builds on earlier validation with an independent depletion construct and rescue of differentiation by RNAi-resistant Samp1. This prior evidence strengthens the biological foundation of the present study. Together, the results identify Samp1 as a useful target for further investigation of the relationship between nuclear organization and muscle development.
Weaknesses:
The study leaves open how the differentiation, spatial, and transcriptional changes are connected. By day three, control cells have formed myotubes, whereas Samp1-depleted cells largely remain myoblasts. This is a consequence of the depletion and does not invalidate the comparison. However, it means that altered chromosome positioning could contribute to the differentiation defect or arise because differentiation has been disrupted. The authors acknowledge that the mechanism remains unresolved; the findings should therefore be interpreted as evidence that these processes depend on Samp1, without establishing that changes in genome positioning drive the transcriptional response.
The earlier rescue of differentiation strengthens the evidence for Samp1's role in myogenesis. The new spatial and transcriptional findings, however, have not been tested by rescue. This leaves some uncertainty about whether these particular effects are specific to Samp1 depletion in the stable cell line used here.
The connection between impaired gene repression and association with the nuclear lamina remains indirect. The published Lamin B1 DamID data describe normal differentiation and do not measure what happens to lamina association after Samp1 depletion. They support the proposed connection but cannot establish it.
The imaging measurements also need a cautious interpretation. The peripheral chromatin index describes relative fluorescence enrichment, rather than directly measuring heterochromatin abundance. The biological meaning of changes such as an index of approximately 1.1 versus 1.3 is therefore not immediately clear. Similarly, the chromosome 8 analysis measures the distribution of chromosome-paint signal within five optical sections rather than the entire chromosome territory. The results support spatial differences, but their magnitude may depend on nuclear geometry, the sampled sections, and image-processing choices whose influence is not fully assessed.
Some conclusions about the transcriptional response go beyond what the statistical comparisons directly test. A significant change during differentiation in control cells and a nonsignificant change in depleted cells do not, by themselves, establish that the responses differ. The comparison between groups at day three provides evidence of expression differences, but does not determine whether the changes from their respective starting points differ.
Reviewer #2 (Public review):
Summary:
Kašnik et al. set out test whether the previously reported requirement for the inner nuclear protein Samp1in myogenic differentiation and its also previously reported role in chromatin organization at the nuclear periphery are functionally linked. The authors confirm using a previously characterized shRNA system in cultured C2C12 cells that Samp1 loss leads to myogenic differentiation defects. Loss of Samp1 results in gene expression changes, including of genes associated with myogenesis and also interfered with the formation of peripheral heterochromatin and peripheralization of chromosome 8 which has previously been reported to be associated with myogenic differentiation. Finally, many of the downregulated genes by Samp1 loss were found in chromatin domains bound by lamin B.
Strengths:
The strength of this study is that is represents a rather straightforward analysis of the loss of Samp1 in myogenic differentiation in vitro. The RNAseq data is informative and is in line with the observed differentiation defects. In addition, the experiments are clearly designed and cleanly performed.
Weaknesses:
The major weakness of the study is that major conclusions are strongly overinterpreted and the degree of advance compared to prior studies is limited.
The study does not - as claimed by the authors - functionally link Samp1 to myogenic differentiation via its role in genome organization. As outlined by the authors there is already considerable evidence for a role of Samp1 in myogenic differentiation. There is also prior data to show that Samp1 loss prevents peripheralization of chromosome 8. The data reported in this study add further data to a possible functional link between these two events, but similar to the currently available data, the evidence remain circumstantial and they fall short of the authors' claim - and stated goal - of directly demonstrating a functional role of Samp1 in myogenesis via genome organization. The presented data does not rule out that the two effects represent unrelated functions of Samp1. The authors have not shown as claimed that Samp1 exerts its effect on myogenesis via genome reorganization.
The most relevant piece of data to point to such a link is the DamID data in Figure 3 showing that many downregulated Samp1-sensitive genes reside in lamin B associated chromatin. However, in this experiment, critical controls are missing to demonstrate that this effect represents an enrichment of Samp1-sensitive genes (see Recommendations).
While the RNAseq data offers the opportunity to speculate on pathways involved in myogenesis, none of the proposed pathways are experimentally tested to validate their functional roles. They remain hypothetical but are presented as demonstrated facts.
Finally, many of the findings are confirmatory in nature, such as the finding that Samp1 inhibits myogenic differentiation, that differentiation is accompanied by chromatin peripheralization or affects MAPK signaling.
Reviewer #3 (Public review):
The study by Kašnik et al investigates the role of Samp1, an inner nuclear membrane protein, in muscle cell differentiation. Samp1 was already known to be required for myogenic differentiation and to influence chromatin organization at the nuclear periphery, but whether these two functions were connected was unclear. The authors set out to address this by asking whether the differentiation defect caused by Samp1 depletion is accompanied by failures in genome reorganization at the nuclear periphery.
Using C2C12 mouse myoblasts with stable Samp1 knockdown, they combine RNA sequencing, immunofluorescence, FRIC analysis, and chromosome 8 DNA-FISH to track both the transcriptional and spatial consequences of Samp1 loss during differentiation. The goal is to show that Samp1 is needed not just for muscle gene activation but also for the peripheral repositioning of chromatin that normally accompanies differentiation, and that these two processes are linked.
The biggest weakness of this study is that it does not explain how Samp1 actually controls chromatin organization at the nuclear periphery. Is it acting through emerin, the LINC complex, or lamin A/C? This is never addressed. At minimum, the authors should test whether a Samp1 mutant that cannot bind emerin or SUN2 can still rescue the chromatin and differentiation defects.
Related to this, only chromosome 8 is examined for repositioning defects. This chromosome was deliberately chosen because it is known to move during myogenesis, so it is not surprising that it fails to reposition upon Samp1 depletion. A chromosome that does not normally reposition should be included to show the effect is specific, and genes that are expected to remain away from the periphery should also be examined to test whether Samp1 depletion causes indiscriminate rather than selective chromatin reorganization defects.
More broadly, the study is largely descriptive. The main finding is that knocking down Samp1 disrupts gene expression and chromatin repositioning, and that these two things correlate. But which comes first? Does the chromatin move and then genes get silenced, or the other way around? This question is central to the paper's argument and is never tested. Without some attempt to address this, the study reads more as an observation than a mechanistic advance.
Author response:
We thank the reviewers for their careful evaluation of our manuscript and their constructive feedback. We appreciate the points raised and will take them into consideration when revising the manuscript.
We plan to clarify and refine certain aspects of the interpretation and discussion, particularly regarding the relationship between Samp1, chromatin organization, and myogenic differentiation. We will also aim to strengthen the study with additional experiments and analyses where feasible, addressing the reviewers’ suggestions to the extent possible within the scope of the study.
We believe these revisions will help improve the clarity and strength of the manuscript.
eLife Assessment
In this potentially useful study, the authors explored the conformational dynamics of the intrinsically disordered protein p53-TAD and a cancer-associated variant by examining its conformation-dependent block of conductance through a MspA nanopore, in conjunction with molecular dynamics simulations. In principle, this combination of single-molecule recordings, hidden Markov modeling, and molecular dynamics simulations can provide a generally applicable platform for studying disordered proteins. While the experimental observations could be described as compelling, there are also issues regarding the apparently low statistics for the observed long-lived pore-blockage events and limited differences between WT and mutant simulations. Therefore, in the current form, the level of evidence is deemed incomplete and requires further validation.
Reviewer #1 (Public review):
The manuscript combines MspA nanopore measurements with molecular simulations to investigate p53-TAD, its cancer-associated N29K/N30D variant, and its interactions with MDM2 and EGCG. The experiments demonstrate a promising approach for distinguishing sequence- and ligand-dependent single-molecule signals. The manuscript is clearly written, and the use of multiple simulation trajectories is a strength. However, the structural and thermodynamic interpretations extend beyond what the current simulations establish. I recommend addressing the following points.
(1) The assignment of low-current states to compact conformations near the constriction, and intermediate-current states to more expanded conformations higher in the vestibule, is plausible but not directly validated (Figures 1-2; main text, pp. 4-9). Current blockage can depend on protein position, orientation, charge distribution, and pore contacts as well as compactness. The simulations do not calculate ionic conduction, and the four structural clusters in Figure 2f do not have an established correspondence to the six experimental current states. Could the authors provide a validated conductance calculation for representative configurations, or explicitly present these structural assignments as hypotheses? In particular, an EGCG-associated current level overlapping S4 does not establish that the ligand stabilizes the same structural ensemble sampled by apo p53-TAD.
(2) Figure 2c-d presents occupancy-derived profiles and regional differences using free-energy terminology. However, the simulations describe finite-duration, constant-force capture trajectories initiated outside the pore, rather than demonstrated equilibrium sampling. Could the authors establish stationarity, evaluate dependence on the analysis window, and explain how the applied bias and normalization are treated? Otherwise, these results should be described as driven occupancy profiles, without interpreting their differences as equilibrium stabilization energies. Similarly, Figure 5 should be clearly identified as a schematic reconstruction rather than a measured molecular free-energy landscape.<br /> The apparent EGCG affinity of 0.43 {plus minus} 0.49 μM also requires cautious interpretation. This value comes from event-frequency measurements at +75 mV and is poorly constrained. Event frequency depends on capture efficiency as well as solution binding populations. At +200 mV, manual voltage reversal censors spontaneous pore-escape times for long-lived complexes; moreover, pore escape and molecular dissociation are distinct processes. Please clarify which kinetic and thermodynamic quantities the assay can establish. The proposed confinement-mediated reduction in the entropic cost of EGCG binding is reasonable as a hypothesis, but is not demonstrated by these measurements.
(3) The SMD protocol distributes a constant force over acidic residues, restrains the pore, and confines p53-TAD within a cylindrical potential (Figure S12; Supplementary Methods, lines 174-198). These approximations are acknowledged, but their influence on the WT-mutant comparison remains unresolved. The double substitution approximately preserves net charge while changing its distribution, and the force prescription treats acidic and basic residues differently. Do the positional and conformational differences persist across pulling strengths, restraint widths, and alternative force assignments? Please also confirm that the simulated pore contains the mutations present in the experimental M2-MspA-NNN construct.<br /> The selected p53-TAD and MDM2 fragments omit full-length p53 regulation and interdomain coupling. This is a construct limitation, rather than a necessary consequence of the narrow constriction, because capture occurs within the wider vestibule. The existing discussion of wider pores and tethering could be extended to clarify this distinction. Phosphorylation-dependent TAD dynamics would be an interesting future application, although the present states cannot be assigned to modification-sensitive conformers without additional evidence.
(4) The mutant exhibits increased intramolecular contacts, but the pore-resident radius-of-gyration differences are only approximately 0.2-0.3 Å, with small standardized effects (Figure 2e; Table S1). Several other descriptors have confidence intervals spanning zero, including the bulk Rg difference. Consequently, the simulations support modest sequence-dependent changes more strongly than a pronounced shift in global compactness. Figure S15 also identifies altered protein-pore contacts, which could contribute directly to positional occupancy and current differences. Could the authors provide convergence analyses, trajectory-level uncertainty, and sensitivity to starting ensembles, and distinguish intrinsic compaction from altered pore association?
(5) The replacement of MDM2-complex events by EGCG-associated events is consistent with inhibition, but differential capture could also change the observed event populations (Figures 3-4). The manuscript acknowledges this possibility on p. 12. Could the authors clarify its implications for the inhibition claim and provide an orthogonal measurement or appropriate controls to distinguish reduced complex abundance from preferential detection? Charge, steric geometry, and pore interactions should be considered without assuming that compaction necessarily explains enrichment.
No new simulations directly investigate EGCG binding or MDM2 competition. Ensemble-based ligand modeling could strengthen the interpretation, but docking poses or short trajectories alone would not establish preferential binding to S4, binding affinity, or conformational selection versus induced fit. The discussion should place the proposed mechanism alongside prior EGCG-p53 studies, including the computational predictions involving Trp23-Lys24 and Pro47-Thr55 reported by Bahena Culhuac and Bello, 2024. If additional modeling is beyond the scope of this work, the claims concerning S4 stabilization, compaction, and MDM2 displacement should be qualified accordingly.
Overall, the experimental platform is promising. The principal revision needed is to distinguish the observed nanopore signal changes from their proposed molecular explanations. Targeted robustness tests and appropriately qualified structural and thermodynamic claims would substantially strengthen the manuscript.
Reviewer #2 (Public review):
Summary:
This manuscript presents an MspA nanopore-based approach for investigating the conformational dynamics and interactions of the intrinsically disordered transactivation domain of p53. The authors resolve multiple current states associated with trapped p53-TAD, identify differences between the wild-type protein and the N29K/N30D variant, and extend the approach to examine MDM2 binding and the effects of EGCG. The combination of single-molecule recordings, hidden Markov modeling, and molecular dynamics simulations provides a useful framework for studying disordered proteins under nanopore confinement.
Strengths:
The experimental observations are compelling, and the manuscript makes a valuable methodological contribution.
Weaknesses:
The analysis and interpretation could be strengthened.
Reviewer #3 (Public review):
Summary:
DeCoeur et al. explored the conformational dynamics of the intrinsically disordered protein (IDP) p53-TAD and a cancer-associated variant by examining its conformation-dependent block of conductance through a nanopore. Using this approach, they identified a preferred conformational ensemble stabilized by the anti-cancer compound EGCG. They supplemented their experimental observations with steered molecular dynamics simulations of the IDP's conformational ensembles. Much of the work appears to recapitulate prior observations for this IDP, and as such, this work reads more like a method development paper. Although the use of nanopores to study IDPs as done here is not completely novel, these sorts of applications remain relatively few, and thus this work is still useful. However, there are several issues with the data and its interpretation that suggest only incomplete evidence for the conclusions. Namely, what appear to be very low statistics for the observed long-lived pore-blockage events on which much of the experimental conclusions seem to be based, and also conclusions from simulations that do not appear to align with the data. Until these issues are addressed, I find some of the conclusions to be only weakly supported by the data.
Strengths:
The use of nanopore current blockage to interrogate the dynamics of IDPs has the potential to be an ideal approach to investigate both mechanisms of variant pathology and drug efficacy.
Weaknesses:
The apparently small dataset may be limiting, and in several cases the authors appear to also over-interpret their results based on small effects.
eLife Assessment
This study offers an important set of findings on a previously unrecognised mechanism by which integrin-dependent trafficking, notably the recycling of CD98hc from endosomes to the plasma membrane, controls the abundance of a major amino acid transporter at the cell surface, with implications for a stable amino acid supply and metabolic homeostasis. The approach and results are convincing in showing the trafficking steps involved, although the experimental analysis is in places incomplete and would benefit from more rigorous approaches, deeper mechanistic characterisation, and stronger links to metabolic homeostasis. With these aspects strengthened, the work would be of interest to cell biologists working on trafficking and amino acid metabolism, as well as to the cancer cell metabolism community.
Reviewer #1 (Public review):
Summary:
Dila Rizalar, Villaseñor and colleagues report the membrane trafficking steps undergone by CD98hc, a major component of the LAT1 amino acid transporter. They found that CD98hc entered cells through Fast Endophilin-Mediated Endocytosis (FEME) and quickly recycled back to the plasma membrane (PM) together with beta1-Integrin to mediate a stable supply of amino acids and support metabolic homeostasis.
Strengths:
The work was planned, executed and analysed to a high standard. A variety of cell models, techniques and assays were performed and combined with mathematical modeling. Appropriate analyses and statistical tests were used to support their conclusions.
Overall, the evidence provided is convincing.
Weaknesses:
(1) The proposed endosomal recycling pathway(s) (positive for Rab9, Rab11 and Rab14) is/are not believed to support recycling fast enough (c. 10-20 min) to explain the very short timeframe (c. 1 min) reported by the authors. This apparent discrepancy should be clarified by the authors, ideally with small additional work.
Notably, the fast Rab4-dependent recycling of TfR directly from Rab5 sorting endosomes has more consistent kinetics (c. 2-4 min, see the classic papers from the Mellman, Maxfield and Zerial labs) and should be investigated or at least discussed.
(2) LAT1 (and thus possibly CD98hc) were previously reported to enter cells through Clathrin-mediated Endocytosis (CME) upon potent PKC activation (PMA) or entry into cell quiescence (Barthelemy and André, Sci Rep 2019 and Kahlhofer J et al. EMBO J 2025). Interestingly, instead of being recycled back to the PM, LAT1 was then routed to lysosomes for degradation.<br /> Does LAT1 co-traffic with CD98hc along the FEME - fast endosomal recycling loop identified by the authors?
If so, the authors could discuss the physiological context in which the whole transporter shifts between the 2 routes (FEME-recycling in steady-state maintenance and CME-Lysosome upon acute signaling? There is a precedent in EGFR for such variation in endocytic fates upon signal strength).
If not, the authors could discuss why the 2 chains would traffic differently (in such a scenario: CD98hc through the FEME-recycling arm and LAT1 through CME-Lysosome).
(3) The study would benefit from mechanistic insights about the molecular details linking CD98hc to FEME.
Specifically, identifying motif(s) and/or adaptor(s) binding to the main FEME adaptors (Endophilin A1, A2 or Bin1) would be a valuable improvement of the work and would provide a strategy to design a point mutant version of CD98hc that would be unlinked to FEME.
Reviewer #2 (Public review):
Summary:
In this manuscript, the authors demonstrated that the amino acid transporter CD98hc (SLC3A2) undergoes fast endophilin-mediated endocytosis (FEME), followed by rapid β1-integrin-dependent recycling back to the plasma membrane. The latter is essential to maintain CD98hc membrane levels and counteract bursts of endocytosis in both endothelial and osteosarcoma cells.
Strengths:
The authors used pharmacological inhibitors, siRNA-mediated downregulation and CRISPR approaches to identify the regulators of CD98hc trafficking in different cell systems, strongly supporting their conclusions. In addition, they complemented their experimental results with computational models, which further validate their conclusions
Weaknesses:
The mechanistic details of how the interaction between CD98hc and b1 integrin determines CD98hc recycling have not been explored in the current version of the manuscript. Moreover, despite the title mentioning metabolic homeostasis, there is limited insight into how changes in CD98hc recycling affect intracellular amino acid content and cell metabolism, and how this impacts cell growth.
Reviewer #3 (Public review):
This paper from the Villasenor lab describes the results of experiments aimed at determining the role of the interaction of β1 integrin with CD98hc on the endo- exocytic trafficking of CD98hc, and its ability to function as an amino acid transporter. The paper demonstrates the following novel and interesting observations:
(1) CD98hc is internalised via fast endophilin-mediated endocytosis (FEME).
(2) A mutant of CD98hc that has been reported to oppose its interaction with β1 integrin (CD98hc ISM) alters CD98hc endo- exocytic trafficking to favour a situation in which CD98hc adopts a more endosomal distribution (than does wild-type CD98hc) following stimulation of FEME with a β-adrenergic agonist.
(3) Driving CD98hc endocytosis with a β-adrenergic agonist reduces uptake of branched-chain amino acids (BCAAs); an effect which is more pronounced following expression of the β1 integrin binding-deficient CD98hc ISM mutant.
General comments:
The findings of this paper are, at their core, very interesting and important and most of the work is well-executed. I consider that the 3 conclusions that I have listed above are solid and interesting (although some additional controls, etc would help consolidate these findings (see below)). However, the authors have vastly overplayed their hand in attempting to use mathematical modelling approaches and (in my opinion) invoking unnecessary working hypotheses concerning endocytic noise for which they have little evidence. I feel that the paper needs to be restructured to simplify the route between the core findings and the interpretation, and to move the mathematical modelling and 'endocytic noise' analyses and hypotheses to the discussion.
Specific points (in order of appearance):
(1) I am puzzled by how the graphs in Figures 1E&F have been derived from primary data and am sceptical of their validity. The curves don't look like anything that has been derived from experimental data. I gather that the curves in Figures 1E&F were generated by the equations described in the 'modelling CD98hc internalisation kinetics..'methods section, but I really cannot see how. Specifically, how can measurements of steady-state plasma membrane (PM) and endosomal pools of CD98 possibly allow the derivation of the APM and ATE operands in the equations in the 'modelling CD98hc internalisation kinetics..' methods section? Moreover, I don't see how this could be achieved in a way that could demonstrate a statistically significant difference in the trafficking of TfnR and CD98hc. The authors cite their previous 2017 Cell Reports paper to support this figure. However, the data in this paper are clearly experimental data (therefore understandable and clear), and the analysis in Vallasenor et al. (2017) does not resemble what has been performed to generate Figures 1E&F. Can the authors please properly clarify and justify this figure?
(2) Figures 1G&H. In the assay described in Figures 1G&H, it doesn't look like the CD98hc mAb is internalised in the control. And with the siRNAs that are purported to oppose CD98 internalisation (ARF6 and endophilin), it looks like CD98 has disappeared completely, rather than accumulating at the PM as one would expect if internalisation were blocked.
(3) In Figure 1H, the authors state that knockdown of endophilin A1 and A2 reduced CD98hc endocytosis, but it doesn't really look like this to me and there are neither quants nor stats on this.
(4) In Figure S1D, the conclusion that knockdown of dynamin or endophilin A1 and A2 decreases intracellular CD98hc-positive vesicles is not apparent to me, and these data need quantification and statistical analysis.
(5) In Figure 1J, the conclusion that addition of dyngo4A reduces internalisation of anti-CD98hc is not statistically supported. Moreover, although β-adrenergic agonists are clearly boosting internalisation of CD98hc in U2OS cells (Figure S1J), this, despite the asterisks (stats), does not appear to be the case in ICE-BECs (Fig. 1J); indeed, the increase is marginal, and the inter-experimental variation is too great to support a statistically significant increase in internalised CD98hc in ICE-BECs.
(6) The authors need to display representative images from experiments leading to the data in Figure 2B.
(7) The data in Figure 2 do not 'show that CD98hc is sorted to recycling endosomes after FEME internalisation' in an integrin-dependent manner. For this, one would need to demonstrate localisation of the CD98hc-β1 integrin complex to Rab11 (or Rab14)-positive compartments and these data are missing.
(8) I cannot find any details of the coordinates or construction of the CD98hc ISM mutant in this paper. I am assuming that this has been informed by the Ingber JCS paper from 2020, but the authors need to describe the construction and the provenance of their mutant. Also, importantly, the authors need to validate that this mutation opposes co-immunoprecipitation of β1 integrin and CD98hc. Or to use some other approach, such as FRET, to show this). OK, some of this has been performed by Ingber's group, but as it this is so important to the authors' conclusions, they need to confirm this. Also, it would be nice if the authors were to use their IP/proteomic approach to compare the interactomes of WT CD98hc and CD98 ISM.
(9) I think that some of the callouts to Figures 3 A-F are not correct in the manuscript. For instance, at the top of the page of the results (no page numbers), the callout to the sentence '...expressing either mGL-CD98hc WT or ISM (Figure 3A-D)' is likely wrong, and this is confusing.
(10) The data in Figures 3A,B,D&E do show that the CD98hc ISM mutant moves from the PM to endosomes more than does WT CD98hc following β-adrenergic stimulation. This is fine. However, I don't agree that the conclusions summarised in Figure 3C&F can be justified from these data. In fact, I do not believe it is necessary or justified to attempt to extrapolate these quite basic immunofluorescence data into an involved kinetic model. It would so improve the paper if these mathematical/kinetic extrapolations were moved to the discussion and presented as discursive points rather than hard-and-fast conclusions.
(11) The data in Figure 4A do not appear to show anything further than those in Figures 3 3A,B,D&E. Indeed, in light of Figure 3 A,B,D&E one would expect the vesicular CD98hc to be increased by the ISM mutation, but I'm not convinced that the amplitude of these so-called oscillations is necessarily different between WT and ISM mutant of CD98hc. All the differences between WT and ISM in Figure 4A&B appear proportionally larger than they are because of the truncated (non-zero baseline) axes in these graphs; a linear axis such as the y-axes in Fig. 4A&B should begin at zero, surely?
(12) In Figure 4D, it doesn't look at all as if the PM localisation of CD98hc and intracellular levels of BCAAs are similar between WT and ISM mutant of CD98hc - as the authors state in their results text.
(13) Why do the authors need to invoke what appear to me to be rather far-fetched and poorly supported arguments around 'endocytic noise' when the data in Figures 3A,B,D&E seem to be sufficiently consistent with those in Figure 4E to support the assertion that altered trafficking of CD98hc influences BCAA uptake?. Surely, it would be sufficient to argue that a mutant transporter that assumes a state which is distributed more endosomally than at the PM would be less efficient in uptake of BCAAs. So why the noisy arguments? It just seems to me to undermine what is potentially quite a nice and interesting paper.
Reviewer #4 (Public review):
Summary:
Rizalar and colleagues present a study where they identified fast endophilin-mediated endocytosis (FEME) as the primary driver of CD98hc internalization in brain endothelial cells. They have characterized trafficking, intracellular localization, and amino acid transport under physiological conditions or metabolic stress using both experimental and mathematical modelling approaches.
Strengths:
The author presents an interesting study. The authors provide experimental and mathematical modeling to dissect integrin-dependent trafficking of CD98hc. They give an interesting role of integrin β1 different from adhesion.
Weaknesses:
(1) Overall, it is difficult to follow which cells were used for each experiment and the reasons. The authors should comment on the choice of osteosarcoma cells for the study, as the paper mainly discusses the choice of brain endothelial cells. Some key experiments in U2OS must be repeated in brain endothelial cells; targets identified in the siRNA screening should be tested in brain endothelial cells to increase the strength of conclusions.
(2) Figure 1B is not clear in this disposition; it seems the same markers were used twice. Additionally, to improve clarity in understanding and reading, I suggest adding the cell line used in the figures.
(3) In Figure S1A, it is difficult to conclude to the presence or absence of colocalization, zoom on ROI should be added as well as a quantification of the colocalization to conclude on the experiment.
(4) Some context on intracellular sorting tubules can be added in the manuscript to improve clarity.
(5) Figure S1C needs to be quantified to see if the mGL does not affect the expression of CD98hc; images could be added to support the absence of impact on cellular phenotype.
(6) It could be worth measuring the localization of the surface pool of CD98hc to increase the strength of the conclusion, maybe by immunofluorescence colocalization between a cell membrane marker and CD98hc, or by pHluorine marker to differentiate cell surface from endosomes.
(7) In Figure 1D, it seems that CD98hc have a different pattern in the upper panel compared to the lower one; it seems that vesicles are larger close to the membrane compared to deeper in the cytoplasm.
(8) Some crucial controls are missing, mainly in siRNA experiments. For all siRNA experiments, efficiency and silencing must be shown by Western blot and quantified (Figure 1G, S1D-E).
(9) In Figure S1D, the authors said "knockdown of ARF6 led to the accumulation of enlarged intracellular CD98hc positive structure". Size and number must be quantified to validate this assumption; I recommend adding an arrow to show structures for clarity. Same comment for the knockdown of dynamin and endophilin, the decrease of intracellular CD98hc positive structure need to be quantified.
(10) In Figure S1G, only one or two cells are positive for GFP; the efficacy seems poor; in this sense, it's difficult to conclude whether they have an impact on internalization.
(11) In Figure 2A, LAMP1 is significantly upregulated in the interactome; how do you interpret this, given that you show in Figure 1B that there is no colocalization.
(12) Figure 2B: CD98hc colocalization with RAB subunits needs to be shown by microscopy in brain endothelial cells.
(13) Overall, it is not clear if it's the integrin that is mutated for the binding domain or if it is the CD98hc binding domain that is mutated. If the integrin was not mutated, this type of experiment needs to be done to confirm the role of integrin. Additionally, immunoprecipitation needs to be done to show the absence or reduction of binding between CD98hc ISM and integrin.
(14) Figure 3 was mainly done on U2OS cells; the integrin mutant needs to be done in at least one brain endothelial cell line. Do you observe the same results between the mutant integrin and KO integrin? The impact of b1 integrin on CD98hc could be tested using integrin b1 blocking antibodies.
(15) The article focuses primarily on β1 integrin; could other integrins be involved? Given that integrins are often overexpressed in cancers, could this increase promote CD98hc internalization and trafficking? Adding this type of experiment could reinforce the conclusion and importance of integrin.
(16) The proliferation of cells re-expressing CD98hc ISM under starvation or not shows a decrease in relative proliferation after 144 h. Could the authors comment on this? Is cell death happening?
(17) The relevance of the results to the cancer field should be discussed in the discussion section, as osteosarcoma cells were used in the majority of experiments.
(18) Integrins are mainly involved in adhesion; do the authors observe the same results between integrin and CD98hc if cells are cultivated on matrices like fibronectin?
Please note, I have mainly reviewed the experimental part of the paper as I am not a specialist in mathematical modelling.
eLife Assessment
The authors make the valuable observation that directional memory during epithelial cell migration is enhanced in collective contexts compared to single-cell migration. While there is solid evidence for some parts of the work, the evidence linking directional memory to adherens junctions and vinculin dimerization remains limited. Key issues are that the Y1065E mutant may affect vinculin beyond its ability to dimerize, and focal adhesions were not directly compared across migration contexts.
Reviewer #3 (Public review):
This manuscript focuses on the presence/origin of directional memory during epithelial cell migration. It starts by analyzing single cells and then moves to more complex systems (confluent layers and scratch assays). The paper first demonstrates that the migration in all of these systems is well-described by persistent random walks which likely emerge from fractional Brownian motion. This is an important demonstration, as it implies orientation memory in the systems. Then the paper proceeds to attempt to discern the origin of this memory and claims to establish key roles for adherens junctions and vinculin dimerization. While for the most part the manuscript is well-written, there are some signification overinterpretations in experimental results. The largest issue is demonstrating the role for vinculin dimerization, which is not a well-studied phenomena inside living cells, as all data is reliant on a single point mutation (Y1065E).
(1) Likely the most significant issue with the manuscript is the interpretation of the vinculin Y1065E variant and the assumption that the only defect the mutations causes is a lack of dimerization. Vinculin dimerization is mediated by a conformational change in the vinculin tail domain induced by F-actin binding (Thompson, FEBS Letters, 2013). Dimerization of the vinculin tail domain has been clearly demonstrated in in vitro systems involving purified proteins, as the authors point out in the manuscript. However, the dimerization of full-length vinculin has not been well characteristic in living cells. There are several reasons to suspect dimerization is potentially not prevalent in cells. For instance, in the presence of other actin binding proteins, there may not be sufficient available binding sites available on neighboring actin filaments to facilitate dimerization. Additionally, pY1065 vinculin and vinculin Y1065E have been associated with increased vinculin activation (Huang, JBC, 2014), so other effects seem possible. While the Y1065E variant clearly has an effect on the tension sensor readout and vinculin dynamics, further experimental evidence is needed to show that these effects are due to lack of dimerization in living cells. Perhaps most importantly, the vinculin tension sensors used in this study have been used extensively in knock-out systems and intermolecular FRET has not been observed (PMID: 20613844, 33335089, 30523675). To justify the definitive claims made in the manuscript, the authors likely need to develop, or employ, an assay for detecting vinculin dimerization in living cells. The authors could choose between intermolecular FRET, proximity labeling assays (i.e. antibodies with DNA for signal amplification), bimolecular fluorescent complementation (i.e. split GFP) based approaches, or some other approach. It should be noted that working with full length vinculin, not just Vt, and designing assay that can incorporate vinculin Y1065 variants (Y1065E and potentially Y1065A/F) would strengthen results. Also, the authors should be aware that the observation of strong dimerization may invalidate the use of FRET-based tension sensors in this system or at least necessitate intermolecular FRET control experiments.
(2) A major conclusion in the manuscript is that in response to overexpression of a specific vinculin constructs, focal adhesions behave the same in single cells, confluent cells, and collectively migrating cells for all the mutants but Y1065E. However, outside of the FRET measurements, there is not much evidence to support this claim. The authors should perform a greater comparison of the focal adhesions between the systems used in the manuscript (single cell, confluent cells, collectively migrating cells). Key measurements would include focal adhesion number per cell, focal adhesion size, focal adhesion orientation, vinculin dynamics (e.g. FRAP), focal adhesion stability, and some indicators of focal adhesion composition. For the last aspect, focusing on focal adhesion components that also have roles in adherens junctions, such as VASP, seems appropriate. Without such characterization, it is an overinterpretation to assume that focal adhesions are the same in each system and, therefore, effects are due to vinculin behavior in the adherens junctions.
Author response:
The following is the authors’ response to the original reviews
Public Reviews:
Reviewer #1 (Public review):
Summary:
In this work, the authors study the migration of isolated cells and of cells in ensembles. They quantify several aspects of the corresponding migration patterns and investigate how these quantities depend on molecules that are known to play an important role in migration. Furthermore, they study the effect of external cues on these migration processes.
Strengths:
The authors provide a clean and uniform setting for comparing the migration of isolated cells and of cells in an ensemble in control and mutant conditions, and in the presence and absence of external cues. This allows for a meaningful comparison between different conditions. In this way, the authors obtain useful data that link the migration of isolated cells to that of cells in collectives.
Weaknesses:
A major weakness of the manuscript is that the authors do not properly introduce the quantities and concepts they are working with. In this way, it is hardly accessible for a reader who does not have a thorough background in cell migration and anomalous transport. In addition, the manuscript uses some notions that are not standard, for example, vinculin or FA stability, which are not properly introduced. Most strikingly, "collective directional memory" is not defined.
We thank the reviewer for pointing this out. We have now added the definition of collective directional memory in the introduction:
“... cells that migrate collectively acquire long-term directional memory, which specifically requires α-catenin in AJs and results from coordination of velocities between neighboring cells. For this reason, we define it as a collective directional memory.” and in the results section:
“... the transition in directional memory observed at confluence are the results of bona fide AJ-mediated intercellular adhesion, but independently of the cadherin subtype. Consequently, we define as a collective directional memory this long-term directional memory that arises from neighbor interactions.”
The definition of anomalous transport is given in the results section too:
“we found that the average Mean Square Displacement as function of lag time ∆t (MSD, see Materials and Methods) was better described by a persistent random walk (Fürth et al. 1920, Uhlenbeck et al. 1930) than anomalous diffusion: the MSD transitioned from ballistic (MSD ∼ ∆t<sup>2</sup>,∆t → 0) to pure diffusion (MSD ∼ ∆t,∆t → ∞) with a persistence time P of the order of 10min, rather than scaling as ∆t<sup>𝛼</sup> with α ̸\= 1” and in the Materials and Methods:
"MSD(∆t) curves were fit with the Fürth equation MSD = 4_D_(∆t−P(1_−exp(−∆_t/P))) where D is the coefficient of diffusion and P the persistence time, as a model of persistent random walk, or with MSD ∼ ∆t<sup>𝛼</sup> (straight line of slope α in ln−ln scale), as a model of anomalous diffusion, with anomalous exponent α > 1 for super-diffusion."
The definition of vinculin and FA stability, which we now specify as “molecular stability” is given in the Materials and Methods section:
“We use the immobile fraction of vinculin and its mutants as a measure of vinculin stability and, by extension of the molecular stability of adhesions, because the stability of other adhesion proteins is regulated by the stability of vinculin (Carisey et al. 2013, Liu et al. 2016, Yonemura et al. 2010).”
The authors infer relationships between different quantities, but they remain qualitative, even though the authors use a language that suggests otherwise. For example, "The combination of Focal Adhesion stability and force transmission from the cytoskeleton predicts the migration speed of single cells" (p 2). I am not sure what is meant by prediction, but this heading suggests that knowledge of FA stability and force transmission yields the migration speed. Reading this line, I expect that if I give you values for FA stability and force transmission, you would give me a value for the migration speed. Such a quantitative mapping is not provided. In fact, it cannot be provided, because - as mentioned before - these quantities are not properly defined, so I would not know how to measure them. I do not even know their units.
We thank the reviewer for suggesting a more quantitative analysis. Initially, we used Boolean logic to build our model. Based on the effects of mutants, vinculin can be either tensed or not, and either stable or not, which yields four combinations that associate with more or less fast migration. We represent the model as a Venn diagram, which we now specify in the results section and figure legends. To complement this simplified approach, we have now added quantitative analyses by fitting changes in single cell speed, in the difference between single and collective cell migration speed, and in the ratio between collective cell migration speed and wound healing rate to the changes in tension, stability, or the gradients of tensions in FAs and AJs based on the quantitative measurements of speed (migration assays), tension (FRET index of vinculin sensors) and stability (immobile fraction of FRAP experiments) re-scaled between their extreme values across vinculin constructs, as explained in the Materials and Methods. This allows us to quantify the contributions of changes in tension, stability, and the gradients of tension to changes in single cell migration speed, in the difference between single and collective cell migration speed, and in the ratio between collective cell migration speed and wound healing rate. As shown in the added Fig. S2I, the change in single cell speed is explained by a linear combination of changes in tension and stability with contributions of 2_/3 and 1/_3, respectively. In Fig. S4J, the difference between single and collective cell migration speed changes with the stability exclusively, and its increased sensitivity to low stability is well captured by a power law. In fig. S5C, the ratio between collective cell migration speed and wound healing rate is set by the difference in mechanosensitivity between FAs and AJs in which 1/3 of that of AJs counts. Overall, these results confirm our initial models.
Furthermore, the authors do interpret some of their results without explaining or justifying the basis for their interpretation. For example, they use the FRET index of vinculin - another notion that is not properly introduced - to make statements about mechanical stress.
We are sorry that we had overlooked this important explanation. We have now added explanations in the main text, and in the Materials and Methods. Briefly, the FRET sensor module inserted in vinculin and its mutants is made of two fluorescent proteins separated by an elastic polypeptidic linker of known elasticity, such that an increase in tension on the protein will stretch the sensor and will decrease the amount of FRET between the two flanking fluorescent proteins. As a result, the lower the FRET, the higher the tension. We cite the original paper that introduced this sensor to allow the reader to get more information (Grashoff et al. 2010). Figures now also include the direction along which the tension increases when FRET changes.
It also seems that the figures could be improved. Some of the sketches are, in my opinion, not helpful. Examples are Figure 3A (how could a cell move while the hexagonal arrangement of the cells is maintained?) or Figures 2F, 4F, and 6F (what do the colored ellipses indicate?). In Figures 1B, 1D, 2A, 2E, 3B, 3D-F, 4A, 4F, 5B-D, it is not clear which lines merely connect data points and which lines are fits to the data.
We thank the reviewer for suggesting these improvements. We have now changed figure 3A for a more realistic drawing of the epithelium. We have specified in figure legends what colored ellipses are (as explained above, they are the Venn diagram categories), and whenever a line was a fit or a curve connecting data points.
Reviewer #2 (Public review):
Summary:
The manuscript by Canever et al was assessed by three Referees at another journal, who brought up a range of critical points. I will not repeat a summary of the work; this can be found in the first-round reviews.
Strengths:
In their revised manuscript, the authors include substantial changes and additional reasoning. Along with their rebuttal letter, I think they make a very convincing case. While the claims are well supported by the analysis, I do not see that the findings need to be universal to be relevant. It might be rather surprising to me if there existed such a universality, in fact. I think that the findings are solid and interesting in their own right and are worthy of publication, especially with the amended discussion in this revision.
Weaknesses:
However, while the more bio-oriented parts are not fully accessible to me, I do have a few points from the data analysis point of view that need amendment. (1) The used mathematical models need to be specified more precisely. First, the authors confuse Levy flights and walks. These are distinct processes in the sense that a Levy flight does not have a finite variance and thus no finite speed. The proper model here would be Levy walks. As in a big body of the literature, both notions are used interchangeably here, while they are distinct processes. Then the authors speak about a "superdiffusive model", for which I do not find a proper definition. There exists an entire range of superdiffusive models, each with a different physical background, so this needs more clarity. The authors may consult one of the standard reviews for more details, e.g., Soft Matter 8, 9043 (2012) or Phys Chem Chem Phys 16, 24128 (2014). Overall, a few equations (maybe in the Supplement) would help to be more specific.
We thank the reviewer for these suggestions. We have changed the main text as follows to address these points:
- "To distinguish between the various walk models that could produce such a super-diffusive behavior (Lévy walk, fractional Brownian motion, or their combinations."
- "we found that the average Mean Square Displacement as function of lag time ∆t (MSD, see Materials and Methods) was better described by a persistent random walk (Fürth et al. 1920, Uhlenbeck et al. 1930) than anomalous diffusion: the MSD transitioned from ballistic (MSD ∼ ∆t<sup>2</sup>,∆t → 0) to pure diffusion (MSD ∼ ∆t,∆t → ∞) with a persistence time P of the order of 10min, rather than scaling as ∆t<sup>𝛼</sup> with α ̸\= 1" and the Materials and Methods as follows:
- "MSD(∆t) curves were fit with the Fürth equation MSD = 4D(∆t−P(1−exp−∆t/P)) where D is the coefficient of diffusion and P the persistence time, as a model of persistent random walk, or with MSD ∼ ∆t<sup>𝛼</sup> (straight line of slope α in ln−ln scale), as a model of anomalous diffusion, with anomalous exponent α > 1 for super-diffusion."
We thank the reviewer for the additional references, we cite the more recent Meroz et al., 2015 for reference on anomalous diffusion.
(2) For fractional Brownian motion, the authors should check the displacement correlation function; it should show slowly decaying, positive correlations. More details on the practical analysis of FBM can be found, e.g., in Phys Chem Chem Phys 27, 14350 (2025). These correlations should decay as a function of the bin time, e.g., as discussed for the opposite case of subdiffusion in Phys Rev E 88, 010101(R) (2013) [cf Fig 3b]. In general, FBM was determined to be a highly relevant process for a number of systems, including amoeba cells at shorter times, see the detailed analysis in Phys Rev Res 4, 033055 (2022). In this paper, there are also different ways to characterise the motion in terms of scaling. Exponents are detailed.
We thank the reviewer for these suggestions and references. We have added explanations in the main text and in the Materials and Methods about the metric we use to characterize fractional Brownian motion together with the suggested reference Phys Chem Chem Phys 27, 14350 (2025).
In summary, we define the angular memory as a modified angular correlation in which we consider the angular deviation θ of each elementary step of duration 10min from its immediate previous step of duration τ equal to the lag time ∆t. Specifically, the angular memory <sub>i,t</sub>(∆t) of a cell i at time t is defined as the cosine of the angle θ<sub>i,t</sub> between consecutive step vectors 
between times t−∆t and t, and 
between times t and t+10min, such that the step duration τ of the first vector and the lag time ∆t before the next are equal. M<sub>i,t</sub>(∆t) = cosθ<sub>i,t</sub> (∆t) = cos 
. Figures display ⟨M<sub>i,t</sub>(∆t)⟩. This metric measures the duration of the past that the cell considers when deciding on its direction. It decays to 0 when ∆t → ∞ for persistent random walkers and remains constant for fractional Brownian motion, similarly to the auto-covariance function of fractional Brownian motion C(∆t/τ) = ((∆t/τ +1)<sup>α</sup> +|∆t/τ −1|<sup>α</sup> −2(∆t/τ) <sup>α</sup>)/2, which is a constant C = 2<sup>α-1</sup> −_1 for all ∆_t \= τ, as in our definition of angular memory (while it decays to 0 when ∆t/τ → ∞, as pointed out by the reviewer) (Metzler et al. 2025). Ballistic motion yields 1 whatever ∆t and uncorrelated random walks, whether anomalous or not, yield 0. For this reason, and also because M<sub>i,t</sub>(∆t) was found more robust to noise than C(∆t/τ), we considered this metric a more compelling discriminant between fractional Brownian motion and other uncorrelated super-diffusive walks.
(Of Note, we have increased the analysed data set in the revised Fig. 1D, 3D,F for the sake of representativeness. The conclusions are unchanged).
(3) Some relevant approaches discussed in literature that should be discussed in the context of this work: eLife 9, e52224 (2020); Rep Prog Phys 86, 126601 (2023); Chaos 35, 023145 (2025). In the context of non-Gaussianity for active particles: Phys Rev E 104, 064615 (2021); New J Phys 25, 013010 (2023).
We thank the reviewer for these interesting references. We have added them in the discussion as follows: "More recently, efforts have focused on developing analysis tools and models to account for heterogeneous migration properties along individual trajectories and among populations (Lemaitre et al. 2023, Khadem et al. 2021, Woszczek et al. 2025, Waigh et al. 2025, Han et al. 2020)." We have also made changes in the main text to specify when and how our data could indicate the presence or the absence of such heterogeneity in the migration of single cells and at confluence.
(4) In the abstract, I am having some issues with the formulation in the sentence: "This directional memory emerges from fractional Brownian motion". It sounds as if FBM were a fully clarified phenomenon. I would prefer some statement along the lines that the data are consistent with such a mathematical modelling approach.
We thank the reviewer for this suggestion. We have changed the formulation to "This behavior is consistent with a fractional Brownian motion that emerges from velocity coordination between neighboring cells with intact intercellular adhesion." After fixing these points, I think the manuscript will clearly warrant being shared.
Reviewer #3 (Public review):
This manuscript focuses on the presence/origin of directional memory during epithelial cell migration. It starts by analyzing single cells and then moves to more complex systems (confluent layers and scratch assays). The paper first demonstrates that the migration in all of these systems is well-described by persistent random walks, which likely emerge from fractional Brownian motion. This is an important demonstration, as it implies orientation memory in the systems. Then the paper proceeds to attempt to discern the origin of this memory and claims to establish key roles for adherens junctions and vinculin dimerization. While for the most part the manuscript is well-written, there are some significant overinterpretations in experimental results. The largest issue is demonstrating the role of vinculin dimerization, which is not a well-studied phenomenon inside living cells, as all data is reliant on a single point mutation (Y1065E). Additionally, the authors seem to be over-interpreting several of the assays; the statistical analysis does not seem to encompass all comparisons made, and the molecular model proposed does not clearly explain the observed results. The discussion could also be strengthened by considering other aspects of vinculin behavior (e.g., vinculin catch bonding) as well as discussing some other recent similar papers.
(1) Likely the most significant issue with the manuscript is the interpretation of the vinculin Y1065E variant and the assumption that the only defect the mutations cause is a lack of dimerization. Vinculin dimerization is mediated by a conformational change in the vinculin tail domain induced by F-actin binding (Thompson, FEBS Letters, 2013). Dimerization of the vinculin tail domain has been clearly demonstrated in in vitro systems involving purified proteins, as the authors point out in the manuscript. However, the dimerization of full-length vinculin has not been well characterised in living cells. There are several reasons to suspect dimerization is potentially not prevalent in cells. For instance, in the presence of other actin-binding proteins, there may not be sufficient binding sites available on neighboring actin filaments to facilitate dimerization. Additionally, pY1065 vinculin and vinculin Y1065E have been associated with increased vinculin activation (Huang, JBC, 2014), so other effects seem possible. While the Y1065E variant clearly has an effect on the tension sensor readout and vinculin dynamics, further experimental evidence is needed to show that these effects are due to a lack of dimerization in living cells. To justify the definitive claims made in the manuscript, the authors likely need to develop, or employ, an assay for detecting vinculin dimerization in living cells. The authors could choose between intermolecular FRET, proximity labeling assays (i.e., antibodies with DNA for signal amplification), bimolecular fluorescent complementation (i.e., split GFP) based approaches, or some other approach. It should be noted that working with full-length vinculin, not just Vt, and designing an assay that can incorporate vinculin Y1065 variants (Y1065E and potentially Y1065A/F) would strengthen results. Also, the authors should be aware that the observation of strong dimerization may invalidate the use of FRET-based tension sensors in this system or at least necessitate intermolecular FRET control experiments.
We thank the reviewer for raising this important point. We agree that we lack direct evidence for the dimerization of vinculin in cells. Therefore, we have replaced “dimerization” with “the ability to dimerize” wherever it is relevant in the revised version of our manuscript, to emphasize that the ability to dimerize does not imply dimerization. The effects of the YE mutant may then be attributed to a lack of proper dimerization, or to any other, yet to be discovered defect dependent on the mutation but independent of the loss of the ability to dimerize. We have added a sentence in the discussion to account for this possibility:
"Whether vinculin dimerization, or another unknown independent function disabled by the YE mutation or phosphorylation of Y1065, is specifically involved in the adhesion and migration phenotypes described above may be addressed in future studies."
We thank the reviewer for suggesting experimental strategies to assess vinculin dimerization in cells. In another work, we have evidenced proximity by intermolecular FRET in FAs between proteins with no known interacting domains (doi: 10.64898/2026.04.22.720102). This raises the possibility that FRET, as well as split-GFP and PLA, the detection ranges of which are equal or larger than FRET, will not be able to discriminate between bona-fide direct interactions and very close proximity in environments such as adhesion complexes where protein densities are very high. As pointed out by the reviewer, this is the major reason why we preferred not to carry the experiments in the present paper in a null background, such that the endogenous vinculin can prevent intermolecular FRET between neighboring sensors, as we explain in the Materials and Methods.
Despite this methodological challenge, indirect evidence is largely in favor of dimerization in cells rather than against it, as explained below.
- While there are indeed many actin binding proteins in cells, the proposed scenario by which they would prevent vinculin dimerization by competing with vinculin for actin binding seems very unlikely to us. Indeed, according to this scenario, vinculin would also fail to be under tension, since it needs to bind to actin for that. We show here, as many others did before, that vinculin is under tension, implying that it is able to bind to actin. Therefore, vinculin-bound actin may subsequently facilitate vinculin dimerization.
- Additionally, vinculin also dimerizes in the absence of actin in solution, albeit at higher concentrations than in the presence of actin, equivalent to at least 0.1 molecule per 100nm² in any 2D section of the volume (doi: 10.1074/jbc.M111.244293, 10.1021/bi801764a). The density of vinculin within adhesions is promoted by the multiple neighboring vinculin-binding sites on its partners (such as talin, in which the neighboring binding sites are occupied nearly all at once upon talin stretch, doi:10.1021/acs.nanolett.6b00650, 10.1039/c5ib00307e), and is likely to reach a molecule per 100nm² in order to accommodate the traction forces that cells exert on their substrates (doi: 10.1021/acs.nanolett.6b04364). This exceeds the threshold concentration for vinculin dimerization in the absence of actin by at least an order of magnitude, and is therefore likely to offer another route for vinculin dimerization in addition to actin binding.
- Furthermore, Huang et al have shown that Y1065 phosphorylation is required for the activation of vinculin, and the YE mutant mimics this behavior, but they have also shown that the activation level is quite similar to that of WT vinculin (compared to the loss of activation of the YA mutant). Additionally, phosphorylation or the YE mutation does not abolish the head-tail interaction of vinculin, as shown by Küpper et al. (doi: 10.1016/j.bbrc.2010.07.110), who also show that YE is less stable in FAs, in agreement with our results. Altogether, these results support a model by which pY1065 is required for vinculin activation but facilitates its inactivation and release from FAs, which is easily explained if dimerization, which antagonizes head-tail interaction, is impaired.
In conclusion, in the absence of identified mechanisms to oppose vinculin dimerization in cells, and to better explain the observed phenotypes at the cellular level, vinculin dimerization in cells remains the most parsimonious scenario. We have added the following sentence to account for this:
"Nevertheless, vinculin dimerization, with its ability to antagonize its head-tail interaction and to promote actin bundling, remains the most parsimonious mechanism that can explain the stabilization of vinculin in FAs and the stiffening of cells described here and elsewhere (Tolbert et al. 2014, Küpper et al. 2010)."
(2) The authors have seemed to assume that FRAP and adhesion stability are interchangeable. To this reviewer’s knowledge, this is not the standard in the field. FRAP informs about molecular dynamics. Stability assays, which probe the spatial position of an entire focal adhesion over time (Zaidel-Bar, JCS, 2007, although other approaches are equally suitable), are typically used for assessing adhesion stability. If the authors wish to make strong claims about the stability of the adhesions, non-FRAP-based assays should be employed. Alternatively, the authors could interpret the FRAP data simply in terms of vinculin dynamics.
We agree with the reviewer. We did not wish to make claims about the positional stability of adhesions as a whole, therefore, we now specify “molecular stability” wherever needed in the revised version of the manuscript, to reflect that we refer to the stability of the molecular components of adhesions. We use the immobile fraction of vinculin and its mutants as a measure of vinculin stability and, by extension of the molecular stability of adhesions, because the stability of other adhesion proteins is dependent on the stability of vinculin, as supported by the various references we cite in the text (Carisey et al. 2013, Liu et al. 2016, Yonemura et al. 2010).
(3) A major conclusion in the manuscript is that in response to overexpression of a specific vinculin construct, focal adhesions behave the same in single cells, confluent cells, and collectively migrating cells for all the mutants but Y1065E. However, outside of the FRET measurements, there is not much evidence to support this claim. The authors should perform a greater comparison of the focal adhesions between the systems used in the manuscript (single cell, confluent cells, collectively migrating cells). Key measurements would include focal adhesion number per cell, focal adhesion size, focal adhesion orientation, vinculin dynamics (e.g., FRAP), focal adhesion stability, and some indicators of focal adhesion composition. For the last aspect, focusing on focal adhesion components that also have roles in adherens junctions, such as VASP, seems appropriate. Without such characterization, it is an overinterpretation to assume that focal adhesions are the same in each system and, therefore, effects are due to vinculin behavior in the adherens junctions.
We thank the reviewer for raising this point. We do not assume or need to assume that FAs are the same in single cells, at confluence or during directed migration. The simplest model that explains our results, however, does imply that the response of FAs to a given construct is the the same whether cells are single, at confluence or involved in directed migration. We explain below our approach:
- First, we show that the migration speeds of WT, FL, TL, T12 but not YE cells at confluence are predictable from their migration speed as single cells (Fig. S4C). Roughly, v<sub>c</sub> ≈ v<sub>s</sub> _−_10(_µ_m/h) for all but YE.
- From here, there are two possible scenarios:
(A) Each construct has the same effect on cell speed in single cells and at confluence through FAs, and AJs provide an additional downregulation of ≈ 10µm/h for all the constructs except YE. This implies that the mutants (except YE) have no effect on cell speed through AJs, since confluence induces the same decrease of ≈ 10µm/h in mutant cells as in FL or WT cells, in which vinculin is not mutated.
(B) One or several mutants have different effects through FAs on cell speed between single cells and at confluence, and there is an additional effect of AJs on migration that depends on the mutant and that compensates for the difference in the effect of the mutant through FAs between single and confluent cells, such as to preserve the observed v<sub>c</sub> ≈ v<sub>s</sub> −10(µm/h). For instance, compared to WT, T12 could increase the speed of 15µm/h in single cells and of 15−xµm/h at confluence through FAs, but decrease it of 10+xµm/h through AJs to compensate. It is already straightforward that this scenario, which adds one unknown parameter per mutant, is the least parsimonious. It would require as many additional pieces of evidence to be considered equally likely as the first scenario.
- If we review the effects of the mutants (except YE), there is no additional evidence in support of the second scenario: overall, mutants have similar effects on tension in FAs of single and confluent cells, as well as in AJs (Fig. 2D, 4D,E, S4G). They also have similar effects on stability in FAs and in AJs (Fig. 2E, 4F).
- As a matter of fact, we do not exclude the possibility that the mutants have different effects in single and confluent cells on other FA properties. In light of the above, however, the principle of parsimony requires that evidence for 1) such differences and 2) their impact on cell speed be brought first before we can consider the second scenario equally plausible.
In summary, there is no need nor other reason for x to be substantially different from 0 for any mutant, and YE is the only mutant for which there is a need for a specific contribution through AJs distinct from the other mutants to explain its difference in speed between single cells and confluent cells. We have now added a quantitative analysis of how this difference in speed changes across mutants (Fig. S4F and Materials and Methods). Results show that a power law provides an excellent description: the speed difference only depends on the molecular stability and not on the tension in AJs, and it is very weakly sensitive to the changes in stability between FL, TL and T12 while very sensitive to the low stability reached by YE. We detail this analysis below in response to Question 6.
(4) What is shown in Figure 3G is not clear. How are P/Po and alpha shown on different areas of the same plot?
We are sorry that our plot was not clear enough. For any given pair of simulation coordinates P<sub>0</sub>/τ and n=πρd<sup>2</sup>, the simulation yields either a persistent random walk or a super-diffusion behavior, based on the Akaike Information Criterion. Simulations show that the vast majority of the points that yield a persistent random walk (in blue) are grouped on the low density side of the plot, and the vast majority of the points that yield a super-diffusion behavior (in orange) are grouped on the high density side of the plot, typical of a system with two distinct phases within the space defined by the simulation parameters. Within each phase, each migration behavior changes as a function of the simulation parameters, these changes are quantified through either P/Po or alpha according to the most likely migration behavior of the phase. Of note, there are 3 regions of slight overlap: very low persistence (bottom line), very high persistence (top line) and the boundary between the two sides of the plot, because the finite number of simulation repeats (5) for each coordinate and their intrinsic stochasticity do not allow us to discriminate the two behaviors when their difference becomes smaller in these regions. Nevertheless, it is possible to determine the most probable behavior for each point and its close neighbors at the cost of some resolution, which results in a boundary that is slightly scaled rather than smooth if one looks closely. We consider that this is sufficient resolution for the purpose of this study. Our procedure is explained in detail in the Materials and Methods section. We have added a reference to this section in the figure legend for clarity.
(5) It seems that an insufficient statistical test was used in many experiments. There are comparisons being made between systems (cell migration speed, FRET index...) that are not directly compared in a statistical test. Statistical tests are limited to differences from control (over-expression of full-length vinculin), and consistent increases or decreases (not quantitative values) are taken as evidence of similarity across systems. It seems that a more rigorous and standard approach would be to use an ANOVA/MANOVA with a suitable post-hoc test to perform all of these.
We thank the reviewer for these suggestions.
- For comparisons within a system (between migration speeds only, or between FRET index only...), we selected a priori the most sensible comparisons for our analysis: between the control FL and mutants (Fig. 2B-E, 4B-F), or between front and back cells for each constructs (Fig. 6C,E). To account for multiple comparisons of FRET to FL, we have now re-analysed our data with non-parametric Kruskal-Wallis and unpaired, two-tailed Dunn’s tests, and applied corrections to the p-values with the Benjamini-Krieger-Yekutieli (BKY) two-stage step-up FDR procedure after a check for positive dependency between comparisons by bootstrapping. To account for multiple comparisons of immobile fractions in FRAP experiments, we applied the same BKY corrections to p-values obtained from fit comparisons. To account for multiple comparisons of FRET between front and back cells, we applied the same BKY corrections to p-values obtained from Mann-Whitney tests since none of the comparisons shared any condition. Overall, the corrections resulted in marginal changes in p-values that did not alter our conclusions.
- For comparisons of effects between systems (migration speed vs FRET index...), MANOVA and related approaches are not suited for comparisons between system-level means with different statistical units between systems. To account for effect sizes, we instead performed Pearson correlations or linear regressions, assuming a linear relationship between effect sizes from different systems (Fig. 6A, S2D-G, S4C,H). This allowed us to identify whether mutants have quantitatively similar effects in different systems and which of the mutants do not. Specifically, this is how we identified the YE mutant as different.
- Finally, to quantitatively support our Boolean models of speed dependence on tension, stability and tension gradients (Fig. 2F, 4G, 6F), which binarized the output of each system for the sake of simplicity (fast/slow, tensed/relaxed, stable/unstable, mechanosensitive/insensitive), we have now performed additional regression analyzes between re-scaled data (as detailed in the Materials and Methods). This allows us to quantify the contributions of changes in tension, stability, and the gradients of tension to changes in single cell migration speed, in the difference between single and collective cell migration speed, and in the ratio between collective cell migration speed and wound healing rate. As shown in the added Fig. S2I, the change in single cell speed is explained by a linear combination of changes in tension and stability with contributions of 2_/3 and 1/_3, respectively. In Fig. S4F, the difference between single and collective cell migration speed changes with the stability exclusively, and its increased sensitivity to low stability is well captured by a power law. In fig. S5C, the ratio between collective cell migration speed and wound healing rate is set by the difference in mechanosensitivity between FAs and AJs in which 1/3 of that of AJs counts. Overall, these results confirm our initial binary models.
(6) It is unclear how a lack of vinculin dimerization at adherences junctions perturbs epithelial migration, but the complete lack of vinculin tail, which can also not dimerize, does not. In other words, how can TL "have no other role in cell migration at confluence than those at FAs as in single cells." Notably, the authors do not include the tailless variation in the schematic model figures. These results should be included and explained.
We are sorry that we lacked clarity. Here was our reasoning:
Both the YE and TL mutants are indeed impaired for dimerization but they differ in two ways:
- YE binds actin and is under tension while TL does not.
- YE is not impaired for head-tail interaction and not stable while TL is.
To determine which of these properties affect cell migration, we first compared YE to the other constructs. Just like YE, FL and T12 also bind actin and are under tension, but the migration speeds of FL and T12 cells at confluence can be predicted from their migration speed as single cells, unlike YE. Therefore, tension is unlikely to explain why YE has an effect on migration speed at confluence not predictable from single cell speed. The irrelevance of tension for the scaling of single to collective cell migration speed is also consistent with the fact that TL lacks tension but still obeys the same relationship as FL and T12 between single and collective cell migration speed. Furthermore, FL and T12 also share with TL their stability in adhesions, unlike YE. The instability of YE is then likely to explain its peculiar effect on migration behavior, because this is the only feature that differentiates YE from all the other constructs, TL included. The Boolean models in Fig. 4G and elsewhere are meant to show within a Venn diagram the main molecular conformations and associated interactions of intact vinculin, and their individual and combined consequences at the adhesion and migration level. We have now specified this is the figure legends of the revised manuscript. There is no 1-to-1 mapping of each mutant to one particular conformation of the intact vinculin. Rather, the effects of the mutations on each conformation are specifically indicated in Fig. S2A.
In support of this model, we have added in the revised manuscript a corresponding quantitative analysis, as mentioned in response to Question 3 and 5. Briefly, we first looked for the optimal linear combination of change in tension and stability that could explain the change in the difference between single and confluent migration speed across vinculin constructs. This step excluded tension as a significant contributor to the speed difference. We then looked for a relationship that was an as simple as a weighted combination of two terms (i.e. with only 1 adjustable parameter) that related the changes in speed difference to those in stability only, and found the power law dependence on stability. These results, which confirm our initial analysis, are shown in Fig. S4F. Notably, they show that 1) none of the constructs have an effect on the speed difference between single and confluent cells through tension, and 2) the changes in stability across TL, FL and T12 have a minimal impact on this difference, only that between YE and the others has an impact, as highlighted by the small exponent of the power law. Overall, this is consistent with a model whereby the migration speed of cells at confluence is largely predictable from their migration speed as individuals for WT, FL, TL and T12 (Fig. S4C), with an effect of the mutants TL and T12 similar to that of the unmutated FL. The most parsimonious explanation is that the effects of those constructs on cell speed at confluence are explained by the same mechanisms as their effects on cell speed in single cells, that is, through their FAs, and that the effects of TL and T12 through AJs are not much different from that of FL, meaning that the mutations have no other role in cell migration speed at confluence than those at FAs as in single cells. We have slightly modified this sentence as above in the text to specify that the lack of effect of these mutations on migration is on migration speed specifically, not that they have no effect whatsoever.
Recommendations for the authors:
Reviewer #1 (Recommendations for the authors):
In general, I would suggest that you properly define all notions that you use. Notably, you should explicitly introduce basic concepts like the Furth model, Lévy flight, fractional Brownian motion, the shape index, and others. It gives you the possibility to introduce your notation. For the readers, it would make your manuscript more accessible as they would not need to look up these notions elsewhere. The following points and questions are not exhaustive, but might serve as a guide to improve your write-up.
We thank the reviewer for this suggestion, we have modified the text in the main manuscript to explain the notions in plain English and have added references to the Material and Methods for the more formal definitions, mathematical formulations and additional explanations whenever appropriate. All the modifications are highlighted in the revised manuscript. We hope this revised version will make our manuscript clearer.
You should give the parameter values for all fits that you provide. Table S1 should give them, if I understand correctly, but in the version I have, this table only provides p-values.
We have now added the fit parameters in Table S1 in the revised version. They are also displayed in Fig. S2D and discussed in the associated text.
Why is vinculin or FA stability related to the immobile fraction in FRAP experiments? I would suggest speaking about vinculin or FA lifetimes as measured by the characteristic recovery time obtained from FRAP. How do you interpret the immobile fraction? Do you think that there is an FA ’core’ that persists until the FA is dissolved? In that case, the immobile fraction would report on the relative core size of the FA.
We indeed consider that this is the amount of protein that remains the longer that is the most relevant to assess the molecular stability of adhesions and its impact on migration, because an adhesion must disassemble entirely to allow cell displacement away from that site.
In Figure 1B, the fits of the exponents are not very convincing, and they extend over as they cover hardly one order of magnitude.
We are sorry that figure was misleading. The extreme slopes of 2 and 1 were not meant to indicate the actual time windows where ballistic and pure diffusion would actually be observed but just the values toward which experimental data would tend at short and long lag times, respectively, for a persistent random walk. We hope the new version of the figure in the revised manuscript will clarify this. Additionally, while the range indeed appears short, it is largely sufficient to discriminate between persistent random walk and pure or anomalous diffusion based on the Akaike Information criterion, which is the quantitative and objective metric we exclusively use throughout this work to discriminate between migration behaviors. The scores range from 0 (pure or anomalous diffusion) to 1 (persistent random walk) and table S1 shows that Fig. 1B (among others) is unambiguously a persistent random walk, even within this limited fitting range.
Figure 2B, D, what do the stars indicate?
Stars indicate p-values. We now explain this in the figure legends in the revised version of the manuscript.
Please explain how the FRET index is related to stress?
We are sorry that we had overlooked this important explanation. We have now added explanations in the main text and in the Material and Methods. Figures now also include the direction along which the tension increases when FRET changes.
"This shows that impaired head-tail interaction, regardless of tension, is sufficient for vinculin stability." Sorry, but I do not understand what you mean.
We are sorry that we lacked clarity. We have modified this sentence as follows in the revised manuscript: “This shows that impaired head-tail interaction, whether associated with impaired actin-binding (TL) or increased tension (T12), is sufficient for vinculin stability with FAs.”
I would avoid the term angular memory. You might use a correlation time, that is, fit the curve to an exponential and take the corresponding characteristic time as a measure for persistence.
We thank the reviewer for this suggestion. The reason why we introduced this metric was not to extract a persistence time, which we already obtained from the MSD, but to provide compelling evidence that the super-diffusive behavior at confluence arises from a fractional Brownian motion, rather than from another type of super-diffusive walk. We now provide in the revised manuscript a more detailed explanation of this metric, why we prefer it to the more conventional velocity auto-correlation function for our purpose, along with its formal mathematical definition, as explained below. The angular memory is a modified angular correlation in which we consider the angular deviation θ of each elementary step of duration 10min from its immediate previous step of duration τ equal to the lag time ∆t. Specifically, the angular memory M<sub>i,t</sub>(∆t) of a cell i at time t is the cosine of the angle θ<sub>i,t</sub> between consecutive step vectors between times t−∆t and t and between times t and t+10min, such that the step duration τ of the first vector and lag time ∆t before the next are equal. M<sub>i,t</sub>(∆t) = cosθ<sub>i,t</sub>(∆t) = cos. This metric measures the duration of the past that the cell considers when deciding on its direction. It decays to 0 when ∆t → ∞ for persistent random walkers and remains constant for fractional Brownian motion, similarly to the auto-covariance function of fractional Brownian motion C(∆t/τ) = ((∆t/τ +1)<sup>𝛼</sup> +|∆t/τ −1|<sup>𝛼</sup> −2(∆t/τ)<sup>𝛼</sup>)/2, which is a constant C \= 2<sup>𝛼-1</sup>−1 for all ∆t \= τ, as in our definition of angular memory (while also decaying to 0 when ∆t/τ → ∞) (Metzler et al. 2025). Ballistic motion yields 1 whatever ∆t and uncorrelated random walks, whether anomalous or not, yield 0. For this reason, and because M<sub>i,t</sub>(∆t) was found more robust to noise than C(∆t/τ), we considered this metric a more compelling discriminant between fractional Brownian motion and other uncorrelated super-diffusive walks. Consistently, we found that the angular memory of single cells decayed to 0 while that of confluent cells did not and remained nearly constant within our experimental time frame.
(Of Note, we have increased the analysed data set in the revised Fig. 1D, 3D,F for the sake of representativeness. The conclusions are unchanged).
"While cells adopt a superdiffusive behavior at confluence, they do so in any possible direction." Sorry, but I do not understand what you mean.
We are sorry we were not clear. We changed the sentence as follows : “While cells adopt a super-diffusive behavior at confluence, they do so in the absence of any external guidance cue, and as a consequence without any preferred direction.”
Reviewer #3 (Recommendations for the authors):
(1) The authors have chosen to use an overexpression system. That means that wild-type vinculin is competing with the mutant vinculin in many cases. Also, the authors have chosen not to study a phospho-deficient form of vinculin or manipulate vinculin dimerization in any other way. Given that these are not the most rigorous experiment set-ups and designs possible, the authors may wish to lessen the strength of some of their conclusions.
This is correct. We have added in the discussion the following sentence to account for the possible contribution of consequences of the YE mutation other than dimerization impairment, and for the presence of the endogenous vinculin:
“Whether vinculin dimerization, or another unknown independent function disabled by the YE mutation or phosphorylation of Y1065, is specifically involved in the adhesion and migration phenotypes described above may be addressed in future studies. Likewise, whether the effects are stronger in the absence of endogenous vinculin may be addressed in a null background.” Additionally, we have modified the text throughout to change "dimerization" into "the ability to dimerize".
(2) Dominant effects have been shown for T12 and VinTL; however, to this reviewer’s knowledge, dominant effects have not been shown for vinculin Y1065E.
This is correct, we now specify it in the revised version.
(3) This statement: "Overall, this is consistent with the idea that vinculin acts primarily as a migration moderator, and this supports that insertion of the sensor module weakens to some extent the auto-inhibition of vinculin, thereby contributing to the dominance of all exogenous constructs over endogenous vinculin, in addition to the effects of their mutations" lacks experimental support. Multiple studies have shown that the effects of the insertion of the tension-sensitive module are small on vinculin, including demonstrations that FRAP recovery is unaltered. To make this statement, the authors should perform comparisons of vinculin and vinculin mutants with and without a tension sensor. If differences are observed, it seems that experiments in the manuscript should be repeated with vinculin mutants not harboring a tension sensor. If the presence of the sensor affects mutant behavior, then it seems unlikely that the claims of the paper are correct. For instance, it would suggest that vinculin dimerization is not the sole driver of the velocity coupling, instead it could be a combination of partial activation and lack of dimerization. Alternatively, the authors could remove or edit this sentence.
We agree that the second part of our statement was incorrect and have removed it. We cannot say that the insertion of the sensor module weakens the auto-inhibition and we agree that there is no difference in turnover compared to a vinculin construct without the sensor.
(4) Vinculin dimerization is drawn incorrectly in many of the schematic figures summarizing results. Vinculin dimerization is dependent on F-actin binding (see work of Sharon Campbell and others). Vinculin is unlikely to form a "dimer" in solution, as shown in these schematics. Also, vinculin dimers are thought to form along the length of two bundled actin filaments, not at the end, as drawn in the schematics.
We thank the reviewer for these suggestions. Sharon Campbell and others have indeed shown that actin binding promotes the dimerization of vinculin tails in solution. They and others have shown, however, that vinculin tails also form dimers in the absence of actin, albeit at higher concentrations, which are equivalent to at least 0.1 molecule per 100nm<sup>2</sup> in any 2D section of the volume (DOI: 10.1074/jbc.M111.244293, 10.1021/bi801764a). The density of vinculin within adhesions is promoted by the multiple neighboring vinculin-binding sites on its partners (such as talin, in which the neighboring binding sites are occupied nearly all at once upon talin stretch, doi:10.1021/acs.nanolett.6b00650, 10.1039/c5ib00307e), and is likely to reach a molecule per 100nm<sup>2</sup> in order to accommodate the traction forces that cells exert on their substrates (doi: 10.1021/acs.nanolett.6b04364). This exceeds the threshold concentration for vinculin dimerization in the absence of actin by at least an order of magnitude, and is therefore likely to offer another route for vinculin dimerization in addition to actin binding. We conclude that vinculin density in adhesions is very likely to allow dimerization in the absence of actin, and the binding to actin additionally promotes dimerization. Due to the omnipresence of actin in adhesions, the dimer unbound to actin certainly accounts for a small fraction of possible vinculin conformations, but this is likely to be very context-dependent and current knowledge is too limited to be more quantitative. We have therefore decided to keep our model simple with all these possible conformations, without indications of their possible proportions.
We agree that the drawing of the interaction with actin does not properly depict the side-binding of actin and we have modified it in the revised version. Nevertheless, vinculin and actin are known to bind by both their sides and ends (DOI: 10.1074/jbc.M110.102830).
(5) Technically, the authors have not shown that conservation of angular momentum is "infinite". It just appears longer than the experimentally accessible terms. The use of term short-lived and long-lived, instead of finite and infinite, might be more appropriate.
We agree, we have changed “virtually infinite” to “long-term” everywhere in the text to avoid confusion.
(6) The conclusions of the manuscript seem to treat vinculin loading and vinculin stability as independently regulated. However, vinculin has been shown to act as a catch, both in vitro (see work of Alex Dunn) and in living cells (see work of Sharon Campbell), suggesting strong coupling of these quantities. A discussion of the points would benefit the manuscript.
We agree that vinculin loading and stability in adhesions depend on each other and thank the reviewer for suggesting these additional references. We formulate our model as follows: “Mechanistically, our results support that the head-tail interaction of vinculin challenges both force transmission and stability of vinculin in FAs, as has been observed in other cell types (Carisey et al. 2013, Humphries et al. 2007, Cohen et al. 2006, Dumbauld et al. 2013, Liu et al. 2016, Omachi et al. 2017)...” This shared dependence and factors that independently affect force transmission or stability, but indirectly feed back to each other, are not mutually exclusive, as we state next: “... and is countered by the actin-vinculin tail interaction, which provides force transmission but is unnecessary for stability, and the ability of the tail to dimerize, which associates with higher stability and is dispensable for force transmission (Fig. 2D,E).”
Such feedback implies that the interaction between actin and the tail of vinculin can promote stability, even though it is not necessary for this (as shown with TL), and that the ability of the tail of vinculin to dimerize can promote force transmission, even though it is not necessary for this (as shown with YE). We have added the following sentence in the revised version to clarify this point:
“As a result of this feedback, stability and force transmission can be further promoted by the interaction of the vinculin tail with actin and its ability to dimerize, respectively, indirectly through their antagonism with the head-tail interaction of vinculin.” Consequently, this opens the possibility that a catch-bond between vinculin and actin also promotes vinculin stability in adhesions, which is indeed shown by the Campbell lab (doi:10.1038/s41467-023-43779-x), thanks to a prolonged antagonism with the head-tail interaction. Interestingly, their results show that the ability to form a catch bond does not appear to oppose the head-tail interaction more than actin binding does (their Fig. S10). This suggests that the actin-vinculin catch bond stabilizes vinculin in adhesions through a mechanism that has yet to be identified. In conclusion, vinculin loading and stability in adhesions are interdependent through the head-tail interaction and through another unknown mechanism that involves the catch bond between vinculin tail and actin. We now include a sentence that refers to this catch bond in the revised manuscript.
(7) Recently, two papers have come out assessing the role of AJs and mechanical coupling during epithelial migration: Noordstra, Dev Cell, 2023, and Shoyer, PNAS, 2023. Comparing the results of this manuscript to those of these publications would strengthen the discussion.
We thank the reviewer for suggesting these references.
Noordstra et al provides a very interesting investigation of the mechanisms whereby the mechanosensitive alpha-catenin down-regulates lamellipodial activity in epithelial sheets that form intercellular contacts during wound closure. This could offer a potential mechanism that would explain why we observe that cells at confluence migrate at a lower speed than single cells. While a minimal lamellipodial activity is indeed required for cell migration, we have previously shown, however, that the down-regulation of lamellipodial activity by the formation of AJs does not decrease cell migration speed (doi:10.1073/pnas.1002662107), but the formation of desmosomes does (doi:10.1242/jcs.146316). We have included these considerations in the discussion of the revised manuscript.
The main finding Shoyer et al is that the phosphomimetic mutant S1033D of vinculin relaxes vinculin tension and reverts it to a close conformation in both AJs and FAs. The mutant affects directed collective cell migration in a manner that preserves the relationship between lateral velocity deviation amplitude and coordination length but consequences on vinculin stability in adhesions, or on single and undirected collective migration are unknown. Thus, it would be worth assessing them to determine whether this mutant affects cell migration through FAs, AJs or both, and how it fits in our model. We have now included this reference in the discussion of our revised manuscript.
[Editors' note: The authors included an author response to reviews from another journal]
Reviewer #1 (Comments to the Authors):
In this manuscript the authors describe that cells in collective movements adopt a superdiffusive behavior to out pace individual cells. This behavior is regulated by cell-cell junctional stability and force transmission. The authors state that speed is regulated by vinculin through mechanosensitivity.
While is makes intuitive sense that cells may move more efficiently collectively as it reduces their exploratory space and therefore increases their efficiency of movement,
We agree that this is an intuitive explanation. However, previous literature had shown that confluent cells may or may not migrate depending on conditions that do not solely depend on the space available per cell, but also involve the intrinsic activity of the cell, its cortical tension, and its adhesion with its neighbors, with sometimes counterintuitive effects (doi: 10.1016/J.CEB.2021.07.011). This was the reason that motivated us to investigate how these various ingredients affected space exploration efficiency on different time scales.
Our results indeed refute the intuition that cells move more efficiently when their exploratory space is reduced by showing that the outcome depends on the time scale considered (Fig. S3B). Specifically, on short time scales (less than 3 hours), the area explored by individual MDCK cells is larger than that explored by MDCK cells at confluence. On a longer time scale (greater than 3 hours), however, the area explored by confluent MDCK cells is larger. This switch is a direct consequence of the change in migratory behavior from persistent random walk to superdiffusion, Moreover, its position in time depends on the cell line: extrapolation of our results on RPE-1 cells suggests that it should theoretically occur after approximately 300hrs, if this time scale was experimentally accessible (Fig. S3F).
…the role of junctions specifically is less clear.
We are sorry that we were not able to clearly convey the roles of junctions. We have substantially rewritten our text to address this and all the changes are highlighted in orange. As summarized in Fig. 6F, junctions have three roles. The first role is on persistence, through velocity coordination between neighbors, the second is on speed, through the stability of junctions, and the third role is on directionality, through the sensitivity of the monolayer to the wound edge.
The first role is evidenced thanks to the comparison of the MSD between single cell and confluent migration assays and the use of the alpha-catenin KD cell line. Alpha-catenin depletion is known to be the most potent disruptor of adherens junctions (DOI:10.1091/mbc.e06-05-0471, , DOI:10.1126/science.aaf7119, (DOI:10.1073/pnas.1002662107, DOI:10.1073/pnas.1119313109), and we show that it significantly alters the superdiffusive behavior that emerges in the confluent migration assay (Fig. 3E,F, 5C). Therefore, junction integrity is critical for the control of cell persistence.
Moreover, alpha-catenin depletion induces a loss of velocity coordination between neighbors (Fig. S3E), which we show through numerical simulations to induce superdiffusion (Fig. 3G). By contrast, E-cadherin KO and vinculin mutants have no effect on the superdiffusion of confluent cells (Fig. 3E, 4A). Therefore, the critical molecular ingredient is the link provided by alpha-catenin to the cytoskeleton that provides junction integrity.
The second role of junctions is evidenced thanks to the comparison of cell speeds between single and confluent migration assays with the vinculin mutants (Fig. S4A). Results show that cell speed is reduced of about 10µm/h by confluence, regardless of the mutant except for YE, whose only difference with other mutants is its lower stability (Fig. 4F). This supports that junction stability, and not the other effects of mutants, controls cell speed (we provide a detailed demonstration in the response to the following question). As expected, junction integrity is required as well, as seen from the higher cell speed of the alpha-catenin KD cell line compared to WT (first MSD point in Fig. 3B, E).
The third role of junctions is evidenced thanks to the comparison between confluent and directed migration assays (Fig. 6A). Results show that the wound healing rate is proportional to cell speed at confluence, regardless of the mutant except for YE, which displays no tension gradient in junctions from front to back cells (Fig. 6C). This supports that such gradient is required for cells to identify on which side is the wound edge. As expected, junction integrity is required as well, as seen from the loss of directional bias of the alpha-catenin KD cell line (Fig. 5F).
The authors chose vinculin as the basis by which to manipulate tensions at cell-cell junctions, but this comes with considerable drawbacks. Namely, since vinculin appears at both cell-cell and cell-matrix junctions, its role and the role of its mutations is not clear here. The authors state that the collective migration speed is related to junctional stability, but because vinculin is also at FA, how can this be concluded?
We apologize for the lack of clarity. We hope that the highlighted changes in the revised manuscript will improve this point. As exemplified above, comparing cell migration between isolated cells and confluent cells is essential to enable us to distinguish between the contributions of AJs and FAs. Indeed, since isolated cells lack AJs, the impact of vinculin mutants on single cell migration can only be explained by their effects on FAs. This is how we first determine the effects of vinculin mutants on migration that depend on FAs. Because confluent cells also have FAs, we expect that the effects of vinculin mutants on the migration of isolated cells will still be present in confluent cells, to which will be added the effects of these mutants on AJs and their consequences on migration, if any.
Therefore, when compared to WT cells, if a given mutant decreases or increases migration speed in individual cells, and does so in confluent cells in the same proportion, then its effects at confluence can be entirely explained by its effects in individual cells, and there are no additional effects of that mutant from AJs. This is indeed what we observe for all mutants except the YE mutant (Fig. S4C), leading us to conclude that none of the vinculin mutants, except the YE mutant, have an effect on migration at confluence that results from AJs. In contrast, the YE mutant has effects on migration at confluence that cannot be explained by its effect on individual cell migration. Therefore, the effects of YE at confluence depend on AJs, whether they result from alterations in AJs, FAs, or both. To distinguish between these scenarios, we proceed by elimination, comparing the effects of YE to those of other mutants on force transmission and adhesion stability, and how these two factors associate with migration speed, as explained below. In FAs, YE alters force transmission differently in individual cells and at confluence, but we already know from Fig. 2 that force transmission in FAs cannot alone explain the speed of migration. This result rules out an indirect effect of AJs on cell migration at confluence through FAs. Furthermore, in AJs, YE affects stability and force transmission, but TL has the same effect on force transmission as YE and we already know that none of the effects of TL on migration depend on AJs (Fig. 3, S4C). This result rules out an effect of force transmission in AJs on migration speed at confluence. We conclude that stability at the AJ level, which is the remaining property specifically impaired by YE, is what regulates migration speed at confluence.
The manuscript's logic and flow are not clear in some places, making the story hard to follow. As one example, the FRAP data, which the authors suggest is used to investigate vinculin's combined role does not help in this capacity as the interpretation and its connection to the bigger story are not clear.
We are sorry again for the lack of clarity. We used FRAP data to evaluate the effects of vinculin mutants on adhesion stability. Indeed, mutants have different effects on adhesion stability (Fig. 2E, 4F). In addition, they also have different effects on force transmission (Fig. 2D, 4D,E). The partial overlap in functional alterations caused by the mutants allows us to test the involvement of the overlapping function (here stability) in the overall migration outcome. For example, if two mutants have a similar effect on adhesion stability but different effects on migration speed (such as TL and T12), we can then rule out that speed results from adhesion stability. Similarly, if two mutants have different effects on stability but a similar effect on speed (such as TL and YE), we can also rule out that speed results from stability. We applied the same reasoning to force transmission to conclude that neither adhesion stability nor force transmission alone is sufficient for cells to migrate rapidly. However, the combination of the two enables rapid migration.
As another example, the information derived from the use of the mutants is not clear in the context of the message in the manuscript since they affect cell-cell and cell-matrix junctions and in some places show results that are counter intuitive and not well-explained, to which the authors admit they are surprising but then do not explain their meaning.
As such, it is very hard to follow the logic with regard to the information resulting from the mutant experiments.
We provide above a detailed break-down of our strategy to analyze the results. We regret that our manuscript did not adequately convey our conclusions and we hope that the new version of the manuscript improves this point.
Proliferation has been shown to play a role in wound healing. Does proliferation change in the various conditions?
This is an important point. The average speed of cells at confluence is approximately 20 µm/h (Fig. 4B), which means that each cell moves approximately its own size in one hour. During this time, assuming a 16-hour cell cycle, 6% of the cells would have divided, each of them likely pushing one of its neighbors a distance equivalent to the size of a cell. Therefore, cell proliferation accounts for at most a few percent of the total cell movement. For this reason, we can assume that growth does not account for a large part of the movement we observe. This is consistent with previous work showing that proliferation does not contribute significantly to wound healing (DOI: 10.1073/pnas.0705062104, DOI: 10.1083/jcb.201207148).
Minor comments:
The authors should provide a better description of the mutants: what does a tailless mutant not bind, or bind differently? More context is needed to help interpret the results. While the mutants have all been published on before, it would be helpful to have more information here so that the manuscript is easier to follow.
We are sorry that the information we provided was insufficient. We have now detailed the mutations to help the reader understand how interactions are altered.
Figure 1A is not necessary. Figure 1 overall is fairly predictable as there have been many papers using the persistent random walk as the best model to single cell migration (dating back to the early 1990's). The authors define a new term, angular memory, which they show decreases with increasing delta t as one would predict.
We acknowledge that persistent random walks have already been observed for individual cells, as in references 3-4 cited in the introduction. Nevertheless, we believe that Figure 1 is important because not all cells migrate as persistent random walkers when isolated. Some migrate in a more exotic manner, resulting in superdiffusive behavior, as in references 5-8 cited in the introduction. Since we observe superdiffusive behavior at confluence (Figure 2), it was therefore necessary to show whether or not single cells were superdiffusive too. We also use this figure to introduce angular memory, a measure that, to our knowledge, has never been used before. According to intuition, it decreases to 0 for persistent random walkers, just as another resembling measure, velocity autocorrelation, would do. However, the angular memory of fractional Brownian walkers does not vanish with increasing delta t (Fig. 3D), while velocity correlation would, just as that of persistent random walkers. This difference makes angular memory much more appropriate for distinguishing between the two migration behaviors, and prompted us to introduce it in the first figure as a reference.
In the wound healing assay, which cells were measured? Leading edge or interior, and does it matter?
Figure 5A shows that cells behave differently depending on their distance from the wound. This is because the traces shown correspond to the first few hours of the movie, during which the cells at the front begin to move first. Figure S5A shows the speed of the cells over time after the wound and indicates that the cells reach a stable speed after approximately 3 to 4 hours. Figure S5B shows the speed of the cells as a function of distance from the wound at steady state. These results show that the speed of the cells no longer depends on the distance from the wound at this stage. As indicated in the “Materials and Methods” section, we only considered time points beyond this stage for subsequent analyses of population-averaged MSD and velocity presented in Figure 5, so the location of cells at the front or rear was irrelevant.
Reviewer #2 (Comments to the Authors):
To migrate cells must spatially explore their environments, a process that is guided by intrinsic signals (adhesive and mechanical properties, etc) and extrinsic (gradient cues) signals. This exploration can occur on the single or multicellular level. In this study, the authors examine the effect of cell-cell interactions, guidance cues, and cell mechanics in the exploratory capacity of MDCK cells. The authors show that cell-cell adhesion provides a "infinite directional memory for migration" and cell speed is dependent upon the focal adhesion stability, cell mechanics, and the mobility of adherens junctions-these processes are modulated by vinculin.
My three major concerns with the manuscript are as follows:
(1) While there is potential new information about the role cell-cell junctions and guidance cues play in cell migration, there is not enough NEW insight presented. Rather the role of vinculin in these processes is expected given what is already known about its ability to control focal adhesion stability, mechanics, and adherens junctions.
We agree that our cell migration results make sense based on the effects of vinculin mutants on the stability and force transmission of adhesions. Nevertheless, we argue that this was not the only possible scenario. Indeed, we find that none of the effects of vinculin mutants on AJs (except YE) have an impact on cell migration (Fig. S4C). One might have expected that the increased stability provided by the TL and T12 mutants would reduce the speed of collective cell migration, just as the YE mutant increased cell speed due to its altered stability. This is not what we found, and this reveals a nonlinear relationship between AJ stability and migration speed that could be investigated more thoroughly in future studies. Another example is that the effects of the mutants on force transmission in AJs do not impact migration speed at confluence but do impact directed collective migration (Fig. 6). One might have expected that vinculin-mediated force transmission in AJs would impact collective migration, whether directed or not.
More importantly, we show that the role of intercellular adhesion in cell migration is more complex than expected. Indeed, it depends on the timescale considered: intercellular adhesion is detrimental to short-term spatial exploration and beneficial in the long term (Fig. S3B). Such a timescale-dependent behavior is impossible to predict from previously known effects of the mutants or other molecular considerations. Furthermore, we show that this behavior can be fully explained by the coordination of velocities between neighbors, which depends on intact connections between AJs and the cytoskeleton via alpha-catenin, but is independent of vinculin mutants that connect AJs to the cytoskeleton in parallel with alpha-catenin. One might have expected these connections to also have an impact on velocity coordination, and thus on spatial exploration, but we show that this is not the case (Fig. 3). Finally, we show that directed collective migration has a negligible impact on cell exploration at our experimental timescale (Fig. 5), whereas we initially expected the wound to make migration more ballistic. This reveals that such a directional signal affects spatial exploration at much longer timescales than expected.
Overall, our results quantify the outcome of competing effects and provide timescales at which one effect outweighs the other in influencing cell migration. We believe this is an original approach that provides substantial new insights into collective cell migration.
(2) The phenotypes of the cells expressing the mutant vinculins varying greatly. These phenotypes are not addressed despite the fact that they could potentially complicate the analyses. For example, there are dramatic differences between focal adhesion numbers and sizes in the cells expressing the different vinculin mutants; cell spreading is also dramatically altered. Likewise, the T12 mutant vinculin has previously been reported to have increased adhesive strength, increased traction forces, and cell spreading. How does this knowledge change the interpretation?
We agree that vinculin mutants may have effects on the size and number of FAs, cell spreading, and traction forces that we do not examine here. These consequences can be explained by the effects of these mutants on force transmission in FAs and on their stability, which we report in our work. They do not affect our interpretations. Here, we provide a predictive model of migration speed based on the combination of two consequences of vinculin function, namely stability and force transmission. An interesting avenue for future research would be to assess whether these combinations can be reduced to a single higherlevel effect of vinculin on the cellular phenotype that would be sufficient to predict migration speed. This work remains to be done, as neither FA size and number, cell spreading, adhesion force, nor traction forces alone are sufficient to predict migration speed.
Along the same lines, it has previously been established that tagged version of vinculin do not efficiently integrate into adherens junctions. Published work from the Nelson laboratory suggests that GFP-vinculins do not localize to cell-cell junctions and work from other laboratories suggests localization occurs only when the endogenous vinculin is silenced.
We are aware that some GFP-vinculin constructs may not localize as well as the endogenous protein at AJs. This is due to the localization of the GFP tag on the head of vinculin and depends on the length of the linker between GFP and the head of vinculin. The longer the linker, the easier the interaction with AJ partners. Unlike these constructs, the vinculinTSMod sensors we use in our work do not carry a GFP on the head and do not suffer from the same limitations.
Furthermore, vinculin recruitment to AJs depends on force, with little or no recruitment when tension on the AJs is relaxed (DOI: 10.1038/ncb2055). Vinculin recruitment has in fact already been used as an indicator of AJ tension in Drosophila (DOI: 10.1038/s41467-01807448-8). Consequently, the amount of vinculin visible at the AJs varies depending on the tension exerted on the AJs, which our results confirm: vinculin is more difficult to detect at the AJs in cells located at the front of a wound than in those located at the back (Fig. 6B), which is consistent with the difference in vinculin tension between front and back cells (Fig. 6C) and to the E-cadherin tension gradient between front and back cells (DOI: 10.1083/jcb.201706013). Overall, these results show that vinculin is not always easy to detect at AJs, but this is due to the properties of vinculin, which the constructs we use reproduce better than previous constructs (see also below).
The images in figure S2 and the prebleach images in figure S4 do not show convincing localization of the mutant vinculins to cell-cell adhesions. This is of special concern given that YE mutant protein hardly has any discernable localization to cell-cell junctions; additionally, none of the mutant proteins were tested for their ability to co-localize with adherens junction components. This raises the question if the parameters being examined and the conclusions drawn from them are affected by a difference in localization.
We agree that the recruitment of vinculin at intercellular contacts may be difficult to see.
Besides force-dependent effects mentioned above, other factors are involved. The images shown in Figures S2 and S4 are from live cells in which cytoplasmic vinculin is still present, and its level proportional to the mobility of vinculin. Indeed, the TL and T12 mutants show a more marked contrast between intercellular contacts and the cytoplasm, which is consistent with their greater stability at AJs (Fig. 4F). Conversely, YE shows lower contrast, which is consistent with the lower stability of this construct at AJs (Fig. 4F). The FL construct lies between the two. As a result, the cytoplasmic content can variably mask vinculin recruitment at the AJs depending on the mutant.
We have now performed additional quantifications of mutant recruitment at intercellular contacts as a function of distance from the basal surface of the cells and relative to their recruitment in FAs, in live cells. Results are shown in the new Fig. S4F. We find that all the constructs are recruited to intercellular contacts with a density that is at most half of that in FAs and that varies along the height. FL shows the highest density, localized more apically, consistent with the localization of an AJ-bound actin belt. The mutants appear to be more homogenously distributed along the height of the lateral surface, which may be explained by their impaired autoinhibition (TL, T12), or mechanosensitivity (YE). This variability also contributes to the difficulty in seeing vinculin recruitment in all cells in a single z-slice.
To confirm the proper recruitment of vinculin constructs to AJs we have performed immunofluorescence against alpha-catenin and phalloidin on each of the stable cell lines. Results are shown in the new Fig. S4D and E. In these experiments, cell permeabilization allows for the release of some of the cytoplasmic pool of vinculin, which highlights the recruitment of all vinculin constructs to intercellular contacts. There, all vinculin constructs colocalize with alpha-catenin and F-actin, as expected. Additionally, images displayed are maximum intensity projections to mitigate recruitment variability along the height.
Overall, our results clearly support the localization of vinculin at intercellular contacts, and the differences between the constructs are consistent with the effects of their mutations.
(3) There is a lack of new mechanistic insight. Conclusions are made about a role of vinculin dimerization. This conclusion appears to be based upon the usage of the mutant version of vinculin Y1065. Did the authors directly measure the ability of this mutant protein to dimerize? Is actin binding also affected.
The binding properties of the Y1065E mutant, including its dimerization and binding to actin, have already been characterized by other researchers (ref. 40 in our manuscript, as well as DOI:10.1111/j.1432-1033. 1997.01136.x or DOI: 10.1016/j.febslet.2013.02.042). We assumed that these properties are now well established and can be used to explain higher-level phenotypes that we show for the first time, to our knowledge.
Reviewer #3 (Comments to the Authors (Required)):
Canever et al. tracked two epithelial cell lines on collagen coated glass and showed that isolated cells (non confluent) move as persistent random walkers, whereas confluent monolayers migrate super diffusive, with long range directional memory. By systematically perturbing adhesion machinery they found that focal adhesion mutations mainly tune the speed of single cell tracks, but cannot create long range memory, while force bearing adherens junctions are essential for the super diffusive regime-genetically perturbing them collapses collective memory. These interesting results identify junctional tension as important to switch epithelial cells/sheets between individual and collective search modes - an important quantitative insight that is of clear relevance to cell biologists.
- The presented data is nicely quantitative and convincing, but I have subtle concerns about the generality of the findings. While the authors show that the differential behavior, they describe is not cell-line specific (MDCK, RPE), there are no experiments evaluating the generality of their conclusions across different matrix conditions. How are the measured migration parameters affected by matrix stiffness? Cell migration on collagen coated glass coverslips is a relatively narrow and artificial condition. How is the collective directional memory expected to behave on softer substrates? The generality of the conclusions could be strengthened by repeating measurements using hydrogels of varying stiff-ness. Further, it should be discussed to which tissues in the body the selected matrix conditions and migration modes plausibly apply.
We agree that the generality of our results and the relevance of glass-rigid substrates is an important point. In vivo, epithelial cells rest on a basement membrane with a typical stiffness of approximately 10 MPa, as demonstrated by experimental evaluations on various tissue explants, including renal glomeruli and Bruch's membrane, which are relevant to MDCK and RPE-1 cells (DOI: 10.1111/j.1742-4658.2007.05823.x, DOI: 10.1172/JCI106898, DOI:10.1038/eye.1987.35), we have added these references in the manuscript to support our experimental strategy. In vitro, the most significant effects of substrate stiffness on FAs and cell migration generally occur at much lower stiffnesses, between 0.2 and 100 kPa, and cell phenotypes generally plateau at levels comparable to those observed on glass, even below 100 kPa (DOI: 10.1242/jcs.133645, DOI: 10.1038/ncb3268, DOI:10.1039/c5ib00307e, DOI: 10.1039/c9sm01893j). Furthermore, substrate stiffness has much more moderate effects on confluent cells than on isolated cells. For example, it has been previously demonstrated that confluent layers of MCF10A epithelium showed no change in velocity coordination in the range of 3 to 65 kPa (DOI: 10.1083/jcb.201207148). Therefore, collagen-coated glass appears to be a reasonable model for the basement membrane. Overall, we believe that we have conducted our experiments under physiological conditions, and that our results apply to a wide range of substrate stiffnesses.
- It would be nice to see how long it takes confluent cell layers to close rectangular wounds of defined size when cells migrate as individual (adherens junctions perturbation) versus collective (wt) (on substrates of different stiffness). Presumably, there should be faster wound closure under the collective regime, at least for simple shaped wounds.
This is an interesting question, which our results indirectly address. In our study, we measured the wound healing speed of the WT MDCK cell line as well as lines expressing mutant vinculin constructs (Fig. 6A). These results show that this speed ranges from 5 to 15 µm/h depending on the construct expressed (and for reasons that we explain in the manuscript). These values make it easy to estimate the time required to close a wound based on its width. For example, it would take 5 hours to close a 100 µm wide wound for the WT cell line, which has a rate of 10 µm/h (on both sides of the wound).
Wound closure for cells with disrupted adhesive junctions has already been documented (DOI: 10.1083/jcb.200910041). The results show that wound closure is indeed slower than with WT cells. Although this previous study does not reveal the underlying causes, our work now shows that there are two factors: weaker directional memory due to impaired intercellular coordination and, in the longer term, an additional lack of sensitivity to the guidance signal provided by the wound.
- Akin to substrate stiffness variation, I am missing experiments that test the effect of cytoskeletal tension on these migration modes. Experiments with Rho kinase or myosin inhibitors could meaningfully broaden the scope of this study.
Rho kinase or myosin inhibitors applied to cells during the time required to assess migration patterns (a movie recorded overnight is necessary to obtain a statistically reliable calculation of MSD over 3 to 4 hours) are likely to affect many other cellular processes in addition to the cytoskeletal tension directly involved in migration. We believe that the accumulation of these effects will make interpretation of the results very difficult. For example, it has been shown that inhibition of ROCK by Y27 promotes healing of corneal endothelial lesions by affecting proliferation through cyclin D and p27 (DOI: 10.1167/iovs.13-12225), or by improving respiration, which would provide the energy necessary for migration (DOI: 10.1096/fj.202101442RR). Consistently, another study on HaCaT epidermal cells confirms that myosin phosphatase accelerates wound healing through proliferation (DOI: 10.1016/j.bbadis.2018.07.013). In contrast, in HUVEC cells, ROCK inhibition significantly impaired the proliferation and migration of vascular endothelial cells in vitro in a dose-dependent manner (DOI: 10.1097/ICO.0000000000000493).
Furthermore, previous studies have highlighted that differential contractility at the subcellular level is important for collective migration (DOI: 10.1038/ncb2133, DOI: 10.1083/jcb.201706013), which is not possible to examine with global activation or inhibition of contractility. This prompts the development of more refined and specific measurement and disruption strategies to assess the respective impact of cytoskeletal tension on cell-cell and cell-matrix adhesion mechanisms. Our work, which uses biosensors to assess how this tension differentially affects cell-cell and cell-matrix adhesions, is a step in this direction. The localized spatio-temporal activation or inhibition of myosin subtypes or Rho GTPase regulators specific to these adhesion structures will likely answer these questions in the future, but we believe that the development and application of these approaches will require a substantial amount of work that goes beyond the scope of our study.
Reviewer #2 (Public review):
Summary:
In this manuscript, the authors aim to assess how the functional organisation of the hippocampus is related to the geometry and neurobiological differences of the hippocampus. In particular, the authors focus on the first three eigenvectors of hippocampal-cortical functional connectivity, based on non-linear dimensionality reduction on resting-state functional MRI data. Furthermore, the work aims to describe changes in these functional axes and their relation to other factors throughout youth and evaluate whether they are predictive of individual variations in cognition.
Strengths:
A major strength of this study is the attempt to replicate key findings across multiple developmental cohorts.
Weaknesses:
The major weaknesses of the manuscript center on gaps in technical transparency and several conceptual inaccuracies. The machine learning methodology used for cognitive prediction is scarce, leaving little means to evaluate whether the behavioral results suffer from data leakage or overfitting. The introduction sets up an oversimplified historical premise regarding the field's understanding and appreciation of hippocampal connectivity, and contains several incorrect references that throw doubt on the argumentation. Additionally, T1w/T2w signal intensity is incorrectly used as synonymous with myelin, despite gold-standard histological validation showing a non-significant correlation between T1w/T2w and myelin staining (Sandrone et al., 2013).
Appraisal of Aims and Conclusions:
The authors partially achieve their aims by illustrating certain age-related changes in hippocampal function; however, the correlative study design is not equipped to examine how these changes are "shaped" by geometry, myelination, or gene expression (especially the latter two). Furthermore, conclusions were often overstated based on small effect sizes.
Context and Field Impact:
This work adds to a growing body of literature focused on gradient-based representations of hippocampal topology. By applying these methods across a wide developmental age bracket, it provides a useful reference point for how the hippocampus and wider cortex interact during maturation. To improve utility to the neuroimaging and cognitive neuroscience communities, the nesting of subfields within the eigenvector topology should be addressed, too.
Reviewer #1 (Public review):
Summary:
The authors studied the development of hippocampal connectivity gradients based on open datasets and performed correlation analyses with other MRI features as well as gene expression information from other datasets. Although the main findings are correlational and cross-sectional, the analyses are overall sophisticated and replicated in several datasets.
Strengths:
The hippocampus is a key region in understanding large-scale brain organization and cognition, and the authors applied advanced and suitable analytics to study its development. The paper is overall well-organized and well-written, and the findings are relevant for studying large-scale brain development.
Weaknesses:
While sophisticated, several of the analyses appear mainly correlational, cross-sectional, and rely on cross-dataset contextualization, which should also be stated as a limitation of the current work.
eLife Assessment
This valuable study compares hippocampal-cortical functional connectivity to various other brain measures and examines their development across youth. It uses sophisticated analyses replicated in multiple datasets, but provides incomplete evidence to support the primary claim that hippocampal-cortical connectivity relates to cognitive maturation. The manuscript would benefit from a more nuanced consideration of the biological basis of some of the derived imaging measures and the limitations of the cross-sectional design. This work will be of interest to neuroimaging specialists and cognitive neuroscientists.
eLife Assessment
This study investigates the temporal dynamics of neural activity preceding self-initiated movements and makes a valuable contribution to this field. To overcome methodological and analytical limitations in previous work the authors introduce a matched control condition in which no response is required to assess when neural activity is predictive of self-initiated movement (using electro- and magnetoencephalography). Applying solid methods and machine-learning analyses, the authors show that a late-stage neural event is most predictive of upcoming movements (~100 ms before the response), whereas earlier neural activity is less informative. This finding contradicts previous work that suggested that earlier build-up of neural activity (e.g., starting 500 ms before the response) reflects the planning and preparation of movement.
Reviewer #2 (Public review):
[Editors' note: this version has been assessed by the Reviewing Editor without further input from the original reviewers.]
Summary:
The authors set out to investigate how well the onset of a self-initiated movement could be predicted at different times prior to action onset. To do so, they collected EEG and MEG data across 15 human participants who watched natural landscape images on a screen. These participants performed active self-initiated movements, or observed passive actions, to have a new image appear. By comparing the neural activity prior to active and time-matched passive actions, the authors found that even though a build-up of neural activity is visible close to 1s prior to action, action onset could only be reliably predicted around 100ms prior to action. These results confirm what was already suggested in previous literature: the commitment to action is only clear from the late stages in the visible neural ramp-up to action onset.
Strengths:
(1) The paper presents a well thought out methodology to assess the predictive value of neural activity prior to a self-initiated movement and passively observed action, while keeping all other experimental factors identical. This methodology can be applied outside the specific scope of this paper as well, in efforts to assess the correspondence of a neural signature with an observed behavior.
(2) The results are a strong confirmation of what was suggested less clearly in previous research (Trevena & Miller, 2010, Consciousness & Cognition; Schmidt et al., 2016, Neuroscience & Biobehavioral Reviews; Travers et al., 2020, NeuroImage).
Comments on revised version.
I thank the authors for addressing my concerns in the Introduction and Discussion of their paper. I don't require any further edits at this point.
Author response:
The following is the authors’ response to the previous reviews
Public Reviews:
Reviewer #2 (Public review):
Summary:
The authors set out to investigate how well the onset of a self-initiated movement could be predicted at different times prior to action onset. To do so, they collected EEG and MEG data across 15 human participants who watched natural landscape images on a screen. These participants performed active self-initiated movements, or observed passive actions, to have a new image appear. By comparing the neural activity prior to active and time-matched passive actions, the authors found that even though a build-up of neural activity is visible close to 1s prior to action, action onset could only be reliably predicted around 100ms prior to action. These results confirm what was already suggested in previous literature: the commitment to action is only clear from the late stages in the visible neural ramp-up to action onset.
Strengths:
(1) The paper presents a well thought out methodology to assess the predictive value of neural activity prior to a self-initiated movement and passively observed action, while keeping all other experimental factors identical. This methodology can be applied outside the specific scope of this paper as well, in efforts to assess the correspondence of a neural signature with an observed behavior.
(2) The results are a strong confirmation of what was suggested less clearly in previous research (Trevena & Miller, 2010, Consciousness & Cognition; Schmidt et al., 2016, Neuroscience & Biobehavioral Reviews; Travers et al., 2020, NeuroImage).
Comments on revised version.
I thank the authors for addressing my concerns in the Introduction and Discussion of their paper. I don't require any further edits at this point.
Recommendations for the authors:
Reviewer #1 (Recommendations for the authors):
(1) Please more concretely define the parameters of 'early' and 'late' time windows in the manuscript so that the reader can better understand the difference between these with respect to the time-course of effects relative to a motor response.
Those changes were addressed in the previous round of revisions (please refer to previous round). Namely, we made the following addition:
“The lower bound of meta-analytic estimates of awareness, at 250 ms before movement onset (Braun et al., 2021), can be used as a pragmatic threshold to classify neural signals as “early” (more than 250 ms before movement onset) or “late” (less than 250 ms before movement onset).”
(2) Code and data do not appear to be uploaded to the linked OSF repository. We understand that this may typically occur after the reviewed preprint is available via eLife; however, please ensure that this is made publicly available. Because such materials were not available, we were unable to verify some of the method's details and went by what was reported in the preprint.
We apologize for the oversight, the link to the code was in the description of the OSF data repository that was linked in the data availability statement. We have now updated the data availability statement to directly reflect links to both the data (OSF) and the code (GitHub):
“Data used in this study are publicly available on the OSF repository: https://osf.io/wm95q/.
Code used in this study is publicly available on GitHub: https://github.com/lucasjeaybizot/Slideshow2025.”
(3) For the sentence: 'Proponents of the early-decision account have taken early brain activity, like the RP, to support the thesis that decisions happen early in the brain, taking these early signals to reflect a cascading process that brings about the forthcoming action.' it would be very informative to cite the papers that propose this account, as only the Schurger et al. (2021) paper is cited in this paragraph.
We have modified the sentence to the following to include the original result (Libet et al., 1983), an influential review that describes this interpretation (Haggard, 2008), and an influential study that applied such interpretation (Fried et al., 2011):
“Proponents of the early-decision account have taken early brain activity, like the RP, to support the thesis that decisions happen early in the brain, interpreting these early signals to reflect a cascading process that brings about the forthcoming action (Libet et al., 1983; Haggard, 2008; Fried et al., 2011).”
(4) In Figure 2C, the units are not displayed in the colour bar. These should be included so that the colours can be better interpreted.
The units have now been added to the figure.
(5) It is unclear why the time-frequency power estimates (and multiple comparisons procedures) extend to an upper limit of 80 Hz (as displayed in Figure 2C), despite the key signal of interest only spanning 10-30 Hz (Mu/Beta range). Extending the frequency range this high results in a larger number of comparisons and a potentially stricter multiple comparisons correction. Please justify why this larger frequency range was used, given this potential drawback.
The rationale for extending the time-frequency decomposition to 80 Hz was to characterize the spectral landscape over the same frequency range that was available to the Haar-AdaBoost classifier, whose features extended up to 80 Hz. The purpose of this analysis was therefore not limited to testing the canonical mu/beta ERD range, but also to visualize whether other frequency-specific activity was present in the broader feature space used by the classifier. We have clarified this rationale in the text:
“We examined this broader frequency range, rather than restricting the analysis to the canonical Mu/Beta bands, to characterize the time-frequency landscape over approximately the same spectral range available to the Haar-AdaBoost classifier, whose frequency-sensitive features extended from 2 to 80 Hz.”
(6) For the time-frequency analysis, it is stated that 'Here, we used complex Morlet wavelets to decompose our data at C3 into 78 logarithmically spaced frequency bands from 2 Hz to 80 Hz with logarithmically decreasing cycle counts'. Do the authors mean logarithmically increasing in this case? Usually, more cycles are used for higher frequencies to achieve a similar time/frequency precision balance across lower and higher frequency wavelets. Fewer cycles for higher frequencies is highly unusual. This should be corrected if it is an error.
We thank the reviewer for catching this important error. We indeed used an increasing cycle count. We have modified the text accordingly by changing the description to “increasing”.
(7) Regarding the sentence: 'We used standard procedures for identifying and interpolating bad channels in the EEG recordings.' Please specify what those procedures are, as they are not necessarily standard across different research groups, and multiple methods exist.
We have modified the section to better describe the procedures:
“We manually identified bad channels and interpolated them using a weighted average of the 5 nearest channels.”
(8) For the ERP analyses of differences between the active and passive conditions, the parameters of the cluster-based permutation test should be specified (e.g., the alpha levels for cluster inclusion and cluster-level statistical significance).
We have added the following to better characterize the cluster-correction parameters:
“Samples exceeding a two-sided uncorrected threshold of were grouped into temporal clusters, and the cluster statistic was defined as the sumα of = 𝑡.-05 values within each cluster. Statistical significance was assessed using 1,000 permutations, with clusters considered α =. 05 significant at a cluster-level based on the two-sided permutation distribution.”
Reviewer #2 (Recommendations for the authors):
(1) I am curious how the current results would compare to an analysis in which pre-action neural activity is compared to baseline and post-action neural activity. In other words, activity at any point in time during which the person was at rest, rather than an explicit time-matched passive action condition. I would expect that neural activity predictive of an upcoming action or intention will look significantly different from the neural activity in time at any other moment at which the person is not intending to act.
We agree that comparing pre-action activity with unconstrained resting activity could be informative and would likely reveal differences associated with preparation for action. However, this comparison would address a somewhat different question from the one targeted here. Our goal was to determine whether activity preceding self-initiated movement contains information specific to the forthcoming movement when compared with a temporally matched condition that controls for elapsed time and anticipation of the slide transition. An arbitrary resting baseline would not provide these controls and could therefore introduce differences unrelated to movement commitment. We agree that such an analysis could be useful in future work.
(2) It would be good to discuss the current results in comparison with those of Trevena & Miller (2010, Consciousness & Cognition).
We have addressed this comment in the previous round of revision:
“Previous work showed that these early ramping signals were not specific to the outcome of a motor decision (Trevena & Miller, 2010). However, because their analyses remained time-locked to decision outcomes, they could not directly test whether the observed slow buildup specifically reflected a commitment-like process.”
(3) The order of reporting throughout the paper is somewhat confusing, as many results are provided before the "Results" section. For example, there is repetition in the "A new approach to self-initiated movement analysis" section, compared to the "Results" section. The argument for why event-locked potentials are not necessarily predictive in value is also repeated multiple times throughout the paper. The paper could benefit from some restructuring to avoid repetition.
We appreciate this point. Some repetition is intentional because the Introduction motivates the methodological problem and previews the logic of the analyses before the formal Results section. We deem this contextual overlap necessary for adequately understanding the rationale for the study at each stage of the manuscript, we have therefore retained the current overall structure.
(4) Is there a particular reason why this paper includes and compares both EEG and MEG data?
The earlier held-out dataset collected at PF included both EEG and MEG, whereas the independent main dataset collected at OC used EEG only. We used the PF EEG data for model selection and parameterization, while the independent OC EEG dataset served as our main test set. Showing that the same overall pattern was present across EEG and MEG also provided an opportunity to assess whether the result generalized across recording modalities.
(5) Is there a particular reason why the AdaBoost results are compared to the outcome of an SVM and Random Forest?
We compared AdaBoost with SVM and Random Forest as part of the model-selection procedure on the held-out PF dataset. Each algorithm was evaluated with a range of feature representations, and Haar-AdaBoost produced the highest AUC in the window immediately preceding slide transition. It was therefore selected for the primary analyses. We additionally compared Haar-AdaBoost post hoc with several standard EEG classifiers, including CSP-LDA and EEGNet, to assess whether the main temporal pattern depended on classifier choice. This supports our claim that lack of early classification is due to non-separability of the data rather than classifier selection.
eLife Assessment
Nui and colleagues tackle several questions concerning the assembly of the heterotetrameric kinesin-1 complex, examining intermolecular interactions between KHC and KLC and KHC with the cargo adaptor TRAK2. Aspects of the study using computational models and biopysical/biochemical analysis to examine the interaction between KHC and KLC and KHC and TRAK2 are useful, alongside an interesting new model for modulation of TRAK2 binding by KLC. The study will be of great interest to the community, but currently it is incomplete, with some claims partially supported.
Reviewer #1 (Public review):
Niu and colleagues tackle several questions concerning the assembly of the heterotetrameric kinesin-1 complex, comprising kinesin heavy chains (KHCs) and kinesin light chains (KLCs), its interaction with adaptor proteins (the mitochondrial adaptor TRAK2), and the implications of these interactions for kinesin autoregulation.
Broadly, the study comprises four parts; strengths and weaknesses are described for each in turn.
First, the authors examine the coiled-coil interface that enables tight binding between KHC and KLC. Binding-site mapping supports previous observations that KHC and KLC interact through the CC3 region of KHC and the amino-terminal coiled-coils of KLC. These regions are modelled using AlphaFold3, and the resulting models are used to inform mutagenesis experiments that identify critical residues within the hydrophobic core of the interface. Although the binding site has previously been reasonably well defined, both biochemically and using computational models, this work extends those findings to individual residues at the interface and provides supporting biophysical analysis. This provides useful experimental validation of the computational models and important tools for the field to enable the selective uncoupling of KHC and KLC.
Second, the authors examine how the mitochondrial adaptor TRAK2 interacts with KHC. Previous studies have defined the critical regions of KHC CC4 and TRAK2 reasonably well, and the authors' biochemical mapping broadly supports these findings. They extend this through biophysical analysis and computational modelling that proposes a 2:2 stoichiometry and an interesting assembly in which CC4 separates to enable formation of a four-helix bundle with TRAK2. This AlphaFold model is supported by biophysical, biochemical, and cell-imaging analyses, including the identification of point mutations that disrupt the interaction. These findings are important because the molecular basis of KHC CC4 interactions with diverse cargo adaptors has remained somewhat opaque, and this study begins to place these interactions on a firmer molecular footing. Some of the principles proposed here may also extend to other KHC-binding adaptors.
Third, and of particular interest, the authors examine the interplay between KLC and TRAK binding to KHC. This is important because previous work has generally favoured the view that mitochondrial transport is independent of KLCs, although some studies disagree. Whether KLCs are excluded from TRAK-containing complexes and vice versa and, if so, how this occurs has remained unclear. The authors provide evidence suggesting that the extended coiled-coil region of KLC, outside the direct high-affinity KHC-binding interface, modulates CC4 to prevent TRAK binding, suggesting that KHC-TRAK and KHC-KLC complexes may be mutually exclusive. Precisely how this mechanism translates into transport regulation within the cell remains unclear, but the effort to establish a structural hypothesis for understanding this interplay is valuable.
Fourth, the authors provide evidence for a novel, weak interaction between CC1 and CC4 that they propose may contribute to kinesin-1 autoinhibition and its regulation by KLC and TRAK-an 'autoinhibitory latch'. I found this part of the work more preliminary because, although the biophysical and biochemical analyses provide some support for such an interaction, unlike other aspects of the study, it is less well supported by computational modelling or mutational analysis. Moreover, there is no direct evidence for how this interaction might function in the context of full-length kinesin-1 autoinhibition, for example through assays of motor activity or conformational state, and the proposed model is not particularly clear with respect to full-length KLC (including its TPR domains) or regulation by MAP7.
Taken together, the study makes a valuable contribution to our understanding of several facets of kinesin-1 assembly, cargo recognition, and regulation. Key strengths lie in the complementary biochemical, biophysical, computational, and cell-imaging approaches used to support most of the key findings, together with the definition of several very useful point mutants that should inform future work.
There are, however, some general weaknesses surrounding unnecessary or overstated claims of novelty and the connection to prior work, and some factual errors. These can be straightforwardly addressed through better integration of the existing literature into the narrative, without detracting from the authors' contributions. Some conclusions appear to extend beyond the available data, and these should be tempered accordingly. Computational models should also be presented more thoroughly and systematically, together with confidence metrics and multiple ranked outputs. The Methods should more clearly describe the constructs and sequences used so that the work can be straightforwardly reproduced, alongside information on replication and additional quantification. Some clarification around measured stoichiometries and the authors' interpretation for some of the short coiled-coil fragments will also be important.
Reviewer #2 (Public review):
Summary:
Niu et al. develop a biochemical pipeline using systematic co-immunoprecipitation from cell lysate coupled with in vitro binding studies to test hypotheses generated based on structure prediction of how the kinesin heavy chain is autoinhibited through intramolecular interactions, as well as interactions with the kinesin light chain. Additionally, the authors provide insight into kinesin heavy chain binding to the cargo adaptor TRAK. Structural understanding of kinesin regulation has been elusive due to apparent flexibility of the motor. The advent of protein structure prediction, validated through cross-linking mass spectrometry (XL-MS), has provided an inroad to understanding kinesin regulation (Tan et al., Weijman et al.). Most recently, Ashaduzzaman et al. have used cross-linking and local refinement to obtain a reconstruction of the autoinhibited KHC:KLC tetramer. The biochemical analysis presented here extends previous studies and has enabled the authors to define the hierarchy of competing KHC interactions with KLC and the TRAK2 adaptor.
Strengths:
While other studies relied predominantly on XL-MS to validate autoinhibitory contacts in the KHC:KLC autoinhibited tetramer, the present study used in-cell interactions to define minimal interacting units, and measured the effects of prediction-guided mutagenesis on their interaction in vitro. This complementary approach stratifies the multitude of predicted autoinhibitory interactions and offers a functional test for reported structures and structure predictions.
Weaknesses:
The work is not appropriately placed in the context of previous literature. The KHC:KLC assembly interface (Figures 1 and 2) was previously predicted by Weijman et al., supported by XL-MS in Tan et al., and subsequently visualized by cryo-EM in Ashaduzzaman et al. Contacts between the regions designated in this study as KHC CC1 and CC4 (Figure 6) were also reported by Tan et al. and Ashaduzzaman et al. Similarly, a structural model of TRAK2:KHC (Figures 3-4) was previously reported and tested by co-immunoprecipitation and mutagenesis by Gladkova et al. Tan et al. additionally mapped binding of the related adaptor TRAK1 to KIF5B using XL-MS. These results should be discussed throughout to help the reader understand whether biochemical observations are consistent with prior work, or whether they hint at previously unknown regulatory mechanisms.
To test the KHC:KLC and KHC:TRAK2 predicted interfaces, mutations of hydrophobic residues were introduced in KHC (Figures 2 and 4), and TRAK2 (Figure 4). However, both KHC and TRAK2 exist as homodimers in solution. Are these mutations specific to destabilizing the heterotetramers, or do they also affect the KHC and TRAK2 homodimers? Such an effect could provide an alternative explanation for reduced binding and impaired mitochondrial recruitment of KIF5C by TRAK2 mutants, such as L217Q/L220Q. Retained mitochondrial localization of TRAK2 mediated by the C-terminus is not a good proxy for functionality of the N-terminal motor-binding region. Mutants should be assessed for stability and dimerization propensity, and their effects fully described.
Several structure-prediction models are introduced without pLDDT and PAE confidence plots. These are necessary to assess structure prediction confidence. Additionally, residue number labels would help orient the reader as to the boundaries of the protein fragments modelled.
Reviewer #3 (Public review):
The authors report an interaction between KHC and KLC1 through the coiled coil 3 of KHC, and then characterise the binding of TRAK2 to KHC through coiled coil 4 of KHC. The reporting of this second interaction is novel. The authors make a number of mutants which change the affinity of TRAK2 for KHC, but it is not clear if it ultimately prevents binding. They then demonstrate through competitive binding that TRAK2 binds more tightly to CC4 than CC1 and therefore can trigger release of Kinesin-1 auto-inhibition. The authors, however, do not show directly that the kinesin is no longer inhibited. This could be tested using negative stain EM, SAXS or motility assays in the future. The experiments are performed with immunoprecipitation pulldowns coupled with western blotting in cells, and ITC with recombinantly purified proteins.
Strengths:
The data showing the binding of KLC1 to KHC and the TRAK2 binding to KHC are interesting.
The ITC data looks very good from the profiles. It seems, however, that the experiments were done once rather than replicated (at least twice, ideally three times to accommodate variation). The size-exclusion chromatography experiments showing competition between different adaptors to KHC are nicely done, and the SEC-MALS is informative in terms of determining the stoichiometry and size of the assembled complexes. The work is performed with recombinant proteins and therefore removes any ambiguity about the interactions in cells.
Weaknesses:
The experiments lack appropriate controls, with some experiments incomplete, and the statements claimed are not supported by the data. There are some problems with figure and data presentation and the number of replicates/lack of quantification. The validation of AlphaFold models and predictions is missing. The cell biology experiments are preliminary.
There is a concern with the cell biology experiments. All experiments are done with transfections in the presence of the endogenous proteins. It is not clear if the proteins are overexpressed and they appear cytoplasmic. Beyond this, it is not possible to claim co-localization of the transfected proteins.
The manuscript would benefit from general editing to improve understanding. The names used to describe the methods are incorrect. This has implications for the solidity of the statements and conclusions of the paper.
Author response:
We thank the editors and reviewers for their careful and constructive evaluation of our manuscript. We are encouraged that the reviewers recognize the value of our biochemical, biophysical, structural-modeling, and cell-based analyses of kinesin-1 holoenzyme assembly, KHC–TRAK2 recognition, and the regulatory interplay between KLC1 and TRAK2. We also appreciate the concerns regarding the relationship of our findings to previous work, the strength of several mechanistic conclusions, validation and presentation of the structural models, and the need for additional controls and quantitative analyses.
We plan to revise the manuscript substantially in response to these comments. Our provisional responses and planned revisions are summarized below.
Reviewer #1
We agree that the novelty of several findings should be more carefully defined in the context of the existing literature. In the revised manuscript, we will better integrate previous biochemical and structural studies of KHC–KLC assembly, TRAK-family adaptor binding to KHC, and kinesin-1 autoinhibition. In particular, we will explicitly compare our findings with the work of Diefenbach et al. [1]., Weijman et al. [2]., Tan et al. [3]., Ashaduzzaman et al. [4]., and Gladkova et al. [5]., and distinguish previously established structural features from the additional residue-level mutational validation and quantitative biophysical analyses provided in our study. We will accordingly revise statements of novelty throughout the manuscript.
We also agree that our current data establish a direct CC1–CC4 interaction and its disruption by TRAK2 and KLC1, but do not directly demonstrate that this interaction is itself autoinhibitory or that its disruption activates full-length kinesin-1. We will therefore distinguish the experimentally observed CC1–CC4 interaction from the proposed regulatory model and temper terms such as “autoinhibitory latch,” “activation switch,” and related mechanistic conclusions where appropriate. We will also expand the Discussion to consider how this interaction may relate to the established roles of full-length KLC and MAP7 in kinesin-1 regulation.
To strengthen the presentation of the AlphaFold3-based models, we will provide model-confidence information, including pLDDT, PAE, and ipTM metrics, specify the exact sequences and residue boundaries used for each prediction, and include additional prediction outputs where informative. We will also more clearly distinguish experimentally supported conclusions from structural hypotheses, particularly regarding the proposed steric and/or conformational effects of the extended KLC1 coiled-coil on the KIF5C CC4 region.
Finally, we will strengthen the quantitative and methodological presentation of the study. We will provide clearer construct information, explicitly report experimental replication, quantify the mitochondrial enrichment and cellular distribution analyses in Figures 4 and 5, and revisit the interpretation of stoichiometries obtained with shorter coiled-coil fragments. Statements implying mitochondrial transport will be revised to reflect that the current cellular experiments directly demonstrate mitochondrial recruitment rather than organelle motility.
Reviewer #2
We agree that the relationship between our study and previous structural work should be made substantially clearer. Although several relevant studies were cited in the original manuscript, the extent to which our findings confirm, extend, or differ from these previous models was not sufficiently explained. We will revise the Introduction, Results, and Discussion accordingly, including explicit comparisons with the structural predictions of Weijman et al. [2]., the XL-MS-supported model of Tan et al. [3]., the recent cryo-EM/integrative model of Ashaduzzaman et al. [4]., and the TRAK–KHC work of Gladkova et al [5].
We also appreciate the concern that mutations introduced into hydrophobic interfaces could affect the intrinsic stability or homodimerization of KHC or TRAK2, rather than selectively disrupting heterotypic interactions. To address this alternative explanation, we plan to evaluate the oligomeric states and stability of key interface mutants using appropriate biophysical approaches, including SEC-MALS and thermal-stability measurements. These results will be incorporated into the revised manuscript and used to reassess the interpretation of the corresponding binding and cellular recruitment experiments.
As noted above, we will also provide substantially more information on the confidence and reproducibility of the AlphaFold3 predictions, including pLDDT, PAE, and ipTM metrics, and will clearly annotate the residue boundaries of all modeled fragments.
Reviewer #3
We agree that the cellular experiments require additional quantification and clearer presentation. We will quantify the mitochondrial enrichment of KIF5C and the effects of KLC1 co-expression on the intracellular distribution of KIF5C and TRAK2 across independent experiments. We will also include appropriate controls for endogenous/background fluorescence, improve image and scale-bar presentation, and revise the corresponding terminology so that the conclusions accurately reflect the experiments performed.
We will strengthen the reproducibility and presentation of the biochemical and biophysical data. Key ITC measurements will be supported by independent repeats and reported with the corresponding binding parameters and variability. For the SEC experiments in Figure 3, we will include SDS-PAGE analysis of relevant fractions and more clearly describe the composition of each protein mixture. Appropriate positive controls will also be added to the Co-IP experiments where requested.
We will further improve the presentation of the structural models by providing confidence metrics and residue annotations, and we will carefully revise the terminology used to describe SEC, SEC-MALS, ITC, and other biochemical approaches. The Abstract and main text will also be edited to more clearly describe the experimental approaches used and to ensure that all mechanistic conclusions are matched to the strength of the available evidence. More broadly, we will revise the manuscript to sharpen its central message while avoiding overstatement. In particular, we will distinguish direct experimental observations—such as KHC–KLC1 assembly, KHC–TRAK2 recognition, KLC1-dependent modulation of TRAK2 engagement, and the CC1–CC4 interaction—from mechanistic models that remain to be tested more directly.
We expect to complete these revisions and associated experiments within approximately 10–12 weeks. We thank the editors and reviewers again for their constructive comments, which we believe will substantially improve the rigor, clarity, and positioning of the study.
(1) Diefenbach RJ, Mackay JP, Armati PJ, Cunningham AL. The C-terminal region of the stalk domain of ubiquitous human kinesin heavy chain contains the binding site for kinesin light chain. Biochemistry. 1998 Nov 24;37(47):16663-70. doi: 10.1021/bi981163r.
(2) Weijman JF, Yadav SKN, Surridge KJ, Cross JA, Borucu U, Mantell J, Woolfson DN, Schaffitzel C, Dodding MP. Molecular architecture of the autoinhibited kinesin-1 lambda particle. Sci Adv. 2022 Sep 16;8(37):eabp9660. doi: 10.1126/sciadv.abp9660.
(3) Tan Z, Yue Y, Leprevost F, Haynes S, Basrur V, Nesvizhskii AI, Verhey KJ, Cianfrocco MA. Autoinhibited kinesin-1 adopts a hierarchical folding pattern. Elife. 2023 Nov 1;12:RP86776. doi: 10.7554/eLife.86776.
(4) Ashaduzzaman M, Tang Y, Okada K, Fried SD, Mckenney RJ, Al-Bassam J. Structural basis of kinesin-1 autoinhibition and its control of microtubule-based motility. Sci Adv. 2026 Jul 17;12(29):eaeg1267. doi: 10.1126/sciadv.aeg1267.
(5) Gladkova C, Paez-Segala MG, Grant WP, Kittisopikul M, Myers SA, Wang Y, Vale RD. A molecular switch for coordinating kinesin and dynein transport of mitochondrial cargo. Science. 2026 Aug 6;393(6811):601-606. doi: 10.1126/science.aeh1475.
eLife Assessment
The authors use a modeling approach to examine the effects of local and global perturbations on the electrophysiological properties of astrocytes: their modeling results suggest that local ionic changes associated with synaptic activity alter the membrane potential in fine distal processes but not in the soma. The work is valuable because it addresses a long-standing question in the field regarding key electrophysiological properties of astrocytes. The conclusions are convincing, but the manuscript would benefit from providing more details about the assumptions underlying the model and from incorporating more realistic astrocyte morphologies into the simulations.
Reviewer #1 (Public review):
Summary:
Nakatani and De Schutter conduct an in-depth computational investigation of astrocyte electrophysiology. Utilizing a conductance-based NEURON model with realistic astrocyte morphology, the study replicates both in vitro and in vivo experimental protocols to characterize local and global depolarizations of the astrocyte membrane. The findings indicate that bath-applied (global) potassium stimuli induce membrane potential changes that follow the potassium equilibrium potential throughout the cell, whereas localized potassium stimulation disrupts this behavior. In the latter case, the local response is governed by the ratio of local conductances. The authors further analyze the observed electrical compartmentalization to identify which molecular components directly affected by synaptic stimuli - such as potassium clearance, glutamate transport, and GABAA and NMDA receptors - could plausibly generate this effect under physiological conditions. The study concludes that perisynaptic astrocyte processes (PAPs) can respond to specific synaptic inputs in an electrically isolated manner, exerting minimal influence on the soma.
The central prediction - that astrocyte processes can be electrically compartmentalized from the soma under local synaptic stimulation, while bath stimulation produces a globally uniform response - is both interesting and surely significant for understanding and measuring astrocyte electrophysiology. However, the manuscript's incomplete model description and disorganized Results section hinder assessment of how well the conclusions are supported. Addressing these issues could render the model a valuable resource for the community and a useful framework for guiding and interpreting voltage-imaging experiments in astrocytes.
Strengths:
(1) The study addresses a timely, long-standing question and may pioneer a multiscale approach to quantifying astrocyte electrophysiology beyond the traditional view of an electrically passive, isopotential glial syncytium. The findings are consistent with recent voltage-imaging evidence of localized depolarization in astrocyte processes.
(2) A key strength of the study is the use of a state-of-the-art, morphologically realistic compartmental model that captures both somatic and peripheral dynamics. This approach addresses spatial and temporal resolutions that are challenging to achieve experimentally.
(3) Simulating both in vitro-like (bath perfusion) and in vivo-like (local, synapse-driven) stimulation protocols within a unified framework may explain discrepancies observed between experimental results obtained under these differing conditions.
(4) The model predictions are benchmarked against experimental voltage-imaging data, and the systematic analysis of individual molecular contributors-including potassium channels, glutamate transporters, and synaptic receptors-yields testable predictions.
(5) The topic is relevant to a broad and multidisciplinary audience, including researchers in glial biology, synaptic physiology, and computational neuroscience.
Weaknesses:
(1) Terminology and clarity
The manuscript relies heavily on jargon and abbreviations, which impedes readability. The central concept of a "Nernstian" depolarization or response, introduced in the title, is unconventional and not explicitly defined. It appears that "Nernstian" refers to a membrane potential that follows the potassium Nernst potential (EK) of the astrocyte. This terminology is ambiguous, as the analysis is restricted to K+ and does not ensure that the astrocyte response excludes contributions from other ion Nernst potentials.
Furthermore, "Nernstian" reflects a modeling choice rather than an alternative such as the Goldman-Hodgkin-Katz formalism. Additional terms in the abstract, including "a variety of biochemicals," "astrocyte peripheries," "isolated peripherals," and "empirical" model, are also ambiguous. Because these terms are central to the main claim, they may confuse readers.
(2) Framing of the research question
The introduction is brief and does not clearly articulate the research question or the manner in which the study addresses it. The motivation, which should draw on cell anatomy (including the gap-junction-coupled syncytium), molecular composition, and stimulation protocol, is underdeveloped, and the knowledge gap is not explicitly identified.
(3) Incomplete model description
Model design, parameters, and stimulation protocols are often not described or justified, and references are missing (e.g., for K2P, TREK1/TWIK1, and the Na+/K+ pumps). In particular:
a) it is not explained how the three ionic species (Na+, K+, Cl−) needed to account for synaptic currents interact and are modeled in the extracellular and intracellular spaces, nor how they contribute to the astrocyte membrane potential;<br /> b) the diffusion formulation (Equation 9) neglects any tangential component of diffusion, an approximation that is left unjustified;<br /> c) gap junctions are modeled as ohmic currents driven by the potential difference between astrocytes, but the network arrangement, location, and distribution of the coupled astrocytes and of the gap junctions in the simulations are not specified;<br /> d) the analytical results (Equations 46-50) are not obviously derived from Equation 16, and the essential derivation steps are missing;<br /> e) it is unclear whether and how the voltage indicator ArcLight is modeled when comparing simulated traces with experimental fluorescence data.
Overall, the simulation setup lacks clarity, which limits reproducibility and impedes evaluation of the study's conclusions.
(4) Organization and rationale of the results
The Results are difficult to interpret owing to unclear organization and an ambiguous rationale. The derivation and application of the scale parameter ΓK (Equation 1) are unclear; in particular, its dependence on axial conductance appears inconsistent with the assumption that intracellular diffusion does not influence PAP responses (cf. van Egeraat et al., Biophys. J. 1993).
(5) Unsupported or missing results
Several statements are not supported by the data presented: some claims could not be matched to any evidence (e.g., L104-106), results described in the text could not be located in the figures (L159-164), comparisons are reported as "data not shown" (L140-143), and the source of the "realistic ECS sizes" (L191-194) is unclear. The authors do not demonstrate the validity of the extracellular K+ perturbation used to mimic action potentials.
(6) Presentation of figures
The figures are often crowded and stylistically inconsistent. Most importantly, in Figures 3, 4, and 6, the rendering of the noisy experimental data obscures the simulation results they are meant to benchmark, making it difficult for readers to assess agreement between model and experiment.
Reviewer #2 (Public review):
Summary:
Nakatani et al. developed a conductance-based, morphologically realistic computational model of an astrocyte soma and its perisynaptic processes (PAPs) to examine how extracellular potassium shapes the astrocytic membrane potential across spatial scales. The model builds on previously published morphological and electrophysiological data, as well as experimental measurements of transporter and ion-channel expression levels. It incorporates the astrocytic potassium channels Kir4.1, the Na⁺/K⁺-ATPase, the K2P leak channels TREK-1 and TWIK-1, and gap-junctional coupling to neighboring processes via connexin-43. Glutamate transporters (GLT-1), NMDA receptors, and GABA_A receptors are also included to examine neurotransmitter-mediated depolarization.
The authors propose that a compartmental model that explicitly distinguishes the soma from the PAPs, provides important insights into realistic potassium uptake and the electrophysiological properties of these distinct compartments.
Strengths:
(1) Clearing extracellular K⁺ is one of the most fundamental functions of astrocytes. Failures of K⁺ buffering are implicated in epilepsy, spreading depolarization, and several neurodegenerative conditions. Yet how K⁺ handling is divided between perisynaptic processes and the rest of the cell, and how this shapes local membrane potential, remains poorly understood. The study gives a mechanistic account. The soma behaves as a low-resistance K⁺ "sink" with little voltage changes for local K⁺ elevations around the PAPs, while producing much smaller, than expected by simply by Nernst equation. The local membrane potential deviates from E_K because current escapes to the rest of the cell. This framing separates physiological, synapse-scale K⁺ handling from global, pathological K⁺ load.
(2) The model described here fills an important gap in computational modelling of astrocytes. While computational models are now central to neuroscience, astrocytes and their interactions with neurons remain relatively underrepresented in mainstream computational modeling. Existing models mostly treat either whole-cell K⁺ buffering or the isolated perisynaptic compartment. This is novel in that it connects the two within one framework and provides a tool to address how astrocytes integrate local and global signals.
(3) Including realistic morphology shows that spatial heterogeneity matters. A key conceptual strength is that the model does not treat the astrocyte as a single, well-mixed, isopotential compartment. Using an EM-informed morphology with perisynaptic astrocytic processes, it shows that stimulus location and spatial extent shape the membrane response. Fine processes, constricted segments, and local conductance ratios generate responses that cannot be predicted from whole-cell properties alone. These questions are difficult to address experimentally because fine processes are below the resolution of conventional electrophysiology, and somatic recordings are limited by space-clamping. An in silico approach is therefore especially valuable, as it allows controlled manipulation of individual variables, such as stimulus location, channel density, and extracellular volume, that cannot be varied independently in tissue.
(4) It is built using standard, open tools, and therefore the model can be extended with ease by future studies. The model is implemented in NEURON, the most widely used simulator for biophysically detailed cells, with mechanisms written in NMODL and driven through Python. It also uses a previously published astrocyte morphology. Parameters are listed, and the authors state that the model files are in a repository. Together, these make the model accessible and reproducible if the repository link is also provided.
(5) The model is anchored to experimental data. Kir4.1 and K2P leak conductance were fitted to experimental current-voltage data (Figure S2), and other channel parameters were constrained by the literature. Notably, the authors avoided channel formulations with built-in mechanisms that fix the resting potential, which are common in earlier astrocyte models. The resting potential and K⁺ responses therefore emerge from the conductance rather than being imposed. The model also reproduces established whole-cell behavior, including Nernstian responses to bath K⁺ application, before being used for new predictions.
(6) The study tests several candidate mechanisms for peripheral depolarization within one framework and offers a way to reconcile conflicting results: K⁺ through Kir4.1, GLT-1, Na⁺/K⁺-ATPase, K⁺/glutamate spillover, and astrocytic GABA_A and NMDA receptors, all of which in principle can produce responses of the observed depolarization under various conditions. The model therefore offers a biophysical explanation for discrepancies in the astrocyte literature.
(7) The work makes several testable predictions:<br /> a) the local [K⁺]o required for a given PAP depolarization;<br /> b) local depolarization depends on the spatial extent of activation;<br /> c) somatic recordings would not detect PAP depolarizations.
Weaknesses:
(1) The Nernst equation itself does not appear to break down; local E_K should still follow it. Rather, the membrane potential deviates from E_K because low-resistance electrical coupling allows current to spread to the rest of the cell. I suggest rewording this statement for accuracy.
(2) The Methods mention an uploaded repository, but no link is provided. The authors should include a public link and deposit the model in ModelDB or a comparable repository.
(3) The deployed morphology is hippocampal, not cortical. But the paper's conclusions are for cortical somatosensory astrocytes based on morphological similarity. That deserves additional explanation, especially since PAP geometry is central to the conclusions.
(4) Astrocytic NMDA receptors best fit the observed depolarizations. However, because the experiments in Armbruster et al. were performed in the presence of APV, an NMDA receptor antagonist, the authors suggest that the responses may involve APV-insensitive NMDA receptor subunits. This explanation would be stronger if supported by evidence that these subunits are expressed in cortical astrocyte PAPs.
(5) For activation of GABA_A receptors on astrocyte PAPs, the source of GABA is unclear, since the stimulated ascending axons are glutamatergic. The model shows these receptors could explain the data, not how they might be activated. This deserves an explanation.
(6) I also recommend adding a discussion paragraph on the functional implications of heterogeneous astrocytic control of potassium homeostasis. Do the results suggest that astrocytes efficiently buffer K⁺ under normal conditions but fail during widespread K⁺ accumulation?
Reviewer #3 (Public review):
Summary:
This study uses computational modeling to examine how local synaptic activity influences astrocyte membrane potential and, in particular, whether the classical Nernstian relationship between extracellular potassium concentration and astrocyte membrane potential is maintained throughout the astrocyte. The authors focus on the possibility that local ionic changes associated with synaptic activity produce membrane potential responses in fine astrocytic processes that differ from those predicted by a global Nernstian relationship. The study addresses an important and longstanding question in astrocyte physiology and provides a potentially useful computational framework for examining how cellular morphology, membrane conductances, and local synaptic inputs shape astrocyte membrane potential. The central conclusion is that Nernstian behavior can break down during local synaptic input, particularly in fine astrocytic processes. The modeling framework could therefore be useful for understanding how the spatial organization of astrocytes influences their responses to local neuronal activity. However, the strength and generality of this conclusion depend on several biological assumptions that are not sufficiently considered in the current model or discussion.
Strengths:
This computational modeling addresses a longstanding question in astrocyte physiology dating back to the 1960s: whether astrocyte membrane potential (VM) follows the Nernstian prediction in response to changes in extracellular K+ concentration ([K+]e). Significant recent progress in astrocyte studies has provided a renewed opportunity to revisit this question. In this regard, the effort to develop a conductance-based NEURON model that incorporates understanding of astrocyte structure and physiology is highly valuable. The model has also shown the potential to simulate how changes in astrocyte membrane Na+, Ca2+, and Cl+ permeabilities, resulting from synaptic inputs, could alter the relative K+ permeability, and therefore the Nernstian behavior of astrocyte VM at different subregions of astrocytes. These questions are important to astrocyte physiology and remain technically challenging to experimentally evaluate. However, several concerns about the omission of recent studies on astrocyte structure and function from the model and the physiological interpretation of the Nernstian behavior of astrocytes need to be addressed before the conclusions can be adequately supported.
Weaknesses:
(1) The premise of the modeling:<br /> The cellular context should be clearly stated. The model is intended to simulate rodent grey matter cortical protoplasmic astrocytes, but not white matter fibrous astrocytes or other specialized astrocyte populations, which differ morphologically and, to some extent, functionally.
(2) The architecture of astrocytes:<br /> This model can be refined by incorporating recent advances in our understanding of astrocyte morphology (PMID: 35283239, PMID: 31550514). For example, (a) the branching pattern of astrocytic processes, particularly the terminal processes (or leaflets), which is a key element of the model; (b) the location of synapses within an astrocytic domain. Specifically, synapses can make contact with virtually every structural component of an astrocyte, from the soma to different levels of processes, including endfeet contacting blood vessels. This anatomical evidence is in sharp contrast to the prevailing perisynaptic astrocytic process (PAP) concept adopted in the modeling; and (c) the existence of a large quantity of reflective astrocytic processes that loop back and form gap junctional coupling at their stemming points that wrap around bundles of axons or large dendrites, forming a scaffold-like architecture (also see PMID:32312988 in your reference list). This may create regions of strong local astrocyte VM response to activity-dependent elevations of [K+]e, which could be an important feature to consider in the modeling.
(3) Astrocyte membrane potential is not an autonomous feature of individual astrocytes:<br /> If the goal is to simulate astrocyte VM, astrocyte syncytial coupling cannot be ignored. Specifically, the very low astrocyte-to-astrocyte pathway resistance (4.2 MΩ) provides strong electrical coupling that equalizes the VM of astrocytes within a syncytium to a uniform level, a property termed astrocyte syncytial isopotentiality (PMID: 26435164; PMID: 30277762). Experimental studies have demonstrated that coupling among 7-9 astrocytes is sufficient to maintain this isopotential state. Thus, the VM of an individual astrocyte is not determined solely by its own K+ conductance and transmembrane K+ gradient, but rather the result of the electrotonic spatiotemporal summation of voltage across the syncytium. A realistic model that seeks to predict astrocyte VM should incorporate syncytial coupling as a fundamental component.
(4) The historical debate about the Nernstian behavior of astrocyte VM in response to [K+]e:<br /> The authors should acknowledge the longstanding debate over the Nernstian behavior of glial VM in response to changes in [K+]e. This relationship was first described in the seminal neuroglia studies of Kuffler and colleagues in 1966, using optic nerve neuroglia, which are likely equivalent to fibrous astrocytes in rodent white matter. They reported a slope of ~58 mV per 10-fold change in [K+]e (PMID: 5966434). This interpretation was subsequently challenged by a series of in vivo studies of neuroglia VM in cortical grey matter, including Pape and Katzman (1972; PMID: 5016004), Ransom and Goldring (1973; PMID: 4805015), and Futamachi and Pedley (1976; PMID: 1276917). Importantly, the Futamachi study provided early in vivo evidence consistent with syncytial isopotentiality, although the existence of the astrocyte syncytium could not be demonstrated experimentally until the later work of Ransom and colleagues in 1981 (PMID: 7248773). This historical literature should be introduced in the Introduction to establish the significance of the present modeling work.
(5) Astrocyte syncytial isopotentiality:<br /> The authors should consider that syncytial isopotentiality is a system-wide feature of both grey and white matter astrocytes. At minimum, this means that somatic astrocyte VM recordings are not consistent with a simple Nernstian prediction of VM in response to changes in [K+]e. Therefore, a model of astrocyte VM should account for syncytial coupling rather than treating VM as a property determined by individual astrocytes.
(6) Distinguish K+ Nernstian behavior from multi-ion permeability and transporter-mediated flux:<br /> Conceptually, the high and near-exclusive K+ permeability of the resting astrocyte VM provides the basis for applying the Nernst equation to describe VM. In the present, Na+ and Ca2+ permeabilities are also introduced without a clear distinction between the conditions under which a Nernst equation remains applicable and those in which it does not. The authors should clearly define Nernstian behavior as referring specifically to K+ permeability. When the Na+ and Ca2+ permeabilities, resulting from the activation of AMPA/NMDA receptors, are involved in astrocyte across-membrane ionic activities, the VM follows the GHK prediction. Additionally, the model includes Na+-dependent transporters, which do not represent channel-mediated permeability. The authors should therefore distinguish channel-mediated ionic permeability from transport-mediated ionic flux when describing and modeling changes in astrocyte VM.
Author response:
We thank the reviewers for the overall positive review that declares our results convincing. We will revise the paper to improve the presentation of the model and results as suggested by the reviewers. The revision will be accompanied by a more detailed response.
eLife Assessment
This important study advances the HIV-1 vaccine development field by establishing six humanized immunoglobulin knock-in mouse models to facilitate the design and evaluation of HIV-1 vaccines. The evidence supporting the claims is compelling, with robust genetic engineering and rigorous baseline sequencing data, while additional assessment of BCR pairing, expression, and repertoire diversity would round out the characterization of these models. These mice offer the field a rich resource to evaluate vaccine candidates that induce broadly neutralizing antibodies against major epitopes of the HIV-1 envelope protein.
Reviewer #1 (Public review):
Summary
In "Mouse models with human antibody repertoires for inducing multiple lineages of HIV-1 broadly neutralizing antibodies", Tian et al. report the production of 6 humanized mouse models to support HIV-1 vaccine design studies aiming to induce different classes of broadly neutralizing antibodies.
These mice were designed to express immunoglobulin genes of human bNAbs targeting four different conserved epitopes of the HIV-1 Envelope while allowing the development of CDRH3 and CDRL3 diversity during gene rearrangements.
Strengths:
(1) The manuscript is clearly written, and findings are clearly presented with self-explanatory figures. The Introduction and Discussion sections provide very valuable context to understand the relevance of these mouse models and are accessible to a general reader. It is appreciated that the data are presented with transparency, highlighting both strengths and weaknesses of the mouse models.
(2) The characterization of the mouse models is overall thorough and rigorous. Immunization data to confirm functionality are not presented; however, it is understandable that this information will be presented in the context of subsequent studies. The authors indicate that these mice have been distributed to different teams.
(3) These mice will be very valuable tools for the HIV-1 vaccine community to evaluate vaccine candidates in a "closer-to-human" setting that allows a certain level of polyclonality while using human immunoglobulin genes.
Weaknesses:
(1) The authors investigate the use of the knocked-in human immunoglobulin genes in heavy chains and light chains and provide the frequency of productive rearrangements using these genes. However, there is no information about the frequency of the right heavy-light chain pairs. This information would be important to calculate the real frequencies of potential bNAb precursors in these mouse models.
(2) The authors use HTGTS-rep-seq to investigate immunoglobulin gene rearrangements at the genomic DNA level. This method is very useful for analyzing gene recombination because it provides information about both productive and non-productive rearrangements. However, this method does not provide information about BCR expression, and it is not optimal for investigating repertoire composition. BCR expression would provide a more rigorous characterization of the mice and confirm functionality.
(3) The goal of this study was to produce models that could provide a diverse bNAb-like repertoire. Specific quantification of the diversity of the repertoire is not provided. The study does not provide information about the sequences that do not use the intended genes.
Overall, the availability of these humanized mouse models will be very valuable for the HIV-1 vaccine community and will provide valuable pre-clinical models to advance the design and testing of vaccine candidates.
Reviewer #2 (Public review):
Tian & colleagues generate and characterize six novel humanized mouse models engineered to express precursor human antibody repertoires for inducing multiple lineages of broadly neutralizing antibodies (bnAbs) against HIV-1. Building on prior "VRC01-class" proof-of-concept work, this manuscript advances the field by demonstrating the value and suitability of these sophisticated models for testing the ability of vaccine candidates to induce bnAbs. By utilizing flow cytometry and HGTS sequencing techniques, the authors demonstrate that the targeted bnAb minigenes successfully undergo physiological V(D)J recombination and junctional diversification. Despite phenotypic heterogeneity in B-cell development, they show that these models reproducibly produce diverse repertoires enriched for functional humanized B-cell receptors expressing knock-in-derived bnAb precursor rearrangements, a subset of which yield the long, correctly charged CDRH3 loops required to target conserved but glycan-shielded bnAb epitopes. This achievement successfully overcomes a major vaccine testing bottleneck by providing a pool of patient-deduced precursor configurations in sufficient amounts to reliably study how subsequent somatic mutation events are induced by candidate vaccine protocols (with the potential exception of the PG9 model). Crucially, the comparative utilization of Kymab data serves as an excellent control to demonstrate the extreme rarity of bnAb rearrangements in fully locus-humanized mice, beautifully illustrating the design limitations of humanized models without adjusting for the much smaller repertoires of mice. Finally, the discussion is nuanced and well-written, including articulating the models' free availability and highlighting why these small-animal systems are complementary-and sometimes essential-in relation to rhesus macaque systems.
The paper's conclusions are robustly data-supported, with the only slight weakness being that the Kymab data were extrapolated rather than derived from actual head-to-head analyses. The comments below predominantly aim to provide further context and clarity-mainly in the abstract, results, and discussion sections-to this already excellent, highly detailed study. Refining the text clarity and phenotypic observations will further polish this impressive manuscript.
(1) Abstract Clarity:<br /> Concluding sentences should incorporate the discussion's explanation: finding natively low-frequency precursors in standard humanized locus mice is statistically prohibitive due to their smaller absolute B-cell compartment. These models bypass this restriction by maximizing relevant segment expression to expand the functional CDR3 search space.
(2) Results (Line 183, BG18 expansion):<br /> The intriguing expanded B1 cell population warrants more conservative commentary regarding genuine autoreactivity versus other possibilities, such as hyperactive baseline signaling via autophosphorylation (as shown by Hassan Jumaa for certain HC/LC clonotypes). While beyond this manuscript's scope, these mechanisms are differentiable via calcium flux assays with general anti-BCR signaling reagents.
(3) Results (p15/Line 324, PCT64/CH235 Vκ Allelic Architecture):<br /> The text states that lacking a recombination cassette caused hVκ3-20 to account for only a fraction of rearrangements. While Figure 1B's general engineering feature (tandem reverse hJκ-RSS 23 and 12 RSS hJκ) implies human Vκ should predominate everywhere, Figure 1C suggests hVκ3-20's low representation in CH235 and PCT64 models stems from being introduced differently-specifically via pre-rearranged minigenes rather than unaltered RSS orientations in an unrearranged cassette. The authors should clarify this point in the text and/or Figure 1.
(4) Results (p14, Clever Engineering of λLC rearranging mice):<br /> Replacing mCκ with hCλ3 is a clever genetic trick and an important proof of concept for stably accommodating a human lambda rearranging light chain cassette within the mouse kappa locus. This excellent methodological highlight deserves prominent recognition as a template for other model developers. I suggest giving this further play in the discussion.
(5) Discussion (Model Comparisons & Baseline Resolution):<br /> The authors aptly compare these humanized mice to macaques; in addition to their superior scalability, they point out that macaques don't recapitulate human-specific bnAb-relevant configurations in certain instances (like lacking the VH1-2 minigene for VRC01-class responses, or utilizing germline D segments that favor V2 apex responses, potentially overestimating human responses). To further enhance nuance, referencing other prior review KI model literature on this topic (some published by these authors) would be beneficial. For context, KI-derived adoptive transfer models offer more potential diversity (thus, 'higher bar' off-target response testing) and frequency control but are restricted to short priming studies. Conversely, the rearranging models in this study support full vaccine regimens and generate multiple variant bnAb precursors to compare potential in vivo response plasticity.
(6) Discussion (Baseline phenotypes):<br /> One limitation of the cellular flow data, albeit beyond the manuscript's current scope, is that the baseline phenotypes predominantly reflect variations of a fixed pair of V(D)J rearrangements, rather than bona fide patient-derived UCA bnAb precursors with correct charges/lengths. To help resolve pre-antigenic functional signaling boundaries and thus potentially aid immunogen design, the discussion could note how future work could aim to use techniques like single-cell transcriptomics to map expressed selection signatures of bnAb precursor paired V(D)J's to rule out anergy or survival/death gene signatures.
(7) Discussion (PG9 Model & HCDR3 Length Constraints):<br /> While the PG9 model's failure to produce bona fide precursors is unsurprising given its long HCDR3, the authors should provide a bit more detail as to whether a subset of PG9 variants with slightly shorter HCDR3s are permissive for bnAb production (analogous to DH511/10E8). Providing the specific HCDR3 sizes generated by this model in relation to recently discovered monkey antibodies-which are shorter/different but function like PG9 bnAbs (e.g., Guenega et al Nature 2026) - would offer excellent context. Such related lineage and/or lineage precursor variants could serve as structural alternatives, and highlighting how these length constraints impact selection provides vital boundary conditions for the immunogen design field when planning germline-targeting regimens.
Author response:
Reviewer #1 (Public review):
(1) The authors investigate the use of the knocked-in human immunoglobulin genes in heavy chains and light chains and provide the frequency of productive rearrangements using these genes. However, there is no information about the frequency of the right heavy-light chain pairs. This information would be important to calculate the real frequencies of potential bNAb precursors in these mouse models.
We acknowledge that our bulk sequencing-based repertoire analysis does not assess the pairing of immunoglobulin (Ig) heavy and light chains. To add this information, we plan to sort IgM<sup>+</sup>IgD<sup>hi</sup> naive B cells from the mouse models and use single-cell RT-PCR to analyze paired heavy and light chains from individual B cells. Since the assay detects Ig transcripts, it will also address comment #2 about the lack of expression data for the human Ig segments.
The main goal of the single-cell RT-PCR experiment is to verify that the human Ig segments in the mouse models can be expressed as B cell receptors (BCRs) in naive B cells. However, because of its low-throughput nature, the assay is unlikely to be sensitive enough for detecting bnAb precursors. As shown in Figure 5G, the frequencies of bnAb-like CDR H3s for DH270, BG18 and PCT64 are in the range of 10<sup>-4</sup>. The frequencies of paired bnAb-like heavy and light chains will be even lower. It will be difficult to robustly measure the frequencies of such rare BCRs even with 10x Genomic single-cell sequencing. For reference, 10x Genomic sequencing of B cells from 10E8 KI mice yielded 1026 heavy and light chain pairs (1). The sample size was adequate for analyzing light chains paired with the pre-rearranged 10E8 heavy chain, which is expressed in most B cells. The sequencing depth will be insufficient for detecting rare bnAb precursors in the rearranging mouse models of this study. Deep bulk sequencing data were used to calculate bnAb precursor frequencies in our previous VRC01 mouse models and in humans and nonhuman primates (2-8). It is possible to enrich bnAb precursors for single-cell analysis by sorting epitope-specific B cells (9-11). However, this method requires sorting probes that exhibit high affinity and selectivity for bnAb precursors. We believe immunization of the mouse models will be the most direct and definitive way to detect bnAb precursors and test their functions in these mouse models. As mentioned in the manuscript, we have already provided the mouse models to several investigators for immunization experiments.
(2) The authors use HTGTS-rep-seq to investigate immunoglobulin gene rearrangements at the genomic DNA level. This method is very useful for analyzing gene recombination because it provides information about both productive and non-productive rearrangements. However, this method does not provide information about BCR expression, and it is not optimal for investigating repertoire composition. BCR expression would provide a more rigorous characterization of the mice and confirm functionality.
It is true that HTGTS-rep-seq does not directly show the expression of productive rearrangements. As outlined in our response to comment #1, we plan to perform single-cell RT-PCR to confirm the expression of the integrated human Ig segments in IgM<sup>+</sup>IgD<sup>hi</sup> B cells.
We have used HTGTS-rep-seq extensively to analyze V(D)J rearrangements. Overall, the outcomes of the repertoire analysis correlated well with functional assays of BCR expression. For example, we identified VRC01-class heavy and light chains in VRC01 mouse models with HTGTS-rep-seq; correspondingly, immunization of the mouse models elicited VRC01-class antibodies (7, 8, 10). As another example, we used HTGTS-rep-seq to characterize V(D)J rearrangements at different stages of B cell development (12). The ratio of productive/nonproductive V(D)J rearrangements increases as B cells mature. The correlation suggests that the productive rearrangements, as detected by HTGTS-rep-seq, are selected by virtue of their expression as BCRs, which are required for B cell maturation and survival.
(3) The goal of this study was to produce models that could provide a diverse bNAb-like repertoire. Specific quantification of the diversity of the repertoire is not provided. The study does not provide information about the sequences that do not use the intended genes.
In our previous analysis of the VRC01 mouse models, we calculated the proportion of unique CDR3s and CDR3 overlap among different animals and used these parameters to represent repertoire diversity (7, 8). We will perform the same analysis for the six mouse models in this study. The V(D)J rearrangements that do not use the human Ig segments are composed of mouse Ig segments, and we will show repertoire analysis of these rearrangements in the revised manuscript.
Reviewer #2 (Public review):
(1) Abstract Clarity:
Concluding sentences should incorporate the discussion's explanation: finding natively low-frequency precursors in standard humanized locus mice is statistically prohibitive due to their smaller absolute B-cell compartment. These models bypass this restriction by maximizing relevant segment expression to expand the functional CDR3 search space.
We will make the suggested change in Abstract.
(2) Results (Line 183, BG18 expansion):
The intriguing expanded B1 cell population warrants more conservative commentary regarding genuine autoreactivity versus other possibilities, such as hyperactive baseline signaling via autophosphorylation (as shown by Hassan Jumaa for certain HC/LC clonotypes). While beyond this manuscript's scope, these mechanisms are differentiable via calcium flux assays with general anti-BCR signaling reagents.
We are currently examining the cause of B1 cell expansion in the BG18 model. Our preliminary data suggest that some of the B1 cells express autoreactive antibodies, but more work is needed to reach a conclusion. Until then, we are open to alternative possibilities, as suggested.
(3) Results (p15/Line 324, PCT64/CH235 Vκ Allelic Architecture):
The text states that lacking a recombination cassette caused hVκ3-20 to account for only a fraction of rearrangements. While Figure 1B's general engineering feature (tandem reverse hJκ-RSS 23 and 12 RSS hJκ) implies human Vκ should predominate everywhere, Figure 1C suggests hVκ3-20's low representation in CH235 and PCT64 models stems from being introduced differently-specifically via pre-rearranged minigenes rather than unaltered RSS orientations in an unrearranged cassette. The authors should clarify this point in the text and/or Figure 1.
The recombination cassette, as diagrammed in Figure 1B, is not in the CH235 and PCT64 models. Instead, the hVk3-20 allele in these models was derived from one of the previous VRC01 mouse models (7). As described in the Luo et al paper (7), hVk3-20 replaces mouse Vk3-7 in the Igk locus, but all the other mouse Vk segments are intact and functional. In this configuration, hVk3-20 rearranges as one of the Vk segments in Igk allele and accounts for about 10% of total Igk rearrangements. We will clarify this point in the revision.
(4) Results (p14, Clever Engineering of λLC rearranging mice):
Replacing mCκ with hCλ3 is a clever genetic trick and an important proof of concept for stably accommodating a human lambda rearranging light chain cassette within the mouse kappa locus. This excellent methodological highlight deserves prominent recognition as a template for other model developers. I suggest giving this further play in the discussion.
We will highlight this design in the discussion.
(5) Discussion (Model Comparisons & Baseline Resolution):
The authors aptly compare these humanized mice to macaques; in addition to their superior scalability, they point out that macaques don't recapitulate human-specific bnAb-relevant configurations in certain instances (like lacking the VH1-2 minigene for VRC01-class responses, or utilizing germline D segments that favor V2 apex responses, potentially overestimating human responses). To further enhance nuance, referencing other prior review KI model literature on this topic (some published by these authors) would be beneficial. For context, KI-derived adoptive transfer models offer more potential diversity (thus, 'higher bar' off-target response testing) and frequency control but are restricted to short priming studies. Conversely, the rearranging models in this study support full vaccine regimens and generate multiple variant bnAb precursors to compare potential in vivo response plasticity.
As suggested, we will include additional references and comparisons of the Ig humanized mouse models with KI mice. The reviewer raised an important point about off-target responses. Adoptive transfer models have the full mouse antibody repertoire. By comparison, the antibody repertoires of the Ig-humanized mouse models are dominated by a select set of human Ig segments, with minimal contribution from mouse antibodies. From this perspective, the adoptive transfer model provides a more diverse range of potential off-target antibodies. On the other hand, the Ig-humanized mouse models may be superior to wild-type mice in producing antibodies against the immunogenic glycan holes in envelopes (13). Reaching these glycan holes may require sufficient CDR H3 length, and based on CDR H3 length distribution, Ig-humanized mouse models could be more capable of recapitulating this type of off-target response than wild-type mice.
(6) Discussion (Baseline phenotypes):
One limitation of the cellular flow data, albeit beyond the manuscript's current scope, is that the baseline phenotypes predominantly reflect variations of a fixed pair of V(D)J rearrangements, rather than bona fide patient-derived UCA bnAb precursors with correct charges/lengths. To help resolve pre-antigenic functional signaling boundaries and thus potentially aid immunogen design, the discussion could note how future work could aim to use techniques like single-cell transcriptomics to map expressed selection signatures of bnAb precursor paired V(D)J's to rule out anergy or survival/death gene signatures.
The reviewer is correct that the surface phenotypes of rare bnAb precursor B cells, such as low BCR levels on anergic B cells, may not be apparent in the bulk flow cytometric analysis. Single-cell transcription signature analysis would be a high-resolution method to analyze bnAb precursor B cells. The technical challenge is low bnAb precursor frequencies, and enrichment of bnAb precursors by immunogen sorting may be the necessary for their detection. We will discuss this point in the revised manuscript.
(7) Discussion (PG9 Model & HCDR3 Length Constraints):
While the PG9 model's failure to produce bona fide precursors is unsurprising given its long HCDR3, the authors should provide a bit more detail as to whether a subset of PG9 variants with slightly shorter HCDR3s are permissive for bnAb production (analogous to DH511/10E8). Providing the specific HCDR3 sizes generated by this model in relation to recently discovered monkey antibodies-which are shorter/different but function like PG9 bnAbs (e.g., Guenega et al Nature 2026) - would offer excellent context. Such related lineage and/or lineage precursor variants could serve as structural alternatives, and highlighting how these length constraints impact selection provides vital boundary conditions for the immunogen design field when planning germline-targeting regimens.
The suggested recent studies (14-17) showed that V2 apex targeting is achievable with relatively shorter CDR H3s (22-25aa), which are within reach in the PG9 mouse model. However, the rhesus antibodies utilize IGHD3-15 with optimal V2 apex targeting motif. It remains to be determined whether hD3-3-based antibodies with comparable CDR H3 length would function similarly. We will incorporate the new information from rhesus immunization studies into the discussion.
References:
(1) Ray R, Schiffner T, Wang X, Yan Y, Rantalainen K, Lee CD, et al. Affinity gaps among B cells in germinal centers drive the selection of MPER precursors. Nat Immunol. 2024;25(6):1083-96.
(2) Habib R, Solieva SO, Lin ZJ, Ghosh S, Bayruns K, Singh M, et al. Deep mining of the human antibody repertoire identifies frequent and genetically diverse CDRH3 topologies targetable by vaccination. Proc Natl Acad Sci U S A. 2026;123(18):e2532810123.
(3) Swanson O, Martin Beem JS, Rhodes B, Wang A, Barr M, Chen H, et al. Identification of CDRH3 loops in the B cell receptor repertoire that can be engaged by candidate immunogens. PLoS Pathog. 2023;19(5):e1011401.
(4) Steichen JM, Phung I, Salcedo E, Ozorowski G, Willis JR, Baboo S, et al. Vaccine priming of rare HIV broadly neutralizing antibody precursors in nonhuman primates. Science. 2024;384(6697):eadj8321.
(5) Steichen JM, Lin YC, Havenar-Daughton C, Pecetta S, Ozorowski G, Willis JR, et al. A generalized HIV vaccine design strategy for priming of broadly neutralizing antibody responses. Science. 2019;366(6470).
(6) Willis JR, Berndsen ZT, Ma KM, Steichen JM, Schiffner T, Landais E, et al. Human immunoglobulin repertoire analysis guides design of vaccine priming immunogens targeting HIV V2-apex broadly neutralizing antibody precursors. Immunity. 2022;55(11):2149-67.e9.
(7) Luo S, Jing C, Ye AY, Kratochvil S, Cottrell CA, Koo JH, et al. Humanized V(D)J-rearranging and TdT-expressing mouse vaccine models with physiological HIV-1 broadly neutralizing antibody precursors. Proc Natl Acad Sci U S A. 2023;120(1):e2217883120.
(8) Tian M, Cheng C, Chen X, Duan H, Cheng HL, Dao M, et al. Induction of HIV Neutralizing Antibody Lineages in Mice with Diverse Precursor Repertoires. Cell. 2016;166(6):1471-84.e18.
(9) Havenar-Daughton C, Sarkar A, Kulp DW, Toy L, Hu X, Deresa I, et al. The human naive B cell repertoire contains distinct subclasses for a germline-targeting HIV-1 vaccine immunogen. Sci Transl Med. 2018;10(448).
(10) Cottrell CA, Hu X, Lee JH, Skog P, Luo S, Flynn CT, et al. Heterologous prime-boost vaccination drives early maturation of HIV broadly neutralizing antibody precursors in humanized mice. Sci Transl Med. 2024;16(748):eadn0223.
(11) Schiffner T, Phung I, Ray R, Irimia A, Tian M, Swanson O, et al. Vaccination induces broadly neutralizing antibody precursors to HIV gp41. Nat Immunol. 2024;25(6):1073-82.
(12) Lin SG, Ba Z, Du Z, Zhang Y, Hu J, Alt FW. Highly sensitive and unbiased approach for elucidating antibody repertoires. Proc Natl Acad Sci U S A. 2016;113(28):7846-51.
(13) McCoy LE, van Gils MJ, Ozorowski G, Messmer T, Briney B, Voss JE, et al. Holes in the Glycan Shield of the Native HIV Envelope Are a Target of Trimer-Elicited Neutralizing Antibodies. Cell Rep. 2016;16(9):2327-38.
(14) Marchitto L, Wagh K, Roark RS, Coleon S, Li H, Skelly AN, et al. Enhanced B cell priming induces broadly neutralizing HIV-1 apex antibodies. Nature. 2026;656(8128):712-22.
(15) Guenaga J, Ádori M, Bale S, Phulera S, Zygouras I, Schleich FA, et al. Vaccination generates broadly cross-neutralizing antibodies to the HIV Env apex. Nature. 2026;654(8119):777-85.
(16) Mishra N, Liang B, Roark RS, Ghosh AR, Callaghan S, Lee WH, et al. Germline-targeting HIV immunogen induces cross-neutralizing antibodies in outbred macaques. Immunity. 2026;59(4):1140-60.e11.
(17) Ma KM, Sutton HJ, Pratap PP, Steichen JM, Carnathan D, Quinn J, et al. HIV broadly neutralizing antibody precursors to the Apex epitope induced in nonhuman primates. Sci Immunol. 2025;10(110):eadt6660.
eLife Assessment
This useful study examines the role of Dynamin-2 in regulating neural progenitor behavior and fate by regulating Notch ligand and receptor function. Whilst the study is of interest because of the importance of these questions, the analyses are incomplete, and there are significant areas requiring additional experimentation in order to provide full confidence in the findings.
Reviewer #1 (Public review):
Summary:
This study explores the role of Dynamin 2 (Dnm2)-mediated clathrin-mediated endocytosis (CME) in the fate determination of neural progenitor cells in the embryonic cerebral cortex, with particular attention to its asymmetric regulation of ligand and receptor functions in the Notch signaling pathway. Using conditional knockout mice, in vitro co-culture systems, in utero electroporation, and luciferase reporter assays, the authors propose that Dnm2 is required for neural stem cell maintenance and that CME plays a dual role in Notch signaling.
Strengths:
The authors systematically demonstrate the critical role of Dnm2 in embryonic cerebral cortical development. They show that Dnm2 is the predominant dynamin isoform expressed in the embryonic central nervous system, and that its conditional knockout disrupts ventricular zone (VZ) architecture, causes premature depletion of radial glia (RG), and leads to precocious neurogenesis, accompanied by periventricular hemorrhage and perinatal death.
Weaknesses:
I believe that some conclusions of this study should still be treated with caution. Specific comments follow:
(1) scRNA-Seq analysis of the Dnm2-cKO cortex at E11.5 and E13.5 is needed, or alternatively, additional LGE-MGE scRNA-Seq should be performed, to examine whether the key gene expression changes are conserved across the MGE-CGE-LGE and cortex.
(2) Regarding Figure 5 (Loss of Dynamin 2 leads to abnormal accumulation of Dll1 and altered NICD distribution in the E11.5 cortex), I do not understand why Dll1 accumulation occurs primarily in the medial cortex.
(3) A detailed schematic model is needed to illustrate the function and mechanism of action of Dnm2.
(4) Dnm2-cKO exhibits extensive cell death and vascular abnormalities. The authors should explain the possible specific causes.
(5) Clinical relevance: Are there specific forebrain phenotypes associated with Dnm2 mutations in humans? What are the possible reasons? This study appears to have limited relevance to Dnm2-related human disease.
Reviewer #2 (Public review):
Summary:
Breuig et al. studied the effect of Dnm2 conditional knockout in the developing neocortex. The authors generated CNS-specific Dnm2 conditional knockout (cKO) mice using Nestin-Cre, Emx1-Cre, and Nex1-Cre drivers. Confirming existing findings of previous research, they reported that Dnm2 was the predominant dynamin isoform in the embryonic forebrain by immunoblot, as well as by analysis of published single-cell and spatial transcriptomic atlases. The authors went on to demonstrate that deletion of Dnm2 in progenitors causes disorganization of the ventricular zone, depletion of Sox2+ progenitors, an excess of Tbr2+ intermediate progenitor cells and Tuj1+ immature neurons, as well as periventricular hemorrhage and perinatal lethality. On the contrary, deletion of Dnm2 in postmitotic neurons produces no detectable phenotype. In the mutant cortex, Dnm2 conditional knockout led to accumulation of Dll1, reduced cleavage of NICD, and reduced Hes1/Hes5 expression. To separate ligand- from receptor-side effects, they utilized a neural stem cell co-culture assay. They tested dynamin function in the in vitro assay using a variety of molecules and probes, including a dominant-negative Dnm2 (K44A), an shRNA, and the tamoxifen-inducible deletion in reprogrammed Dnm2-floxed neural stem cells. They further performed postnatal electroporation with Hes5-EGFP and Dll1-mOrange2 reporters and an in vivo dual-luciferase assay to test the effect of receptor side effects. In summary, they showed that inhibiting or downregulating Dnm2 increases Notch signaling levels, while the involvement of Dll1 remains unclear depending on the experimental condition.
Strengths:
The topic of the manuscript, which is the molecular mechanisms and the signal transduction of neural stem cell maintenance, is of fundamental importance to the field of neurodevelopment. The conditional knockout strategy is compelling.
Weaknesses:
The central claims of the current manuscript are not sufficiently supported by experimental results. The rigor of the experiments conducted also needs to be significantly improved.
(1) Quantification and statistical analyses of the core in vivo phenotypes are essential to establish the central claim of progenitor depletion and premature neurogenesis. Only representative images of Sox2, Ki67, Tbr2, Ngn2, Tuj1, and cleaved Caspase-3 are shown (Figures 2, 3, 4, S2, S3), as is the EM observation of increased clathrin-coated vesicles (Figure 3G-H).
(2) The authors report that "the brain of mutant animals appears indistinguishable from control littermates". However, based on the representative images of Figure 2, at 11.5 the disorganization of the neocortex and the disruption of the radial columns have already manifested. The authors should carefully describe the aspects of the brain that they deemed indistinguishable between control and mutant animals, and by what criteria.
(3) The transition from Dnm2 conditional knockout (cKO) to Notch signaling appears somewhat abrupt. The authors should provide additional experimental evidence or literature precedents to logically bridge this gap.
(4) The claim that dynamin inhibition in receptor-expressing cells increases Notch signaling is not supported by the comparisons shown. In Figure 7D, the significant receptor-side comparison is K44A-receiver + Dll1 versus K44A-receiver without ligand, which tests whether ligand-driven activation survives dynamin inhibition, not whether it is enhanced. The in vivo results are marked p = 0.07 (Fig. 8F), p = 0.07 (Fig. 8L), and n.s. (Fig. 8Q), yet the text states the increase was significant, and the Figure 8 title asserts it. This is the paper's most novel claim and the basis of the ligand/receptor dichotomy in the title and Abstract. As stated, it is not supported at conventional significance.
(5) Since the prominent phenotype of Dnm2 cKO was a reduction in radial glial cells, the subsequent reduction of signaling molecules (Hes1, Hes5, etc.) may be due to the reduced progenitor pool, rather than cell-autonomous reduction of Notch signaling. Quantitative results, such as the ones in Figure 6, should take changes in cellular composition into consideration and normalize to the reduced proportion of apical progenitors.
(6) The authors stated that Notch ligands "accumulate at the cell surface," but all immunostaining appears to be on permeabilized tissue, and no internalization assay is reported. The proposed mechanism is a block in ligand endocytosis. Accumulated Dll1 in permeabilized sections is equally consistent with intracellular retention in stalled endocytic intermediates, which the EM in Figure 3G-H might actually favor.
(7) Inconsistency in the cortical regions examined (Anterior vs posterior, medial vs lateral, etc.). Are the observations and experiments conducted in the same region of the cortex? If so, which region and why?
(8) Statistical tests should be described in sufficient detail in the methods section. The number of samples or replications needs to be reported for all experiments. The "Imaging and Quantification" section also did not provide enough information (the authors only stated that "Cell counts were performed on matched cortical regions across genotypes and developmental stages"). Figure legends report significance (Figs. 6B, 6C, 7D, 7F, 8F, 8L, 8Q, S6) without specifying what was compared or how in the corresponding main text or methods.
(9) The main text attributes the cNICD (Val1744) result to Figure 5A-B, but the Figure 5 legend describes the red channel as "NICD (red; total Notch intracellular domain)," which is the antibody that the main text explicitly distinguishes. Figs. 4 and 5 appear to show overlapping Dll1/NICD data.
eLife Assessment
This valuable study provides a comprehensive metabolic and multi-omics characterization of the Ts66Yah mouse model of Down syndrome, identifying widespread metabolic, inflammatory, and molecular alterations that may help explain metabolic features observed in individuals with Down syndrome. The evidence supporting the authors' conclusions is solid, based on extensive physiological, molecular, and systems-level analyses in both male and female mice, although interpretations of sex-specific metabolic phenotypes and cross-model comparisons would benefit from additional clarification. The work will be of interest to researchers studying Down syndrome biology, metabolism, and the development and application of preclinical disease models.
Reviewer #1 (Public review):
Summary and Strengths:
In this manuscript, the authors continue their efforts to characterize metabolism-related phenotypes and molecular changes in Down syndrome mouse models, with a focus on Ts66Yah in the present study. Because the phenotypes described are relevant to metabolic abnormalities observed in individuals with Down syndrome, these findings are important. In addition, Ts66Yah is a relatively new model generated by removing the triplicated region of Mmu17 from Ts65Dn, which is not syntenic to Hsa21. Thus, Ts66Yah retains the triplicated region of Mmu16 that is syntenic to a region of Hsa21, carried on an extra chromosome.
Weaknesses:
(1) Figure 3 and Figure 10: The authors should offer a possible explanation for why the GTT and ITT differences between Ts66Yah and wild-type males observed under chow conditions are no longer evident under HFD conditions. The statement in the Abstract that "both male and female Ts66Yah mice share a common phenotype of pronounced glucose intolerance and insulin resistance" appears stronger than the data support. Under chow conditions, impaired insulin sensitivity by ITT is observed in males but not females. Moreover, although the Abstract states that glucose intolerance and insulin resistance are further exacerbated by HFD, the GTT and ITT abnormalities under HFD appear to be predominantly observed in females, whereas the genotype differences are no longer evident in males. The Abstract should therefore be revised to more accurately reflect these sex- and diet-dependent effects.
(2) Figure 7B: In the female liver PCA plot, PC1 is labeled as explaining 13.2% of the variance, whereas PC2 explains 21.6%. Since principal components are ordered by decreasing explained variance, PC1 should explain greater variance than PC2. The authors should verify whether the axis labels or percentages are correct.
(3) Figure 14 legend: A cutoff of log2FC > 0 was used, whereas Figure 8 uses a log2FC cutoff of 0.5. The authors should explain the rationale for using different thresholds in the two analyses.
Reviewer #2 (Public review):
Summary:
The authors of Saqib et al "Deep phenotyping and multi-omics analyses reveal systems-wide metabolic dysregulation in a refined trisomy mouse model of Down syndrome" were testing the hypothesis that metabolic dysregulation in Ts66Yah Down syndrome model, as compared to euploid mice, would replicate metabolic dysregulation in individuals with DS. The authors then compared the results from Ts66Yah mice in the present study mostly with results from Ts65Dn and Dp1Yey mice, where they had reported metabolic differences in previous studies.
Strengths:
Strengths of the methods and results include multiple phenotypes examining potential metabolic differences, analysis of male and female mice separately with appropriate numbers of mice (power), and the ability to compare the results to other DS mouse models, which is very useful in selecting appropriate DS mouse models to test metabolic and other related phenotypes.
The authors have shown that Ts66Yah mice differ from euploid mice in many of their metabolic parameters. Male and female Ts66Yah mice have differing phenotypes in specific metabolic parameters, but the overall result of glucose intolerance, insulin resistance, altered inflammatory cytokines, and dysregulated serum and liver metabolomes is shared between male and female mice.
Weaknesses:
Weaknesses include how the data are presented. There are some questions about methodology that leave some doubt in the analyses. The comparisons of Ts66Yah to Ts65Dn and Dp1Yey mice are less convincing, mostly due to oversimplifying the genetic constructs of DS mouse models.
Reviewer #3 (Public review):
The study by Saqib M. et al. aimed to develop a refined murine model for the study of Down syndrome. The authors developed a Ts66Yah murine line by removing triplicated genes using CRISPR technology from the previously developed Ts65Dn. In this study, the authors provide a deep metabolic profiling of the murine model, studying highly metabolic tissues (liver, skeletal muscle and adipose depots) in both sexes.
The major strength of the study is the large variety of metabolic assessments in males and females, identifying key dimorphic features in the model. However, this could also be considered a point of weakness since this is not a hypothesis-driven study, but rather a descriptive/exploratory study.
Due to the thorough description of the model and the metabolic profile, this study will be very useful for people working on research in Down syndrome. However, there is a lack of information related to cognition, behaviour, and other key features relevant to Down syndrome that would need further study.
eLife Assessment
This study provides a valuable analysis of T cell receptor repertoire organization in paired primary melanomas and sentinel lymph nodes from treatment-naive patients. The evidence for differences in repertoire architecture between these compartments is solid, but the evidence linking TCR sequence features and sequence-based communities to specific tumor antigens, particularly neoantigens, is incomplete. Additional benchmarking and robustness analyses of the community-based approach, together with more cautious interpretation of antigen specificity, would strengthen the study.
Joint Public Review:
Kitanovski and colleagues compare TCRβ repertoires from paired primary tumors and sentinel lymph nodes from 24 treatment-naive melanoma patients. The paired design and comprehensive repertoire analyses are important strengths, and the study provides solid evidence for differences between these compartments, including reduced diversity and greater clonal dominance in primary tumors.
The principal limitation is that antigen specificity is inferred from bulk TCRβ sequencing rather than directly demonstrated. Database annotation and sequence similarity are useful for generating hypotheses, but neither an annotated β-chain nor membership in a sequence-similar community establishes antigen specificity. In particular, the absence of a database match should not be interpreted as evidence of neoantigen recognition. Claims concerning neoantigen specificity should therefore be moderated or supported by additional evidence.
The ClustIRR analysis is potentially useful for identifying relationships among predominantly private TCR repertoires, but its added value and robustness require further evaluation. Benchmarking against conventional repertoire-overlap or alternative TCR-similarity approaches, testing sensitivity to key clustering parameters, and determining whether the results are driven by highly abundant communities would substantially strengthen the analysis. HLA genotype and the non-independence of pairwise repertoire comparisons should also be considered when interpreting inter-individual repertoire similarity.
Overall, the study provides a valuable dataset and convincing evidence for spatial differences in melanoma TCR repertoire architecture, while the biological interpretation of individual TCRs and communities in terms of defined antigen or neoantigen recognition remains less well supported.
eLife Assessment
This important study investigates how the hippocampus distinguishes between reactive escape and anticipatory withdrawal during approach-avoidance conflict in rats performing a naturalistic decision-making task. Solid evidence supports the main finding that hippocampal neuronal representations differ during different types of defensive behaviors. The study will be of interest to researchers studying memory, navigation, and decision-making in the presence of competing rewards and threats.
Reviewer #1 (Public review):
Damphousse, Calvin and Redish recorded from dorsal hippocampus in rats running past a robotic predator to reach food. They re-analyzed the cohort from Calvin et al. (2025, PLoS Biol) and added a new cohort. They compared three behaviors: retreats right after an attack, mid-track aborts (MTAs) where the rat turns back without being attacked, and mid-track continues (MTCs) where the rat pauses and then carries on. The authors report that retreats and MTAs, though both end at the nest, differed in speed and in decoded position. During retreats, decoding stayed near the attack zone. During MTAs, it was weighted more toward the nest and the far end of the track. Pauses that ended in an abort showed more decoding of the attack zone than pauses that ended in continuing, and a similar difference was seen on the way out, before the pause.
The main contribution is conceptual. Separating fear-like escape from anxiety-like withdrawal within the same task is a real step beyond showing that the hippocampus represents threat. The task places the robot, the attack threshold and the reward at different points on the track, which makes the decoding easier to interpret. Figs. 5-7 use the same analysis layout, so the comparisons are easy to follow. The main patterns are seen in both cohorts, even though the cohorts differ in several procedural details.
The revision tightens the abstract, clarifies how epochs were defined and normalized, and acknowledges that the results are correlational. Overall, the data support the main conclusion that hippocampal representations differ between reactive and anticipatory responses to threat. The paper should interest people working on hippocampus, defensive behavior and approach-avoidance conflict, and the task offers a good way to study how threat and reward are weighed against each other.
Reviewer #2 (Public review):
Summary:
This manuscript extends previous work from Calvin et al. and examines hippocampal representations during approach-avoidance conflict in a robotic predator foraging task. The paradigm itself is very interesting and addresses an important but relatively understudied question in the navigation and foraging literature: how the brain balances risk versus reward during goal-directed behavior. While hippocampal representations of positively valenced goals and future intentions have been extensively studied, much less is known about how these representations evolve during risk-reward tradeoffs involving threat.
The authors use a relatively simple and interpretable decoding approach together with thoughtful behavioral comparisons to ask whether future behavioral outcome can be read out from hippocampal activity before behavior diverges. The most compelling comparison is between mid-track aborts (MTAs) and mid-track continues (MTCs), where the animals initially exhibit very similar pause behavior but ultimately either abort or continue the trajectory. The authors show that decoded location during these pauses differs prior to the overt manifestation of the behavioral decision, suggesting that hippocampal representations may reflect evolving internal evaluation processes during approach-avoidance conflict.
Strengths:
A major strength of the work is the behavioral paradigm itself. This type of risk-reward conflict task is relatively uncommon in the hippocampal navigation literature and provides a rich framework for examining defensive decision-making during naturalistic foraging behavior.
The decoding analyses are also relatively simple and easy to interpret. Rather than relying on highly complex modeling approaches, the authors use straightforward comparisons of decoded spatial representations across behavioral conditions, making the results accessible and conceptually clear.
Another strength is the use of behavioral controls to isolate comparisons between related behaviors. In particular, the comparison between MTAs and MTCs is compelling because the animals exhibit similar pause states before the behavioral outcomes diverge. This provides a useful framework for asking whether hippocampal activity reflects future behavioral outcome before the decision is overtly expressed.
Overall, the study asks an interesting question using a novel paradigm and provides evidence that hippocampal representations during approach-avoidance conflict may reflect future behavioral trajectory.
Weaknesses:
The main weakness is that many of the reported effects are relatively subtle and are not sufficiently controlled for differences in speed, trajectory structure, and other behavioral variables across conditions. While the subtraction plots (green versus purple decoding differences) appear visually striking, the actual effect sizes are fairly small, making it difficult to assess how robust or behaviorally meaningful these differences are.
Relatedly, many of the most interesting questions in this task concern how behavior unfolds dynamically within a trial, yet much of the analysis averages across events and trajectories. As a result, potentially important aspects of the behavior may be obscured.
In particular, the manuscript would benefit from richer characterization of the animals' actual movement trajectories and spatial strategies. Because the analyses rely heavily on linearized position, it is difficult to determine whether animals behave differently in two-dimensional space across conditions. For example, during continued approaches, do animals preferentially hug the wall opposite the robot? Do different behavioral conditions show distinct lateral occupancy or trajectory structure? These types of analyses would make the behavioral interpretation substantially more compelling.
Reviewer #3 (Public review):
Summary:
The study reanalyzes data from a previously published cohort together with an additional cohort to investigate hippocampal activity during approach-avoidance conflict. Unlike many prior studies that isolate reward- or threat-based learning, this task requires animals to evaluate reward and threat concurrently. The central finding is that hippocampal representations differ between hesitant behaviors that lead to approach versus avoidance outcomes, with representations of the attack zone more likely during pauses preceding abort decisions. This is an important extension of prior work on hippocampal activity and deliberation, suggesting that the hippocampal content may help shape the eventual outcome.
Strengths:
All behavioral findings are replicated independently across cohorts, making the behavioral results highly convincing. The design is robust, and the task is especially valuable for studying approach-avoidance conflict. The behavioral paradigm is complex and rare, and neuronal recordings in such a paradigm are of great value.
The major strength of the study is the comparison of neural activity during hesitant behavior leading to different outcomes; namely, pauses followed by the animal aborting the approach (mid-track aborts), and pauses followed by the animal committing to the approach (mid-track continues). Hippocampal activity differed between the two pauses: the attack zone was more likely to be represented during mid-track aborts. The same effect was observed on the journey before the pause: even before the animal hesitates, hippocampal activity before a pause that led to a mid-track abort was more likely to represent the attack zone than hippocampal activity before pauses that led to continued approach. This analysis suggests that hippocampal content before and during deliberative behavior is predictive of the animal's decision.
Weaknesses:
The interpretation of the retreat-related decoding results is less clear. The study compares two sets of retreating behavior: on the one hand, retreat after being attacked, and on the other hand, retreat after hesitation in the absence of an attack (a mid-track abort). Hippocampal activity represents the attack zone more after the animal is attacked. However, these two retreating behaviors originate from different spatial locations: retreats always start past the "attack threshold", while mid-track aborts always start before this threshold. Given that hippocampal decoding is strongly location-dependent, this difference in position makes the neural decoding results difficult to interpret. The increased representation may be due to differences in physical location, rather than distinct processing of immediate threat and an anticipatory return state.
Comments on revised version.
The revised manuscript is indeed more clear, but the study is largely unchanged.
Author response:
The following is the authors’ response to the original reviews.
Public Reviews:
Reviewer #1 (Public review):
Summary:
This study by Damphousse, Calvin, and Redish investigates how the hippocampus represents competing future outcomes during approach-avoidance conflict. Using an ethologically relevant robotic predator foraging paradigm, the authors aimed to dissociate hippocampal activity associated with reactive defensive responses (escape) from that linked to anticipatory withdrawal decisions. The central finding is that dorsal hippocampal representations differentiate these two modes of defensive behavior within a single naturalistic assay. Specifically, the authors show that attack-triggered retreats and mid-track aborts differ in movement dynamics and hippocampal spatial decoding despite sharing a common behavioral endpoint, that hippocampal representations during pauses predict subsequent behavioral outcomes, and that these representational biases emerge before overt behavioral divergence. The main importance of the study lies in moving beyond viewing the hippocampus as merely encoding spatial location or threat salience, instead suggesting that hippocampal ensemble activity dynamically tracks and differentially weights threat-related, reward-related, and safety-oriented future states to bias behavior before overt action occurs.
Strengths:
The study has several notable strengths. First, the behavioral decomposition into retreats, mid-track aborts, and mid-track continues is rigorous and provides a highly interpretable analytical framework. Second, replication across two independent cohorts - despite differences in arena configuration, robot design, and extinction procedures - meaningfully strengthens confidence in the robustness of the findings. Third, the unified reanalysis pipeline across cohorts reflects strong analytical discipline, and the Bayesian decoding framework is well-suited to addressing the central representational questions. Fourth, the ethological relevance of the robotic predator paradigm is a major advantage, allowing the authors to examine a richer repertoire of defensive and decision-related behaviors than is possible in conventional fear-conditioning assays. Overall, the experiments are well designed, the data are clearly presented, and the findings make a valuable contribution to understanding how the hippocampus supports decision-making under threat.
Weaknesses:
The study is technically strong, but a few modest revisions would further enhance it.
(1) First, the abstract mentions extinction and reinstatement effects, but neural analyses focus primarily on the attack phase. It would be helpful to clarify or adjust the abstract accordingly.
(2) Second, some interpretive language ("guide," "bias") leans toward causal phrasing. Given the correlational data, using "predict" or "correlate with" would be more precise.
(3) Third, given the relationship between running speed and hippocampal theta, considering speed-related contributions to decoding differences would be useful.
(4) Fourth, reporting turnaround positions for mid-track abort and continue trials (Figure 7) would provide helpful context.
(5) Fifth, a figure comparing stimulated vs. non-stimulated sessions in cohort 2 would support the claim that closed-loop stimulation had no measurable effect.
(6) Finally, reporting effect sizes for key decoding comparisons would add clarity.
Reviewer #2 (Public review):
Summary:
This manuscript extends previous work from Calvin et al. and examines hippocampal representations during approach-avoidance conflict in a robotic predator foraging task. The paradigm itself is very interesting and addresses an important but relatively understudied question in the navigation and foraging literature: how the brain balances risk versus reward during goal-directed behavior. While hippocampal representations of positively valenced goals and future intentions have been extensively studied, much less is known about how these representations evolve during risk-reward tradeoffs involving threat.
The authors use a relatively simple and interpretable decoding approach together with thoughtful behavioral comparisons to ask whether future behavioral outcomes can be read out from hippocampal activity before behavior diverges. The most compelling comparison is between mid-track aborts (MTAs) and mid-track continues (MTCs), where the animals initially exhibit very similar pause behavior but ultimately either abort or continue the trajectory. The authors show that decoded location during these pauses differs prior to the overt manifestation of the behavioral decision, suggesting that hippocampal representations may reflect evolving internal evaluation processes during approach-avoidance conflict.
Strengths:
A major strength of the work is the behavioral paradigm itself. This type of risk-reward conflict task is relatively uncommon in the hippocampal navigation literature and provides a rich framework for examining defensive decision-making during naturalistic foraging behavior.
The decoding analyses are also relatively simple and easy to interpret. Rather than relying on highly complex modeling approaches, the authors use straightforward comparisons of decoded spatial representations across behavioral conditions, making the results accessible and conceptually clear.
Another strength is the use of behavioral controls to isolate comparisons between related behaviors. In particular, the comparison between MTAs and MTCs is compelling because the animals exhibit similar pause states before the behavioral outcomes diverge. This provides a useful framework for asking whether hippocampal activity reflects future behavioral outcome before the decision is overtly expressed.
Overall, the study asks an interesting question using a novel paradigm and provides evidence that hippocampal representations during approach-avoidance conflict may reflect future behavioral trajectory.
Weaknesses:
The main weakness is that many of the reported effects are relatively subtle and are not sufficiently controlled for differences in speed, trajectory structure, and other behavioral variables across conditions. While the subtraction plots (green versus purple decoding differences) appear visually striking, the actual effect sizes are fairly small, making it difficult to assess how robust or behaviorally meaningful these differences are.
Relatedly, many of the most interesting questions in this task concern how behavior unfolds dynamically within a trial, yet much of the analysis averages across events and trajectories. As a result, potentially important aspects of the behavior may be obscured.
In particular, the manuscript would benefit from richer characterization of the animals' actual movement trajectories and spatial strategies. Because the analyses rely heavily on linearized position, it is difficult to determine whether animals behave differently in two-dimensional space across conditions. For example, during continued approaches, do animals preferentially hug the wall opposite the robot? Do different behavioral conditions show distinct lateral occupancy or trajectory structure? These types of analyses would make the behavioral interpretation substantially more compelling.
More generally, while the results are suggestive and interesting, the relatively small decoding differences and substantial behavioral confounds make it difficult to conclude that the observed effects reflect distinct internal evaluative or threat-related states.
Reviewer #3 (Public review):
Summary:
The study reanalyzes data from a previously published cohort together with an additional cohort to investigate hippocampal activity during approach-avoidance conflict. Unlike many prior studies that isolate reward- or threat-based learning, this task requires animals to evaluate reward and threat concurrently. The central finding is that hippocampal representations differ between hesitant behaviors that lead to approach versus avoidance outcomes, with representations of the attack zone more likely during pauses preceding abort decisions. This is an important extension of prior work on hippocampal activity and deliberation, suggesting that the hippocampal content may help shape the eventual outcome.
Strengths:
All behavioral findings are replicated independently across cohorts, making the behavioral results highly convincing. The design is robust, and the task is especially valuable for studying approach-avoidance conflict. The behavioral paradigm is complex and rare, and neuronal recordings in such a paradigm are of great value.
The major strength of the study is the comparison of neural activity during hesitant behavior leading to different outcomes, namely, pauses followed by the animal aborting the approach (mid-track aborts), and pauses followed by the animal committing to the approach (mid-track continues). Hippocampal activity differed between the two pauses: the attack zone was more likely to be represented during mid-track aborts. The same effect was observed on the journey before the pause: even before the animal hesitates, hippocampal activity before a pause that led to a mid-track abort was more likely to represent the attack zone than hippocampal activity before pauses that led to continued approach. This analysis suggests that hippocampal content before and during deliberative behavior is predictive of the animal's decision.
Weaknesses:
The interpretation of the retreat-related decoding results is less clear. The study compares two sets of retreating behavior: on the one hand, retreat after being attacked, and on the other hand, retreat after hesitation in the absence of an attack (a mid-track abort). Hippocampal activity represents the attack zone more after the animal is attacked. However, these two retreating behaviors originate from different spatial locations: retreats always start past the "attack threshold", while mid-track aborts always start before this threshold. Given that hippocampal decoding is strongly location-dependent, this difference in position makes the neural decoding results difficult to interpret. The increased representation may be due to differences in physical location, rather than the distinct processing of immediate threat and an anticipatory return state.
We thank the reviewers for their careful, constructive, and fair assessment of our manuscript, “Hippocampal representations differentiate reactive and anticipatory responses during foraging under threat.” We were pleased that the assessment characterized our study as important and the evidence supporting its main finding as solid. We also appreciate the reviewers’ recognition of the strengths of the behavioral paradigm and the conceptual contribution of distinguishing hippocampal representations associated with reactive and anticipatory responses to threat.
We carefully considered the reviewers’ comments and do not feel that the manuscript requires extensive revision. The reviewers suggested several thoughtful additional analyses, including further behavioral controls and explicit predictive analyses. We agree that these would be informative and intend to carry several of these directions forward with future datasets. However, for the present manuscript, the suggested analyses would extend the scope of the work but are unlikely to substantially alter the central observations, and we have therefore chosen not to undertake them for the present manuscript.
We have only made very minor changes to the text and figures, corresponding to what we might do in a manuscript that had been “accepted” under the old-style journal process. We have
- Revised the abstract to more accurately reflect the neural analyses presented.
- Softened the interpretive language throughout to avoid implying causality where the data establish associations, and explicitly acknowledged this limitation in the discussion.
- Clarified the explanation of the timing and location of robot attacks relative to the attack threshold.
- Clarified the definitions and explanations of the temporal segmentation of MTA and MTC pauses.
- Clarified the explanations of the outbound epoch used in the MTA-MTC comparison.
- Expanded the methods describing temporal and spatial normalization of behavioral epochs.
- Moved the headers on figures 5-7 as suggested by the reviewers to make them more clear.
With these revisions, we believe the manuscript now accurately reflects what can be concluded from the current data and appropriately acknowledges the limitations raised during review.
We sincerely thank the editor and reviewers for their thoughtful engagement. Their comments have improved the clarity and precision of the manuscript and provided valuable directions for future work.
eLife Assessment
This study provides important findings on the neural circuits underlying dishabituation of the olfactory avoidance response in Drosophila. The data as presented provide convincing evidence that the dishabituation involves distinct pathways from habituation. They show that reward-activated dopaminergic neurons provide input for within-modal dishabituation, while punishment-activated dopaminergic neurons provide input for cross-modal dishabituation. The work will interest neuroscientists, particularly behavioral neuroscientists working on habituation, neural circuits, and the dopaminergic system.
Reviewer #1 (Public review):
Summary:
Charonitakis and co-authors characterize dishabituation in adult flies, where they use olfactory habituation to octanol, then dishabituate the flies with disruptions of electric shock or yeast odors. They systematically investigate the neurotransmitters and neural circuits involved in dishabituation and figure out a lot about how this process works in the brain, as an independent circuit. I like the paper and I like the very structured approach to figuring out the problem.
Strengths:
The introduction nicely sets the stage for the work presented, bringing in knowledge from other organisms and motivating the study.
The results section lays out a logical set of experiments, using a common set of behavioral assays in many flies exposed to thermogenetic or optogenetic manipulation. The paper systematically figures out the necessity and/or sufficiency of specific brain regions and neurotransmitters, culminating in a new understanding of how the important process of dishabituation works.
I like the bar graph representation for the data throughout, with the helpful icons - if a paper figures are going to be 90% bar graphs it helps when they are super clear like this. And I like how the all the parts build up to the conclusion in the last figure, nicely summarizing the thorough characterization of dishabituation.
Weaknesses:
There are no major concerns, and I appreciate the addition of a new figure that details the habituation / induced dishabituation protocol, along with additional discussion of dishabituation, and framing the topic for a more general audience.
Reviewer #2 (Public review):
This is an interesting study in Drosophila comparing potentially differential requirements for subsets of Kenyon Cells (KCs) and Dopaminergic neurons (DANS) in olfactory dishabituation driven by either a novel odor ("homosensory") or footshock ("heterosensory). The authors measure olfactory aversion to Octanol (OCT) in a T-maze, induce olfactory habituation with 4-minunute prior exposure to OCT, and use either brief yeast odor (YO) or footshock (FS) to achieve dishabituation. The major observation that YO-mediate dishabituation is mediated by reward activated DANs (PAM cluster), while FS mediated dishabituation is mediated by punishment-activated PPL-DANs is generally solid and convincing. Also convincing are experiments showing the involvement of KCs in the pathway for YO and FS induced dishabituation, and the argument that KCs drive DAN activation that causes dishabituation, though not experimentally shown, is more than reasonable. The work is significant because, as the authors take pains to point out, circuits and pathways for dishabituation have been very lightly studied and clear identification of dopaminergic neuron subsets in dishabituation achieved to different means, represents unique and interesting progress.
Comments on revised version.
I am appreciative of the authors' effort to address the various concerns and queries of the reviewers. The revised work, which more clearly positions the work, better specifies the conclusions reached and their relevance to the broader field, is a very interesting and useful addition to the field.
Author response:
The following is the authors’ response to the original reviews.
Public Reviews:
Reviewer #1 (Public review):
There are no major concerns, but some material could be added for clarity and to make the work more accessible to a more general scientific audience.
We will add text for clarity and to make the work more accessible to a general audience per this comment and similar suggestions of the other reviewers.
(1.1) A figure clearly showing the habituation protocol and the use of the dishabituators would be a good addition, even if the procedure has been done before and is cited. There can always be readers who are seeing this for the first time.
We do think this is a good idea as the time scales of the experiment will be clearly marked as well and we plan to generate one in the revised manuscript.
(1.2) It would also be nice to comment on other ways dishabituation can happen (for example, when the stimulus is removed for a short time and returns) and what their time scales are.
If the stimulus is withheld, spontaneous recovery occurs, a process distinct from dishabituation and worth exploring on its own. In a previous publication (Semelidou et al. eLife 2018;7:e39569), we have shown that in this habituation paradigm with 4 min exposure either to the aversive Octanol, or the attractive Ethyl Acetate, spontaneous recovery occurs on or after 6 minutes after the habituated stimulus is withheld. This contrasts the immediate effect of the single dishabituating stimulus, delivered for a few seconds at the end of exposure to the habituator. Granted that per Thomson (Neurobiol Learn Mem. 2009), spontaneous recovery is a characteristic of habituation, we will work this point in the text.
(1.3) And more generally, the paper could perhaps improve by making a stronger case for why the results are important not just for flies but for neuroscience in general.
Thank you for the encouragement. We will try to rationally generalize our findings.
Reviewer #2 (Public review):
(2.1) However, the claim that this represents a fundamental difference between homosensory and heterosensory pathways for dishabituation is overstated.
We had no intention of stating more than the fact that footshock and yeast odor dishabituators relay these stimuli to the mushroom bodies via distinct dopaminergic neurons, hence differentiating distinct dishabituating stimuli via the mechanosensory (footshock) and olfactory (yeast odor) modalities as they engage the mushroom bodies. As the reviewer suggests we will use more measured and specific language to state the above.
(2.2) The introductory section does not adequately present current broad models for habituation and dishabituation.
This was not done intentionally, but rather because we aimed at a less extended introductory section and ostensibly this resulted in brief and possibly inadequate presentation of current habituation models. We will present a much more detailed introduction and detail of habituation and dishabituation models in the revised manuscript (Also see reply to point 3.5 below).
(2.3) There are many different time scales, even for Drosophila olfactory habituation. These, as well as potential underlying mechanistic differences, need to be acknowledged; any claim should be specifically qualified for the time scales being studied here.
We understand and appreciate the point of the reviewer, as well as its significance and we will address this both in the revised text, but also by the paradigm figure we will add as stated above (point 1.1), where the time scales will be explicitly included and emphasized.
(2.4) Additionally, there are several unclear, vague, and inaccurate sections and statements. A more careful, precise, and considered presentation of current views, as well as more measured claims of the impact of the findings, would substantially enhance my enthusiasm.
We will address these concerns of course, though pointing out the specific offending parts would ascertain addressing them thoroughly. As stated above, we will incorporate current views in the introduction and when discussing our results and their impact.
Reviewer #3 (Public review):
(3.1) The key issue is that the main concepts of this manuscript appear to be based on a misunderstanding/misinterpretation of the literature. As the authors set out to settle the debate "whether the novel dishabituating stimulus elicits sensitization of the habituated circuits, or it engages distinct neuronal routes to bypass habituation reinstating the naïve response", it seems that the authors based their investigation on the premise that "sensitization" is mediated by a facilitatory process within the S-R pathway, and "dishabituation" by a facilitatory process outside the S-R pathway. This is not the status quo in the field, particularly with the prevailing theory like the Dual-Process Theory.
We appreciate the reviewer’s comment and the opportunity to clarify the conceptual framework of our work. Our intention was in fact to test the Groves and Thomson hypothesis (Neurobiol Learn Mem. 2009), in our olfactory habituation system. As such, dishabituation could have been the result of a facilitatory process within the S-R pathway, or from mechanisms outside of it. Our experimental design allowed to distinguish these possibilities and our results clearly show that dishabituation involves circuitry outside the S-R pathway. We do thank the reviewer for pointing out that we have not articulated clearly this intention and we will take care to communicate this effectively in the revised manuscript.
(3.2) The original version of Dual-Process Theory (Groves and Thompson 1970, but also see Thompson 2008, Neurobiol Learn Mem) already hypothesized that habituation happens within the specific S-R pathway, and sensitization occurs separately in an "organism-wide" state system that modulates the output of all S-R pathways.
As mentioned above, we are aware of the Dual-Process hypothesis. In fact, our data demonstrate that activity outside the olfactory S-R pathway, engaging novel neuronal circuits, mediates dishabituation. Unlike habituation, these circuits mediating dishabituation include at minimum, the mushroom bodies, the dopaminergic system and the APL neurons. In our view this does not support the “organism-wide state” system, but rather particular circuits that in agreement with the Groves and Thomson hypothesis, are outside the S-R pathway and modulate its behavioral output. We will work these concepts in the discussion section of the revised manuscript.
(3.3) Dishabituation is recognized by the Dual-Process Theory as sensitization (organism-wide facilitation) manifested on top of existing habituation (depressed S-R pathway). This notion has been supported by a wide range of studies, including cat spinal cord reflex (e.g. Spencer et al. 1966) and work in Aplysia on heterosynaptic facilitation for both sensitization and dishabituation. Therefore, simply showing that the newly identified facilitatory pathways are outside the S-R habituation pathway is insufficient to demonstrate dishabituation.
We respectfully disagree with the concluding sentence here. In all of our experiments, we observe a clear recovery of olfactory avoidance after exposure to the footchock, or yeast odor dishabituators. Moreover, the dishabituators are emulated by (photo)activation of particular neuronal circuits and the recovery of olfactory avoidance is blocked when these circuits are silenced. Regardless of whether this recovery is classified as dishabituation via sensitization or another facilitatory process, the key point is that the habituated response is reliably reinstated contingent upon the dishabituating stimulus. We believe this meets the established criteria for dishabituation.
(3.4) As behavioral facilitation of a habituated response can be achieved by dishabituating (specific recovery of the S-R pathway) and/or superimposed sensitizing (organism-wide) processes, dishabituation and sensitization of this olfactory response must be first dissociated; however, the study provided no evidence for the dissociation. Without this piece of evidence, the claim of this paper that the newly identified pathways mediate dishabituation is not fully supported.
We agree with the reviewer that we have not provided specific evidence dissociating dishabituation and sensitization of the particular olfactory response beyond the evidence implicating particular circuitry in the outcome of facilitation of the olfactory response.
It should be noted that in photoactivation of the implicated circuitries in naïve flies, we do not observe enhanced octanol avoidance, suggesting that activation of these circuits alone does not induce sensitization. Moreover, our results show that neither footshock nor yeast odor drive an organism-wide sensitization, as silencing specific circuits was sufficient to block dishabituation—something that would not be expected if a global sensitization process was responsible of reinstating the olfactory response.
Nonetheless, we will also attempt to dissociate sensitization from dishabituation using mutants previously reported deficient in sensitization (Duerr and Quinn, PNAS 1982), assuming these mutants retain normal olfactory habituation.We will also try sensitization protocols in the case of within-modal dishabituation to further clarify the underlying mechanisms. In principle, this includes using diluted Octanol as the habituating stimulus and attempt dishabituation with concentrated octanol.
(3.5) The literature review of this manuscript has some discrepancies. In the introduction, the authors wrote "initial studies in Aplysia were consistent with the "dual-process theory" (Groves and Thompson 1979), where response recovery due to dishabituation appeared to result from sensitization superimposed on habituation, thus driving reversal of the attenuated response (Carew, Castellucci et al. 1971, Hochner, Klein et al. 1986, Marcus, Nolen et al. 1988, Ghirardi, Braha et al. 1992, Cohen, Kaplan et al. 1997, Antonov, Kandel et al. 1999, Hawkins, Cohen et al. 2006)." Hochner 1986 and Marcus 1988 in fact indicated otherwise. Hochner 1986 suggests that dishabituation and sensitization involve different molecular processes, while Marcus 1988 showed that dishabituation and sensitization have different behavioral characteristics. Therefore, the authors' statement is not supported by the cited literature.
We are grateful to the reviewer for pointing out these significant discrepancies, consequent of multiple rounds of edits followed by our own oversight. These important publications for this manuscript will be referenced properly in the revised manuscript.
eLife Assessment
The authors study global neuropeptide changes in C. elegans mutants lacking functional BMP, using this as a viable and fertile animal model that exhibits consistent dyspepsia and chronic dietary restriction. The findings of this study are valuable as they provide a global picture of neuropeptide responses to starvation and detail a circuit from a sensory neuron to lipid metabolism in the gut. This work provides evidence that nutrient stress resulting from impaired food processing engages NLP-10-, NPR-35-, FLP-18-, and NPR-4-dependent signalling to promote fat utilization. However, the evidence is incomplete, as the order of action of the neuropeptides is weak.
Reviewer #1 (Public review):
Summary:
This manuscript describes how defective food grinding due to loss of TGF-beta signalling leads to fat loss via specific neuropeptides and their cognate receptors in C. elegans. They define transcriptional mechanisms from the ADL sensory neuron to neuropeptide relays from neurons to the gut, wherein enzymatic effectors mediate fat metabolism. The genetic evidence presented to support this neuropeptide circuitry is based largely on transcriptional changes detected in bulk mRNA, cell-type-specific rescue, and lipid measurements using Nile Red staining in permeabilized/fixed animals and with lipidomics. The findings are of interest to researchers in neuropeptide regulation of physiology.
Strengths:
(1) The multifaceted analysis of dbl-1 and sma-6 provides good evidence that the primary defect leading to metabolic phenotypes is due to pharyngeal function.
(2) The transcriptional survey of neuropeptide expression under fasting and in TGF-beta pathway mutants will be a useful resource for many labs studying neuropeptides in food sensing.
(3) Outlining the process from dietary restriction to fat metabolism in the gut provides a big-picture view that establishes physiological relevance.
(4) Identifying new functions for specific neuropeptides and their receptors in fat metabolism.
Weaknesses:
Evidence that links NLP-10-NPR-35 signalling to FLP-18-NPR-4 is weak, and primarily rests on overlaps in npr-35 and flp-18 expression, as well as rescue of flp-18 under the npr-35 promoter. It remains possible that FLP-18 acts from those neurons, but is not activated by NPR-35 signalling, since the mechanisms downstream of NPR-35 are unclear in this context. Strengthening this link would require showing a change in FLP-18 activity; for example, showing npr-35-dependent changes in flp-18 transcription in npr-35-expressing neurons.
The expression patterns of the neuropeptides and receptors that guide the sites of cell-specific rescue experiments are from well-fed animals. However, it remains a possibility that starvation alters neuropeptides and their receptors are expressed, which would produce a different circuit.
Different food levels produce different physiological and transcriptional responses. The manuscript largely assumes that starvation is equivalent to food reduction when they are physiologically distinct states.
Reviewer #2 (Public review):
Summary:
This manuscript investigates how animals adapt their metabolism to conditions of reduced nutrient availability. Using C. elegans, the authors identify a neuropeptide signaling network that promotes intestinal lipid mobilization when nutrient acquisition is compromised. The study proposes that nutrient stress activates signaling involving NLP-10, NPR-35, FLP-18, and NPR-4, ultimately leading to increased utilization of stored triglycerides.
A major strength of the work is the combination of multiple experimental approaches, including genetic analyses, neuron-specific rescue experiments, starvation paradigms, and lipid measurements. Together, these experiments support the conclusion that the identified neuropeptide pathway contributes to the regulation of energy stores during nutrient stress. The observation that similar signaling components function in both the mutant and starvation contexts strengthens the physiological relevance of the findings.
Strengths:
The revised manuscript is substantially improved and presents a coherent model linking sensory and interneuronal signaling to peripheral metabolic responses. The data convincingly establish functional roles for the identified neuropeptide components in controlling fat storage and mobilization.
Readers should keep in mind that the exact relationships among pathway components are not completely resolved. While the genetic evidence supports placement of NLP-10, NPR-35, FLP-18, and NPR-4 within a common regulatory framework, the data are somewhat less definitive regarding the precise order of signaling events and whether all interactions are direct. The authors appropriately acknowledge some of this complexity, and a clearer distinction between experimentally demonstrated versus inferred connections would further strengthen the presentation.
Overall, this is a carefully executed and biologically interesting study that advances our understanding of neuroendocrine control of metabolism during nutrient limitation.
Weaknesses:
No weaknesses were noted by this reviewer in the revision. My main comment is that whilst I appreciate the authors' efforts to refine the signaling model in response to earlier comments, the pathway architecture remains somewhat complex, particularly regarding the relationships among NLP-10, NPR-35, FLP-18, and NPR-4. I encourage the authors to make one final pass through the text and model figure to clearly distinguish interactions that are directly supported by experimental evidence from those that remain inferred based on genetic relationships. This would help readers accurately interpret the strength of evidence supporting each step of the proposed pathway.
Reviewer #3 (Public review):
Summary:
In this manuscript, Liu et al. used C. elegans animals lacking functional BMP as a model to dissect the neuropeptide signaling pathways that lead to BMP-associated fat loss. First, they established the BMP mutants as a model of dyspepsia that results in chronic dietary restriction. They then used this model to identify neuropeptide genes that are upregulated and required for BMP-associated fat loss. Next, they sought to identify the receptors for these neuropeptides and to place the genes in a genetic and cellular pathway. They also showed that the same neuropeptides and receptors contribute to fat loss in diet-restricted C. elegans. They conclude that they have identified a neuropeptide signaling cascade in which impaired pharyngeal function activates sensory neurons, which then signal through interneurons to enhance triglyceride hydrolysis in the intestine.
Strengths:
In this manuscript, Liu et al. characterized the pharyngeal muscles and the intestinal bacterial context of animals lacking functional BMP. They showed a correlation between BMP-associated fat loss and pharyngeal impairments that could lead to insufficient food grinding. This finding suggests that C. elegans lacking functional BMP could serve as a model for dyspepsia and chronic dietary restriction. Moreover, the authors analyzed previously published gene expression data to identify neuropeptides that are overexpressed in animals lacking BMP. They then used mutants for these neuropeptides and their receptors to determine which of these genes contribute to BMP-associated fat loss. Finally, they showed that the same neuropeptides and receptors also contribute to fat loss in diet-restricted C. elegans.
Weaknesses:
(1) The authors use double mutants between BMP pathway mutants and neuropeptide or neuropeptide receptor mutants to argue that the reduced fat levels of BMP mutants are suppressed. However, with the exception of unc-31, the fat content of the corresponding single neuropeptide or neuropeptide receptor mutant is not shown. Without these, the double-mutant phenotype cannot be interpreted as suppression, since it could reflect an independent effect on fat storage. To show that the effects are related to BMP signaling, the authors should compare wild-type animals to neuropeptide mutants in a sma-6(+) background. Similarly, the fat content of ADL-ablated animals is not shown in an otherwise wild-type background.
(2) In several places, the authors use cell-specific expression of neuropeptides to conclude that a given neuropeptide acts in a particular cell type (e.g., Figure 4B,C; Figure 5C,D). However, because neuropeptides are secreted and can act at a distance, expression in a non-native cell type may rescue the phenotype. These experiments therefore show sufficiency but do not establish the endogenous site of action. Cell-specific knockout or knockdown, or expression from control promoters in unrelated cell types, would help address this.
(3) The authors report that expressing npr-35 from two promoters with non-overlapping expression each fully restores the low-fat phenotype of sma-6; npr-35 mutants. This is surprising for a receptor and is difficult to interpret.
(4) The authors identify neuropeptide genes co-expressed with npr-35 and follow up on flp-18. The rationale for restricting the search to co-expressed neuropeptides may bias hypothesis generation. Similarly, the authors show that flp-1 and flp-4, in addition to flp-18, contribute to BMP-associated fat loss, but only flp-18 is examined further because of its overlapping co-expression with npr-35, again creating a bias in the investigation of the role of neuropeptide signaling in BMP-associated fat loss.
(5) It is difficult to interpret the experiment in Figure 5E, where the authors conclude that npr-4 is required in flp-18-overexpressing animals. Without a demonstrated fat-loss phenotype from overexpressing flp-18, a requirement for npr-4 in those animals cannot be meaningfully assessed.
Overall, the authors convincingly identify neuropeptides and receptors that are overexpressed and that contribute to fat loss. However, several points make it difficult to fully support the proposed model and the linear pathway from the pharynx to the intestine through ADL sensory neurons and interneurons. These include the absence of phenotypic analysis of the single mutants, the challenges of interpreting cell-specific rescue experiments when studying neuropeptide function, the fact that additional neuropeptides causing BMP-associated fat loss are identified but not studied further, and results that do not necessarily point to a linear pathway (e.g., the triple mutant sma-6; nlp-10; npr-35 has a stronger phenotype than the double mutants with sma-6, the fat loss caused by overexpression of nlp-10 in a sma-6 background is not suppressed by npr-35, and the flp-18 overexpression experiment is difficult to interpret). The reviewer therefore suggests that the linearity of this pathway, and the sites of action of the neuropeptides and receptors, should be presented more cautiously or supported by additional data.
Author response:
eLife Assessment
The authors study global neuropeptide changes in C. elegans mutants lacking functional BMP, using this as a viable and fertile animal model that exhibits consistent dyspepsia and chronic dietary restriction. The findings of this study are valuable as they provide a global picture of neuropeptide responses to starvation and detail a circuit from a sensory neuron to lipid metabolism in the gut. This work provides evidence that nutrient stress resulting from impaired food processing engages NLP-10-, NPR-35-, FLP-18-, and NPR-4-dependent signalling to promote fat utilization. However, the evidence is incomplete, as the order of action of the neuropeptides is weak.
We thank Reviewing Editor, Senior Editor and the three reviewers for their constructive comments. We are encouraged that the reviewers regard the central finding that impaired pharyngeal function engages neuropeptide signaling to modify intestinal lipid metabolism as physiologically meaningful and supported by multiple lines of evidence. At the same time, we recognize that the reviews converge on a central concern: the evidence for the order of neuropeptides’ action is weak. The reviewers also suggest us to distinguish more clearly between what our experiments demonstrate and what we infer, to present the pathway as less strictly linear, and to be more explicit about the limits of cell-specific rescue for secreted peptides. We take these points seriously, and we are exploring possible revisions in light of them.
Public Reviews:
Reviewer #1 (Public review):
Summary:
This manuscript describes how defective food grinding due to loss of TGF-beta signalling leads to fat loss via specific neuropeptides and their cognate receptors in C. elegans. They define transcriptional mechanisms from the ADL sensory neuron to neuropeptide relays from neurons to the gut, wherein enzymatic effectors mediate fat metabolism. The genetic evidence presented to support this neuropeptide circuitry is based largely on transcriptional changes detected in bulk mRNA, cell-type-specific rescue, and lipid measurements using Nile Red staining in permeabilized/fixed animals and with lipidomics. The findings are of interest to researchers in neuropeptide regulation of physiology.
Strengths:
(1) The multifaceted analysis of dbl-1 and sma-6 provides good evidence that the primary defect leading to metabolic phenotypes is due to pharyngeal function.
(2) The transcriptional survey of neuropeptide expression under fasting and in TGF-beta pathway mutants will be a useful resource for many labs studying neuropeptides in food sensing.
(3) Outlining the process from dietary restriction to fat metabolism in the gut provides a big-picture view that establishes physiological relevance.
(4) Identifying new functions for specific neuropeptides and their receptors in fat metabolism.
We thank R1 for an accurate summary and for recognizing the value of this work. We also note R1’s characterization of our evidence as transcriptional and genetic. We agree with this description and will explicitly state the scope of our claims in future versions, clarifying that this work defines a functional hierarchy among genes and cells but does not provide biochemical, electrophysiological, or secretion-based evidence for direct molecular interactions.
Weaknesses:
Evidence that links NLP-10-NPR-35 signalling to FLP-18-NPR-4 is weak, and primarily rests on overlaps in npr-35 and flp-18 expression, as well as rescue of flp-18 under the npr-35 promoter. It remains possible that FLP-18 acts from those neurons, but is not activated by NPR-35 signalling, since the mechanisms downstream of NPR-35 are unclear in this context. Strengthening this link would require showing a change in FLP-18 activity; for example, showing npr-35-dependent changes in flp-18 transcription in npr-35-expressing neurons.
We agree that this is the weakest link in our model. At present, this rests mainly on two observations: the overlap between npr-35 and flp-18 expression reported in public databases, and the ability of npr-35p::FLP-18 (or ttx-3p::FLP-18) to restore fat levels. These observations cannot exclude the possibility that FLP-18 acts from these neurons, or even from additional neurons, without being activated by NPR-35 signaling, and that the FLP-18 induction reflects the nutritional state rather than receptor activity.
To test this directly, we will begin by measuring flp-18 transcript levels in two comparisons: (1) well-fed npr-35(lf) versus well-fed sma-6(lf); npr-35(lf), and (2) well-fed npr-35(lf) versus fasted npr-35(lf). In addition, transcriptional levels of nlp-10, flp-1, and flp-4 can be compared in these worms. If NPR-35 activity is required for flp-18 induction when nutrients are limiting, removing npr-35 should blunt or reduce this induction in the sma-6-mutated or nutrient-deprived backgrounds; If NPR-35 activity is not required, flp-18 induction should not be affected. A third possibility is that NPR-35 amplifies FLP-18 signaling, which would not conflict with the nutrient-sensing role of FLP-18 per se. Regardless of the outcome, we will rewrite this section to clarify the step from NLP-10 to FLP-18.
The expression patterns of the neuropeptides and receptors that guide the sites of cell-specific rescue experiments are from well-fed animals. However, it remains a possibility that starvation alters neuropeptides and their receptors are expressed, which would produce a different circuit.
This is a fair point, and we accept it as a limitation of the present study. The expression patterns that guided our rescue experiments were defined in well-fed animals, using published datasets (Beets et al., 2023; Ripoll-Sánchez et al., 2023; Ghaddar et al., 2023) and our own transcriptional reporters under well-fed conditions. We cannot exclude that nutrient status alters the expression of these peptides and receptors.
We plan to examine intestinal npr-4p::GFP in well-fed and starved animals in the near future. We could also attempt to image neuronal npr-35p::GFP, but this would be technically challenging given mosaic expression and issues with worm orientation. In the Discussion of future versions, we will state this caveat and will be careful to describe our rescue experiments as identifying the cells in which a gene is sufficient to support the phenotype under the conditions tested, rather than as confirming the circuit in the starved animals.
Different food levels produce different physiological and transcriptional responses. The manuscript largely assumes that starvation is equivalent to food reduction when they are physiologically distinct states.
We agree, and we regret that our wording implied equivalence. Our intention was not to treat all nutrient-limited states as identical but to ask whether a common pathway is engaged. The observation that the same components are required under each of these energy-deficient conditions supports the physiological relevance of the pathway. We will rewrite the relevant sentences to convey this intent and to avoid implying that different nutritional states are equivalent. We acknowledge the limitation that our design cannot determine the degree of energy deficit required to engage the neuropeptides, or whether the response is graded across milder and more severe restriction.
Reviewer #2 (Public review):
Summary:
This manuscript investigates how animals adapt their metabolism to conditions of reduced nutrient availability. Using C. elegans, the authors identify a neuropeptide signaling network that promotes intestinal lipid mobilization when nutrient acquisition is compromised. The study proposes that nutrient stress activates signaling involving NLP-10, NPR-35, FLP-18, and NPR-4, ultimately leading to increased utilization of stored triglycerides.
A major strength of the work is the combination of multiple experimental approaches, including genetic analyses, neuron-specific rescue experiments, starvation paradigms, and lipid measurements. Together, these experiments support the conclusion that the identified neuropeptide pathway contributes to the regulation of energy stores during nutrient stress. The observation that similar signaling components function in both the mutant and starvation contexts strengthens the physiological relevance of the findings.
Strengths:
The revised manuscript is substantially improved and presents a coherent model linking sensory and interneuronal signaling to peripheral metabolic responses. The data convincingly establish functional roles for the identified neuropeptide components in controlling fat storage and mobilization.
Readers should keep in mind that the exact relationships among pathway components are not completely resolved. While the genetic evidence supports placement of NLP-10, NPR-35, FLP-18, and NPR-4 within a common regulatory framework, the data are somewhat less definitive regarding the precise order of signaling events and whether all interactions are direct. The authors appropriately acknowledge some of this complexity, and a clearer distinction between experimentally demonstrated versus inferred connections would further strengthen the presentation.
Overall, this is a carefully executed and biologically interesting study that advances our understanding of neuroendocrine control of metabolism during nutrient limitation.
Weaknesses:
No weaknesses were noted by this reviewer in the revision. My main comment is that whilst I appreciate the authors’ efforts to refine the signaling model in response to earlier comments, the pathway architecture remains somewhat complex, particularly regarding the relationships among NLP-10, NPR-35, FLP-18, and NPR-4. I encourage the authors to make one final pass through the text and model figure to clearly distinguish interactions that are directly supported by experimental evidence from those that remain inferred based on genetic relationships. This would help readers accurately interpret the strength of evidence supporting each step of the proposed pathway.
We are grateful to R2 for the positive assessment and for the constructive suggestions. We understand that the remaining concerns related to clarity and presentation. We agree that the pathway architecture remains complex and that readers should be able to distinguish experimentally connections from inferred ones. We will revise the text and the model figure accordingly. For example, future versions could use solid connectors for data-supported steps and dashed connectors for inferred steps.
We agree with R2 that readers should keep in mind that the exact relationships among pathway components are not completely resolved, and we agree and appreciate R2’s comment that “…the data are somewhat less definitive regarding the precise order of signaling events and whether all interactions are direct”, which echoes the suggestions from R1. We will address these points as outlined in our response to R1.
Reviewer #3 (Public review):
Summary:
In this manuscript, Liu et al. used C. elegans animals lacking functional BMP as a model to dissect the neuropeptide signaling pathways that lead to BMP-associated fat loss. First, they established the BMP mutants as a model of dyspepsia that results in chronic dietary restriction. They then used this model to identify neuropeptide genes that are upregulated and required for BMP-associated fat loss. Next, they sought to identify the receptors for these neuropeptides and to place the genes in a genetic and cellular pathway. They also showed that the same neuropeptides and receptors contribute to fat loss in diet-restricted C. elegans. They conclude that they have identified a neuropeptide signaling cascade in which impaired pharyngeal function activates sensory neurons, which then signal through interneurons to enhance triglyceride hydrolysis in the intestine.
Strengths:
In this manuscript, Liu et al. characterized the pharyngeal muscles and the intestinal bacterial context of animals lacking functional BMP. They showed a correlation between BMP-associated fat loss and pharyngeal impairments that could lead to insufficient food grinding. This finding suggests that C. elegans lacking functional BMP could serve as a model for dyspepsia and chronic dietary restriction. Moreover, the authors analyzed previously published gene expression data to identify neuropeptides that are overexpressed in animals lacking BMP. They then used mutants for these neuropeptides and their receptors to determine which of these genes contribute to BMP-associated fat loss. Finally, they showed that the same neuropeptides and receptors also contribute to fat loss in diet-restricted C. elegans.
We thank Reviewer 3 for a detailed assessment on this manuscript.
Weaknesses:
(1) The authors use double mutants between BMP pathway mutants and neuropeptide or neuropeptide receptor mutants to argue that the reduced fat levels of BMP mutants are suppressed. However, with the exception of unc-31, the fat content of the corresponding single neuropeptide or neuropeptide receptor mutant is not shown. Without these, the double-mutant phenotype cannot be interpreted as suppression, since it could reflect an independent effect on fat storage. To show that the effects are related to BMP signaling, the authors should compare wild-type animals to neuropeptide mutants in a sma-6(+) background. Similarly, the fat content of ADL-ablated animals is not shown in an otherwise wild-type background.
We accept this criticism. We are going to add experiments on wild-type and related single-mutant strains in the near future. Specifically, we will report the fat content of nlp-10(tm6232), npr-35(ok3258), flp-18(tm2179), flp-1(sy1599), flp-4(sy1606), and npr-4(tm1782), and of ADL-ablated animals, in the sma-6-functioning background. We will also survey published datasets for independent measurements of fat content in these strains, if any exist.
(2) In several places, the authors use cell-specific expression of neuropeptides to conclude that a given neuropeptide acts in a particular cell type (e.g., Figure 4B, C; Figure 5C, D). However, because neuropeptides are secreted and can act at a distance, expression in a non-native cell type may rescue the phenotype. These experiments therefore show sufficiency but do not establish the endogenous site of action. Cell-specific knockout or knockdown, or expression from control promoters in unrelated cell types, would help address this.
We agree with the underlying point of this criticism. Expression of a secreted peptide from a heterologous promoter demonstrates that the peptide can act when produced in that cell; however, it does not establish where the peptide acts under physiological conditions. We are grateful to the reviewer for these suggestions and have carefully considered both approaches. While we fully appreciate their potential value, we believe it is more appropriate to regard them as complementary rather than decisive, and we would like to explain our reasoning in detail.
The first approach, cell-specific knockdown, may or may not affect peripheral lipid catabolism in this system, because (1) the number of peptides capable of influencing fat storage is large, as shown in Figure 2 – figure supplement 1D, E, and (2) the expression pattern of a receptor or peptide gene may be altered under nutrient-deprived conditions (as in comments of R1). The axis we describe could be buffered by ligand redundancy, as our data on FLP-1 and FLP-4 show. Under such redundancy, a negative result from knocking down a single peptide in a group of cells would not exclude a physiological role for that peptide in that cell. The second approach, expression from a control promoter in an unrelated cell, might also be complicated by the point that R1 raises: if nutrient status alters the expression of peptides and their receptors, it becomes difficult to specify a cell type that is reliably “unrelated” to lipid regulation in the relevant physiological state. Therefore, we will modify our manuscript to describe the rescue experiments consistently as demonstrating sufficiency and discuss this limitation frankly.
(3) The authors report that expressing npr-35 from two promoters with non-overlapping expression each fully restores the low-fat phenotype of sma-6; npr-35 mutants. This is surprising for a receptor and is difficult to interpret.
We thank the reviewer for raising this point, and we realize that our description was not sufficiently precise. The two promoters drive expression in different subsets of npr-35-expressing neurons; however, in both cases npr-35 is overexpressed, as is typical of C. elegans transgenic lines. A non-exclusive explanation also seems plausible to us: the two cell populations (nmr-1-positive & ttx-3-positive) may converge on a shared downstream output, which is the release of FLP-18 from these interneurons, so that restoring receptor function in either population restores the ultimate output. We will revise the text to present both possibilities and to describe the result in terms of sufficiency.
(4) The authors identify neuropeptide genes co-expressed with npr-35 and follow up on flp-18. The rationale for restricting the search to co-expressed neuropeptides may bias hypothesis generation. Similarly, the authors show that flp-1 and flp-4, in addition to flp-18, contribute to BMP-associated fat loss, but only flp-18 is examined further because of its overlapping co-expression with npr-35, again creating a bias in the investigation of the role of neuropeptide signaling in BMP-associated fat loss.
We accept that restricting our attention to peptides co-expressed with npr-35 introduced a bias, and we will state this openly. We would like to note two considerations that may help contextualize this bias. First, flp-18 showed the strongest transcriptional induction among the peptides we examined (Figure 2 - figure supplement 1D), and this was an important reason for prioritizing it. Second, our subsequent analysis was not confined to co-expressed peptides: flp-1 and flp-4 are also required for BMP-associated fat loss, while flp-1 shows little overlap with npr-35 and does not appear to be released from npr-35-expressing neurons (Figure 5 - figure supplement 1D). The co-expression therefore served as an entry point rather than as a filter. We will clearly describe this rationale in the text, present the broader set of candidate peptides identified in our screen, and note that additional peptides are likely to contribute to fat regulation. We hope that the dataset will support work in this direction, both in our laboratory and in other laboratories.
(5) It is difficult to interpret the experiment in Figure 5E, where the authors conclude that npr-4 is required in flp-18-overexpressing animals. Without a demonstrated fat-loss phenotype from overexpressing flp-18, a requirement for npr-4 in those animals cannot be meaningfully assessed.
We assume that the reviewer is referring to Figure 5G, and we apologize for the confusion in this panel. This confusion may have arisen from the way we arranged the columns. The experiment was designed to ask whether npr-4 is required for overdosed FLP-18 to function (with overexpression ensured by sma-6(lf)-induced upregulation and the ttx-3p::FLP-18 transgene). Therefore, a clearer way to organize these columns would be to swap the third and fourth columns, which we will do in the revised version.
We nonetheless agree with the reviewer’s underlying point. Because ttx-3p::FLP-18 alone does not reduce fat in the already fat-deficient sma-6(lf), the assay has limited dynamic range. So, a requirement for npr-4 cannot be established from this comparison alone. We will reorder the columns, correct the legend, and present the panel as consistent with npr-4 acting downstream of flp-18. We would also place greater emphasis on the comparisons in Figure 5B, in which sma-6(lf); npr-4(lf) retained more fat than sma-6(lf); flp-18(lf), an outcome expected if the receptor acts downstream of the ligand and if additional ligands converge on NPR-4.
Overall, the authors convincingly identify neuropeptides and receptors that are overexpressed and that contribute to fat loss. However, several points make it difficult to fully support the proposed model and the linear pathway from the pharynx to the intestine through ADL sensory neurons and interneurons. These include the absence of phenotypic analysis of the single mutants, the challenges of interpreting cell-specific rescue experiments when studying neuropeptide function, the fact that additional neuropeptides causing BMP-associated fat loss are identified but not studied further, and results that do not necessarily point to a linear pathway (e.g., the triple mutant sma-6; nlp-10; npr-35 has a stronger phenotype than the double mutants with sma-6, the fat loss caused by overexpression of nlp-10 in a sma-6 background is not suppressed by npr-35, and the flp-18 overexpression experiment is difficult to interpret). The reviewer therefore suggests that the linearity of this pathway, and the sites of action of the neuropeptides and receptors, should be presented more cautiously or supported by additional data.
We are grateful for this synthesis, which we take as the central conceptual comment of the review. We agree that our data do not support a strictly linear pathway, and we will revise both the text and the model figure accordingly.
The two observations the reviewer cites, that the triple mutant sma-6; nlp-10; npr-35 retains more fat than either double mutant, and that removing npr-35 does not offset the fat loss caused by NLP-10 overexpression, point in the same direction: NLP-10 likely acts through receptor(s) in addition to NPR-35. We consider this plausible on independent grounds. Because nlp-10 is induced when nutrients are limiting, an additional receptor might be engaged only when NLP-10 rises above its basal level, or only when the nlp-10 expression domain expands under dietary restriction. Either scenario would explain why such a receptor has not emerged from interaction screens carried out under standard feeding conditions.
In the revision, we will describe the axis as a genetic framework in which the NLP-10/NPR-35 and FLP-18/NPR-4 modules are related, with branch points at both the ligand and receptor levels, and we will mark the connections that remain inferential. We will also note that the conservation of NPR-35 across nematodes (Beets et al., 2023; Golinelli et al., 2024) may provide an entry point for identifying additional NLP-10 receptors, and we would welcome works in that direction.
Recommendations for the authors:
Reviewer #1 (Recommendations for the authors):
The manuscript could provide more nuanced and accurate context in several areas:
(1) Work from the Ashrafi Lab (Greer et al., 2008, Cunningham et al., 2014) has described roles for daf-7 (another TGF-beta) in fat metabolism and should be cited to provide a more comprehensive introduction.
Thanks! We will cite the work from the Ashrafi laboratory on the DAF-7 branch of TGF-β signaling and on fat metabolism in the Introduction of the revised manuscript.
(2) The interpretation of the unc-31 phenotype is not as straightforward as the text indicates. unc-31 does not only affect DCVs, unc-31 mutants also have SV defects. Moreover, DCVs also release TGF-beta and monoamines. Thus, an unc-31 phenotypes does not necessarily implicate neuropeptides.
We thank the reviewer for this comment and apologize for the confusion. In fact, in addition to analyzing unc-31(lf); sma-2(lf), we analyzed the fat levels of egl-3(lf); sma-2(lf) and egl-21(lf); sma-2(lf), and the phenotypes caused by the loss of egl-3 or egl-21 were similar to those caused by unc-31(lf) in these BMP mutants. We omitted these data from the previous version for simplicity, and we acknowledge that we wrongly overlooked the non-DSV roles of UNC-31/CAPS. We will revise this section by adding additional data on egl-3 and egl-21.
(3) In lines 132-134, the logic of the argument is not clear, since it does not indicate how sma-3 rescue in the pharynx is linked to sma-6 in the ADL neurons.
We agree with the reviewer’s comment and will remove the wording related to sma-3 rescue from this sentence in the revised manuscript.
(4) Lines 271-273 indicate that BMP mutants are mildly long-lived. This point is only partially correct. Depending on the food level, daf-7 mutants can be long-lived, have no phenotype, or be short-lived.
Thank you. We will rewrite the section on lifespan so that it reflects the lifespan of daf-7 mutants and BMP mutants properly.
This manuscript could also be improved by clarifying or acknowledging several technical gaps/weaknesses:
(1) For Supplemental Figure 2d, it is not clear if there is any correction for multiple hypothesis testing when assaying multiple neuropeptides. This point should be clarified.
We acknowledge that we did not perform any multiple hypothesis testing on the data shown in this figure panel in the previous version. Due to technical limitations, we were unable to compare all qRT-PCR groups within a single 96-well plate. The qRT-PCR assays were performed separately for each gene, with double mutant and single mutant worms collected in the same-day manner. Accordingly, these qRT-PCR results are not suitable for comparisons between different genes, but they do allow valid comparisons between the double mutant and the corresponding single mutant within the same target gene. We will clarify this point in the revised manuscript.
(2) Strains used in this work have only been outcrossed twice, which leaves 25% of mutations. The parent strains with tm and ok alleles were isolated after mutagenesis and are expected to contain thousands of mutations. Two outcrosses imply that several hundred unlinked mutations remain, which weakens the genetic evidence.
Because most of the data were obtained from double mutants (e.g., sma-6(lf); nlp-10(lf), sma-6(lf); npr-35(lf)), we performed at least one additional round of outcrossing by introducing the single mutants into the BMP mutants. We therefore expect that the background mutation load in these double mutants is lower than 25%. As we plan to do more experiments on the single mutants, we will first increase the number of outcrosses before measuring fat levels.
(3) More details are needed on the methods related to fluorescence microscopy, especially since most of the manuscript's conclusions rest on quantitative fluorescence. It is important to describe the light source and whether there was any calibration to control for variable fluorescence due to the equipment.
Thank you. We will do so. All Nile Red imaging was performed on the same microscopic system, which is not a confocal microscope but an upright widefield microscope, which captures the full depth of field along the z-axis (rather than a single optical section). In the revised manuscript, we will expand the fluorescence microscopy methods to describe the illumination source, the acquisition settings, and the measures taken, etc.
Reviewer #2 (Recommendations for the authors):
Minor Comments
(1) The finding that NLP-10 overexpression alone does not produce a strong phenotype in wild-type animals is intriguing. A brief discussion of how this observation fits into the proposed model would be useful for readers.
We agree with the reviewer that NLP-10 overexpression alone does not produce a strong phenotype in wild-type animals, which is intriguing. We have not yet performed a detailed mechanistic analysis of this phenomenon. We speculate that animals likely possess metabolic homeostatic mechanisms that buffer against perturbations in NLP-10 levels. As R1 suggested, it is also possible that dietary restriction or starvation conditions may broaden the expression range of nlp-10 or its receptor(s). We will discuss this possibility and its relationship to our proposed model in greater depth in the revised manuscript.
(2) Some figure legends describing complex genetic comparisons could benefit from additional explanatory language highlighting the key conclusion from each experiment. This would improve accessibility of the paper.
Figure legends will be improved in the revised version.
Overall Recommendation:
The manuscript has been substantially strengthened through revision, and the remaining issues are largely matters of clarity and presentation. I am supportive of the paper following minor revision.
Thank you again for your feedback. Revisions will include updated data and improved wording in future versions.
Reviewer #3 (Recommendations for the authors):
(1) dbl-1 mutants are mentioned in the Introduction without explanation. The authors should briefly describe what dbl-1 encodes and how it relates to the BMP pathway studied here.
We apologize for this confusion and will introduce DBL-1 properly in the Introduction by identifying it as the C. elegans BMP ligand that signals through the SMA-6/SMA-2/SMA-3/SMA-4 pathway.
(2) Lines 19, 60, and 238: It is unclear what exactly the authors mean by "mutant-specific effects," "mutant-specific phenomenon," and "mutant-specific mechanism."
We will replace the term “mutant-specific” with alternative wording. By this term, we meant “specific to BMP mutants”, as opposed to a general response to nutrient insufficiency.
(3) Lines: 56-57 The authors state that "the ADL sensory neurons release neuropeptide NLP-10... to promote the secretion of... FLP18". To make this statement, the authors need to use a neuropeptide secretion assay (e.g., the coelomocyte release assay) to show their claim or tone down the language.
We will tone down the language since we do not measure peptide secretion directly.
(4) ADL ablation in the sma-6; nlp-10 double mutant could help support the authors' hypothesis.
We agree that ADL ablation in sma-6(lf); nlp-10(lf) could in principle provide additional information. However, we do not consider this experiment essential for the revised manuscript. Our main concern is that metabolic regulation is mediated by many peptides with partially redundant and compensatory functions. Therefore, the outcome of ADL ablation in sma-6(lf); nlp-10(lf) would be difficult to interpret unambiguously. We will acknowledge this limitation in the Discussion of the revised manuscript.
(5) Figure 3D: The authors perform an ADL-specific rescue experiment to show that expression of sma-6 in ADL does not rescue the fat content defect of sma-6 mutants. Because this is a negative result, it is hard to interpret unless the authors first show that the construct is indeed expressed in ADL (e.g., by co-expressing a fluorescent reporter from the same transcript, such as in an operon with sma-6) and, second, that they can rescue the sma-6 phenotype by expression in the pharynx, especially if they want to support the model shown in Figure 3E.
We will either image ADL neurons expressing srh-220p::sma-6::sl2::gfp, using DiI filling as an ADL landmark, or test whether pharyngeal expression of SMA-6 rescues the phenotype.
(6) Figure 5G: the figure legend describes a different experiment than the one shown in the figure.
We will correct the legend for Figure 5G.
Minor points:
(1) Line 106: change "verify" to "verified".
(2) Line 134: change "might have worked" to "might work".
(3) Line 148: replace the ";" at the end of the phrase with a full stop.
(4) Figure 5- figure supplement 1E: the title reads "npr-4 is required in flp-10 overexpressed animals". The authors presumably mean "nlp-10-overexpressing animals".
We thank the reviewer for pointing out these flaws and will make corrections in subsequent versions.
eLife Assessment
This important study records basal ganglia output neurons in the substantia nigra pars reticulata of head-fixed mice performing self-timed reaching and licking tasks. Contrary to simple predictions of action selection and inhibitory gating by basal ganglia output, the authors report that most substantia nigra pars reticulata neurons modulate congruently across reaching, licking, and reaches to different targets. Optogenetic identification of substantia nigra pars reticulata neurons projecting to the central and lateral, orofacial superior colliculus shows that these congruent signals are routed to both output zones. Evidence from the recordings is convincing, while that from the projection-specific experiments is solid but rests on few animals and would be strengthened by increased power and clearer criteria for unit identification.
Reviewer #1 (Public review):
Summary:
The manuscript describes results from a series of experiments designed to test different facets of the gating model and potential alternative explanations. In most respects, the results are squarely inconsistent with the gating model. Some of the authors' attempts to test the model force the authors to define the model in greater detail than even the model's proponents typically do. That is a revealing exercise in itself, which is well worth bringing to the field's attention. The manuscript is well written for the most part, and figures illustrate results clearly.
Strengths:
The gating model states that pauses in the activity of a select subset of SNr neurons disinhibit (i.e., "open the gate" to) a specific desired action, whereas a larger population of SNr neurons increase their firing so as to suppress other potentially conflicting actions. Selection of different actions is performed by the generation of pauses and increases in activity in different subsets of SNr neurons. Contrary to that prediction, the authors report that SNr neurons that show a reduction in firing for one action tend to show reductions in activity for other actions as well and, similarly, increase-type neurons tend to show increases for all of the actions examined. The authors tested the generality of these observations across a series of experiments that compared activity between markedly different actions (moving forelimb vs. tongue), similar actions (different directions of forelimb movement) and several forms of self-directed and spontaneous movement. Moreover, the authors examined the activity of anatomically identified subsets of SNr neurons that innervate forelimb and orofacial sub-regions of the superior colliculus. Combined, the series of experiments described here provide overwhelming evidence against any straightforward "simple" form of the gating model of basal ganglia function.
Weaknesses:
The most important limitations of the manuscript actually arise from additional ambiguities of the gating model - ambiguities that may be worth mentioning in the manuscript. For example, how granular is the action selection process supposed to be? Perhaps the SNr circuits examined in the present study are selecting "actions" defined at a level that the authors did not examine. This is not a major limitation, however, given the concerted effort invested in testing different levels of granularity and contexts. This relates to another ambiguity of the model in that basal ganglia output does not affect actions directly, but only indirectly via its influences on downstream multi-synaptic neural circuits.
Another ambiguity of the model is how to interpret the activity of neurons that don't respond. In other words, are comparisons of decrease- and increase-type responses the appropriate comparisons to make to test the model? The present study compares decrease- vs. increase-type responses, but perhaps no-response and increase-type neurons should be lumped together. After all, the simplest version of the action selection hypothesis (cf. Hikosaka) considers the high baseline tonic activity of SNr neurons to be action suppressive by itself, without an additional task-related increase. A comprehensive action selection model should include an explanation of the relative roles of non-response and increase-type responses. The current manuscript would benefit from a discussion of this ambiguity and a rationale for the decisions the authors made to make the model testable.
A third ambiguity of the model is how to interpret graded changes in SNr firing rate. It comes as no surprise that the authors observed slow ramp-like changes in SNr firing rate prior to movement initiation in the absence of an immediate sensory trigger. The authors interpret these ramp-like changes as being equivalent (from the perspective of the gating model) to the abrupt changes in firing rate observed around the time of initiation for movements that are sensory-triggered. But the simple gating model presents basal ganglia output as having only three states: no-change, pause, and increase. Implied in the model is that these changes in firing rate influence not only what action is performed but also when it is performed. It is not clear that slow ramping responses qualify as "open the gate" disinhibitory temporally-abrupt signals. The manuscript could provide more explanation concerning the position the authors take to make ramping responses fit into the gating model.
Reviewer #2 (Public review):
The manuscript by Hou et al. tests the hypothesis that SNr neurons' activity is inhibited during specific movements, allowing the selection of specific movements. They do so by performing electrophysiological recordings in SNr, in head-fixed mice trained to perform self-paced reaching or self-paced licking. They first show that SNr neurons present ramping activity hundreds of milliseconds before reaching, with the vast majority of cells presenting positive increases in activity in these rampings. Then they show that the activity of these cells is related to the velocity (and ramping) of movements, independently of the timing to initiate. They verify these findings in spontaneous reaching. To then test the activity of the same cells in self-paced reaching vs. licking, they found that the vast majority of cells present congruent firing between conditions (not opposite firing as expected). Further, they ran controls asking whether dips in activity are related to specific movements, checked if the specific cells that are correlated with lick cycles respond differently in self-paced reaching vs. licking, asked whether there could be differences in activity in reaching to different targets, to finally "opto-tag" SNr- orofacial vs SNr-arm cells and asked whether these differentiate their spiking in reaching vs. licking, concluding that a very minority of SNr neurons distinguished between movements while the vast majority presented congruent modulations among movements.
Strengths:
The study is a tour de force, testing the general assumption of action selection being reflected in SNr spiking.
Weakness:
Even though the authors set up an amazing set of experiments to test the mentioned hypothesis, their final opto-tagging experiment needs to be screened to better support their conclusion.
Reviewer #3 (Public review):
Summary:
The authors aimed to answer the question of whether the basal ganglia, specifically neurons in the substantia nigra pars reticulata (SNr), directly select actions. This was done by recording the activity of SNr neurons, as well as neurons in the superior colliculus (a major recipient of SNr projections), during cued and self-directed actions, specifically recording the same SNr units during clearly distinct actions.
Strengths:
As the major output nucleus of the rodent basal ganglia, the SNr is a key location to determine if single-cell activity within the basal ganglia is specific to the selection or suppression of specific actions. This choice also provides a natural transition to the activity/roles of downstream motor nuclei that receive basal ganglia output. Multiple definitions of "action" were addressed throughout this manuscript, which is essential for an ambiguous concept like "action selection." They convincingly demonstrate that a "simple action selection hypothesis" in which there is a 1:1 correspondence between SNr single-unit activity and specific actions is incompatible with their data. The manuscript was clearly written and well-organized.
Weaknesses:
While the authors systematically address multiple definitions of an "action," all of the actions in this work are variations of either licking or reaching. The similar SNr activity during different types of reaches has several potential interpretations, which the authors enumerate and present arguments for why their interpretation (SNr does not implement action selection) is the most plausible. However, the study does not quite fully eliminate these possibilities. Finally, while the authors essentially eliminate the possibility of a 1:1 correspondence between SNr units and specific actions, it remains possible that SNr selects actions through a more complex population coding scheme. Indeed, they demonstrate that they can distinguish lick and reach actions by examining population SNr activity.
Appraisal of whether authors achieved their aims/evidence supports conclusions:
The authors achieved their aims of determining whether individual SNr neurons select for specific actions. This includes multiple definitions of "actions," with some actions differentiated by muscle group and body part (licking vs reaching), target, or timing. The evidence provided largely supports the authors' claims, though the low number of subjects in the causal optogenetic experiments makes interpretation of those results more difficult.
Discussion of likely impact of the work in the field:
This work will provide a foundation for future research into more nuanced dissections of basal ganglia circuitry, specifically between the SNr and its downstream target nuclei.
Author response:
Thanks to the reviewers, reviewing editor and senior editor for their assessment of our manuscript. We summarize and address reviewer comments and suggestions below:
(1) The granularity of action selection is ambiguous (Reviewers 1 & 3)
(2) Consider neurons not modulated by movement (Reviewer 1)
(3) How to interpret ramp-like vs. abrupt changes in pre-movement firing rates (Reviewer 1)
(4) Population encoding could be considered a form of action selection (Reviewer 3)
(5) Improve the sample-size and screening of optotagging results (Reviewers 2 & 3)
(1) The granularity of action selection is ambiguous (Reviewers 1 & 3)
The reviewers astutely pointed out a critical ambiguity of the original action selection model: on what level are actions differentiated in the basal ganglia? Is it activation of specific muscles, or concerted movements of multiple body parts, or even complex sequences of behavior? Anatomically, the organization of striatonigral input as well as nigrocollicular pathways appear to be highly somatotopic (Foster et al., 2021; Lee et al., 2020), in the sense that there are parallel circuits that separately project to orofacial or skeletomotor regions of the superior colliculus (Benavidez et al., 2021). This might suggest an upper bound for the granularity of action selection in SNr, manifest at the level of tongue/jaw vs. limbs, different directions of reaching, or reaching vs. retraction. We tested these permutations and found little evidence for sign reversals in SNr responses. Nonetheless, one could always parse “action” into even finer components, or combine them into more complex ensembles or sequences of movements. It was exactly this logic that led us to test more and more movements, starting with different directions of reaching and culminating in the unbiased examination of spontaneous movement (Fig. 5, 6, 7). Nonetheless, we of course could not exhaustively exclude all movement possibilities/permutations, and our arguments are limited to the sets of actions that we tested. But we also point out that the granularity argument could be extended ad infinitum, to the point of being practically unfalsifiable! In the end, we will lay this logic out more clearly in the revised Discussion. (On the other hand, it is exactly this “granularity” conundrum that led us away from trying yet another movement paradigm, and toward the idea of examining functional connectivity between the SNr and SC, as outlined below.)
(2) Consider neurons not modulated by movement (Reviewer 1)
Reviewer 1 pointed out that unmodulated SNr neurons could be grouped with increase neurons, since they could maintain or increase inhibitory output in order to inhibit unwanted movements. Upon further inspection, we found 56/427 (13%) SNr units decreased firing for reaching but not for licking; conversely, 12/427 (3%) units decrease firing for licking but not for reaching; while 334/427 (78%) units increased firing or were unmodulated for both reaching and licking. In our original analysis excluding unmodulated SNr units, the proportions of units with congruent (86%) vs. incongruent (14%) signs of modulation for reaching and licking did not differ drastically when we included unmodulated units (78% vs. 16%). Nonetheless, this is a valid point, and we will include this in the revised Results and Discussion.
(3) How to interpret ramp-like vs. abrupt changes in pre-movement firing rates (Reviewer 1)
Reviewer 1 questioned how the slow ramp-like modulation of SNr firing relates to the disinhibitory gating model. We adopted the self-timed/spontaneous movement task not only to separate sensory vs. motor activity, but also due to clinical observations in Parkinson’s disease that patients are more debilitated for self-initiated rather than cued movements. We observed early pre-movement neural modulation in SNr during self-initiated movements, whereas Mink and Thach (1991) argued that pallidal activation was too late for movement initiation – albeit only in the context of cued movements. Slow ramp-like changes SNr firing does not contradict action selection per se, since modulation of SNr firing might need to reach some threshold level for a desired movement to occur, even in the context of action selection. On the other hand, these ramp-like changes might also represent a generic movement initiation signal without specifying which action to take. We pointed out the existence of ramp-like modulation during self-initiated movements, but the ramp-like nature of SNr modulation is not fundamentally incompatible with the gating model. We will clarify this point in the revised Discussion.
(4) Population encoding could be considered a form of action selection (Reviewer 3)
The reviewers also pointed out that while action selection is largely absent in single SNr neurons, licking vs. reaching could be efficiently decoded from population activity. Reviewer 3 reasonably asked: could population encoding be a form of action selection?
The answer, of course, is yes – but by the same token, any motor area that displays selectivity (motor cortex, thalamus, cerebellum, brainstem, etc.) could be argued to play a role in action selection. So what’s special about the basal ganglia? In this regard, it’s important to re-examine the basis for the action-selection hypothesis. The original model was put forward by Mink & Thach in 1991 to explain the paradoxical observation that most inhibitory pallidal output neurons increase rather than decrease firing during movement. The authors speculated that this increase population could inhibit unwanted movements, while the smaller decrease population could disinhibit the desired movement. The model has been tweaked here and there through the years, but the fundamental core of the idea has persisted, with little supporting evidence, presumably because it provides an explanation for the paradoxical increase neurons – an interesting explanation to be sure. However, if neurons that increase firing do so to inhibit an unwanted movement, then the simplest “other side of the coin” explanation is that they should decrease firing when that same movement is desired. Yet this sign-reversal motif was exceedingly rare among the SNr neurons in our study. Thus the critical question remains: why do so many pallidal output neurons increase firing with movement? We hope that one useful outcome of our study will be to stimulate others to think about pallidal responses outside of the action-selection framework.
Nonetheless, one could certainly argue that action selectivity is a prerequisite for action selection, and it may be worth asking how the SNr selectivity could contribute to this role. The SNr projects broadly to multiple brainstem and thalamic motor nuclei, which in turn affect specific movements. One could certainly devise wiring schemes whereby SNr outputs is parsed out among those downstream targets to mediate action selection. But these schemes are not exclusive to the SNr; they could pertain to any motor area with selectivity, nor do they provide an explanation for the paradoxical increase neurons. Nonetheless, R3’s thoughts likely reflect those of other readers, and we will try to lay out our logic more clearly in the revised Introduction and Discussion.
(5) Improve the sample-size and screening of optotagging results (Reviewers 2 & 3)
We used an optotagging technique to identify SNr neurons projecting to arm vs. orofacial SC and demonstrated that – during both reaching and licking – SNr neurons predominantly increased their inhibitory output regardless of their putative projection target. We think these results have important implications for the role of SNr output. However, the reviewers pointed out some critical limitations of our optotagging approach. Typically, to rule out polysynaptic confounds, light-evoked spikes are expected to have very low latency (typically a few milliseconds, but highly dependent on light power and opsin efficacy). However, it is unclear where the threshold for spiking latency should be drawn. Instead, to rule out polysynaptic confounds in our study, we relied on the fact we observed extremely few (<5%) silenced SNr neurons during optogenetic stimulation of a predominantly GABAergic region. Because SNr neurons send collateral GABAergic projections to other SNr neurons, optogenetic activation of a neighboring neuron should synaptically inhibit (rather than excite) a neuron of interest. Another possible source of contamination would have come from light leaking to into SC, which could have polysynaptically excited SNr neurons through the collicular-thalamic-striatal pathway. In this case, direct pathway striatal projection neurons should also suppress SNr firing. However, we did not observe any light-evoked inhibition among hundreds of putative SNr neurons, perhaps because viral labeling of neurons was sparse enough that polysynaptic inhibition was rare and largely unable to overcome optogenetic excitation. Therefore we did not suspect polysynaptic contamination among our results. Nonetheless, even after we had restricted our analysis to SNr units with low (<median) optogenetic responses latencies, the fraction of SNr units negatively modulated for reaching or licking was still extremely low (4% for reach among putative arm-SC projecting neurons; 8% for lick among putative orofacial-SC projecting neurons). We plan to collect additional data to address concerns regarding the smaller sample size.
References:
Benavidez, N. L., Bienkowski, M. S., Zhu, M., Garcia, L. H., Fayzullina, M., Gao, L., Bowman, I., Gou, L., Khanjani, N., Cotter, K. R., Korobkova, L., Becerra, M., Cao, C., Song, M. Y., Zhang, B., Yamashita, S., Tugangui, A. J., Zingg, B., Rose, K., . . . Dong, H.-W. (2021). Organization of the inputs and outputs of the mouse superior colliculus. Nature Communications, 12(1), 4004. https://doi.org/10.1038/s41467-021-24241-2
Foster, N. N., Barry, J., Korobkova, L., Garcia, L., Gao, L., Becerra, M., Sherafat, Y., Peng, B., Li, X., Choi, J.H., Gou, L., Zingg, B., Azam, S., Lo, D., Khanjani, N., Zhang, B., Stanis, J., Bowman, I., Cotter, K., . . . Dong, H.-W. (2021). The mouse cortico–basal ganglia–thalamic network. Nature, 598(7879), 188194. https://doi.org/10.1038/s41586-021-03993-3
Lee, J., Wang, W., & Sabatini, B. L. (2020). Anatomically segregated basal ganglia pathways allow parallel behavioral modulation. Nature Neuroscience, 23(11), 1388-1398. https://doi.org/10.1038/s41593-020-00712-5
eLife Assessment
The study makes a valuable contribution to understanding the development of reflexes responsible for gaze stabilization. Using a combination of behavioural, electrophysiological and imaging approaches and analyses in developing Xenopus tadpoles, the findings provide a convincing demonstration that maturation of the angular vestibulo-ocular reflex in larval tadpoles involved increased modulation of silent but not spontaneously firing abudens motor units during horizontal semicircular canal activation, as well as increased myelination and innervation of the lateral rectus motor nerve. The findings will set the stage for future investigations to uncover the mechanistic underpinnings of the current observation.
Reviewer #1 (Public review):
Summary:
Movements, including reflexes, can undergo changes during development. This is the case for the angular vestibulo-ocular reflex in many species, including the Xenopus laevis frog. This study aimed to study the underlying mechanisms by which this reflex matures during development of larval Xenopus laevis using electrophysiology and electron microscopy. Their electrophysiological recordings of the motor axons of the abducens motor nerve during evoked angular vestibulo-ocular reflex reveal that a subpopulation of abducens motoneurons, those that are silent rather than spontaneously active, increase their firing rate and modulation during the angular vestibulo-ocular reflex during development. Furthermore, electron microscopy imaging of the developing abducens motor nerve shows increased myelination. Light microscopy reveals increased innervation of the lateral rectus muscle
Strengths:
Recordings during the evoked angular vestibulo-ocular reflex enable assessment of activity of motoneurons during the movement. Recordings done at specific developmental stages reveal convincing changes in modulation of firing of silent motor units that are distinguished by their high-amplitude spikes in nerve recordings. Microscopy, both electron and light, provide strong evidence for increased myelination of the abducens motor nerve and innervation of the lateral rectus muscle.
Weaknesses:
The authors did not provide data showing why the silent motor units became more responsive with development. Is it related to maturation of the sensory inputs to these neurons or changes in intrinsic properties?
Reviewer #2 (Public review):
Summary:
The manuscript by Pain et al. explores the functional maturation of key oculomotor populations responsible for gaze stabilization. The work is motivated by our poor understanding of this process in establishing appropriate reflexes during development. The authors focus on two types of abducens motoneurons implicated in the maturation of angular vestibulo-ocular reflexes (VOR) in amphibian tadpoles, namely spontaneous motor units that provide baseline control and silent motor units recruited only during head motion. This is a particularly good choice given the well-defined and evolutionarily conserved functions of these populations and the ease with which they can be tracked during changes in angular VOR as tadpoles develop.
The authors first use semi-intact larval preparations and both step and sinusoidal horizontal stimuli to characterize developmental improvements in angular VOR, which include faster, more robust compensation of head rotation and higher frequency tuning. Next, to explore the central versus peripheral origins of maturational changes, the authors use galvanic vestibular stimulation of the vestibular nerve (which should be independent of canal size) and BrdU-pulse chase to track neurogenesis, which strongly link improvements in VOR to changes in existing circuits, rather than the addition of new ones. To provide a more comprehensive assessment of what these functional changes may entail, they then perform recordings of changes in motor neuron discharge patterns during development specific to horizontal motion. By combining with sophisticated spike sorting analysis, these data reveals improvements in head velocity encoding by silent subtypes, while head positioning by spontaneous subtypes is relatively consistent during development. Finally, electron microscopy analysis of the abducens motor nerve is used to demonstrate that improvements in encoding are accompanied by broad changes in peripheral myelination and innervation.
Together, the data argue that silent abducens motor units show the largest changes in recruitment in response to sinusoidal head rotations, which is accompanied by increases in myelination and signal fidelity. Ultimately, the work suggests that improvements in oculomotor reflexes during development are accomplished by non-uniform changes within defined oculomotor populations, which is novel and interesting.
Strengths:
Overall, the data are very convincing, the analysis is thorough, the findings will be of interest to those in the VOR field, and also set the stage for future investigations aimed at uncovering the mechanistic underpinnings of this cool observation. I only have a few comments to consider for improvement.
Weaknesses:
Major.
(1) A bit more clarity is needed.
For non-specialists, it would be helpful in the introduction (e.g., Line 52) to spell out what the functional differences are to begin between spontaneous and silent subtypes. The contribution to head positioning versus head velocity is raised later on, but bringing it up earlier would be helpful to understand the functional implications of differences up front.
(2) Figure call outs.
These should be double-checked. I found one that did not seem appropriate on line 332. In addition, the reference to Figure 3A and the spike rate are not illustrated.
(3) Spelling and grammar.
It would be helpful to proofread more carefully for typographical errors, particularly in terms of using singular versus plural and the appropriate articles. Most are minor (those that I found listed below), but in one case the lack of clarity was particularly problematic. On Line 518: 'Altogether, past studies and our present work suggest the existence'. Consider revising this sentence; it reads now that motoneurons mature 'conjointly'; however, this does not seem to capture the main point here, which is that they are not maturing at the same time. Do the authors mean staggered or conjointly with the behaviours they control?
(4) Interpretation.
In the discussion, it would be good to be clearer that myelination will not explain the changes in firing pattern, but could explain the behavioural changes. Ideally, this would be linked to the different types, but I appreciate it is not easy to decipher that from EM.
eLife Assessment
This important study characterizes wing flicking in sexually immature Drosophila melanogaster females as a rejection response behavior that produces patterned acoustic signals and suppresses male courtship. The authors combine behavioral and acoustic measurements with manipulations of sensory input and targeted neuronal populations to provide solid evidence for how sexual maturation is signaled. With some clarification, additional information, and a more cautious interpretation of the findings in the discussion, this study will be of interest to neurobiologists and ethologists studying sexual maturation and behavior and will open up opportunities to address questions across several distinct areas, including sexual dimorphism, sexual maturation, and even the evolution of sexual signals.
Reviewer #1 (Public review):
The authors investigate a previously described behavior of sexually immature female Drosophila melanogaster, wherein immature females flick their wings in response to male courtship. They record, analyze, and characterize the auditory signals generated by these wing-flicks and show that these sounds reduce male courtship song and copulation attempts. The study combines several techniques including behavioral observations, acoustic analysis, developmental experiments, sensory manipulations, and neuronal manipulations to investigate the stimulus dependence and neural basis of this behavior. Amidst all the literature on sexual communication, this study uniquely addresses an interesting question on communication during sexual maturation, providing valuable insights on why and how sexual maturation could be signaled.
A major strength of the study is its behavioral and acoustic measurements with manipulations of sensory input and identified neurons. Comparative analysis of immature wing-flick, male-male aggression, and male courtship song was equally interesting. The developmental analysis also provides a clear temporal relationship between female maturation, wing flicking, and mating receptivity. Together, these experiments provide compelling evidence that immature female wing flicking is a reproducible behavior in response to male courtship, and that it significantly reduces male copulation attempts. Identifying a defined population of neurons that coordinate this response behavior further supports this study.
However, it must be pointed out that some of the broader conclusions extend beyond what is directly demonstrated. In particular, while the evidence supports the interpretation that wing flicking functions as a rejection or deterrence signal from sexually immature females, the designation of this behavior as a signal of "asexuality" and the proposed evolutionary relationship between female wing flicking and male agonistic wing flicking are less firmly established. Furthermore, I think the similarity in acoustic structure and the involvement of related neuronal populations is not sufficient to conclude a) that the two behaviors share an evolutionary origin or b) that one behavior evolved through modification of the other. Similarly, proposed explanations concerning energetic costs, predation risk, and adaptive benefits should be presented as hypotheses rather than conclusions unless directly tested. These interesting points can be included in a subsection "Ideas and Speculation", to clearly distinguish directly demonstrated effects and evolutionary or mechanistic interpretations.
In conclusion, the study contributes to our understanding on communication during sexual maturation.
Reviewer #2 (Public review):
Summary:
This study characterizes wing flicking in sexually immature female Drosophila as a transient behavior that produces patterned acoustic signals and suppresses male courtship. The authors show that wing flicking is restricted to a short developmental window, is influenced by multiple components of male courtship and juvenile hormone-dependent maturation, and depends on Dsx+ pC1 neurons. They further show that wing flicking shortens male song bouts and reduces copulation attempts, and propose that it functions as a rejection signal associated with sexual immaturity. Overall, this is an interesting and valuable study that provides new insight into a little-characterized stage of female sexual maturation. The core characterizations are compelling, and some additional clarification and refinement of the mechanistic interpretations would further strengthen the manuscript.
Strengths:
The work expands our understanding of female sexual behavior and highlights sexual maturation as an active transition between distinct behavioral states rather than simply the acquisition of receptivity. The comparison with the post-mating unreceptive state is especially informative. It also brings richness to the understanding by combining behavioral, developmental, and neural analyses and interpretation. The data showing how male behaviors are influenced by female wing flicks are particularly compelling and represent a critical aspect of behavioral characterization that is often overlooked.
Weaknesses:
Event-level analysis of behaviors such as singing, chasing, tapping, licking, or attempted copulation would help clarify the sensory and social context of female wing flicks. The neural conclusion also requires some caution. The split-GAL4 reagent used to assign the phenotype to pC1a labels two pC1 neurons per hemisphere in this study, whereas pC1a is generally described as a single bilateral pair. Given the functional heterogeneity of pC1 subtypes, the current data do not fully exclude a contribution from another pC1 subtype. While the comparison with male agonistic wing signals is informative and interesting, the rejective wing flicks of immature females may be more directly comparable to the previously reported wing flicks and song produced by young and mature males in response to male courtship, which are not mentioned or explored in this study.
Reviewer #3 (Public review):
Summary:
Female receptivity to male courtship is regulated by many factors, including development and internal status. In mature, mated females, male courtship triggers ovipositor extrusion, which prevents intromission. In immature flies, however, rejection occurs through another behavior: immature wing flicking. How this behavior is engaged and regulated is unclear. Here, Bernet & Philipsborn advanced in this matter by defining behavioral and neuronal keys related to this phenotype. The authors show that immature wing flicking depends on pheromones and acoustic cues from the males to be triggered. Additionally, the authors identify the pC1a cluster as key for this behavior.
Strengths:
(1) In contrast to the regulation of ovipositor extrusion, immature wing flicker, while observed, has not been fully characterized. This is a significant step forward in our understanding of the ethology around female receptivity.
(2) The behavioral approaches that the authors used to determine the nature and correlation of immature wing flicker with male courtship are very convincing.
Weaknesses:
While the manuscript has clear strengths, particularly in describing immature wing flicking behavior, several analyses make some conclusions difficult to assess. This is especially troublesome since some of these assertions are the main points the authors highlight as significant findings. In particular:
(1) The authors describe immature female wing flicking as resembling that produced by mature males during agonistic encounters. While I see some similarities, this assessment is subjective. In Figure 1D, the authors present comparative metrics without statistics; thus, it is difficult to assess the validity of the results. Implications, however, can be direct: do females produce acoustic waves to mimic males and deter mating partners (or vice versa)?
(2) From Figure 3F, it is difficult to conclude that the arborization pattern does not differ between mature and immature flies (here, referred to as a 5h vs 10d difference). This may be an artifact of the representative images, but the arborization at 10d appears less dense, particularly around the PI area. If this difference is consistent, this could also reflect neuronal maturation contributing to this phenotype, which is particularly relevant to what the authors discuss in lines 309-339. Differences in synaptic partners might potentially explain the contrasting action of pC1 neurons in immature vs mature females (PMCID: PMC11386958).
(3) The experiments presented do not exclude the possibility that other pC1 neurons, in addition to pC1a neurons, affect the behavior. As written (line 363), the statement indicating that pC1a is the only one affecting immature female flicking is then incorrect.
eLife Assessment
This work presents a computational resource for identifying and classifying chloroplast outer envelope membrane β-barrel proteins (COB) across a broad range of plant species. By substantially expanding the available sequence space for this proteome, the COB database will serve as a potentially important resource for researchers working in diverse areas, including chloroplast protein import, outer envelope membrane biology, and membrane protein evolution. Although the computational framework and broad comparative analyses are generally solid, the limitations of the negative training set and the similarity-based sequence expansion approach may require further implementation. In addition, some of the broader biological implications need to be more carefully addressed, as they remain to be validated experimentally.
Reviewer #1 (Public review):
Proctor et al. developed the COB (chloroplast outer-envelope beta-barrel) database, a resource for identifying, classifying and structurally analysing chloroplast outer envelope beta-barrel proteins, built by combining sequence search, machine-learning classification, structural prediction and similarity-based database expansion. The scale and accessibility of the resource is a genuine strength, and it will be useful to researchers working on chloroplast protein import, outer envelope membrane biology and membrane protein evolution. However, several of the manuscript's central claims are not adequately supported by the data presented.
First, the classifier's negative training set consists entirely of non-transmembrane beta-barrel proteins, so its reported accuracy does not demonstrate reliable discrimination from other protein classes (like alpha-helical membrane proteins). The similarity-based expansion that substantially grew the database demonstrates only internal consistency among related sequences, not independent validation, and errors in the seed set will propagate through it. By design, this approach also misses divergent beta-barrels, as the author's own recovery of OEP40 and TOC159 outside the classifier illustrates. This is important because the authors identify the evasion of divergent barrels from these types of tools as a gap in the field. For these reasons, the database should not be described as "comprehensive".
Second, the structural conclusions and the biological implications made substantially outpace the evidence. The reported prevalence of "open" barrel conformations is, for now, a feature of the AlphaFold predictions, not a demonstrated biological phenomenon, and the authors themselves note these may be artefactual. Similarly, the hybrid-barrel analysis is built almost entirely around TOC159-family proteins, which are heavily overrepresented in the Arabidopsis input set. The manuscript does not test whether this apparent "universal" interaction is anything more than a compositional artefact of that enrichment, and most of the novel interactions have only moderate confidence scores. As framed, both the title and multiple statements in the text assert a general topology of "open and hybridized" for this group of proteins that only experimental data can establish.
Third, the evolutionary comparisons between Streptophyta and Chlorophyta, and between chloroplast and bacterial beta-barrels, are presented as robust differences but are not supported by any formal statistical treatment of proteome size, taxonomic sampling or phylogenetic non-independence.
Overall, the database and classifier are a valuable contribution for further exploration and experimental validation of chloroplast outer envelope membrane beta-barrel proteins, but the manuscript's broader structural and evolutionary claims need to be substantially tempered, and the title changed, to match what the data actually show.
Reviewer #2 (Public review):
Summary:
The manuscript presents COB, a computational database for the systematic identification and classification of chloroplast-associated beta-barrel proteins across Viridiplantae. The authors combine evolutionary contact-map features with machine-learning classification, sequence-based searches and structural predictions to generate a large collection of candidate beta-barrel proteins. In addition to established OEP families, they identify two previously uncharacterized groups, BUL16 and BUL22, analyze the evolutionary distribution and structural diversity of beta-barrel proteins, and propose a number of novel hybrid beta-barrel assemblies involving members of the TOC159 family.
Strengths:
A major strength of the study is the establishment of a dedicated and comprehensive resource for chloroplast beta-barrel proteins. These proteins have historically been difficult to identify and predict reliably, and the broad taxonomic coverage of COB should therefore be highly useful for the chloroplast and plant membrane biology communities.
The manuscript also goes beyond database construction by exploring structural and evolutionary properties of the identified proteins. In particular, the analyses of beta-strand number, open versus closed barrel conformations, taxonomic distribution, and the identification of BUL16 and BUL22 generate several interesting hypotheses that may stimulate future experimental work.
The authors are generally appropriately cautious in acknowledging limitations of the structural predictions, particularly with respect to open and hybrid barrel conformations.
Weaknesses:
The principal limitation of the study is that it is entirely computational. This is not necessarily problematic for a database/resource paper, but several of the most interesting biological conclusions remain speculative and are not independently validated.
This is particularly relevant for the proposed hybrid beta-barrel assemblies. While the known TOC75-TOC159 interaction provides a biologically established example, the predicted interactions of TOC159-family proteins with OEP24, OEP80 and TGD4 are more difficult to reconcile with the known functions of these proteins and currently lack experimental support.
A second conceptual issue concerns the newly identified BUL16 and BUL22 groups. The authors explicitly state that these proteins cannot confidently be assigned to the chloroplast outer envelope, yet they are subsequently analyzed together with established OEP families. This makes it difficult to distinguish between evidence for a beta-barrel fold and evidence for chloroplast outer-envelope localization.
eLife Assessment
This valuable study presents a computational model of meta-learning that bridges two known neural mechanisms of learning: through activation dynamics and through synaptic plasticity. The evidence that the model is a good match to data from a mouse perceptual decision-making task is convincing, but the evidence that it also matches data from a widely studied human multistep decision-making task is incomplete. This work will be of broad interest to neuroscientists and cognitive scientists studying reinforcement learning and decision making.
Reviewer #1 (Public review):
Maeda et al. propose an algorithm for meta reinforcement learning (Meta RL) that they term "hybrid deep reinforcement learning" (HDRL), which combines two distinct learning mechanisms: one that occurs through continual synaptic plasticity (via backpropagation-through-time), and the other that occurs through recurrent dynamics (i.e., in-context learning). The standard approach to Meta-RL, which I will refer to here as "RL2", is one that involves learning only through recurrent dynamics; this model has found recent application to understanding meta reinforcement learning in the brain. One downside of RL2 as a model of meta-learning in the brain is that, in learning only through recurrent dynamics, RL2 essentially throws the baby out with the bathwater by assuming synaptic plasticity does not have a role in moment-by-moment learning. This study therefore makes a valuable contribution in showing how both synaptic plasticity and recurrent dynamics can be combined in a single model. Importantly, this model does not require an explicit mechanism for mediating between these two learning mechanisms; rather, the authors show that the task (and meta-task) itself determines which mechanisms the model comes to rely on. They then show that their HDRL model can better explain differences in behavioral strategies across conditions of a perceptual decision-making task, and use in silico perturbations to establish that these differences in strategies are due to the distinct contributions of their model's two learning mechanisms. Finally, the authors provide evidence that the neural population activity of mice recording during each task has a consistent structure as the activity of artificial neurons in their HDRL model, but not to the standard RL2 model, which always relies on learning through recurrent activity. This study offers an important extension of the standard method for Meta RL, to one that is both more biologically plausible (i.e., because it seems unreasonable for plasticity to only occur periodically as it does in RL2) and more computationally interesting (i.e., as it explains how these two distinct learning strategies may be deployed automatically based on task demands).
The paper's conclusions seem fairly well supported by their analyses; however, there were a few details that need to be clarified.
(1) There are many architectural differences between the H-DRL vs RL2 model, which precludes a true comparison between the two models solely on the addition of Weight RL. The implementation of H-DRL is relatively stripped down compared to RL2, in terms of using Elman rather than LSTM, standard RL vs A2C, only positive activations allowed, and Q-learning instead of policy gradient. Because of these implementation details (none of which seem essential to the 'real' difference between these models), it is difficult to know exactly how these models differ. E.g., if the differences in behavior are in part due to the differences in architecture between HDRL and RL2 rather than to the differences in learning mechanisms.
(2) The conclusion that mouse behavior is more consistent with HDRL than with RL2 seems likely but is not sufficiently supported by the analyses. One thing I wondered was whether all model instances were truly converged; RL2 will of course take much longer than HDRL to learn since it is only updated per session. The authors' claim that mouse behavior is more like HDRL than RL2 seems to be based on the BIC comparison in Figure 2f, but insufficient detail is given to understand what this analysis is. Figure 2e, by comparison, seems to present all three models as similar (though this comparison is difficult to make given the different y-axes).
(3) The interpretation of the two-step task feels incomplete, and potentially in conflict with the interpretation of HDRL in the rest of the paper. When trained on the two-step task, HDRL is presented as reproducing a mixture of model-based and model-free behavior. What does the RL2 model do in this context? Later, on the perceptual decision-making task, HDRL seems to be presented as being model-based-like (and using recurrent activity) in one context and model-free-like (and using plasticity) in the other. Is it the case that in the two-step task, HDRL is simply doing both? As is, it's unclear to me how I should interpret the analyses related to the two-step task.
Reviewer #2 (Public review):
Strengths
The manuscript claims that standard methods for deep recurrent RL ("meta-RL"; Wang et al., 2017) provide a poor match to behavior data from humans and animals in many situations, including the two-step task and a mouse behavior dataset previously published by these authors. This negative story is solid and convincing.
The manuscript refines this model by introducing an online variant of deep recurrent RL. This is a good idea for many reasons. One compelling one, which the authors could choose to emphasize more, is that neurobiology does not respect a strict separation between "online" periods during which weights do not change and "offline" periods in which they do. Both weights and activities are changing all the time. Another is that the behavior literature provides abundant reason to believe in short-timescale innate reward-seeking mechanisms, which offline batch-based RL does not provide.
The manuscript shows that this online deep recurrent RL can provide a good match to their mouse behavior data. This involves a 2AFC perceptual decision-making task with strong temporal dependencies between the stimuli. In one condition, the stimulus on a particular trial is very likely to be identical to the one on the previous trial and only rarely switches; in another, the stimulus is very likely to differ, and only rarely repeats. Mice find the first condition much easier to learn, perhaps because of an innate win-stay/lose-switch bias which is helpful in the first but harmful in the second.
It also provides evidence that online deep recurrent RL can provide a good match to neural data from the OFC in this task. In the "repeat" condition, their model learns the task largely in weights, which is possible because the weight learning rule is hard-coded to be win-stay/lose-switch. In the "alternate" condition, the model must learn it in activities. Broadly consistent with this, mice in the "alternate" condition, but not the "repeat" condition, show strong task-based representations in the OFC.
Taken together, these results are a meaningful advance in applying the "NeuroAI" research program to the case of trial-by-trial reward learning.
Weaknesses
(1) The key innovation of this manuscript is replacing batch-based recurrent deep RL (which they call "meta RL"), in which weights are updated once per batch via backpropagation through time, with a version of online recurrent deep RL in which weights are updated after every trial.
It is difficult for readers to understand how the version of online deep RL being used works, how it compares to other possibilities, and why it was chosen. The manuscript describes it as truncated backprop through time (TBPTT, equation 9). It is not clear from the equation what the truncation window is. If it is only a single trial, then it is unclear how a network trained in this way manages to do the representation learning necessary to do anything model-based at all.
Various other techniques for real-time recurrent learning exist, many under the heading of "biologically plausible" deep learning. These include techniques involving synaptic eligibility traces (e.g., Murray et al., 2019; Bellec et al., 2020) and possibly others.
The manuscript calls this "Hybrid Deep RL". This name is confusing to me and may be confusing to readers. If this method is a "hybrid" between deep RL and some other thing, it's not clear to me what that other thing might be. It seems rather to be a variant of deep RL. I'd suggest a name that's clearer about that and more specific, perhaps "Online" or "Real-time" Deep RL, or similar. The manuscript also describes it in places as a "two-module" system, which is confusing to readers because it has one module: a single RNN with a single learning rule.
The manuscript would benefit greatly from a more careful discussion of these issues and a positioning of the model within the literature. What is lost when giving up BPTT? (long-timescale representation learning, which is often crucial for RL). What is gained? (learning rules that are local in time and space, and therefore more plausible as mechanistic models of biological learning). Where can we expect this method to fail? (tasks with temporal dependencies that are meaningfully longer than the TBPTT window). What alternatives exist? (eligibility trace methods, etc)
(2) The manuscript claims that their model "reproduced mixed strategies of humans in a two-step task". This portion of the paper is brief and unconvincing.
The key diagnostic plot, the trial-history regression in figure 1e, shows behavior that looks almost entirely model-free. There is little, if any, separation between the lines for "common" and "uncommon" within either of "rewarded" or "unrewarded". The bar plot does show an apparent difference, but these one-trial-back analyses are subject to many confounds.
Human studies involving this task typically involve each participant performing only a single session lasting ~100s of trials. When participants are run for multiple sessions, behavioral strategies change, becoming substantially more "model-based" (Economides et al., 2015). If I understand correctly, the model here is trained for 100 sessions of 100 trials, and the data we are shown comes only from the last 25. A fairer comparison would be to show only the first, or first few, of these sessions.
I imagine that H-DRL is unlikely to show much of a model-based pattern within the first session, as it will require averaging over many trials to learn to sculpt activity dynamics appropriately. By the same token, I am skeptical that H-DRL is plausible as a mechanistic account of the human two-step task - instead, I think it is much more likely that explicit working memory mechanisms are involved (e.g., Otto et al., 2013). H-DRL may be a better model of the two-step task as implemented in animals (Miller et al, 2017; Miranda et al., 2020; Akam et al., 2021).
The manuscript claims that human subjects on the two-step task typically use a mixture of both model-based and model-free strategies, and that neurobiology supports a division between the two. Both of these points are controversial, at least: da Silva and Hare (2020) present a compelling argument that humans are primarily model-free, and neural recordings dating back to the very first two-step task paper (Daw et al., 2011) show overlapping signals.
I do think it might be possible to address this with more careful writing and further simulations. But I do not think this is necessary. The rhetorical role of the two-step task in this manuscript seems really just to be illustrating that H-DRL can exhibit different online learning strategies in activities and in weights. This point is made far more convincingly by the analysis of the mouse perceptual decision-making task. I'd recommend simply dropping the two-step task, and its associated claims, from the manuscript.
Reviewer #3 (Public review):
Summary:
This paper proposes and tests a modified reinforcement learning model that can express different decision-making strategies which spontaneously emerge when performing a task and match real animal behavior. A recurrent neural network (RNN) is trained in the style of prior work on meta reinforcement learning (e.g. Wang et al. 2016, 2018), where an initial algorithm is used to train the network and a second algorithm emerges later from the recurrent dynamics of the network. These are termed weight-RL and recurrent-RL, respectively. By making several adjustments to the meta-RL training procedure, the authors find that their version (called hybrid deep reinforcement learning, or H-DRL) shows a better match to real animal behavior and orbitofrontal cortical activity dynamics. Overall, this paper is an improvement on the prior meta-RL model, but it is unclear what neuroscientists should take away from this.
Strengths:
(1) The authors introduce a model, H-DRL, whose main strength is that it can better approximate animal behavior and neural activity on specific tasks compared to its predecessor, meta-RL. This could potentially be important for interpreting the algorithms underlying animal learning.
(2) The changes that have been made from meta-RL are not overly complex.
(3) The model is assessed using two types of decision-making tasks that are similar to many of the tasks used in rodent behavioral and systems neuroscience.
(4) In addition to behavioral data, RNN unit activity is compared directly with single unit orbitofrontal cortical activity. This makes the model appealing for neuroscience applications.
Weaknesses:
(1) In the introduction, the authors mention that other extensions of meta-RL have been tried. How does H-DRL compare with these other modifications?
(2) The authors try to convey the impact of their model in the introduction by stating that, unlike meta-RL, "it embodies a fundamentally different theoretical claim in the context of neuroscience". It is not clear what the fundamentally different theoretical claim is. While meta-RL implements weight updates over longer timescales, H-DRL implements weight updates after each trial. Is this the fundamentally different theoretical claim? If so, that does not seem like a novel concept for behavioral and systems neuroscience.
(3) It is not clear what inspired the specific changes to the meta-RL training procedure. Were they decisions based on incremental trial and error to intentionally reproduce the empirical behavioral and neural results, or were they based on prior logic?
(4) Regarding the inconsistency between OFC neurons and H-DRL in previous outcome encoding (Figure 5c), it would be helpful to speculate on why this happens because it seems to be the only place where H-DRL fails.
(5) Figure 5e seems tangential to the overall message of the paper. The paper is about H-DRL as a superior approach to modelling real behavior and neural activity compared to meta-RL. But this panel, which is the last panel in the paper, is not directly relevant because there is no comparison with the model.
eLife Assessment
This important study provides an accessible comparative atlas of gene activity in sensory hair cells across four vertebrate species. The evidence is convincing, supported by comparative analyses and experimental validation, although technical differences between datasets and unclear definitions of cell maturity limit some interpretations. The resource will be of interest to researchers studying hearing, balance, sensory cell development and regeneration, and evolutionary biology.
Reviewer #1 (Public review):
Summary:
"A comparative single-cell transcriptomic atlas for diverse populations of vertebrate hair cells" by Basu et al. aims to define pan-hair cell enriched (HCE) genes across hair cell subtypes from four vertebrate species through integration of published scRNA-seq, SMART-seq, and single-nucleus RNA-seq datasets. Pan-HCE genes are weighted using AUC values, where AUC is used to identify genes enriched in hair cells relative to supporting cells. The resulting dataset is made publicly available, allowing researchers to compare gene expression across species. The dataset integration and creation of a public resource are clear strengths for the inner ear community.
Strengths:
This manuscript presents a valuable cross-species resource and a substantial integration effort.
Weaknesses:
Weaknesses addressed by the authors in the text include the lack of human cochlear scRNA-seq comparisons to mouse/chicken cochlear hair cells. While the rationale that the datasets were not available at the time of investigation is acceptable, the use of human cochlear datasets would be the ideal experimental comparison to mouse and chicken cochlear hair cells.
Reviewer #2 (Public review):
In this manuscript, the authors investigate transcriptional diversity and similarities between mechanosensory hair cells across different species to provide an atlas of hair-cell-enriched genes.
Their analysis encompasses publicly available and a few newly generated single-cell RNA-seq data from human, mouse, chick and zebrafish, in total comprising 43 populations of hair cells from different organs (cochlea, vestibular sense organs, lateral line) and different stages. They first compare the transcriptome of each hair cell population with that of supporting cells from the same data to find hair-cell-enriched genes. To do this, they use a score (AUC) that reflects how well a gene separates both cell populations; this approach identifies about 5000 genes enriched in at least one of 29 mechanosensory populations. Grouping these genes by their enrichment pattern gives 22 clusters of co-regulated genes, many species- or organ-specific and a few shared by most hair cells.
Compared with previously published lists of hair cell genes, they define 884 "pan-HCE" genes enriched in most hair cells across species. This set is enriched for stereocilia and deafness genes. In situ hybridization is presented to support enrichment for 7 of 10 tested genes.
Adding 14 immature populations splits the list into early-and-sustained, transient and late genes. The authors then use these to assess the maturity of hair cells from cochlear organoids, regenerating chicken basilar papilla, reprogrammed fibroblasts, and developing zebrafish, mouse and human tissue. All data are released through a Shiny app and the gEAR portal.
Overall, this is a useful resource for the community, presented in an exploitable format. It is timely; the inclusion of many datasets is very valuable, and the comparison of HCs vs SCs from each organ is a good approach. I have some concerns regarding the clarity in describing the analysis, whether some findings could reflect differences between datasets, and how cell 'maturity' is defined.
Major comments:
(1) The data were generated by different labs, different methods and sequencing platforms. Hierarchical clustering of the data splits cells mainly by species. It is not clear to me that the authors have ruled out that this indeed reflects biology or could be due to technical differences. This needs to be discussed. Is it possible to show that the results hold for data generated with the same platform?
(2) The calculation of enrichment scores needs clarification. In the methods, it says that AUC values below 0.5 were set to 0.5 and then supporting cell-enriched genes were set to 1 − AUC. How do supporting cell-enriched genes like ID2 end up with values like 0.11 (Figure 2C)? Please describe the calculation step by step; how were skipped genes because of low expression handled?
(3) It is not clear how duplicated genes were handled when assigning orthologues; the approach allows one-to-many. How were they combined into a single score? The results say that each human gene was linked separately to each ortholog?
(4) The reference gene list for known hair cell genes seems to overlap with genes reported to be enriched; this seems to be a circular argument. Among the hand-curated list are stereocilia-associated genes, which are then reported to be enriched. Is it possible to test enrichment using gene sets that were not used to choose the threshold?
(5) The authors need to define how they classify cells as mature, maturing, and immature - there are no definitions at all. Provide these and appropriate references. Mouse P7 cells are classified as mature, which they are not. Later data depend on this classification, so it needs to be justified properly.
(6) The authors assess maturity of hair cells in organoids; it seems that this depends on the % of genes in each set (mature, maturing, immature) that pass the cutoff. Has this been compared to a random or control set?
(7) Verification of hair cell enrichment by HCR/RNAscope: This verification is not very convincing for most genes tested. 30% of the genes tested fail, while genes described to be HC-specific are also expressed in surrounding cells. E.g. Figure 5C: both Apba1 and Pcsk5 seem to be expressed in hair cells as well as the surrounding cells; Figure 5D: both genes are not expressed in all hair cells. Figure 5 supplement 2B: the images of the organ of Corti are too low resolution and magnification to identify hair cells reliably. It is not possible to see if these genes are indeed expressed in hair cells. If the authors claim HC enrichment, they need to provide quantification of expression levels e.g. HCs vs SCs and show negative controls.
(8) A new single-cell dataset from human embryos has just become available; van der Valk et al., 1026, Nat. Neuroscience. This dataset fills an important gap in their current analysis, and it would be very good to integrate it.
(9) The figure legends are often a bit short and do not explain the figures well; please review and expand.
Reviewer #3 (Public review):
Summary:
The authors generated a cross-species single-cell transcriptomic atlas of inner ear and lateral line organs across four species (human, mouse, chicken, zebrafish) and 11 developmental stages. By comparing mechanosensitive hair cells to surrounding supporting cells, they identified hair cell-enriched (HCE) genes, conserved transcriptional programs, and stratified developmental markers. They used these core and temporally stratified gene sets to evaluate hair cell identity and maturation state in models of avian regeneration, mouse cochlear organoids, and transcription factor-reprogrammed fibroblasts. A publicly available web-based resource (HRP Hair Cell Gene Explorer and gEAR portal) was created to facilitate field-wide data exploration.
Strengths:
(1) Comprehensive Multi-Species Benchmark:
Cross-species analysis spanning 43 hair cell populations and 11 developmental stages establishes a rigorous reference dataset for benchmarking hair cell identity across development, cell biology, and regenerative models.
(2) Biological Relevance:
The pan-HCE gene set captures 25% of all known human deafness genes, confirming its strong predictive value for hair cell-specific function and disease mechanisms.
(3) Methodological Rigor & Robust Thresholding:
The use of area under the receiver operating characteristic curve (AUC) provides a platform-independent metric. Testing four calculation strategies (combining 29 mechanosensitive vs. 23 mature populations with unweighted [ES1] vs. species-balanced [ES2] weighting) and taking their intersection ensures robust gene selection without species bias.
(4) Temporal Stratification & Application:
Categorizing pan-HCE genes into early, sustained (ES-HCE), late (L-HCE), and transiently enriched (T-HCE) sets enables tracking of maturation state across different animal and cellular models, proving effective in evaluating regenerating avian hair cells, mouse cochlear organoids (Lgr5^+ lineage), and Gfi1/Atoh1/Pou4f3/Six1-reprogrammed fibroblasts.
(5) Experimental Validation:
Key pan-HCE genes were independently validated across species using spatial transcriptomics, RNAscope, and Hybridization Chain Reaction (HCR).
(6) Community Utility:
Broad accessibility via an interactive ShinyApp and gEAR portal integration facilitates customized field-wide analysis.
Weaknesses:
Weaknesses were considered minor.
(1) Interspecies Homology Resolution:
Machine learning-based clustering (MetaNeighbor) showed that factors such as age, species, and sensory modality often override subtype identity, failing to resolve fine-grained cross-species homologies for specific subtypes (e.g., distinguishing mouse inner vs. outer hair cells relative to chicken tall vs. short hair cells).
(2) Species Imbalance & Mapping Artifacts:
Despite weighting schemes (ES2), teleost genome duplications (1-to-many ortholog mapping in zebrafish) complicate cross-species signal normalization. This remains an overarching issue in many fields when performing cross-species analysis.
(3) Different mechanisms of regeneration:
Converted fibroblasts induced only a percentage of pan-HCE genes (28.7%) and very few transient genes (9.3%). This could reflect incomplete programming towards a fully mature native hair cell state, as the authors suggest. It is also possible that direct programming via transcription factor expression or Lgr5+ cell conversion could employ mechanisms distinct from those used in normal development or in regenerative species.
eLife Assessment
This important study presents an efficient approach to storing and retrieving memories in large language models based on complementary learning systems theory, and highlights the gap between storing information and using it flexibly. The presented evidence for the efficacy of the technical implementation of the memory system is convincing. However, support for the proposed parallels with certain aspects of complementary learning systems theory and with biological memory systems more generally remains incomplete. This work will be of broad interest to cognitive and computational neuroscientists studying memory, as well as researchers developing memory systems for artificial intelligence.
Reviewer #1 (Public review):
Summary:
This study develops a large-language-model implementation of the complementary learning system (CLS) framework for continual episodic learning. The authors employ a pretrained open-weights model to serve as a neocortex-like semantic knowledge store and introduce a novel hippocampus-like memory system in which each episode is assigned its own sparse rank-2 masked LoRA adapter, eliminating interference at encoding. A separate semantic embedding model provides a retrieval key for each episode, and winner-take-all routing selects the adapter whose stored embedding best matches an incoming cue. The reported storage efficiency of approximately 2-3 trainable parameters per token suggests that this one-adapter-per-episode scheme could scale to large numbers of memories, although its total storage requirements also include the embeddings used for content-addressable retrieval.
A key finding is that successful encoding of an episode does not guarantee flexible access to its contents. Although a correctly selected adapter can reconstruct its stored episode, the model performs poorly when asked to answer questions about that episode directly. The iRAG procedure addresses this by inserting the reconstruction into the base model's context. Consolidation through replay of the original passages enables reconstruction without adapters, but direct question answering remains substantially less successful, even with richer training objectives. Overall, the paper provides a technically impressive implementation of episodic memory in LLMs. However, several aspects of its learning and consolidation procedures significantly limit its correspondence to CLS, as discussed below.
Strengths:
A major strength of this work is the concrete parametric implementation of a CLS-like division of labor within a modern transformer network. The episodic memory system incorporates key properties of the biological hippocampus, including pattern-separated encoding and content-addressable retrieval. The interaction between the selected adapter and the LLM provides an appealing functional analogue of hippocampal-cortical reinstatement, in which activation of a sparse episodic trace biases distributed cortical processing so as to reconstruct the mnemonic content of an episode.
A second strength is that the model cleanly separates interference during episodic storage from interference during retrieval. By assigning memories to non-overlapping parameter traces, individual episodes remain faithfully stored even at large memory loads. Performance declines for more strongly overlapping memories can therefore be attributed largely to competition during retrieval. This distinction shows that successful pattern separation does not by itself guarantee reliable access to stored memories.
A final strength is the careful optimization and evaluation of the episodic memory system. The authors introduce a masked LoRA that substantially reduces parameter cost relative to standard LoRA approaches, systematically identify the most parameter-efficient placement of these adapters, and show that storage cost scales approximately linearly with episode length, with similar results across two base models. They also test the framework on two datasets with quite different memory structures: largely independent, topically diverse Wikipedia passages and more strongly correlated autobiographical episodes with recurring people, places, and themes.
Weaknesses:
(1) The model implements the two learning regimes motivated by CLS: pattern-separated episodic learning and slower learning in shared weights. However, it does not implement the central proposed interaction between them, whereby hippocampal traces reinstate experiences that serve as training events for neocortical consolidation during offline periods (e.g., Káli & Dayan 2004, Nat. Neuro.; Singh et al. 2022, PNAS; Sun et al. 2023, Nat. Neuro). Instead, consolidation here relies on replaying the retained ground-truth passages and therefore does not depend on the episodic memory system at all. The authors could either demonstrate consolidation driven by replay from the episodic adapters or more explicitly discuss this simplification and its implications for the CLS comparison.
(2) The authors characterize the episodic system as implementing "fast learning" or "rapid storage". However, encoding involves repeated optimization on the same passage, with up to 1,000 epochs per attempt and potentially multiple retraining attempts. This raises questions about its correspondence to the rapid, single-shot learning that characterizes the hippocampal component of CLS and is supported by evidence in the trisynaptic pathway (e.g., Nakazawa et al. 2003, Neuron; Nakashiba et al. 2008, Science). The authors should (a) report the distribution of optimization steps, retraining attempts, and mask sizes required for successful encoding on the two datasets and (b) clarify this distinction when relating the model to biological hippocampal learning. They could also examine whether larger adapters reduce the optimization required, clarifying the trade-off between storage efficiency and encoding speed.
(3) Relatedly, it is unclear whether there is meaningful variability in the number of optimization steps required across episodes and whether characteristics of the passages explain this variability. For example, do passages containing less predictable content require more training and/or larger adapters? One way to examine this would be to consider the frozen base model's pre-learning perplexity for each passage and test whether this predicts optimization steps to criterion or final parameter count, controlling for passage length. Such an analysis could help clarify what determines the difficulty and resource cost of forming a new episodic memory.
(4) The consolidation results raise concerns about the model's capacity for stable lifelong learning, as incorporating new memories comes at a cost to existing knowledge. The presented consolidation for 1000 memories produces substantial degradation of the base model's existing capabilities, particularly under the conditions that most improve direct access to consolidated knowledge. In a lifelong setting, subsequent consolidation would presumably continue to cause interference not only for pre-existing semantic knowledge but also for memories incorporated during earlier consolidation cycles. The alternation between NREM and REM sleep has been proposed as one biologically motivated solution to this problem, with NREM supporting integration of recent hippocampal memories into cortex and REM helping to reactivate and protect pre-existing cortical knowledge (Norman et al. 2005, Neural Networks; Singh et al. 2022, PNAS). The authors already note that a lifelong system would need to interleave encoding and consolidation and determine which memories to replay; it would also be valuable to discuss how previously consolidated knowledge and broader base-model capabilities could be protected across repeated cycles of consolidation.
(5) It appears that the consolidation conditions reported in Fig. 9 might differ substantially in training exposure. As described in Supplementary Note 5, "passage-only" uses one example per episode, "paraphrase augmentation" uses ten, and the "knowledge instruction" and "masked reconstruction" conditions add many further training examples. Because training is specified in epochs, could the authors clarify whether these conditions are matched for optimizer steps or training tokens? If not, it would be helpful to determine how much of the improvement in direct QA and accompanying loss of base-model capabilities reflects the specific objectives versus greater training exposure. A step- or token-matched control would help disentangle these possibilities.
(6) The reported storage efficiency of approximately 2-3 parameters per token appears to refer only to the masked LoRA adapters, despite each episode also requiring a stored 1,024-dimensional embedding for routing. The authors appropriately note that this index is part of the memory system's storage cost and is in fact substantially larger than the adapter store. It would therefore be useful to incorporate the routing embeddings into the reported per-token storage estimate, or otherwise report the total storage cost per episode/token alongside the adapter-only figure.
Reviewer #2 (Public review):
Summary:
The paper asks whether the classic complementary learning system (CLS) can be instantiated in an LLM.
The authors first show that continual fine-tuning and three regularization remedies (L2, EWC, on-policy distillation) fail to encode new episodes without memory interference. They then propose to use gated LoRA instead: one dedicated low-rank adapter per episode for encoding, with winner-take-all selection among adapters based on embedding similarity for retrieval, and finally replays stored episodes in batches to fine-tune the base weights for consolidation.
The evaluation covers storage capacity, end-to-end recall at up to 1,000 episodes, question answering (via a two-stage "internal RAG" and via direct parametric access), and consolidation, on two datasets: Wikipedia paragraphs and an autobiographical dataset, and two LLMs: Llama-8B and Qwen-8B.
Strengths:
(1) The attempt to instantiate a CLS in an LLM is very interesting, timely, and potentially useful for AI.
(2) The use of LoRA as the episodic encoding mechanism is efficient: they showed that LLMs can store 1,000 episodes with a small amount of parameters (< 0.01% relative to the base model).
(3) The evaluation is careful. The negative results on regularization are informative. An LLM judge is validated against other metrics, with very informative failure examples (e.g., the "Man on the Run" case, where teacher-forced accuracy is 0.91 but every fact is fabricated).
Weaknesses:
I think the main thing is that the two systems, at this point, do not richly interact:
(1) The episodic component is not meaningfully involved in consolidation. In this model, consolidation is batch re-training with replay, not transfer. The consolidated model is trained from the raw passages - the authors state that it "relearns the same episodes into the shared base weights by batch replay of the original passages, rather than transferring parameters from the episodic adapters."
(2) The influence of semantic memory on episodic encoding is not deeply investigated. For example, empirically, schema-consistent information is acquired rapidly (Tse et al., 2007). It would be informative to understand if this model shows a similar effect. Overall, showing that this method works for a pre-trained LLM is a strength, but it would be very exciting to know how that impacts encoding.
(3) This might be out of scope, but I still wonder, eventually, how this kind of system would decide when to use episodic vs semantic memory. This is relevant to the author's points that the low-rank adaptor can be freed after consolidation, which is true, but to do that autonomously, the system needs to know whether a memory is fully consolidated. Overall, I think this would be an interesting future direction.
Reviewer #3 (Public review):
Summary:
In this manuscript, the authors implement a complementary learning system within a pretrained LLM. Each new episode is stored in its own small adapter, a routing step selects the right adapter at recall, and a slower consolidation stage transfers the stored episodes into the base weights. The main contributions are a demonstration that this fast store retains a thousand episodes without the catastrophic forgetting seen under vanilla sequential fine-tuning, a retrieval scheme in which a stored episode is regenerated into the context window so the model can answer questions about it, and a consolidation procedure that transfers the episodes into shared weights while leaving the model's general capabilities largely intact. The authors also show that neither the adapters nor the consolidated weights support direct question answering on their own, which separates storing a memory from being able to access it.
Strengths:
The paper is technically sound and, to my knowledge, one of the first efforts to make the connection between complementary learning systems and the LLM plus adapter configuration explicit and then build it end to end. It is very clearly written and honest about negative results. The authors have built a system in which the fast and slow systems and the transfer between them are all explicit and measurable, on top of semantically rich pretrained representations, which is something memory research has underutilized in the LLM era.
Weaknesses:
My main comment is that the paper should be positioned more explicitly. The framing should be clearer if the paper is more of an ML contribution or a neuroscience contribution (it is both, but there should be an emphasis). Routed adapters/loras are already popular, and many studies have addressed continual learning in the ML literature (gated LoRA and related methods), and the authors should be upfront about this in the introduction rather than only in the discussion to help readers understand the framing earlier. Previous studies on adapters do not diminish the paper, because in my opinion its value is the neuroscience framing, and that is where I think the paper has not explored sufficiently. The classic phenomena that CLS was proposed to explain are cited but not directly tested, which I think is a missed opportunity, since most of them can be tested with materials already in the paper. Below are my main suggestions:
The original 95 CLS paper proposes a slow cortical system because integrated knowledge supports inferences/generalizations that the episodic memory system cannot, and people need time and often sleep to form such generalizations. The paper's adapter loads one episode at a time, but the consolidated model might be able to extract the core rules or schemas of the episodic memories. If consolidation can support inference/generalizations that retrieval cannot, that is the strongest possible argument for the two-system design.
Additionally, in the original CLS proposal, replay is interleaved so that new memories are transferred into cortex alongside old knowledge. Here, consolidation trains on the new episodes only, and the general capability of the LLM degrades accordingly. The effect of rehearsal or pseudo-rehearsal, and the interference with existing knowledge close to the consolidated content, are not measured.
In human memory, repeated visits to the same scenario are a core aspect of episodic memory: you visit your office many times a day with slight changes in context, and episodic memory has to keep track of these visits while knowing which differences should be memorized. In the paper's design, a repeated episode would create a second adapter with a near-identical embedding, and a slightly paraphrased episode would become a new memory. Human memory studies show memory strengthens or updates the existing trace when revisitation happens, as in reconsolidation. The current design also rules out any overlap between adapters, giving a hippocampal module that does pattern separation but not the "pattern completion" half.
Two well-established findings in the consolidation literature are worth examining: consolidated memories keep the gist and lose specific details (Winocur and Moscovitch), and new information consolidates much faster when it fits something already known (Tse et al. 2007's schema study in rodents). The paper's judge scores each character episode on separate details, and the supplement contains an example in which a question about one episode makes the consolidated model reconstruct a different episode from years earlier with the same theme, which resembles a classic "false memory". The current setup in the paper could examine these false memory structures in a much more nuanced way.
I believe experiments tied to the memory, psychology, and cognitive science literature along these lines would greatly strengthen the study, and could make this paper a "template" for how the field can study effects that have been observed in humans but have lacked a concrete and powerful enough modern modeling system.
eLife Assessment
Loss of POLQ function is synthetically lethal with BRCA1 or BRCA2 deficiency, as is pharmacological inhibition of its motor (ATPase) or polymerase activity. This important manuscript validates an inhibitor of the DNA polymerase theta (POLθ) motor domain in biochemical and cell-based assays and provides evidence of its efficacy against homologous recombination-deficient tumors in preclinical and clinical settings, particularly in combination with PARP inhibitors. The experiments are rigorous, the evidence is convincing, and the results are appropriately interpreted. This work will be of considerable interest to the ovarian cancer research community.
Reviewer #1 (Public review):
Summary:
This study describes targeting Poltheta Helicase with a new drug RP-3467. The properties of the drug and a clinical response in a gBRCA1 Ovarian Cancer Patient with a mutation in TP53BP1 are described.
Strengths:
This is an important and interesting set of observations.
Weaknesses:
(1) The response of the patient with a TP53BP1 mutant tumor is of interest but requires more clarification before solid conclusions can be drawn. The presence of the TP53BP1 mutation after but not before olaparib treatment is suggestive of selection, but in the absence of demonstration of a biallelic loss, this is conjecture. It could be a co-selected passenger mutation. The authors should report the TP53BP1 variant allele fraction, LOH, and orthogonal validation should be performed by targeted deep sequencing or ddPCR. This would establish the prevalence before and after selection and enhance the interpretation of the functional effects of the mutation observed.
(2) To further strengthen the proposed mechanism, the authors should consider knocking in the mutation (or a similar LOF mutation) into an appropriate cell line and establish resistance to olaparib and sensitivity to the combination, as well as each drug individually. This will address the latter question of whether it is the poltheta inhibitor or the combo that is responsible for the clinical response.
(3) In the animal experiments, "synergy" is overstating the situation. Without an interaction analysis, this is an enhanced combination activity. Also, several of the animal experiments have a small N. It would be better to show the growth of tumors in individual mice.
(4) The authors propose that Poltheta inhibition could prevent the emergence of reversion mutations. This is not addressed in the paper, and if they want to suggest this, they should provide some evidence.
Reviewer #2 (Public review):
Summary:
The authors set out to assess the in vitro potency of a new POLQ helicase inhibitor, assess its in vitro BRCA1/2 synthetic lethal properties, and confirm how it performs in vivo in PDX and cell line xenografts. The authors also show an anecdotal patient response with a PARP inhibitor + POLQ inhibitor combination.
Strengths:
This is a novel POLQi of the helicase activity. The anecdotal clinical response in a heavily pre-treated patient is interesting.
Weaknesses:
Most of these issues could be addressed by changes to the main text, i.e. by the authors describing limitations/caveats to interpretation, etc.
(1) The clinical anecdote is interesting but underanalyzed. It's difficult, though, to know one way or the other whether the PARPi+ POLQi treatment effect (only partial response) has anything to do with the 53bp1 mutation. There is a causative sign that the 53bp1 mutation could be the cause of PARPi resistance (the increase in 53bp1 mutation allele frequency in PARPi-naïve vs PARPi-resistant tumors), but no similar sign to show that the 53bp1 mutant tumor clone is targeted by POLQi treatment. For example, does 53bp1 mutation allele frequency fall in ascites or ctDNA during the course of treatment, or is the 53bp1 mutation lost in a tumor taken at the end of treatment?
(2) Although the relevant work is cited, the authors should be more explicit in the introduction and state that there is already evidence from 2 papers that Combined loss of BRCA1 and 53bp1 causes enhanced sensitivity to small-molecule POLQ helicase or polymerase inhibitors.
(3) Figure 4. Not exacerbating existing PARPi myelosuppression is great, but the authors have to acknowledge that the myelosuppression is still present when POLQ+ PARPI is used. Single-agent POLQi doesn't give myelotoxicity, but it also does not give strong anti-tumor efficacy. i.e. it would be fair for the authors to acknowledge this.
(4) The increase in reticulocytes when mice are treated with POLQi could be a sign that there is an effect on myeloid lineages, but the animals are responding by increased production of this young myeloid lineage. Alternatively, this reticulocyte increase could drive or be a sign of pathology (polycythaemia vera, etc.). It would be good to see a commentary on this in the Discussion.
(5) Re Figure 4. A therapeutic advance would be to see that, compared to other PARPi combinations such as platinum + PARPi or temozolomide + PARPi, a POLQ + PARPi combination doesn't give the same myelosuppression that these combinations do (by analogy, combined platinum + PARPi gives curative responses in mice but also wipes out the adaptive immune system). The authors need to add data showing the platinum+ PARPi positive control, or at the very least add a limitations of the study section to the manuscript to explain this fairly.
(6) Figure 2 might give the impression that the new POLQi gives a similar, if not greater, BRCA2 synthetic lethal effect than a PARPi (e.g. 2d-f). It would help readers to see the full saruparib and olaparib dose-response curves, not (as per 2 A, B, C) just the single point estimates.
(7) Why was DOTC42510 not profiled for PARPi?
(8) Figure 2f shows error bars on an n=2 dataset - one cannot estimate variance from only an n=2 dataset range.
(9) Is there a statistical analysis on Figure 2G?
(10) Figure 3. Olaparib 25 mg/kg QD is well below the MTD in mice. This makes the olaparib response in HCT116 BRCA2 k/o xenografts look modest when compared to the combinations.
Reviewer #3 (Public review):
Summary:
The authors seek to validate the effectiveness of an inhibitor of the Pol theta helicase-like domain ATPase activity in its ability to ablate MMEJ activity in cell line models. They then confirm the inhibitor is synthetic lethal in a variety of BRCA1/2 deficient cancer models (and one PALB2 model). Finally, they report apparent effectiveness in treatment of a single patient that failed therapy with PARPi due to 53BP1 deficiency, a hypothesized special use-case for targeting pol theta.
Strengths:
It is a clear bench-to-bedside translation of mechanistic-oriented basic science to a clinical therapy. The experiments are largely rigorously performed. The approaches employed are impressively diverse, from in vitro to cell-line model, to preclinical model, to clinical application. The interpretations made are well supported by the available data.
Weaknesses:
The addition of a few missing details would improve the usefulness to the reader. Additionally, the rationale behind some approaches could be clarified.
Author response:
Public Reviews:
Reviewer #1 (Public review):
Summary:
This study describes targeting Poltheta Helicase with a new drug RP-3467. The properties of the drug and a clinical response in a gBRCA1 Ovarian Cancer Patient with a mutation in TP53BP1 are described.
Strengths:
This is an important and interesting set of observations.
We thank the reviewer for their positive assessment and appreciate their recognition of the significance and novelty of our findings.
Weaknesses:
(1) The response of the patient with a TP53BP1 mutant tumor is of interest but requires more clarification before solid conclusions can be drawn. The presence of the TP53BP1 mutation after but not before olaparib treatment is suggestive of selection, but in the absence of demonstration of a biallelic loss, this is conjecture. It could be a co-selected passenger mutation. The authors should report the TP53BP1 variant allele fraction, LOH, and orthogonal validation should be performed by targeted deep sequencing or ddPCR. This would establish the prevalence before and after selection and enhance the interpretation of the functional effects of the mutation observed.
We thank the reviewer for this important comment. We agree that identifying a TP53BP1 mutation after olaparib treatment does not, by itself, establish whether the mutation was directly selected under therapeutic pressure or represents a co-selected event, and we do not intend to make that distinction. Instead, we report that whole-genome sequencing (WGS) identified the TP53BP1 mutation in the tumor collected at the second debulking surgery. Importantly, an additional biopsy obtained at disease progression during the POLAR trial revealed a TP53BP1 variant allele frequency of 77%, with FACETS analysis identifying copy-neutral loss of heterozygosity (CN-LOH). These findings (new data in the paper) provide additional evidence consistent with biallelic inactivation of TP53BP1 and support the biological relevance of this alteration. Nevertheless, we acknowledge that these observations do not establish whether TP53BP1 loss directly drove acquired olaparib resistance, and we clarified this distinction in the revised manuscript.
(2) To further strengthen the proposed mechanism, the authors should consider knocking in the mutation (or a similar LOF mutation) into an appropriate cell line and establish resistance to olaparib and sensitivity to the combination, as well as each drug individually. This will address the latter question of whether it is the poltheta inhibitor or the combo that is responsible for the clinical response.
We thank the reviewer for this thoughtful suggestion and agree that establishing the contribution of TP53BP1 loss to treatment sensitivity is important for interpreting the clinical response. We would like to highlight, however, that this genetic relationship has been investigated in several published studies. Genetic depletion of POLQ has been shown to be synthetic lethal with 53BP1 or Ku70 deficiency (Wyatt et al., Molecular Cell, 2016; Feng et al., Nature Communications, 2019). More directly, Zatreanu et al. (Nature Communications, 2021) demonstrated that loss of 53BP1 or Shieldin components (SHLD1/SHLD3) in BRCA1-deficient cells confers resistance to olaparib while retaining sensitivity to the POLθ inhibitor ART558 as a single agent. These findings are further supported by Krais et al. and Ronson et al. (Nature Communications, 2023), who demonstrated POLθ dependency in BRCA1/53BP1-deficient backgrounds. Collectively, these studies provide strong genetic and pharmacological support for our proposed mechanism and suggest that POLθ inhibition itself may have contributed substantially to the observed clinical response. While we acknowledge that the relative contributions of RP-3467 and olaparib cannot be definitively distinguished in this patient, we believe that the existing literature provides a strong mechanistic foundation for interpreting this clinical observation, without necessitating additional genetic modeling in the context of the present study.
(3) In the animal experiments, "synergy" is overstating the situation. Without an interaction analysis, this is an enhanced combination activity. Also, several of the animal experiments have a small N. It would be better to show the growth of tumors in individual mice.
Although we believe that minimally in BRCA2-deficient CDX models (e.g. Fig. 2a,b) the term synergy is appropriate (no/minimal single-agent activity of RP-3467 and tumor regressions in combination), we replaced the term ‘synergy’ with ‘enhanced combination activity’ due to lack of a formal interaction analysis. Data from individual animals are included as Source Data.
(4) The authors propose that Poltheta inhibition could prevent the emergence of reversion mutations. This is not addressed in the paper, and if they want to suggest this, they should provide some evidence
We thank the reviewer and agree that the manuscript presents no data bearing on reversion mutations. The discussion should not have implied otherwise. We have rewritten relevant paragraph so that the idea is presented explicitly as an untested hypothesis arising from the microhomology signatures reported at reversion breakpoints by others.
Reviewer #2 (Public review):
Summary:
The authors set out to assess the in vitro potency of a new POLQ helicase inhibitor, assess its in vitro BRCA1/2 synthetic lethal properties, and confirm how it performs in vivo in PDX and cell line xenografts. The authors also show an anecdotal patient response with a PARP inhibitor + POLQ inhibitor combination.
Strengths:
This is a novel POLQi of the helicase activity. The anecdotal clinical response in a heavily pre-treated patient is interesting.
We thank the reviewer for their positive assessment and are pleased that they recognize the novelty of targeting the POLθ helicase activity and the potential clinical significance of the response observed in a heavily pretreated patient.
Weaknesses:
Most of these issues could be addressed by changes to the main text, i.e. by the authors describing limitations/caveats to interpretation, etc.
(1) The clinical anecdote is interesting but underanalyzed. It's difficult, though, to know one way or the other whether the PARPi+ POLQi treatment effect (only partial response) has anything to do with the 53bp1 mutation. There is a causative sign that the 53bp1 mutation could be the cause of PARPi resistance (the increase in 53bp1 mutation allele frequency in PARPi-naïve vs PARPi-resistant tumors), but no similar sign to show that the 53bp1 mutant tumor clone is targeted by POLQi treatment. For example, does 53bp1 mutation allele frequency fall in ascites or ctDNA during the course of treatment, or is the 53bp1 mutation lost in a tumor taken at the end of treatment?
We thank the reviewer for this suggestion. While longitudinal monitoring of the TP53BP1-mutant clone would be informative, such data are unavailable. A biopsy obtained at progression on the POLAR trial revealed a TP53BP1 variant allele frequency of 77%, with copy-neutral LOH by FACETS analysis. Importantly, extensive preclinical evidence establishes synthetic lethality between 53BP1 deficiency and POLQ inhibition (Wyatt et al., 2016; Feng et al., 2019; Zatreanu et al., 2021; Krais et al., 2023), supporting a contribution of POLθ inhibition to the clinical response. Nevertheless, determining whether the response was driven by POLθ inhibition alone or its combination with olaparib will require evaluation in a larger clinical cohort.
(2) Although the relevant work is cited, the authors should be more explicit in the introduction and state that there is already evidence from 2 papers that Combined loss of BRCA1 and 53bp1 causes enhanced sensitivity to small-molecule POLQ helicase or polymerase inhibitors.
Agreed, and adopted. We changed the introduction to state this prior evidence explicitly on first mention, naming both studies and specifying that the effect has been demonstrated for inhibitors of Polymerase theta.
(3) Figure 4. Not exacerbating existing PARPi myelosuppression is great, but the authors have to acknowledge that the myelosuppression is still present when POLQ+ PARPI is used. Single-agent POLQi doesn't give myelotoxicity, but it also does not give strong anti-tumor efficacy. i.e. it would be fair for the authors to acknowledge this.
We agree with the reviewer and have revised the text to more accurately reflect these findings. While POLθ inhibition does not exacerbate PARPi-associated myelosuppression, it does not eliminate the hematologic toxicity of PARPi treatment. We also acknowledge that POLθ inhibitor monotherapy has limited antitumor efficacy in the models tested, highlighting the importance of combination treatment and the need to balance efficacy with tolerability.
(4) The increase in reticulocytes when mice are treated with POLQi could be a sign that there is an effect on myeloid lineages, but the animals are responding by increased production of this young myeloid lineage. Alternatively, this reticulocyte increase could drive or be a sign of pathology (polycythaemia vera, etc.). It would be good to see a commentary on this in the Discussion.
We agree, and updated the language in Results, page 9, to reflect this possibility.
(5) Re Figure 4. A therapeutic advance would be to see that, compared to other PARPi combinations such as platinum + PARPi or temozolomide + PARPi, a POLQ + PARPi combination doesn't give the same myelosuppression that these combinations do (by analogy, combined platinum + PARPi gives curative responses in mice but also wipes out the adaptive immune system). The authors need to add data showing the platinum+ PARPi positive control, or at the very least add a limitations of the study section to the manuscript to explain this fairly.
We thank the reviewer for this suggestion. The substantial myelosuppression associated with combining PARP inhibitors with platinum-based chemotherapy or temozolomide has been well documented in preclinical and clinical studies, including the references cited by the reviewer (Rajan et al., 2012; PMID: 24827126). We agree that a direct comparison would be informative. However, given that these toxicities are already established in the literature, we believe that repeating these experiments is not essential to the present study. We have instead clarified that our findings demonstrate that RP-3467 does not exacerbate olaparib-associated myelosuppression under the conditions tested, without implying that the combination is free of hematologic toxicity or has been directly demonstrated to be safer than other PARPi combinations.
(6) Figure 2 might give the impression that the new POLQi gives a similar, if not greater, BRCA2 synthetic lethal effect than a PARPi (e.g. 2d-f). It would help readers to see the full saruparib and olaparib dose-response curves, not (as per 2 A, B, C) just the single point estimates.
The ZIP synergy matrices in Extended Data Figure 2 were acquired across a full concentration grid for both agents, and therefore already contain the complete single-agent dose–response for olaparib and for saruparib in each line.
(7) Why was DOTC42510 not profiled for PARPi?
We apologize for the poor signposting that made this data hard to find. DOTC24510 cells were profiled with olaparib: the combination growth data appear in Extended Data Figure 2c and the corresponding ZIP synergy plot in Extended Data Figure 2d. The main-text sentence reporting the olaparib combination does name DOTC24510 alongside the two colorectal lines, but cites only the main-figure panels, which do not include this line.
(8) Figure 2f shows error bars on an n=2 dataset - one cannot estimate variance from only an n=2 dataset range.
We thank the reviewer for pointing this out. We agree that an n=2 dataset is insufficient for a reliable estimate of variability. We revised Figure 2f to show the data points without error bars and clarify the sample size in the figure legend.
(9) Is there a statistical analysis on Figure 2G?
The data originate from two independent experiments, so we did not consider it appropriate to perform a formal statistical analysis. We note, though, that the data were consistent between the two replicates and that a robust difference was observed between the combination and single-agent arms.
(10) Figure 3. Olaparib 25 mg/kg QD is well below the MTD in mice. This makes the olaparib response in HCT116 BRCA2 k/o xenografts look modest when compared to the combinations.
We used 25 mg/kg to highlight the strong combination potential between RP-3467 and olaparib. That said, Fig. 3b shows that even when a PARP inhibitor (saruparib in this case) is used at its maximally efficacious dose (PMID: 35929986), the RP-3467 combination has the potential to outperform PARPi monotherapy.
Reviewer #3 (Public review):
Summary:
The authors seek to validate the effectiveness of an inhibitor of the Pol theta helicase-like domain ATPase activity in its ability to ablate MMEJ activity in cell line models. They then confirm the inhibitor is synthetic lethal in a variety of BRCA1/2 deficient cancer models (and one PALB2 model). Finally, they report apparent effectiveness in treatment of a single patient that failed therapy with PARPi due to 53BP1 deficiency, a hypothesized special use-case for targeting pol theta.
Strengths:
It is a clear bench-to-bedside translation of mechanistic-oriented basic science to a clinical therapy. The experiments are largely rigorously performed. The approaches employed are impressively diverse, from in vitro to cell-line model, to preclinical model, to clinical application. The interpretations made are well supported by the available data.
We sincerely thank the reviewer for their thoughtful and encouraging assessment of our work. We are particularly pleased that the reviewer recognizes the translational significance of the study, the breadth and rigor of the experimental approaches, and the strength of the conclusions. We appreciate these positive comments and the constructive suggestions, which have helped us further strengthen the manuscript.
Weaknesses:
The addition of a few missing details would improve the usefulness to the reader. Additionally, the rationale behind some approaches could be clarified.
eLife Assessment
This study provides a valuable mapping of how genomic variation from a large collection of Klebsiella pneumoniae clinical isolates shapes bacterial fitness across 214 conditions. Solid data describe that fitness is strongly condition-dependent and that features can be advantageous under selection. Though the predictive, plasmid causality and GWAS association findings would be further strengthened by additional analyses ruling out population-structure confounding, these findings are of value for our understanding of the power of genotype to predict phenotype for this human pathogen.
Reviewer #1 (Public review):
This manuscript by Zhou, Williams, Millner, and colleagues reports findings from a large-scale, multicenter study of Klebsiella pneumoniae (Kp) isolates. These isolates were sequenced using short-read sequencing and extensively characterized using a variety of field-appropriate tools, such as Kleborate, AMRFinderPlus, and PlasmidFinder. Then, the authors measured the growth of these Kp isolates across many (>200) environmental conditions they pre-determined as important for making human-centric fitness conclusions. These genotype and phenotype data were integrated, and the authors deployed an association-based strategy to identify genetic determinants of the observed fitness effects. Finally, the authors built models to predict fitness outcomes using the associated genetic determinants. From these data, the authors conclude that fitness is "an emergent property" of a given Kp isolate's genetic background and environmental constraints, with a nod toward the future development of biomarkers to predict fitness outcomes using genotype alone.
The primary strength of this study is the integration of two large datasets, one genotypic and the other phenotypic. This approach is novel in its scale (genotypic, geographic, etc.), which adds to the strength of the overall conclusions. These conclusions, for the most part, are consistent with the data. As such, this study can act as an example to other studies tying genotype to phenotype, especially in the context of careful genetic analysis and rational phenotypic design. A secondary strength is that the findings are supported by other, smaller-scale studies such that there is a growing consensus as to the general fitness determinants employed by Kp that are potentially driving its increased global importance as a bacterial pathogen.
The primary weakness of this study is that some confounding factors are not taken into account or are deemphasized. The primary confounding factor is that the associations between genetic signatures of fitness are not fully considered. This is especially important given that an important driver of Kp fitness is its accessory genome, which is heavily influenced by mobile genetic elements and plasmids. The use of short-read sequencing leads to an inherent inability to tie fitness effects to such genetic elements, which may lead to masking or overinterpretation of particular genotype-phenotype relationships. For example, nearly all of the gene-level determinants of condition-specific fitness and pleiotropic genes are known plasmid-associated genes; however, it is unclear if these specific genes are driving fitness or if they are merely signatures of other associated genetic elements. This broadly weakens the conclusions and translatability, and in the absence of long-read sequencing to resolve genetic architecture, it is an unresolvable weakness. A second is the over-interpretation of the variables used to assess bacterial growth, which is the primary fitness metric. Colony opacity can be influenced by Kp capsule production, which is a highly variable trait that is associated with environmental conditions. Finally, it is unclear whether clonal isolates are excluded to avoid overinterpretation of fitness outcomes.
Reviewer #2 (Public review):
Summary:
The authors mapped context-dependent fitness of more than 1,400 clinical K. pneumoniae isolates across 214 antibiotic, chemical, environmental, metal, envelope and nutrient conditions, integrating fitness measurements with whole-genome sequencing to build a genotype-phenotype-fitness framework and a machine learning pipeline for predicting bacterial fitness across environments.
Strengths:
The scale of the dataset is remarkable both in terms of isolates and conditions tested. The central findings that fitness is strongly condition-dependent and that resistance-associated features help under selection are well supported. Resource accessibility (ColonyExplorer portal, code, trained models) is excellent, allowing future integration of different genetic backgrounds and environmental contexts.
Weaknesses:
(1) Normalisation uses a plate-level random effect and plate-median scaling, whereas within-plate (position) effects and their correction are not described, although typically required in such high-density arrays.
(2) Hypermucoid or heavily encapsulated isolates may form smaller but denser colonies. The area-opacity correlation (Figure S3A) does not exclude this, so the virulence-associated costs in selected conditions (Figures 2F-G, S5A) could be partly measurement-driven.
(3) Plasmid effects are attributed causally from observational data. Fitness effects are inferred from short-read replicon calls (Figure 3) and may be confounded by lineage and co-occurring resistance determinants.
(4) The study does not apply linear mixed model GWAS on the continuous fitness phenotype, relying instead on median dichotomization with a non-mixed model that is vulnerable to population structure (especially in a highly clonal species) and could exacerbate apparent separation, inflating association counts.
(5) GWAS-derived features appear to have been selected using all isolates rather than within each training fold. Test-set labels thus influenced which features entered the models, biasing correlation, coverage and interval estimates upward. Although mitigating overfitting, nested hyperparameter tuning and lineage-stratified tests do not remove this bias. Because of this, I do not find the claim that genomic variants predict fitness beyond population structure adequately supported yet.
With the points above addressed, I believe this work would be highly valuable for prioritising genes and variants for functional validation and for generally studying fitness trade-offs in hospital pathogens.
eLife Assessment
This important study provides a contribution to retinal regeneration research by using overexpression of pro-neural factors to reprogram fetal human RPE cells into retinal neurons. The authors provide solid evidence of fetal RPE reprogramming into neural and photoreceptor-like states using scRNA-seq and imaging validation. There remains concerns regarding comparisons between the effectiveness of different combinatorial transcription factor codes.
Reviewer #2 (Public review):
Summary:
This is an interesting study that explores how human RPE could be used as source for new retinal neurons. This is a welcome addition to the field of retinal regeneration which is currently focused almost exclusively on the regenerative capacity of Muller glia cells. The line of inquiry is firmly rooted in findings from amphibian and embryonic chick model systems and advances a fetal human retina RPE based screening system as a rich resource for insights into human RPE biology, including as a potential stem cell source.
The authors investigate the potential of fetal human RPE cells to be reprogrammed into retinal neurons using overexpression of pro-neural factors. While this is a critical knowledge gap in the field of retinal regeneration with significant promise for developing regenerative therapies, several methodological concerns impact the interpretation of results. Firstly, while the authors sought to evaluate factors that enhance RPE reprogramming when co-expressed with ASCL1, nearly all co-expression constructs tested failed to achieve appreciable expression of ASCL1, leaving a central hypothesis of this study largely untested (Major concern 1). Second, although the authors were able to detect a cluster of photoreceptor-like cells in their screen, they were unable to identify which reprogramming construct generated this cluster (Major concern 2). Finally, an essential control that definitively demonstrates the value of combinatorial transcription factor reprogramming is missing (Major concern 3). In summary, the authors establish a valuable new paradigm for culturing and reprogramming fetal human RPE, and even more importantly, demonstrate successful reprogramming to neural fates. However, the discussion and interpretation of results needs to be modified significantly to make it clear that (i) the outcome of many co-expression paradigms remains effectively unknown/untested due to failed over-expression of ASCL1, and that (ii) the reprogramming construct giving rise to photoreceptor-like cells could not be conclusively identified from their initial screen.
Strengths:
(1) Powerful new screening system advanced for exploring the regenerative potential of human feta RPE cells.
(2) Co-expression vector system for testing additive effects of proneural transcription factors
Weaknesses:
Major concerns:
(1) The authors executed a screen for combinations of factors that can enhance ASCL1-mediated reprogramming of RPE into retinal neurons. However, the expression level of ASCL1 was remarkably low in virtually all co-expression paradigms (see Fig. 3C). Notably, the reprogramming combination with the highest potency (ASCL1 + NEUROD1) was also the one exhibiting the highest level of ASCL1 expression. The "failed" reprogramming of most of the co-expression constructs (ASCL1+LMO1, ASCL1+EZH2, and ASCL1+RAX2) is potentially a false negative resulting from low transgenic expression of ASCL1.
(2) The authors interpretation is that the overexpression of NeuroD1 and Ascl1 generated a new cluster that expressed markers of photoreceptors such as RXRG and RCVRN (see Figure 3D). However, there does not actually appear to be any overlap between the ASCL1+NEUROD1 cluster (orange dots, left panel) and the cells expressing markers of photoreceptors (yellow/green/purple?/black? dots, right panel; yellow being ASCL1-EZH2, green being ASCL1, purple being ASCL1-and black being control - though color coding here is admittedly somewhat confusing). Thus, the photoreceptor-like cluster of interest actually seems to correspond to gray cells that were unmapped / exposed to an unknown programming cocktail. So, it remains completely unknown which reprogramming construct generated this cluster.
(3) To conclusively establish the additive role of NEUROD1 in reprogramming, it would be prudent to compare ASCL1 + FA directly to ASCL1+NeuroD1+FA. This control was not included but is needed for a more complete interpretation of results.
Comments on revised version.
We thank the authors for their thoughtful responses to our concerns and for modifying a figure to improve clarity. However, we were disappointed to see that the responses to our major concerns were not incorporated into the revised manuscript, as we believe they provide context that is essential to proper interpretation of the data. Minimal revisions of the current manuscript would address our first two major concerns. This could be achieved by adding two additional sentences, for example incorporating the sentences below which are excerpted from the authors response to our concerns:
The following excerpt from the authors response would address major concern 1 if incorporated into the relevant section in the main text of the manuscript:
"Insufficient ASCL1 expression in the combinations might have led to lack of reprogramming, representing a false-negative result."
And this excerpt would address major concern 2:
"The farthest subcluster with highest RCVRN/RXRG expression was not assigned ASCL1-NEUROD1 identity."
Author response:
The following is the authors’ response to the original reviews.
Public Reviews:
Reviewer #1 (Public review):
Summary:
In this study, the authors identified transcription factor combinations capable of inducing retinal neuronal programs in cultured fetal human retinal pigment epithelial (RPE) cells. Using a pooled screening strategy, single-cell RNA sequencing, lineage barcoding, and immunohistochemical analyses, they identified ASCL1 and NEUROD1 as an effective combination for inducing retinal neuron-associated transcriptional states. This work aims to advance the development of therapeutic approaches for retinal regeneration by exploring the plasticity of RPE cells.
Strengths:
A major strength of the study is the comprehensive experimental design. The combination of transcription factor screening, lineage tracing, single-cell transcriptomics, and molecular validation provides a detailed characterization of the cellular responses to reprogramming factor expression.
Weaknesses:
All experiments were performed using fetal human RPE cells. Because fetal RPE remains relatively immature and retains proliferative capacity, it remains unclear to what extent the observed responses reflect true reprogramming of differentiated RPE cells versus activation of developmental plasticity already present in fetal tissue. The absence of adult human RPE controls limits assessment of the generality and translational relevance of the findings.
We agree with the assessment that the reprogramming of fetal RPE due to its potential fate plasticity poses a different challenge than the adult. In fact, embryonic chick RPE will only undergo neural trans differentiation when treated with FGF before embryonic day 4. Therefore, we specifically focused on RPE at early developmental stages for our experiments. It is interesting in that light that the ages of fetal human retina are far more advanced in their development than the E4 chick, yet still have the capacity for neural reprogramming; while it will be interesting to explore in the future how long during fetal development this plasticity exists, we were limited by the fetal ages that were available to us. Moreover, the facility that supplied our tissues (Birth Defects Research Lab) are no longer able to use NIH funds to distribute fetal tissues as they have for the last >40 years, making any future studies with primary fetal tissues much more difficult.
Reviewer #2 (Public review):
Summary:
This is an interesting study that explores how human RPE could be used as a source for new retinal neurons. This is a welcome addition to the field of retinal regeneration, which is currently focused almost exclusively on the regenerative capacity of Müller glia cells. The line of inquiry is firmly rooted in findings from amphibian and embryonic chick model systems and advances a fetal human retina RPE-based screening system as a rich resource for insights into human RPE biology, including as a potential stem cell source.
The authors investigate the potential of fetal human RPE cells to be reprogrammed into retinal neurons using overexpression of pro-neural factors. While this is a critical knowledge gap in the field of retinal regeneration with significant promise for developing regenerative therapies, several methodological concerns impact the interpretation of results. Firstly, while the authors sought to evaluate factors that enhance RPE reprogramming when co-expressed with ASCL1, nearly all co-expression constructs tested failed to achieve appreciable expression of ASCL1, leaving a central hypothesis of this study largely untested (Major concern 1). Second, although the authors were able to detect a cluster of photoreceptor-like cells in their screen, they were unable to identify which reprogramming construct generated this cluster (Major concern 2). Finally, an essential control that definitively demonstrates the value of combinatorial transcription factor reprogramming is missing (Major concern 3).
In summary, the authors establish a valuable new paradigm for culturing and reprogramming fetal human RPE, and even more importantly, demonstrate successful reprogramming to neural fates. However, the discussion and interpretation of results needs to be modified significantly to make it clear that (i) the outcome of many co-expression paradigms remains effectively unknown/untested due to failed over-expression of ASCL1, and that (ii) the reprogramming construct giving rise to photoreceptor-like cells could not be conclusively identified from their initial screen.
Strengths:
(1) Powerful new screening system advanced for exploring the regenerative potential of human fetal RPE cells.
(2) Co-expression vector system for testing additive effects of proneural transcription factors.
Weaknesses:
Major concerns:
(1) The authors executed a screen for combinations of factors that can enhance ASCL1-mediated reprogramming of RPE into retinal neurons. However, the expression level of ASCL1 was remarkably low in virtually all co-expression paradigms (see Figure 3C). Notably, the reprogramming combination with the highest potency (ASCL1 + NEUROD1) was also the one exhibiting the highest level of ASCL1 expression. The "failed" reprogramming of most of the co-expression constructs (ASCL1+LMO1, ASCL1+EZH2, and ASCL1+RAX2) is potentially a false negative resulting from low transgenic expression of ASCL1.
We agree with the reviewer that insufficient ASCL1 expression in the combinations might have led to lack of reprogramming, representing a false-negative result. We do not rule out that individual, rather than pooled, expression vectors of the “failed” varieties might prove effective. A future version of such a screen might need a control to test proper vector functionality of each member in the pool. However, the expression of ASCL1 alone from a vector in the same pool worked well. Additionally, the expression of ASCL1 from the ASCL1-NEUROD1 vector was not nearly as high as from ASCL1 alone and still led not only to reprogramming but also to a shift in the gene expression landscape as seen in 4D. This shift is consistent across experiments and vector constructs.
(2) The authors' interpretation is that the overexpression of NeuroD1 and Ascl1 generated a new cluster that expressed markers of photoreceptors such as RXRG and RCVRN (see Figure 3D). However, there does not actually appear to be any overlap between the ASCL1+NEUROD1 cluster (orange dots, left panel) and the cells expressing markers of photoreceptors (yellow/green/purple? /black? dots, right panel; yellow being ASCL1-EZH2, green being ASCL1, purple being ASCL1-and black being control - though color coding here is admittedly somewhat confusing). Thus, the photoreceptor-like cluster of interest actually seems to correspond to gray cells that were unmapped/exposed to an unknown programming cocktail. So, it remains completely unknown which reprogramming construct generated this cluster.
We apologize for the confusion in Figure 3D, which stemmed from lack of proper labeling.
The right panel shows the expression of photoreceptor genes in a feature plot, not the vector assignment. The color code is different between the left most panel (which shows vector assignment) and the other two panels (which show gene expression). To make this more clear, we have now separated this into a panel D and a panel D’, and added a caption as well as a scale bar for the D’ panel.
The reviewer is correct that the farthest subcluster with highest RCVRN/RXRG expression was not assigned ASCL1-NEUROD1 identity. The inherent sparsity of scRNAseq experiments and the small size of that particular cluster made definite identification challenging. Furthermore, vector assignment is tied to vector and thus barcode expression. It is conceivable that as cells matured towards a photoreceptor-like cells, they silenced the vector insertion locus but initiated marker expression. Additionally, these photoreceptor-like cells were assigned to the same cluster (neur, Fig. 3B) as the positively identified ASCL1-NEUROD1 cells by unsupervised clustering, express NEUROD1 (Suppl. Fig. 2D) and seem to follow the same trajectory. This suggests that they stem from ASCL1-NEUROD1 overexpression. Moreover, the follow-up experiments showing the expression of the relevant genes in the reprog4 cluster (Fig. 4) and the corresponding stainings (Fig. 5) provide evidence that ASCL1-NEUROD1 overexpression causes expression of RCVRN.
(3) To conclusively establish the additive role of NEUROD1 in reprogramming, it would be prudent to compare ASCL1 + FA directly to ASCL1+NeuroD1+FA. This control was not included but is needed for a more complete interpretation of results.
We thank the reviewer for this suggestion. Such a control would indeed clarify the role of FA treatment on reprogramming with ASCL1 alone. However, to address the additive role of NEUROD1, we presented in this study the comparison of ASCL1-NEUROD1 with ASCL1 alone. Even without the addition of FA, we noted a shift in expression pattern (Fig. 4D) and the formation of neurogenic clusters (Suppl.Fig. 4B) with the addition of NEUROD1. Moreover, as noted above, it is no longer possible for us to obtain tissue from the BDRL to carry out additional experiments at this time.
Recommendations for the authors:
Reviewer #1 (Recommendations for the authors):
(1) The antibody section should mention the dilution used and how primary antibodies were validated (in the current study, or cited from previous literature).
We have added the dilutions of the used primary antibodies. They are targeted against well-known markers in the field and have been validated by our lab and others.
(2) A supplementary table summarizing and comparing pro-photoreceptor genes detected vs functional photoreceptor genes would be helpful to readers.
We think that the heatmap in 4C provides a good overview of detected genes of interest.
(3) The manuscript would benefit if figures have scale bars and marker labels above or below the images, but not superimposed (so as not to block the gene signal in some cells). (Figure 1C).
We changed the labels in 1C and added a scale bar to be more consistent with the other figures in the manuscript.
(4) Figure 2B and 5B (neurogenic dome) are missing scale bars.
We added scale bars in 2B and 5B as well as the supplemental figures.
(5) Plasmid design: dual promoters without insulators should be addressed (MCV and Ef1a). If possible, an RPE65-specific promoter to drive ASCL1-NEUROD1, to show that short temporal expression is not capable of inducing the desired reprogramming.
We agree that an RPE-specific promoter would not only gauge the strength and duration that is necessary for reprogramming, but would also support translational efforts. While we have addressed that question somewhat with the HES1-ASCL1 construct in our initial experiments, more in-depth investigation to choose an optimal promoter will be necessary in the future.
Reviewer #2 (Recommendations for the authors):
(1) The first sentence of the abstract is identical to the third sentence of the introduction. Rephrase to avoid redundancy. There are currently a few promising approaches for treatment: for example, gene therapy to augment or replace mutated genes has proven successful in preclinical studies, and some of these therapies are moving towards the clinic [2], [3].
We rephrased the sentence to:
Current developments in gene therapy, which aims to deliver healthy copies of mutated genes, show promising results and are increasingly becoming available in the clinic.
(2) Panel 1C would be more informative if images were enlarged/zoomed in ~4x or higher.
We have enlarged the panels as requested by the reviewer.
(3) Panel 1D reports size/area differences for cultures at a variety of time points. Data should thus be represented as a line chart to reflect whether cell size is changing over time, rather than pooling data points from disparate time points into a single group.
We agree that such a chart would provide a measure of RPE maturation over time; however, our focus is the relationship between cultured and native RPE. As there seems to be little spread in the measurements between ages, and the range of ages between cultured and native RPE is comparable, we believe that the current chart represents a valid visualization.
(4) Panel 2D would be clearer if cluster labels were moved to the top of the figure (such that they sit above the columns corresponding to those clusters). It could be further clarified by adding category labels to the marker genes (i.e., RPE markers, ascl1 targets, Neural markers, etc.).
We thank the reviewer for this suggestion. We shifted the cluster labels and created new labels for the gene groups.
(5) No comma necessary here: Using transcription factor binding site analysis on these peaks, revealed an enrichment of members of the ETS family like ELK1/4 and ETV1/4.
Thank you for your attention to detail. We have removed that comma.
(6) Add dashed box to Panel 3D to indicate region of detail displayed in panels to the right (as was done in Panel 3C).
We have added a dashed box in panel 3D.
(7) Distracting typos in the legend of Figure 3.
We have corrected the typos.
eLife Assessment
This important study combines crystallographic fragment-based screening with fluorescence-polarization competition assays to demonstrate that the SARS-CoV-2 accessory protein ORF9b, and its interface with the mitochondrial import receptor TOM70, is accessible to small-molecule binding. This is a result with practical implications beyond coronavirus biology for efforts to restore interferon signalling and the antiviral response. The structural evidence is convincing, resting on an unusually large body of ligand-bound crystal structures whose fragment hotspots are independently corroborated by orthogonal biophysical approaches including fluorescence polarization and SPR. This study will be of interest to the broad readership of eLife, spanning structural biology, drug discovery, and coronavirus communities.
Reviewer #2 (Public review):
Summary:
The authors investigate chemical strategies to disrupt the interaction between the SARS-CoV-2 accessory protein Orf9b and the host mitochondrial receptor Tom70, an interaction implicated in suppression of type-I interferon responses. They employ two discovery approaches: a crystallographic fragment screen against the Orf9b homodimer and a high-throughput fluorescence polarization screen for compounds that compete with Orf9b binding to Tom70. The study identifies fragment-binding hotspots on Orf9b, develops lipidated analogs that stabilize the Orf9b homodimer, and discovers Tom70-binding compounds with low micromolar activity that inhibit Orf9b binding in vitro.
Strengths:
Impressive amount of work using a variety of complementary approaches and methods to validate binding (FP, SPR, and computational modelling and SAR). The combination of crystallographic fragment screening on Orfb9 and HTS on Tom70 provides two independent routes for perturbing the Orf9b-Tom70 interaction. The structural work seems to be of very high-quality. The fragment campaign is extensive, yielding a substantial number of fragment-bound structures and identifying biologically meaningful binding hotspots on Orf9b.
Finally, the screen results in reporting useful chemical starting points. Although potency remains modest, the study provides tractable scaffolds and a clear framework for future optimization.
Comment on revised version.
I don't have any further comments. The authors have addressed my suggestion by adding a statement for future work in the discussion.
Author response:
The following is the authors’ response to the original reviews.
Public Reviews:
Reviewer #1 (Public review):
Summary:
The authors used FBDD screening to identify numerous compounds interacting with the ORF9b dimer. They expanded the original fragment hit, soaked the derivatives into the crystals and confirmed their binding poses, and showed that the derivatives bind the target with higher affinity. The authors further targeted the ORF9b binding site on TOM70, and used a fluorescence polarization-based (FP) assay to screen a compound library and obtained several hits. Structure-activity relationship (SAR) optimization yielded hit analogs that have higher binding affinity to TOM70.
Strengths:
(1) The study adopted novel drug design strategies, including stabilizing ORF9b homodimer to prevent it from binding TOM70, and blocking ORF9b from binding TOM70 by screening compounds that compete with ORF9b for binding TOM70.
(2) The work established a feasible high-throughput screening assay. This FP-based assay screened ~50,000 compounds, from which two hit compounds were further optimized to yield analogs with higher binding affinity.
We thank reviewer #1 for their thoughtful assessment of our work and their constructive feedback.
Weaknesses:
(1) The study lacks functional assays to evaluate whether the ORF9b-stabilized compounds or TOM70 binding compounds could affect IFN inhibition caused by ORF9b or virus infection.
We thank reviewer #1 for their feedback on this subject. We agree that cell-based assays measuring the impact of these compounds on IFN stimulation/suppression would be a valid next step. We have expanded the discussion section to include potential future experiments that could be conducted to probe this:
“Although we have not assessed the permeability of these compounds and their ability to restore IFN in the context of cell-based assays, one way this could be accomplished is by adapting a previously used reporter cell line. Specifically, this assay uses a reporter cell line that has a Lucia reporter under the control of the IFN-β/ISG56 promoter that is expressed upon mock infection with synthetic dsRNA. When WT Orf9b is co-transfected with the reporter, IFN is suppressed, and, we have shown that phosphomimetic mutants of Orf9b that block its interaction with Tom70 restores IFN signaling. If these compounds are sufficiently permeable, then, for the lipidated-compounds targeting the Orf9b homodimer, we would expect the Orf9b monomer: homodimer equilibrium to shift towards the homodimer due to the stabilizing effect these lipidated-compounds have on the homodimer conformation leading to an increase in IFN signaling. Similarly, for compounds targeting Tom70, we would expect that these compounds block the Orf9b monomer from binding to Tom70 to produce a similar effect.”
(2) There is a lack of experimental evidence to reveal the binding mode of lipidated-compounds with ORF9b homodimer.
- We thank reviewer #1 for their feedback. We agree that we were unable to provide direct evidence that our lipidated-compounds are bound to the homodimer through the central lipid-binding channel. Experimental structures of our compounds bound to the Orf9b homodimer would have strengthened our functional studies, however, as we described in the manuscript pertaining to our initial crystallographic fragment screen:
“Although we have shown that it is possible to refold the Orf9b homodimer to eliminate the lipid from the central channel, we found that apo-Orf9b homodimer crystals were far less tolerant to DMSO and on average diffracted to resolutions worse than 2.5Å. Importantly, we found that microseeding helped increase crystal reproducibility and yielded larger crystals on average. Therefore, for our crystallographic fragment screening campaign we elected to use the lipid-bound protein to identify additional potential binding sites that could be exploited to stabilize the Orf9b homodimer.”
- We did undertake efforts to co-crystallize our lipidated-compounds with refolded apo-Orf9b homodimers but were unable to generate crystals of sufficient quality to collect X-ray data. Therefore, we relied on both our fluorescence polarization assay and prediction models like Chai1 to investigate if the expected behavior could be recapitulated. We felt this was reasonable for three reasons.
- First, as we explain in the discussion, none of these compounds or structures were publicly available for models like Chai1 to train on which prevents data leakage:
“None of these structures were publicly available on the PDB prior to these structural predictions, therefore, they were not available for Chai1 to train on. The placement of the lipidated-analogs in the central channel was in line with the anticipated binding mode, albeit, with discrepancies between the predicted and structurally observed portions of the analog compounds.”
- Second, as we outlined in the fluorescence polarization assay, we tested these compounds in two control experiments that investigated whether the changes in FP signal over time could also be explained by the compounds binding to Tom70 directly, or, by acting on the obligate Orf9b monomer which was generated by truncating the N and C-termini:
“We first considered whether the lipidated-analogs had any binding activity for Tom70 on its own and whether changes in FP signal were due to binding to the Orf9b homodimer. We tested the compounds on their own in the presence of the pre-equilibrated Tom70:Orf9b-FITC complex and observed no significant decrease in FP signal indicating that the lipidated-analogs do not directly bind to Tom70 to drive a decrease in FP signal (Figure 2—figure supplement 2A). We next tested whether these compounds acted on the Orf9b homodimer specifically or if they could inhibit the Orf9b monomer from binding to Tom70 by incubating compounds with an Orf9b peptide that acts as an obligate monomer. When chasing with the pre-equilibrated Tom70:Orf9b-FITC complex, we observed no significant differences in kinetics of Orf9b peptide binding compared to the peptide on its own which showed a fast single-exponential kinetic decay in signal suggesting that these compounds do not act on the Orf9b monomer to inhibit binding to Tom70 (Figure 2—figure supplement 2B)”
(3) There is a lack of experimental evidence to reveal the binding mode of HTS hits or analogs for TOM70.
We thank reviewer #1 for their feedback. We agree that experimental structures of our HTS compounds bound to Tom70 would provide more insights into how these compounds bind to Tom70 (and their precise location) in ways that the predicted complexes from models like Chai1 were unable to do. At the time of drafting this manuscript, we had not yet identified suitable conditions for structurally resolving Tom70 with our HTS compounds. Therefore, we elected to rely on two orthogonal methods of predicting where our HTS compounds are bound to probe whether the predicted binding sites could plausibly block Orf9b from binding. In the time since releasing this manuscript, one group had been able to crystallize apo-Tom70 which may serve as a suitable platform for future structural studies investigating these compounds and their exact binding mode to Tom70 to enable structure-guided optimisations of these initial hits.
(4) Overall, none of the compounds shown in the paper have promising potency warranting further development; their binding affinity is limited to the micromolar range.
We thank reviewer #1 for their feedback. We agree that these compounds should be viewed as starting points for potential future optimization on targets where there were no previous existing ligands. The molecules serve to validate that it is possible to develop small molecules that target Tom70’s C-terminal domain which may be relevant to future studies investigating Tom70’s biology.
Reviewer #2 (Public review):
Summary:
The authors investigate chemical strategies to disrupt the interaction between the SARS-CoV-2 accessory protein Orf9b and the host mitochondrial receptor Tom70, an interaction implicated in suppression of type-I interferon responses. They employ two discovery approaches: a crystallographic fragment screen against the Orf9b homodimer and a high-throughput fluorescence polarization screen for compounds that compete with Orf9b binding to Tom70. The study identifies fragment-binding hotspots on Orf9b, develops lipidated analogs that stabilize the Orf9b homodimer, and discovers Tom70-binding compounds with low micromolar activity that inhibit Orf9b binding in vitro.
Strengths:
An impressive amount of work using a variety of complementary approaches and methods to validate binding (FP, SPR, and computational modelling and SAR). The combination of crystallographic fragment screening on Orfb9 and HTS on Tom70 provides two independent routes for perturbing the Orf9b-Tom70 interaction. The structural work seems to be of very high-quality. The fragment campaign is extensive, yielding a substantial number of fragment-bound structures and identifying biologically meaningful binding hotspots on Orf9b.
Finally, the screen results in reporting useful chemical starting points. Although potency remains modest, the study provides tractable scaffolds and a clear framework for future optimization.
We thank reviewer #2 for their thoughtful assessment of our work and their constructive feedback.
Weaknesses:
General comment:
(1) Although there is already an incredible amount of data presented, one limitation of this study is the lack of cellular validation - do these drugs enter cells, restore interferon signalling, reduce viral loads, or alter Orfb9 localization?
We thank reviewer #2 for their feedback. See also the feedback from Reviewer 1. We have not assessed the permeability of these compounds in the context of cell-based assays and IFN signaling and agree that cell-based assays measuring the impact of these compounds on IFN stimulation/suppression would be a valid next step. We have expanded the discussion section to include potential future experiments that could be conducted to probe this:
“Although we have not assessed the permeability of these compounds and their ability to restore IFN in the context of cell-based assays, one way this could be accomplished is by adapting a previously used reporter cell line. Specifically, this assay uses a reporter cell line that has a Lucia reporter under the control of the IFN-β/ISG56 promoter that is expressed upon mock infection with synthetic dsRNA. When WT Orf9b is co-transfected with the reporter, IFN is suppressed, and, we have shown that phosphomimetic mutants of Orf9b that block its interaction with Tom70 restores IFN signaling. If these compounds are sufficiently permeable, then, for the lipidated-compounds targeting the Orf9b homodimer, we would expect the Orf9b monomer: homodimer equilibrium to shift towards the homodimer due to the stabilizing effect these lipidated-compounds have on the homodimer conformation leading to an increase in IFN signaling. Similarly, for compounds targeting Tom70, we would expect that these compounds block the Orf9b monomer from binding to Tom70 to produce a similar effect.”
(2) The logic of locking Orfb9 as a dimer is that the monomer binds Tom70 - thus, a more stable dimer means less monomer. In Figure 2, the Orfb9 homerdimer stabilization by compounds should reduce binding affinity to Tom70. A direct binding experiment measuring reduced Tom70 binding with compound treatment would better strengthen this claim.
We thank reviewer #2 for their observation and agree that compounds targeting the Orf9b homodimer conformation should block binding to Tom70 by preventing Orf9b homodimer dissociation into Orf9b monomers. As schematized in Figure 2D, we described how our fluorescence polarization assay measures the binding of Orf9b to Tom70. It provides insight into how compounds that bind to the Orf9b homodimer may slow the rate of homodimer dissociation. In addition, we have previously demonstrated in a separate publication that there are clear differences in the Orf9b homodimer dissociation kinetics when using either lipid-bound or lipid-free Orf9b homodimers as the competitor due to the stabilizing effect the lipid has on the homodimer conformation.
Reviewer #3 (Public review):
Summary:
This paper attempts to and succeeds in demonstrating that Orf9b is able to bind small molecules using X-ray fragment screening, SPR and FP assays. Exploration of sites from the fragment screening is performed along with fragment linking with inter-dimer lipid moieties.
Strengths:
The experimental work looks strong and well performed. The interpretation of the data is appropriate and was often validated through orthogonal methods and follow-up compounds. The use of Tom70 to find binders that might disrupt interactions between Orf9b and Tom70 is elegant.
Weaknesses:
The use of Chai-1 to predict co-folded structures with binding molecules was not properly described - no mention of this in the methods. It was not commented on whether the compounds which were found were attempted to be co-crystallised. If they were but negative data was collected (didn't crystallise, didn't diffract or no additional density was found), then this needs to be stated.
We thank reviewer #3 for their feedback. We have revised the manuscript to clearly point out that beyond our initial attempts to crystallize apo-Orf9b homodimer to enable fragment soaking, we also attempted to both soak our lipidated-compounds into apo-Orf9b homodimer crystals as well as co-crystallize the compounds which did not yield crystals.
“We attempted two strategies of crystallizing our lipidated-compounds with the apo-Orf9b homodimer to structurally resolve the complex. First, we attempted to soak our compounds into apo-Orf9b homodimer crystals but observed poor tolerance to DMSO which quickly disintegrated the crystals after minutes. Second, we attempted to co-crystallise the lipidated-compounds after refolding the Orf9b homodimer to generate apo-Orf9b homodimers but could not identify conditions that yielded crystals upon re-screening.”
We have also revised the methods and materials section to better explain the use of Chai1 to co-fold both Orf9b and Tom70 with their respective compounds.
“Chai1’s web server was used for co-folding either Orf9b or Tom70 with their respective hits. For the Orf9b homodimer, the sequence for a single chain was supplied to represent chains A and B in the homodimer and the ligands were supplied as SMILES strings. In the case where two copies of the ligand were investigated, the number of copies of the ligand was increased to two. Co-folding was performed using MMSeqs2 for the MSA and template searching to ensure the correct folding of the homodimer conformation. Tom70 co-folding was performed by supplying the sequence for hTom70 as a single chain and the compounds as SMILES strings. Co-folding was performed using MMSeqs2 for the MSA with template searching.”
Recommendations for the authors:
Reviewer #1 (Recommendations for the authors):
(1) It is disappointing that the authors screened binders for ORF9b homodimer, identified hits and derivatives binding at sites 1 & 2, measured the binding affinity to the target, but found no effects on stabilizing ORF9b homodimer. The FP competition assay was used to assess the efficacy of the compounds; maybe the authors could use alternative methods like Thermal Shift Assay (TSA) assays to detect some subtle changes in stability.
We thank the reviewer for their comments and agree that the lack of stabilizing activity observed for the site 1 and 2 analogs was disappointing. However, we do feel that the lipid-binding central channel may be a viable alternative site target that may also yield insights into the role of lipid binding in Orf9b biology. We agree that orthogonal methods such as TSA could be insightful for probing the stabilizing effects of the compounds on the Orf9b homodimer.
(2) To better illustrate the results of the FP kinetic assay to investigate whether lipidated-compounds binding to the Orf9b homodimer had effects on the stability of the Orf9b homodimer and Orf9b-Tom70 equilibrium, I recommend moving the results shown in Supplemental Figure 4 to a main figure and merging these results with Figure 2CD.
We think that the eLife publication platform provides appropriate visibility for the supplementary figures and that Figure 2D does the best job of illustrating the effects on kinetics for the initial reader.
(3) Have the authors tried to co-crystallise their top-rated lipidated-compounds with ORF9b homodimer to see where the lipid part of the compound could extend into the central channel of ORF9b homodimer?
We have revised the manuscript to clearly point out that beyond our initial attempts to crystallize apo-Orf9b homodimer to enable fragment soaking, we also attempted to both soak our lipidated-compounds into apo-Orf9b homodimer crystals as well as co-crystallize the compounds which did not yield crystals.
“We attempted two strategies of crystallizing our lipidated-compounds with the apo-Orf9b homodimer to structurally resolve the complex. First, we attempted to soak our compounds into apo-Orf9b homodimer crystals but observed poor tolerance to DMSO which quickly disintegrated the crystals after minutes. Second, we attempted to cocrystallize the lipidated-compounds after refolding the Orf9b homodimer to generate apo-Orf9b homodimers but could not identify conditions that yielded crystals upon re-screening.”
(4) Have the authors tried to co-crystalize hit compounds or hit analogs resulting from SAR with the TOM70 protein, to really observe the binding mode of these compounds? Only using in silico methods is not satisfactory.
We initially attempted to develop a crystallization system for Tom70 that would enable us to structurally resolve our bound compounds, but, at the time of writing the manuscript, had not identified a system that yielded crystals. However, another group has published a structure of apo-human Tom70 (https://pmc.ncbi.nlm.nih.gov/articles/PMC13075491/) - we sent them compounds and were unable to identify convincing density by soaking. We agree that a future study in a different crystal system or by pursuing co-crystallization could be fruitful.
(5) Although the authors improved the binding affinity of HTS hits for TOM70 using SAR, there is a lack of functional assays to evaluate whether the HTS hits or analogs could affect IFN inhibition caused by ORF9b or virus infection (cell-based or animal-based assays).
We have expanded the discussion section to include potential future experiments that could be conducted to probe this:
“Although we have not assessed the permeability of these compounds and their ability to restore IFN in the context of cell-based assays, one way this could be accomplished is by adapting a previously used reporter cell line. Specifically, this assay uses a reporter cell line that has a Lucia reporter under the control of the IFN-β/ISG56 promoter that is expressed upon mock infection with synthetic dsRNA. When WT Orf9b is co-transfected with the reporter, IFN is suppressed, and, we have shown that phosphomimetic mutants of Orf9b that block its interaction with Tom70 restores IFN signaling. If these compounds are sufficiently permeable, then, for the lipidated-compounds targeting the Orf9b homodimer, we would expect the Orf9b monomer:homodimer equilibrium to shift towards the homodimer due to the stabilizing effect these lipidated-compounds have on the homodimer conformation leading to an increase in IFN signaling. Similarly, for compounds targeting Tom70, we would expect that these compounds block the Orf9b monomer from binding to Tom70 to produce a similar effect.”
(6) Figure 4 doesn't have A, B, or C indicated in the figures.
We have revised Figure 4 to include A, B, and C.
Reviewer #2 (Recommendations for the authors):
(1) Page 5 line 23: The authors could comment on the biological relevance of the lipid-bound version of Orfb9 and how it could associate with membranes.
We thank the reviewer for their comments. Currently, it is unknown what role lipid-binding plays on Orf9b’s function beyond the stabilizing effect it has on the homodimer conformation. While it has been shown that Orf9b co-localizes at the outer mitochondrial membrane through Tom70 association, it has not been shown that Orf9b itself binds to membranes. While this does not exclude the possibility that the Orf9b homodimer could directly bind to membranes, we suspect a more plausible explanation is that Orf9b homodimers can bind to lipid-like metabolites within cells such as palmitate which we had previously shown.
(2) Figure 3F: These SPR sensorgrams are not all to the same y-axis scale.
We thank the reviewer for their comment. The sensorgram axis scale is reflective of three separate SPR experiments on three capture points for the ligand. Therefore, differences in ligand capture on each surface will produce slight differences in the total measured response. The axes are scaled to make interpretation easier. We are not quantitatively comparing between treatments, but noting the qualitative behavior of each individually as the concentrations change.
(3) Discussion: The authors could comment on the next steps, ligands ADME, solubility, permeability as the compounds are at low uM binding affinity, with increased concentrations, comment on selectivity/specificity.
Given that these are still “hits”, we think a discussion of these properties (which we study intently in the lab!) is premature and beyond the scope of this manuscript.
Reviewer #3 (Recommendations for the authors):
(1) The authors should carefully read the manuscript as there are numerous typos and duplicated words across the whole document and supplementary data.
We have gone through the manuscript and the supplementary material and corrected the typographical errors and duplicated words that we found, including a repeated sentence in the crystallization methods and a repeated word in the purification methods. We have also renamed the supplementary figures to the eLife figure supplement style. We expect copy editing at the VOR stage to catch anything that remains.
(2) Details of how Chai-1 was run, including constraints, should be given in the Materials and Methods section.
We have revised the methods section to describe how Chai-1 was used for both Orf9b and Tom70 co-folding predictions. See above.
(3) There was no comment on whether the compounds which were found were attempted to be co-crystallised. If they were, but negative data were collected (didn't crystallise, didn't diffract, or no additional density found), then this needs to be stated.
We have revised the manuscript to clearly point out that beyond our initial attempts to crystallize apo-Orf9b homodimer to enable fragment soaking, we also attempted to both soak our lipidated-compounds into apo-Orf9b homodimer crystals as well as co-crystallize the compounds which did not yield crystals.
“We attempted two strategies of crystallizing our lipidated-compounds with the apo-Orf9b homodimer to structurally resolve the complex. First, we attempted to soak our compounds into apo-Orf9b homodimer crystals but observed poor tolerance to DMSO which quickly disintegrated the crystals after minutes. Second, we attempted to co-crystallize the lipidated-compounds after refolding the Orf9b homodimer to generate apo-Orf9b homodimers but could not identify conditions that yielded crystals upon re-screening.”
eLife Assessment
By combining previously established mathematical models, this valuable study demonstrates that the opposite direction of surface contraction waves in Xenopus embryos and starfish oocytes can in principle arise from the same underlying mechanism. The modeling results, showing the effect of nuclear size and position, cell size, and nuclear Cdk1 concentration, are convincing and plausible but have not yet been experimentally verified.
Reviewer #1 (Public review):
[Editors' note: this version has been assessed by the Reviewing Editor without further input from the original reviewers.]
Summary:
The main goal of this manuscript is to develop a mathematical model of the regulation of cortical dynamics by Cdk1 activity to explain why, in some embryos (e.g., Xenopus), surface contraction waves are believed to move in the same direction as Cdk1, while in other embryos (e.g., starfish) they are believed to move in the opposite direction.
Strengths:
(1) The paper addresses a very important question.
(2) The mathematical model is sensible and suggests that the different relationship between Cdk1 and surface contraction waves might arise from the different behavior of the mitotic entry wave and the mitotic exit wave.
(3) The authors propose a mechanism by which the wave observed at mitotic exit might not passively follow the trigger wave observed at mitotic entry'
(4) The proposed mechanism is a potential explanation of the observed differences.
(5) The proposed mechanism is centered on different dynamics between the nucleus and the cytoplasm, highlighting the potential importance of the nucleus (and nuclear size) in organizing cortical dynamics.
Weaknesses:
(1) The proposed mechanism works if the activity in the nucleus is much higher than the high activity (high state of the bistable system) of the cytoplasm. So, as the wave propagates across the cytoplasm, the activity around the nucleus remains higher, which potentially causes a delay in the onset of Cyclin B-Cdk1 degradation in the region around the nucleus compared to the surrounding cytoplasm. This effect happens over a typical length scale, and if such a length scale is comparable to embryo size, this becomes the predominant mechanism. However, such a mechanism should exist near the nucleus independently of embryo size. So, it seems that for embryos where the wave back and wave front should travel together, nuclear activity must be adjusted not to be much higher than cytoplasmic activity. Is there any experimental evidence that the overall activity of Cdk1 is higher in the nucleus than in the cytoplasm?
(2) The authors give the existence of trigger waves as a fact. While the predominant view is that such waves exist in the first cycle of the Xenopus embryos (however, this is from measurement of the cortical contractions, so a bit circular for this paper), it is unclear if waves exist in the starfish embryo, so the potential explanation that the starfish embryo simply has different Cdk1 dynamics cannot be ruled out.
Reviewer #2 (Public review):
Summary:
Large oocytes show prominent waves of cortical contractions. Previous works combining experiments and computational modeling have shown that the waves are driven by gradients of CDK1 kinase activity that trigger excitable Rho activity patterns on the cortex. This present work combines two previously published mathematical models for CDK1 activation and Rho activation, respectively. They show that the models combined can explain diverse shapes of cortical contractions observed in different species and at various stages of development. This shows how the same molecular machinery can generate diverse patterns dependent on the size of the system and the size and position of the cell nucleus.
Strengths:
(1) Carefully done modeling work providing a simple and elegant explanation for a complex cellular behavior.
(2) Very nicely illustrated, simulations can be directly compared to previous experimental observations.
(3) Explains observations made in different model systems, providing a unifying model.
Overall, I find this work important, as it shows that combining models of the CDK1 gradient and Rho activation modules can explain the surface contraction waves observed in oocytes. Strikingly, it elegantly explains the differences seen between different experimental systems. While previously these were considered a 'controversy', modeling shows that the differences are simply a consequence of the difference in the size of the oocytes. In addition, the model makes several intriguing predictions that can be tested in future experiments.
Reviewer #3 (Public review):
Summary:
Using realistic mathematical models, Cebrián-Lacasa et al. address the relationship between waves of activation of Cyclin B-Cdk1 that propagate through the cytoplasm of large (~1 mm) oocytes and fertilized eggs and surface contraction waves (SCWs) driven by Rho GTPase activity in the cell cortex. They present numerical simulations of the underlying reaction-diffusion equations that account in broad strokes for both the expected behavior of 'fronts' of Cdk1 activation (that propagate at constant velocity from the nucleus-the source of Cdk1 activity-to the cell cortex) and the unusual behavior of 'backs' of Cdk1 inactivation (that may propagate either away from or towards the nucleus, or exhibit simultaneous inactivation throughout the cytoplasm). They also model Rho GTPase activity in the cortex as an excitable system that propagates SCWs (target patterns, spiral waves, and more complicated patterns). When Cdk1 is activated in the cortex, it phosphorylates and inhibits the RhoGEF, Ect1, which suppresses SCWs by reducing Rho GTPase activity. As the wave-back of Cdk1 inactivation moves across the cortex, Rho GTPase activity recovers abruptly, and SCWs reappear as 'phase waves' whose speed and directionality are determined by the underlying cytoplasmic Cdk1 signal.
Strengths:
As a theoretical examination of an interesting and puzzling aspect of early embryonic development, this study shares the same strengths and weaknesses as all mathematical and computational approaches to molecular cell biology. The mathematical models are precise formulations of the underlying assumptions of the authors (which are quite reasonable in this reviewer's opinion), and the analysis and computational results are dependable consequences of the molecular mechanisms the authors have in mind. The model is expertly analyzed, and the results are both reliable and intriguing. The results are discussed in light of experimental evidence. Because the authors' methods and results suggest novel-and sometimes counterintuitive-avenues for experimental research, this paper is likely to have a significant impact on the field of Rho GTPase signaling in oocytes and early embryos, and perhaps in other cells as well.
Weaknesses:
Like all mathematical models, the underlying assumptions can be critiqued as neglecting this -or-that 'crucial' effect (e.g., mechanical coupling via cortical tension or cytoplasmic flow, as the authors acknowledge), and the highly technical methods of analysis and simulation can be unfamiliar and off-putting to experimental cell biologists. For those who take the time to understand this paper, it may change the way they think about the coupling of cell cycle control (Cdk1 activation and inactivation) and cell surface contraction waves.
Author response:
The following is the authors’ response to the original reviews.
We thank the editors and reviewers for their careful and constructive assessment of our manuscript. Based on their comments and suggestions, we have revised the manuscript extensively to improve accessibility, clarify model novelty and the physical intuition behind the wave directionality, and correct and expand figures, units, equations, parameter definitions, and references.
The main changes in the revised manuscript are that:
- The abstract and main text now explicitly describe the work as coupling established reaction–diffusion models of Cyclin B–Cdk1 signaling and the excitable Rho–actin cortex, rather than presenting the individual modules as newly developed models.
- The governing equations, model assumptions, coupling, and key parameters are presented in the main text, with clarified units and a complete parameter table.
- Figures 1 and 2 were redesigned to make the geometry, kymographs, wave-front/wave-back definitions, and coexistence of travelling-wave-like and diffusive dynamics more intuitive.
- The diffusion-coefficient units and the conversion used to calculate total active nuclear Cyclin B–Cdk1 were corrected; the affected plots were regenerated and total active nuclear content is now defined explicitly in Eq. 7.
- The Introduction and Discussion now better distinguish established Cdk1–cortex coupling from the contribution of this study, incorporate work in Drosophila and zebrafish, and distinguish experimentally supported observations from model predictions.
- The cortical dynamical regimes and the role of Ect2 were clarified, supplementary figures were revised, and the suggested classic references were added.
eLife Assessment
This valuable study combines previously established mathematical models to investigate why cortical waves in starfish and Xenopus embryos propagate in opposite directions. The modeling results are solid and plausible, but remain experimentally untested. Improving the presentation and discussion of the results could make the study more accessible to a wider audience.
We appreciate this assessment and have focused the revision on exactly these points. The manuscript now makes the established origin of both model modules explicit, provides the core equations and parameters in the main text, gives a more intuitive explanation of the front–back mechanism, redesigns the key figures, and expands the discussion of experimental support, limitations, and testable predictions. The work remains theoretical, and we now state that limitation more clearly rather than implying experimental validation. See Abstract, Introduction, Model, and Discussion in the revised manuscript.
Public Reviews:
Reviewer #1 (Public review):
Summary:
The main goal of this manuscript is to develop a mathematical model of the regulation of cortical dynamics by Cdk1 activity to explain why, in some embryos (e.g., Xenopus), surface contraction waves are believed to move in the same direction as Cdk1, while in other embryos (e.g., starfish) they are believed to move in the opposite direction.
Strengths:
(1) The paper addresses a very important question.
(2) The mathematical model is sensible and suggests that the different relationship between Cdk1 and surface contraction waves might arise from the different behavior of the mitotic entry wave and the mitotic exit wave.
(3) The authors propose a mechanism by which the wave observed at mitotic exit might not passively follow the trigger wave observed at mitotic entry'
(4) The proposed mechanism is a potential explanation of the observed differences.
(5) The proposed mechanism is centered on different dynamics between the nucleus and the cytoplasm, highlighting the potential importance of the nucleus (and nuclear size) in organizing cortical dynamics.
We thank the reviewer for this positive assessment. The revision now makes the central question and the scope of the contribution more explicit: the individual Cdk1 and cortical modules were already previously established, while the present study asks how their coupling, together with cell geometry and nuclear Cdk1 content, can generate distinct wave-back dynamics and cortical responses. The revised manuscript now also foregrounds the mechanistic distinction between activation fronts and wave backs and uses this distinction to organize the Results and Discussion sections. See Abstract, Introduction, Results and Discussion in the revised manuscript.
Weaknesses:
(1) The proposed mechanism works if the activity in the nucleus is much higher than the high activity (high state of the bistable system) of the cytoplasm. So, as the wave propagates across the cytoplasm, the activity around the nucleus remains higher, which potentially causes a delay in the onset of Cyclin B-Cdk1 degradation in the region around the nucleus compared to the surrounding cytoplasm. This effect happens over a typical length scale, and if such a length scale is comparable to embryo size, this becomes the predominant mechanism. However, such a mechanism should exist near the nucleus independently of embryo size. So, it seems that for embryos where the wave back and wave front should travel together, nuclear activity must be adjusted not to be much higher than cytoplasmic activity. A better discussion of the discovered process and its implications would strengthen the paper. It requires careful reading to understand what, in hindsight, is a rather simple explanation. Is there any experimental evidence that the overall activity of Cdk1 is higher in the nucleus than in the cytoplasm?
We agree that the mechanism needed a more intuitive explanation. The revised Results now separate two simultaneous processes. The activation front propagates outward with an approximately constant trigger-wave-like velocity, whereas excess active Cyclin B–Cdk1 associated with the enriched nuclear pool also spreads diffusively. This diffusive component exists locally around the nucleus irrespective of cell size. What depends on effective system size is whether that diffusive length scale is large enough, relative to the cell, to shape the wave back over a substantial fraction of the domain and ultimately at the cortex. This is why nuclear content and effective system size act together. The revised Fig. 2B and the accompanying text now explain this competition explicitly. We also revised the Discussion to distinguish experimental plausibility from quantitative validation: active Cyclin B–Cdk1 has been observed to accumulate in nuclei before surrounding cytoplasmic activation in frog egg extracts, nuclei can specify where mitotic waves begin, and Cyclin B enrichment near the nuclear region has been reported in starfish oocytes. However, whether nuclear active Cyclin B–Cdk1 reaches the level required for complete wave-back reversal remains a model prediction, which we now state explicitly. See Results, Fig. 2B and caption (p. 6), Methods and Eq. 7 in the revised manuscript.
(2) While the fact that Cdk1 can enslave cortical dynamics is clearly shown in the model, this is expected from the literature. There are systems where the enslavement of cortical and bulk actomyosin contractility to Cdk1 activity has been more clearly demonstrated (Drosophila and zebrafish embryos), as well as shown to have clear functions (nuclear positioning and ooplasmic segregation).
We agree. The revised manuscript no longer presents Cdk1 control of cortical dynamics as a novel result. We expanded the introduction to discuss Drosophila and zebrafish and to note the established functions of Cdk1– actomyosin coupling in nuclear positioning and ooplasmic segregation. We now frame the contribution of the present work as explaining how established Cdk1-to-cortex coupling can transmit geometry-dependent cytoplasmic waveforms into different cortical directionality and reactivation regimes. See Introduction and Discussion in the revised manuscript.
(3) The writing could be improved. The authors make some claims of originality that seem a stretch, e.g., in the abstract, they say: "we develop a reaction-diffusion model of Cyclin B-Cdk1 signaling in spherical cells with localized nuclear activation", but they essentially use a previous model with a few numerical tweaks. The figures are sometimes mislabelled or not explained, and some of the units seem wrong.
We agree and revised the framing throughout. The Abstract now states that we “couple established reaction–diffusion models of Cyclin B–Cdk1 signaling and the excitable Rho–actin cortex.” The Introduction similarly states that we are not developing a new model, and a dedicated Model section now gives the equations, assumptions, coupling, and original sources of the two modules. We also revised figure labels and captions, added schematic panels, corrected the diffusion units to μm<sup>2</sup>/s, corrected the conversion used for total active nuclear Cyclin B–Cdk1, and regenerated the affected plots. The control variable is now defined explicitly as total active nuclear content in Eq. 7 and reported in pmol. See Abstract, Introduction, Model, Fig. 1 and caption (p. 3); Fig. 2 and caption (p. 6); Methods, especially Eq. 7 and Table 1 (p. 16) in the revised manuscript.
(4) The authors give the existence of trigger waves as a fact. While the predominant view is that such waves exist in the first cycle of the Xenopus embryos (however, this is from measurement of the cortical contractions, so a bit circular for this paper), it is unclear if waves exist in the starfish embryo, so the potential explanation that the starfish embryo simply has different Cdk1 dynamics cannot be ruled out.
We agree that the evidence is not equivalent across systems. The revised manuscript avoids treating a starfish intracellular Cdk1 trigger wave as an experimentally established fact and instead presents the cross-species comparison as an interpretation within the model framework. We distinguish the model prediction from the available experimental observations and explicitly state that the nuclear active Cyclin B–Cdk1 level required for complete waveback reversal remains to be tested. Thus, different underlying Cdk1 dynamics in starfish are not excluded by our study. See Introduction, Results and Discussion in the revised manuscript.
Reviewer #2 (Public review):
Summary:
Large oocytes show prominent waves of cortical contractions. Previous works combining experiments and computational modeling have shown that the waves are driven by gradients of CDK1 kinase activity that trigger excitable Rho activity patterns on the cortex. This present work combines two previously published mathematical models for CDK1 activation and Rho activation, respectively. They show that the models combined can explain diverse shapes of cortical contractions observed in different species and at various stages of development. This shows how the same molecular machinery can generate diverse patterns dependent on the size of the system and the size and position of the cell nucleus.
Strengths:
(1) Carefully done modeling work providing a simple and elegant explanation for a complex cellular behavior.
(2) Very nicely illustrated, simulations can be directly compared to previous experimental observations.
(3) Explains observations made in different model systems, providing a unifying model.
We thank the reviewer for this assessment. We have revised the manuscript so that this scope is explicit from the outset: the model modules are established, while the contribution is the analysis of their coupling and the identification of geometry, total nuclear Cdk1 content, and cortical recovery kinetics as controls that generate qualitatively different behavior within one framework. See Abstract, Introduction, Results and Discussion in the revised manuscript.
Weaknesses:
(1) Purely theoretical work, no experimental validation.
We agree that direct experimental validation would strengthen the work. We have not added a new experimental dataset in this revision. Instead, we have made the experimentally testable status of the conclusions explicit. The revised discussion identifies total nuclear active Cyclin B–Cdk1 content and effective system size as controls of wave-back directionality, states that the nuclear enrichment required for complete reversal remains to be tested, and identifies perturbation of actomyosin contractility while monitoring Cdk1/Rho dynamics as one way to probe the role of omitted mechanics. We therefore present the work as a theoretical framework that generates concrete predictions, not as experimentally validated proof of the mechanism. See Discussion section in the revised manuscript.
(2) Adopts previously published models more or less 'as is', without detailed re-evaluation and re-assessment, or without developing them further.
We agree that the individual modules are adopted without substantial re-derivation or extension, and the revised manuscript now mentions this directly. We now discuss the models in the main text, mentioning the Yang and Ferrell / Puls Cdk1 model and the Michaud et al. cortical model, giving the equations and coupling, and then the Methods sections states that parameter values are retained from the original publications. Our aim is not to claim a new formulation of either module, but to analyze what emerges when they are coupled in a common geometry. That analysis yields the front–back asymmetry, geometry-dependent reversal of the Cdk1 wave back, and distinct cortical reactivation regimes. See Introduction, Model, Methods, and Results in the revised manuscript.
Overall, I find this work important, as it shows that combining models of the CDK1 gradient and Rho activation modules can explain the surface contraction waves observed in oocytes. Strikingly, it elegantly explains the differences seen between different experimental systems. While previously these were considered a 'controversy', modeling shows that the differences are simply a consequence of the difference in the size of the oocytes. In addition, the model makes several intriguing predictions that can be tested in future experiments.
We thank the reviewer. In revising the manuscript, we were careful not to reduce the mechanism to cell size alone: the model shows that directionality depends jointly on effective cell size and total nuclear Cdk1 content, which is set by nuclear size and nuclear concentration. We now mention this combination consistently and make the associated predictions explicit in the Discussion. See Results and Discussion in the revised manuscript.
Reviewer #3 (Public review):
Summary:
Using realistic mathematical models, Cebrián-Lacasa et al. address the relationship between waves of activation of Cyclin B-Cdk1 that propagate through the cytoplasm of large (~1 mm) oocytes and fertilized eggs and surface contraction waves (SCWs) driven by Rho GTPase activity in the cell cortex. They present numerical simulations of the underlying reaction-diffusion equations that account in broad strokes for both the expected behavior of 'fronts' of Cdk1 activation (that propagate at constant velocity from the nucleus-the source of Cdk1 activity-to the cell cortex) and the unusual behavior of 'backs' of Cdk1 inactivation (that may propagate either away from or towards the nucleus, or exhibit simultaneous inactivation throughout the cytoplasm). They also model Rho GTPase activity in the cortex as an excitable system that propagates SCWs (target patterns, spiral waves, and more complicated patterns). When Cdk1 is activated in the cortex, it phosphorylates and inhibits the RhoGEF, Ect1, which suppresses SCWs by reducing Rho GTPase activity. As the wave-back of Cdk1 inactivation moves across the cortex, Rho GTPase activity recovers abruptly, and SCWs reappear as 'phase waves' whose speed and directionality are determined by the underlying cytoplasmic Cdk1 signal.
Strengths:
As a theoretical examination of an interesting and puzzling aspect of early embryonic development, this study shares the same strengths and weaknesses as all mathematical and computational approaches to molecular cell biology. The mathematical models are precise formulations of the underlying assumptions of the authors (which are quite reasonable in this reviewer's opinion), and the analysis and computational results are dependable consequences of the molecular mechanisms the authors have in mind. The model is expertly analyzed, and the results are both reliable and intriguing. The results are discussed in light of experimental evidence. Because the authors' methods and results suggest novel-and sometimes counterintuitive-avenues for experimental research, this paper is likely to have a significant impact on the field of Rho GTPase signaling in oocytes and early embryos, and perhaps in other cells as well.
We thank the reviewer for this thoughtful assessment. We have preserved the mathematical formulations and made the assumptions, mechanisms, and experimentally testable consequences easier to identify. In particular, the revised Results section emphasizes the distinct mechanisms governing the activation front and wave back, the revised figures provide a more intuitive visual explanation, and the Discussion separates experimentally supported observations from model predictions. See Model, Results, and Discussion in the revised manuscript.
Weaknesses:
Like all mathematical models, the underlying assumptions can be critiqued as neglecting this -or-that 'crucial' effect (e.g., mechanical coupling via cortical tension or cytoplasmic flow, as the authors acknowledge), and the highly technical methods of analysis and simulation can be unfamiliar and off-putting to experimental cell biologists. The paper is a difficult read, even for an experienced theoretician. For those who take the time to understand this paper, it may change the way they think about the coupling of cell cycle control (Cdk1 activation and inactivation) and cell surface contraction waves.
We agree on both accessibility and model scope. To improve accessibility, we added the core equations and parameters to the main text, introduced more qualitative descriptions of the dynamical regimes, and redesigned Fig. 2B to show the physical mechanism schematically. We also expanded the discussion of the limitations. We did not add mechanical coupling to the model in this revision; instead, the manuscript now states explicitly that showing biochemical signaling to be sufficient within the model does not demonstrate that cortical tension, cytoplasmic flow, or surface deformation are unnecessary in living cells. These effects are identified as additional important factors that can affect the dynamics and as targets for future perturbation experiments. See Model, Results, Fig. 2B and caption (p. 6), Discussion in the revised manuscript.
Recommendations for the authors:
Reviewer #1 (Recommendations for the authors):
(1) Carefully proofread the manuscript, Figures, and Figure Legends to make sure that statements are accurate. Please describe in more detail what is shown in every Figure panel. The units of diffusion in Figure 2C appear wrong, total activity is not really explained in the text, and it is a strange use, as the activity changes over time; the units of total activity are millimolar, which seems unrealistic. The Figure 1D heatmap does not explain what is plotted. Intensity is neither defined nor is it clear how the first wave front was drawn; the second appears to be convincing. The curvature of what physical object is plotted in Figure 1D, and again, how are the wavefronts on those Figures drawn?
We carefully reviewed the manuscript and all main and supplementary figures. Figure 1 was redesigned to show the model in 3D, 2D, and 1D and to make the coordinates used in the kymographs explicit. The revised Fig. 1 caption now states that the experimental color scales correspond to Cdk1-reporter intensity (top) and local membrane curvature (bottom), and that the dotted lines mark the tracked front and back. These overlays are intended as visual guides in the adapted experimental kymographs rather than quantitative fits used in our model analysis. Figure 2 was also redesigned and its caption expanded. The diffusion coefficient is now reported in μm<sup>2</sup>/s. In addition, we identified and corrected an error in the conversion of total active nuclear Cyclin B–Cdk1. The corrected control variable is total active nuclear content, reported in pmol and defined explicitly in Eq. 7; the affected main and supplementary plots were regenerated. See Fig. 1D and Fig. 1 caption (p. 3); Fig. 2 and caption (p. 6); Results, Methods, Eq. 7, Table 1 (p. 16) and Supplementary Figs. S3–S4 in the revised manuscript.
(2) Provide a better intuitive explanation of the proposed mechanism.
We substantially rewrote the Results section describing geometry-controlled directionality. The revised explanation first establishes the front–back asymmetry and then gives an intuitive description: a constant-speed activation front coexists with a slower diffusive spread of the elevated nuclear pool; because diffusion lags behind the front, distant regions can relax earlier, creating a gradient that drives an inward wave back. Revised Fig. 2B depicts these processes schematically. See Results, especially Fig. 2B and caption (p. 6) and Discussion in the revised manuscript.
(3) Reduce claims of novelty when not justified. I think that there is a potentially interesting idea in this paper, but also a lot of known facts that are revisited, and being unclear about what was known and what is new does not help the manuscript.
We agree. The revised Abstract and Introduction sections explicitly say that we couple established models rather than develop new individual modules. The Model section cites the original sources of each module, and the Introduction distinguishes established biological coupling from the question addressed here. We reserve the main contribution for the consequences of the coupled analysis: front–back asymmetry, the joint control by nuclear Cdk1 content and effective system size, and the resulting cortical response. The discussion also mentions the key unvalidated step as a testable prediction. See Abstract, Introduction, Model and Discussion in the revised manuscript.
(4) The interface of Cdk1 activity with actomyosin contractility (and cortical dynamics) is much better understood in other systems, and discussing those systems will improve the impact of the paper.
We expanded the Introduction to include Drosophila and zebrafish and to note the established functions of Cdk1-controlled actomyosin dynamics in nuclear positioning and ooplasmic segregation. This broader context makes clear that Cdk1–cortex coupling was already established and that our specific question is how cytoplasmic wave geometry shapes the downstream cortical response in large oocytes and eggs. See Introduction section in the revised manuscript.
(5) Be clearer about what the goal of this paper is and keep the focus on that. In my opinion, there is too much writing spent describing known facts.
We revised the Introduction to move more directly from the biological puzzle to the central question and then into the coupled model. The manuscript now explicitly states that we ask whether the opposite SCW directionality can follow from geometry and nuclear Cdk1 content rather than organism-specific cortical mechanisms, and then ask how the resulting cytoplasmic waveforms are transmitted to the cortex through Ect2 inhibition. We also reduced repeated background material and moved the core mathematical specification into a dedicated Model section. See Introduction, Model and Results in the revised manuscript.
(6) A discussion that the Cdk1 spatiotemporal patterns remain to be analyzed in both Xenopus and starfish embryos would be appropriate.
We added this limitation to the discussion in a way that distinguishes what is known from what remains to be measured. The revised text summarizes observations supporting nuclear enrichment and wave initiation but also says that whether nuclear active Cyclin B–Cdk1 reaches the levels required for complete wave-back reversal remains experimentally to be tested. We also avoid presenting the starfish intracellular Cdk1 wave as established fact. See Introduction, Results, and Discussion in the revised manuscript.
Reviewer #2 (Recommendations for the authors):
(1) I have one main suggestion. I would find it important that the authors present the models in more detail in the main text. In principle, even if they adopted previously published models, the reader should not need to go back to those original papers to find the equations and parameters. Readers should not even need to dig deep into the supplemental material. For example, the main equation used to model the CDK1 gradient should be shown in the main text with critical parameter values listed. For example, the text mentions that the diffusion coefficient of CDK1-Cyclin B has a critical importance (which I very much agree with), but I could not easily find the values used and still had to guess the units.
We agree. We added a dedicated model section to the main text containing the governing reaction-diffusion equations for both modules and the Cdk1–Ect2 coupling. The variables and roles of the principal parameters are defined there, while Table 1 lists the standard parameter values and units, including the diffusion coefficients used for the cytoplasmic regimes. The total active nuclear Cyclin B–Cdk1 control variable is defined explicitly in Eq. 7 in the Methods section. See Model, Methods, Table 1 (p. 16) and Eq. 7 in the revised manuscript.
(2) Overall, the manuscript is well written and nicely illustrated. However, the text could be more concise and, as explained above, show more equations rather than trying to describe them in words. Some of the illustrations could also be larger and possibly shown at higher contrast. Finally, the subtitles could be more imaginative, describing the main finding rather than simply stating what has been done.
We revised the manuscript throughout to reduce repeated background material and to replace lengthy verbal descriptions of the core model with equations in the main text. We standardized figure text and redesigned several panels, especially Figs. 1 and 2, to improve readability and interpretability. We did not replace every subsection title, but we have tried to use clear descriptive titles (with a message where possible). We revised several headings in the Results section so that they communicate the main finding directly, for example “Front–back asymmetry: traveling fronts and gradient-controlled backs,” “System size and diffusion modulate gradient effectiveness,” and “Phase diagram of Cdk1 waves.” See Introduction, Model, Results headings and text, Figs. 1– 2 and captions; supplementary figures in the revised manuscript.
(3) The starfish work focuses on meiosis I, not meiosis II.
We reviewed the stage-specific wording throughout. Rather than replacing every reference with “meiosis I,” we removed unnecessary stage labels where the argument concerns starfish oocyte meiosis more generally, and we use stage-specific wording only when it is supported by the cited experiment. In particular, Fig. 1 now labels the experimental panel simply as starfish meiosis rather than meiosis II. See Fig. 1D and caption (p. 3); Introduction and Results in the revised manuscript.
Reviewer #3 (Recommendations for the authors):
(1) Figure 1D. Right: experimental kymographs. What are 'intensity' and 'curvature'? I suppose 'intensity' refers to Cdk1 activity and 'curvature' to local curvature of the cell membrane. If so, Cdk1 activity seems to propagate as a wave around the cortex from the animal pole, but (I suppose) these are phase waves rather than trigger waves, because the nucleus is closer to the animal pole, so the cytoplasmic trigger waves reach the cortex sooner there and later at larger values of phi. In the starfish oocyte, the curvature 'bumps' at the animal pole and the equator seem to be stationary. Maybe I am misinterpreting the data; in any case, it might help readers to guide their reading of this data.
We clarified both the labels and the interpretation. The revised Fig. 1 caption explains the plotted experimental quantities as Cdk1-reporter intensity (top) and local membrane curvature (bottom), and mentions that the dotted lines mark the tracked front and back. We also distinguish the cytoplasmic propagation from the cortical phase-wave readout. Because the nucleus is off-center, the distance from the nuclear source to the cortex varies with angular position, so the radial Cdk1 solution is sampled at different times around the cortex. The Numerical methods now state explicitly that this geometry produces the angle-dependent arrival times shown in Fig. 1D. See Fig. 1D and Fig. 1 caption (p. 3); Model geometry and Numerical methods in the revised manuscript.
(2A) In panels (ii) and (iii), what is the meaning of the hyperbolic dotted line labelled 'underlying diffusion'?
The original schematic was replaced by the reworked Fig. 2B. The revised schematic and results now separate the outward trigger-wave-like front from the slower diffusive spread of the elevated nuclear pool. The “underlying diffusion” component is therefore no longer an unexplained fitted-looking curve; it is presented as the diffusive process that establishes the gradient controlling the wave back. See Fig. 2B and caption (p. 6) and Results in the revised manuscript.
(2B) This panel is very confusing. What is the significance of all the different colored lines? What is the meaning of '◄Time'? This panel needs more explanation. Also, fix 'increaseded'.
We agree that this was confusing. We now redrew Fig. 2B as a two-dimensional schematic with a consistent activity color scale and explicit labels for the phases and dynamical processes. The ambiguous time-arrow presentation was removed, and the typographical error was corrected. See Fig. 2B and caption (p. 6) and Results in the revised manuscript.
(2C) Units of diffusion constant are wrong; should be μm2/s. Also, please explain (perhaps in context of Figure S3) how you calculate 'units of active nuclear CycB-Cdk1 (mmol)' in terms of total [CycB-Cdk1] in the nucleus, radius of the nucleus, and model parameter γ.
The diffusion coefficient is now reported consistently in μm<sup>2</sup>/s. We also identified and corrected an error in the conversion to total nuclear amount. Eq. 7 now defines the total active nuclear Cyclin B–Cdk1 content from the enriched nuclear active concentration and nuclear volume; γ scales the nuclear concentration relative to the cytoplasmic baseline. The corrected quantity is reported in pmol, not mmol, and the corresponding main and supplementary plots were regenerated. Supplementary Fig. S3 shows that varying γ or nuclear radius collapses onto this single control variable. See Fig. 2D and caption (p. 6); Results, Methods, (Eq. 7), Table 1 (p. 16) and Supplementary Fig. S3 in the revised manuscript.
(2D) Please explicitly indicate the speed of the wave front, because it is difficult to read, given the scale of the vertical axis.
Agreed. The revised Fig. 2 caption now explicitly states that the wave-front velocities are fitted by a quasiconstant velocity of approximately 23 μm/min. See Fig. 2D caption (p. 6) and Results in the revised manuscript.
(3) Box 2. C. You might indicate that the Rho-actin dynamics are oscillatory for 0.59 < α < 1.02 and excitable for α slightly larger than 1.02 and (I suppose) slightly smaller than 0.59. Furthermore, I suppose (given the results in Figure 4) that the system does not propagate waves for α {less than or equal to} 0.5. Somehow, the authors should indicate how wave propagation is suppressed for small values of α.
We clarified the bifurcation structure and the low-α suppression of cortical activity. We did not adopt the proposed regime labels verbatim, because the spatial model supports propagating waves across a broader active window than the homogeneous oscillatory interval alone would suggest. The revised text therefore reports the two Hopf bifurcations at α = 0.59 and α = 1.02, states directly that the cortex is quiescent for low α, and reports that propagating waves persist across 0.59 < α < 1.26 up to the wave instability at α = 1.26. This directly explains how sufficiently strong Cdk1-mediated reduction of effective Ect2 suppresses propagation. See Box 2C (p. 9) and Results in the revised manuscript.
(4) Lines 273 and 294. "These differences arise...".
Corrected. The relevant Results text was rewritten during the broader restructuring, so the original wording no longer appears. See Revised Results.
(5) Lines 380, 381. "...the Hopf bifurcation (at α = 0.59) that separates...".
We did not retain the sentence in its original form. Instead, the revised text now identifies both Hopf bifurcations explicitly (α = 0.59 and α = 1.02), states the wave-propagating window, and then describes how Cdk1derived inhibition moves the cortical system across the bifurcation during suppression and recovery. We believe this is clearer than adding the parenthetical phrase to the old sentence. See Results and Box 2C (p. 9) in the revised manuscript.
(6) Table 1. Include a value for α_0.
Done. Table 1 now lists α<sub>0</sub> = 1 (dimensionless), and the Methods state that α = α<sub>0</sub> = 1 before nuclear envelope breakdown. See Table 1 (p. 16) and Methods in the revised manuscript.
(7) Line 522. What does 'implemented manually' mean?
We replaced the ambiguous wording. The Numerical methods now state that both models were integrated with an explicit second-order Runge–Kutta method implemented in custom code; the cytoplasmic and two dimensional cortical simulations use finite differences, and the three-dimensional cortical simulations use a finite element method implemented in FEniCSx. See Numerical methods in the revised manuscript.
(8) Figure S1. A. The first green dot should be labelled γ = 1. B. Left panel is γ = 1, I suppose.
Corrected. Revised Supplementary Fig. S1 explicitly labels the baseline γ = 1 condition and the enriched γ = 1.5 condition. Its caption also explains that γ scales the nuclear concentration above the cytoplasmic baseline while preserving the active-to-total ratio. See Supplementary Fig. S1 and caption; Model in the revised manuscript.
(9) Figure S2. What is X_0? How is it related to γ?
We addressed this by removing the X<sub>0</sub> notation rather than adding another conversion. The revised supplementary analysis uses γ directly as the nuclear enrichment factor, defined in the main Model section and illustrated in Supplementary Fig. S1. The revised Supplementary Fig. S2 now focuses on the fitted wave-front and wave-back trajectories for different γ values. See Model, lines 123–132; Supplementary Figs. S1–S2; Methods, lines 620–638.
(10) References. Emphasis is on post-2010 literature. One classic paper on SCWs is mentioned: Hara (1971). The classic publication on Cdk1-driven mitotic waves in fruit fly embryos is Foe & Alberts (1983) J Cell Sci 61:31. The first theoretical prediction of Cdk1 trigger waves in syncytial cells was given by Novak & Tyson (1993) J Theor Biol 165:101. Citing Deneke et al. (2016) Dev Cell 38(4) could also be considered.
We added the suggested literature. Novak & Tyson (1993) is now cited for the early theoretical prediction of trigger-wave behavior; Foe & Alberts (1983) is cited in the discussion of cell-cycle/cortical waves; and Deneke et al. (2016) is cited in the Introduction both in the mitotic-wave context and in the broader discussion of Cdk1–actomyosin coupling. See Introduction and reference list in the revised manuscript.
eLife Assessment
Using genome databases, the authors performed solid bioinformatic analyses to trace the genomic history of the clinically relevant Staphylococcus aureus tetracycline resistance plasmid pT181 over the last seven decades. They discovered that this element has transitioned from a multicopy plasmid to a chromosomally integrated element, and the work represents a valuable demonstration of the use of publicly available data to investigate plasmid biology and inform clinical epidemiology. This work will appeal to researchers interested in staphylococcal evolution, plasmid biology, antimicrobial resistance and bacterial genome structure.
Reviewer #1 (Public review):
[Editor's note: this version has been assessed by the Reviewing Editor without further input from the original reviewers.]
The study provides a robust bioinformatic characterization of the evolution of pT181. My main criticism of the work is the lack of experimental validation for the hypotheses proposed by the authors.
Previous comments on the study:
(1) One potential reason for the decline in pT181 copy number over time may be a high cost associated with the multicopy state. In this sense, it would be interesting if the authors could use (or construct) isogenic strains differing only in the state of the plasmid (multicopy/integrated). With this system, the authors could measure the fitness of the strains in the presence and absence of tetracycline, and they could be able to understand the benefit associated with the plasmid transition. The authors discuss these ideas, but it would be nice to test them.
(2) It would be interesting to know the transfer frequencies of the multicopy mobilizable pT181 plasmid, compared to the transfer frequency of the plasmid integrated into the SSCmec element (which can be co-transferred, integrated in conjugative plasmids, or by transduction).
(3) One important limitation of the study that should be mentioned is that inferring pT181 PCN from whole-genome data can be problematic. For example, some DNA extraction methods may underestimate the copy number of small plasmids because the small, circular plasmids are preferentially depleted during the process (see, for example, https://www.nature.com/articles/srep28063).
Comment from the Reviewing Editor:
In the revised version the authors have addressed comment 3 from Reviewer #1 through additional analysis, finding that the decline in copy number over time remains significant after controlling for instrument used, type of DNA extraction technique, and sequencing centre. The suggestion in Comment 1 from Reviewer #1 has not been acted upon as the authors state that this additional experimental work "goes beyond the scope of the paper" and is being explored for a future study. Similarly, the suggestion in Comment 2 from Reviewer #1 has not been acted upon as the authors state that "inferring these rates from natural sequence data is not feasible in this case given the low heterogeneity of the plasmid sequence. A laboratory-based experimental study could not address the real transfers we observe over the course of decades, as in vitro S. aureus transfer rates are often not good proxies for in vivo".
Reviewer #2 (Public review):
Summary:
The authors performed bioinformatic analyses to trace the genomic history of the clinically relevant pT181 plasmid. Specifically, they:
(1) tracked the presence of pT181 across different S. aureus strain backgrounds through time. It was first found in one, later multiple strains, though this may reflect changes in sampling over time.
(2) estimated the mutation rate of the chromosome and plasmid.
(3) estimated the plasmid copy number of pT181, and found that it decreased over time. The latter was supported by two sets of statistical analyses, first showing that the number of single-copy isolates increased over time, and second, that the multicopy isolates demonstrated a lower PCN over time.
(4) reported the different integration sites at which pT181 integrated into the genome.
As a caveat, they mentioned that identical plasmid sequences have variable plasmid copy numbers across different genomes in their dataset.
Strengths:
This is a very solid, well-considered bioinformatic study on publicly available data. I greatly appreciate the thoughtful approach the authors have taken to their subject matter, neither over- nor underselling their results. It is a strength that the authors focussed on a single plasmid in a single bacterial species, as it allowed them to take into account unique knowledge about the biology of this system and really dive deep into the evolution of this specific plasmid. It makes for a compelling case study. At the same time, I think the introduction and discussion can be strengthened to demonstrate what lessons might be drawn from this case study for other plasmids.
Weaknesses:
The finding that the pT181 copy number declined over time is the most interesting claim of the paper to me, and not something that I have seen done before. While the authors have looked at some confounders in this analysis, I think this could be strengthened further in a revision.
For the flow of the storyline, I also think the estimation of mutation rates (starting L181) and integration into the chromosome (starting L255) could be moved to the supplement or a later position in the main text.
Clearly, the use of publicly available data prevents the authors from controlling the growth and sequencing conditions of the isolates. It is striking that they observe a clear signal in spite of this, but I would have loved to see more discussion of the metadata that came with the publicly available sequences and even more use of that metadata to control for confounding.
Comment from the Reviewing Editor:
Reviewer #2's comment about the weakness due to "confounders in this analysis" has been addressed in the revision through additional analysis using metadata to control for confounding.
Author response:
The following is the authors’ response to the original reviews.
eLife Assessment:
Using genome databases, the authors performed solid bioinformatic analyses to trace the genomic history of the clinically relevant Staphylococcus aureus tetracycline resistance plasmid pT181 over the last seven decades. They discovered that this element has transitioned from a multicopy plasmid to a chromosomally integrated element, and the work represents a valuable demonstration of the use of publicly available data to investigate plasmid biology and inform clinical epidemiology. This work will appeal to researchers interested in staphylococcal evolution and plasmid biology.
Thank you, we agree with this overview. We also think this work is interesting to people interested in antimicrobial resistance and bacterial genome structure.
Public Reviews:
Reviewer #1 (Public review):
The study provides a robust bioinformatic characterization of the evolution of pT181. My main criticism of the work is the lack of experimental validation for the hypotheses proposed by the authors.
Comments on the study:
(1) One potential reason for the decline in pT181 copy number over time may be a high cost associated with the multicopy state. In this sense, it would be interesting if the authors could use (or construct) isogenic strains differing only in the state of the plasmid (multicopy/integrated). With this system, the authors could measure the fitness of the strains in the presence and absence of tetracycline, and they could be able to understand the benefit associated with the plasmid transition. The authors discuss these ideas, but it would be nice to test them.
We agree that the relative fitness of integrated versus multicopy plasmids is interesting and a costly multicopy state could explain the transition of independent pT181 replicons to chromosomal integration. This is a project we are exploring for a future study. However, we think that this additional experimental work goes beyond the scope of the paper.
(2) It would be interesting to know the transfer frequencies of the multicopy mobilizable pT181 plasmid, compared to the transfer frequency of the plasmid integrated into the SSCmec element (which can be co-transferred, integrated in conjugative plasmids, or by transduction).
We agree with the reviewer that this is an interesting question. However, we think inferring these rates from natural sequence data is not feasible in this case given the low heterogeneity of the plasmid sequence. A laboratory-based experimental study could not address the real transfers we observe over the course of decades, as in vitro S. aureus transfer rates are often not good proxies for in vivo (McCarthy et al.). In addition, we do not know what is moving the integrated plasmid. pT181 could be moved by a phage or plasmid, so we are uncertain what the correct experiment would be to explore this.
(3) One important limitation of the study that should be mentioned is that inferring pT181 PCN from whole genome data can be problematic. For example, some DNA extraction methods may underestimate the copy number of small plasmids because the small, circular plasmids are preferentially depleted during the process (see, for example, https://www.nature.com/articles/srep28063).
We have further investigated the role of technical variation in influencing measured PCN. The decline in copy number over time within our longest-collected strain remains significant after controlling for a number of technical variables: instrument used, type of DNA extraction technique, and sequencing center. In addition, plasmid copy number estimations between extraction techniques are not significant when comparing contemporaneously collected samples. We have added these supporting analyses to our results (L328-333) and methods (L612-634).
Reviewer #2 (Public review):
Summary:
The authors performed bioinformatic analyses to trace the genomic history of the clinically relevant pT181 plasmid. Specifically, they:
(1) tracked the presence of pT181 across different S. aureus strain backgrounds through time. It was first found in one, later multiple strains, though this may reflect changes in sampling over time.
(2) estimated the mutation rate of the chromosome and plasmid.
(3) estimated the plasmid copy number of pT181, and found that it decreased over time. The latter was supported by two sets of statistical analyses, first showing that the number of single-copy isolates increased over time, and second, that the multicopy isolates demonstrated a lower PCN over time.
(4) reported the different integration sites at which pT181 integrated into the genome.
As a caveat, they mentioned that identical plasmid sequences have variable plasmid copy numbers across different genomes in their dataset.
Strengths:
This is a very solid, well-considered bioinformatic study on publicly available data. I greatly appreciate the thoughtful approach the authors have taken to their subject matter, neither over- nor underselling their results. It is a strength that the authors focussed on a single plasmid in a single bacterial species, as it allowed them to take into account unique knowledge about the biology of this system and really dive deep into the evolution of this specific plasmid. It makes for a compelling case study. At the same time, I think the introduction and discussion can be strengthened to demonstrate what lessons might be drawn from this case study for other plasmids.
Weaknesses:
The finding that the pT181 copy number declined over time is the most interesting claim of the paper to me, and not something that I have seen done before. While the authors have looked at some confounders in this analysis, I think this could be strengthened further in a revision.
We have investigated this issue further in revisions. The decline in copy number over time within our longest-collected strain remains significant after controlling for a number of technical variables. In addition, plasmid copy number estimations between extraction techniques are not significant. We have mentioned these updates in the results (L327-333) and provided more explanation for the extraction technique in the methods (L612 - 634).
For the flow of the storyline, I also think the estimation of mutation rates (starting L181) and integration into the chromosome (starting L255) could be moved to the supplement or a later position in the main text.
We have chosen to keep the structure of the results largely the same.
Clearly, the use of publicly available data prevents the authors from controlling the growth and sequencing conditions of the isolates. It is striking that they observe a clear signal in spite of this, but I would have loved to see more discussion of the metadata that came with the publicly available sequences and even more use of that metadata to control for confounding.
We have further investigated the role of technical variation in influencing measured PCN. The decline in copy number over time within our longest-collected strain remains significant after controlling for a number of technical variables: instrument used, type of DNA extraction technique, and sequencing center. In addition, plasmid copy number estimations between extraction techniques are not significant when comparing contemporaneously collected samples. We have added this to our results section (L327-333) and explain further in the methods (L612 - 634). Unfortunately, the metadata is incomplete for many isolates, limiting our ability to control for other possible confounders.
Recommendations for the authors:
Reviewer #1 (Recommendations for the authors):
I would recommend doing some experiments to test the fitness effects of pT181 in multicopy and integrated into an isogenic S. aureus strain. Simple growth curves with and without tetracycline would be extremely informative.
We think that this is an interesting scientific question but is beyond the scope of this paper.
Reviewer #2 (Recommendations for the authors):
Major:
(1) L160 and Figure 1:
To what extent is this decrease in pT181 prevalence related to a delay between isolate collection and sequence upload to public databases? In my own work, I have found an average of 2 years difference in the sequence collection and sequence upload date of GenBank sequences. This could explain the decreasing numbers in recent years and may possibly skew relative proportions of pT181 containing/or plasmid-free strains.
There are fluctuations in collection rates over the years which definitely do not correspond to actual fluctuations in S. aureus prevalence in the population. We would expect delays in uploads in both plasmid-containing and plasmid-free strains, so we would not anticipate a decline in estimated plasmid prevalence due to this delay; however, it could be somewhat noisy for more recent years.
The large peak of S99.5_27 around 2009 seems oddly reminiscent of targeted sampling during a hospital outbreak, or so? Did you do any checks on where these genomes came from? Could this affect any of the downstream results?
The 2008 peak in S99.5_27 comes from a single BioProject (PRJEB4140), which contributed approximately 1,000 samples to our analysis. Based on the BioProject description, this BioProject aimed to characterize diversity in S. aureus ST239 diversity in Thailand. We performed the single-copy analysis with this strain excluded and still got the same result that integration has increased over time, so we feel that this is acceptable and should not unduly impact downstream results. We have clarified the results section to more explicitly state that this is the sampling peak that was excluded (L306-311).
(2) Figure 3: If I understand the analysis correctly, all mutation rate and phylogenetic analyses were conducted based on comparison to a single reference genome. This can bias results: https://academic.oup.com/mbe/article/31/5/1077/2925634 . Please comment.
Yes, SNPs were calculated relative to a reference genome. We understand this could bias results. However, it is standard practice in the field and we do not think it changes the main findings of the paper.
We have taken the reference-free SKA alignment and the SKA alignment mapping to a reference plasmid and calculated the SNP distances between all plasmid pairs with the tool snp-dists (Seemann et al.). For given plasmid pairs, the distributions are different by Wilcoxon signed-rank test with continuity correction (p < 2.2 x 10<sup>-16</sup>) though this is due in part to the large sample size for pairwise comparison. The median and modal difference between methods is 0 SNPs detected. The average is 0.1 more SNPs detected in the reference-free method than the mapping method. We have added this analysis to the methods section (L505-511).
(3) I am not fully convinced by the analysis in Figure 6 (i.e. that the pT181 copy number has declined over time). It is rather confusing how the authors account for strain and bioproject-related influences on their primary regression (L298-306).
We have edited the explanation of the GLMM in the results section (L320-323). We have moved some of the explanation to the methods section (L592-605).
(a) What does this analysis look like without S99.5.2?
The slope is still negative and significant with S99.5_2 excluded. We have added this to the results (L327-328).
(b) In line 296, it is made clear that 2 separate analyses were carried out, but in the subsequent paragraphs (starting 298 and 315), it would be helpful to repeat that each looks at only one component.
We have edited this section to explicitly state that separate regressions were carried out for each component (L298-300).
(c) Figure 6 should include error bars that reflect the different number of samples per collection year (please also comment on how this influences the power/validity of the linear regression!).
We have modified Fig. 6A to show samples collected per year as the color of the point. Samples collected per year can also be viewed in Fig. 1. In addition, we have changed to logistic regression to assess single-copyness as a function of collection year. The error bar on the regression curve shows the SE range. While there are many more samples collected in recent times than in the more distant past, we still think the signal is real. It does put more emphasis on each early sample. However, the genomic data matches early descriptions of tetracycline resistance in S. aureus being plasmid-mediated, not chromosomal (Gillespie et al.).
(4) Strains that were isolated in the 60s were presumably sequenced at a much later point in time. Were they grown and isolated from their respective host bacteria following similar protocols as present-day isolates? Growth conditions can have massive effects on PCN, and selective PCR amplification on the relative number of plasmid-related reads. How do you account for this? [Figure 7/8].
Yes, samples that were collected in the 1960s were sequenced roughly 50 years later. The date of collection could impact storage conditions for the isolate. However, they were grown up for sequencing and actually sequenced much closer together and should have the same growth conditions as isolates collected at a more recent date. Underlying our model of PCN is that it is a property of the isolate and will not fluctuate over time in storage.
Minor:
(1) This study is a nice demonstration of the type of genomic epidemiology that can be conducted now that genomes are sequenced and shared more widely. This is already somewhat common for clinically relevant sequence types of pathogenic bacteria, but I have not seen it much for plasmids. This could be nice to highlight. Stressing this more clearly in the abstract and introduction may broaden the audience of this work.
We have described our approach as genomic epidemiology in the first paragraph of the introduction (L72).
(2) More generally, the introduction is quite winding. A lot of interesting information is mentioned, but without a clear narrative structure as to why this is important for the study. Some editing may strengthen the message of the paper and invite more people to read it. For instance, starting each paragraph with a topical sentence about the general message the paragraph should convey. Yet, this is entirely a style/personal taste remark, and I leave it to the authors to decide whether to make these changes.
We have made edits to restructure the text.
(2a) Similar comment for the discussion (possibly some discussion could be put in the supplement too).
We have made edits to restructure the text, but have chosen to leave the content in the Discussion section. Additional references to previous research have also been added (L386-388).
(3) L125: Could you include a reference for this statement?
We have added a reference for this statement (L109).
There is interesting existing work on the evolution of genes on multi-copy plasmids (and the evolution of copy-number itself), which seems more pertinent here than the segregation mechanisms of low-copy-number plasmids.
We have chosen to keep the general background on plasmid copy number regulation and segregation, to keep the paper approachable to readers who have no background on plasmid biology. We have added additional references to work on evolvability of multicopy plasmids and their copy number (L117-120).
(4) The references are a bit of a mess. Certain entries are in all caps, others are missing titles, etc. I would recommend looking this over again carefully.
We have resolved this issue.
(5) L187 CDS is used in line 187 but not defined as a term until line 201.
We have defined CDS with first usage (L188).
(6) L189: Do you have any estimate of the error around 1.15 mutations/Mb/year from the linear regression?
We have added standard error for the 1.15 mut/Mbp/year from the regression to the results (L191): SE = 0.3 mut/Mbp/year.
(7) Figure 2: It would be helpful to have a visual indication of which areas the authors consider hypervariable.
We have added a black dash underneath the hypervariable regions on the plot.
(8) Figure 5: The two colors of grey are hard to distinguish.
The grey for shared genes has been made darker to better distinguish them.
(9) L285: Why do you write "could assume the copy number of the larger element"? What other options are possible?
We have expanded on this section to elaborate on the ways that copy number could be changed by cointegration, and added an additional citation to support these claims (L276-280).
(10) Figure S1: What does this scale mean? What are the units? This figure is uninformative as it stands. I would possibly consider labelling the tips.
We have modified the figure and legend to indicate that the scale is in substitutions per site. We are providing these figures to give some idea of scale and shape, showing how star-like they are. Unfortunately, tip labels are illegible on these trees given the large number of samples.
References
Gillespie, M. T., et al. “Antibiotic Susceptibilities and Plasmid Profiles of Nosocomial Methicillin-Resistant Staphylococcus aureus: A Retrospective Study.” Journal of Medical Microbiology, vol. 17, no. 3, June 1984, pp. 295–310. DOI.org (Crossref), https://doi.org/10.1099/00222615-17-3-295.
McCarthy, Alex J., et al. “Extensive Horizontal Gene Transfer during Staphylococcus aureus Co-Colonization in Vivo.” Genome Biology and Evolution, vol. 6, no. 10, 2014, pp. 2697–708, https://doi.org/10.1093/gbe/evu214.
Seemann, Torsten, et al. Snp-Dists. https://github.com/tseemann/snp-dists.
eLife Assessment
This important manuscript reports exceptionally preserved, millimeter-scale segmented bilaterian microfossils from the early Cambrian Fortunian Kuanchuanpu Formation in South China. The authors propose that these non-biomineralized animals were preserved through microbial pseudomorphs rather than direct phosphatization, filling a critical taphonomic and evolutionary gap in early animal evolution. The physical preservation and high-resolution imaging of the fossils are convincing, but the theoretical grounds for establishing "true segmentation" versus superficial annulation/homonomy remain underdeveloped and need revision and further context. The new fossils and their taphonomy will be of broad interest to the paleontology community and those interested in early animal evolution.
Reviewer #1 (Public review):
Summary:
This manuscript reports exceptionally preserved, millimeter-scale segmented bilaterian microfossils from the Fortunian Kuanchuanpu Formation in South China. The authors propose that these non-biomineralized animals were preserved through microbial pseudomorphs rather than direct phosphatization, filling a critical taphonomic and evolutionary gap in early animal segmentation.
Strengths:
This paper filled the critical taphonomic and evolutionary gap in early animal segmentation.
Weaknesses:
Extremely similar work has already been published by Xian et al., 2026, PNAS, with specimens from absolutely the same formation, and they also proposed that these animals are segmented. So in this context, this work's significance is eclipsed.
Reviewer #2 (Public review):
Summary:
The manuscript by Yang and authors presents an interesting set of new microfossils from the early Cambrian of South China. These are interpreted as segmented bilaterians, and so may be the oldest such examples in the fossil record, with implications for our understanding of early animal evolution. Further, the preservation of these fossils as microbial pseudomorphs is a unique pathway to fossilize such important organisms.
Strengths:
The fossils are impressive and arguments for their affinities as segmented bilaterians are convincing. Discussion of somewhat novel preservation is detailed and helpful to understand this taphonomic mode.
Weaknesses:
It needs to be clear what the authors' definition of a "segment" is and how that matches (or not) other interpretations. Arguments by Hannibal and Patel (2013, EvoDevo) convincingly define all segmented organisms based on polarity within repeated units of bilaterally symmetrical animals, but others may disagree. The manuscript would benefit from discussion of this in the introduction, laying out a hypothesis for what we might expect to find as evidence of the earliest segmented animals and then describing these new fossils within that context.
As it currently reads, the definition of a segment is not presented until the discussion and is somewhat ambiguous, seemingly referring to any organism with bilateral symmetry and repeated units. However, modern (e.g. tardigrades) non-segmented animals exhibit these features but are not considered truly segmented. This is critical as various Precambrian fossils (Arborea (Dunn et al., 2019, Palaeontology); Cloudina (Schiffbauer et al., 2020, Nature Communications) Dickinsonia (Gold et al., 2015, Evolution and Development); Parvancorina (Darroch et al., 2017, Biology Letters); Spriggina (Evans et al., 2026, Scientific Reports); Yilingia (Chen et al., 2019, Nature) etc.), have been argued to possess bilateral symmetry ad have bodies composed of repeated units. These older examples should be acknowledged and discussed in detail. Stating that "the identity of these [Precambrian] organisms is controversial" (line 51) is not justification for the main claims of this paper, and it should be clear what distinguishes the new fossils from previously proposed evidence of earlier segmented bilaterians. I believe there is a case to be made in support of the authors' claims, but this context is vital to make that argument convincingly. It is also worth discussing evidence of segmented bilaterians in the early Cambrian trace fossil record, such as Treptichnus pedum (Kesidis et al., 2019, Proceedings B) and Rusophycus (Daley et al., 2018 PNAS), that may predate or be roughly coeval with these fossil remains.
Author response:
Reviewer #1 (Public review):
Summary:
This manuscript reports exceptionally preserved, millimeter-scale segmented bilaterian microfossils from the Fortunian Kuanchuanpu Formation in South China. The authors propose that these non-biomineralized animals were preserved through microbial pseudomorphs rather than direct phosphatization, filling a critical taphonomic and evolutionary gap in early animal segmentation.
Strengths:
This paper filled the critical taphonomic and evolutionary gap in early animal segmentation.
Weaknesses:
Extremely similar work has already been published by Xian et al., 2026, PNAS, with specimens from absolutely the same formation, and they also proposed that these animals are segmented. So in this context, this work's significance is eclipsed.
Thanks the reviewer for the constructive comments. In Xian et al., 2026, PNAS, the taphonomic mechanism has not been explained in details and in our manuscript we propose the mode that how microbes pseudomorph the entire bodies of micro-animals. In addition, the specimens we presented in this manuscript demonstrate the abundance and diversity of segmented bilaterians in the basal Cambrian. In the revision, we will majorly discuss the evolution of body segment based on our new materials and add a new figure to complement the comparison of microbial pseudomorphs vs Ediacaran-type and Burges-Shale type.
Reviewer #2 (Public review):
Summary:
The manuscript by Yang and authors presents an interesting set of new microfossils from the early Cambrian of South China. These are interpreted as segmented bilaterians, and so may be the oldest such examples in the fossil record, with implications for our understanding of early animal evolution. Further, the preservation of these fossils as microbial pseudomorphs is a unique pathway to fossilize such important organisms.
Strengths:
The fossils are impressive and arguments for their affinities as segmented bilaterians are convincing. Discussion of somewhat novel preservation is detailed and helpful to understand this taphonomic mode.
Weaknesses:
It needs to be clear what the authors' definition of a "segment" is and how that matches (or not) other interpretations. Arguments by Hannibal and Patel (2013, EvoDevo) convincingly define all segmented organisms based on polarity within repeated units of bilaterally symmetrical animals, but others may disagree. The manuscript would benefit from discussion of this in the introduction, laying out a hypothesis for what we might expect to find as evidence of the earliest segmented animals and then describing these new fossils within that context.
As it currently reads, the definition of a segment is not presented until the discussion and is somewhat ambiguous, seemingly referring to any organism with bilateral symmetry and repeated units. However, modern (e.g. tardigrades) non-segmented animals exhibit these features but are not considered truly segmented. This is critical as various Precambrian fossils (Arborea (Dunn et al., 2019, Palaeontology); Cloudina (Schiffbauer et al., 2020, Nature Communications) Dickinsonia (Gold et al., 2015, Evolution and Development); Parvancorina (Darroch et al., 2017, Biology Letters); Spriggina (Evans et al., 2026, Scientific Reports); Yilingia (Chen et al., 2019, Nature) etc.), have been argued to possess bilateral symmetry ad have bodies composed of repeated units. These older examples should be acknowledged and discussed in detail. Stating that "the identity of these [Precambrian] organisms is controversial" (line 51) is not justification for the main claims of this paper, and it should be clear what distinguishes the new fossils from previously proposed evidence of earlier segmented bilaterians. I believe there is a case to be made in support of the authors' claims, but this context is vital to make that argument convincingly. It is also worth discussing evidence of segmented bilaterians in the early Cambrian trace fossil record, such as Treptichnus pedum (Kesidis et al., 2019, Proceedings B) and Rusophycus (Daley et al., 2018 PNAS), that may predate or be roughly coeval with these fossil remains.
Thanks for the valuable suggestions and references from the reviewer. In the revision, we will introduce the definition of segment and former researches in the introduction part as suggested. And we will discuss the comparisons of our specimens with the Precambrian fossils mentioned above in more detail. The major difference between our materials and Spriggina and Yilingia is the undisputed bilateral symmetry. While the paired appendages distinguished the Cambrian fossils from Parvancorina, Cloudina, and Dickinsonia. Since the pseudomorphing mechanism caused the lost of internal structures of the original, the well-developed paired appendages and outer segment could be the major evidence of their identity. In addition, we will make a more detailed introduction and discussion on the trace fossils of during the Ediacaran-Cambrian transition.
eLife Assessment
This study provides valuable insights into the architecture of the tripartite attachment complex in Trypanosoma brucei and how it may preserve mitochondrial genome anchoring during basal body movement. Cryo-electron tomography and structural modelling provide solid support for the observed variability in filament length and basal body positioning, although the proposed mechanism involving p197 repeat orientation awaits direct molecular validation. These findings will interest researchers studying mitochondrial DNA inheritance, organelle organization, and the cell biology of kinetoplastid parasites.
Reviewer #1 (Public review):
Summary:
In this manuscript, the authors employed Cryo-ET to visualize the structural organization of the tripartite attachment complex or the TAC, which is a unique structure connecting the flagellar basal body to the mitochondria in Trypanosoma brucei, a protozoan parasite causing human sleeping sickness in sub-Saharan Africa. The TAC is essential for segregation of the mitochondrial DNA, or kDNA, via the connection to the segregating basal bodies during the cell cycle. Although molecular characterization of the TAC and its components has yielded insights into the function of the TAC, the structural organization of the TAC, especially the largest protein p197 within the TAC, remains poorly defined. Using intact zoid cells, the authors provided, for the first time, a high-resolution image of the cytoplasmic part of the TAC in its native context, resolved the basal body, the mitochondrial membranes, the exclusion zone filament connecting the BB to the outer mitochondrial membrane, and quantified the geometry of the region surrounding the TAC. These observations led to a proposed role of p197 as a length-variable connector to reconcile stable kDNA anchoring with the mechanical demands of BB movement. Overall, the work advances our understanding of the structure and function of p197 within the TAC in trypanosomes.
Strengths:
A major strength of the manuscript is the application of Cryo-ET in imaging the structure of the cytoplasmic part of the TAC, with a focus on the largest protein p197, which provided novel insights into its molecular function. Trypanosome cells are too thick for Cryo-ET, and the authors adopted intact zoid cells for Cryo-ET, allowing the resolution of the TAC structure in native conditions.
Weaknesses:
While this is the first attempt to image the TAC by Cryo-ET, which provided new insights into the potential role of p197 in mediating the connection between the BB and the mitochondria, there is a lack of mechanistic understanding of how p197's length is regulated and how the varying length of p197 coordinates with basal body movement. Also, it is unclear why the cell needs to regulate the length of p197, given that the main function of p197 is to connect the BB and the mitochondria for the segregation of kDNA via the BB.
The proposed length variation for p197 is an interesting idea, but it is unclear what exactly the authors meant by saying that p197 is length variable. Do they mean that the central repeats in p197 can fold/extend to variable lengths, or do they mean that p197 with variable lengths contains different numbers of the central repeats? This should be clarified.
The current work can be strengthened by performing some molecular and/or biochemical experiments to assess the function of p197 to complement the Cryo-ET-based structural characterization reported in the manuscript. Expression of some smaller-sized p197 proteins with varying numbers of central repeats in p197 RNAi cells will provide evidence for the proposed function of p197 length variation.
Reviewer #2 (Public review):
While the aim of this study is interesting and the methods are advanced, the novel insight and conclusions rather feel limited at present, particularly considering the literature already available on the functional role of p197. There are some notable methodological issues in quantifying the EZF that need to be addressed.
The claims accompanying p197 and its role in EZF tethering to the BB and pBB to the OMM, as well as its effect on mitochondrial membranes, as displayed in Figure 3, should be augmented via additional comparative measurements with tdzoid cell tomograms, given that tomograms were generated for tdzoid cell lines, but are only utilised in a limited manner for Figures S4 and S5. This represents an underutilised opportunity to use this technique to contrast BB, pBB and mitochondrial membrane separation in a comprehensive manner beyond what previous studies could achieve.
The authors need to justify specifically why different methodologies were employed for minimum filament length measurements between NP-40 wild-type cells and non-detergent-treated Zoid cells, preventing these datasets from being broadly comparable.
The reliance on mutant cell lines (primarily zoids) for conclusions needs to be acknowledged to a greater extent, such as when comparing results with previous works (Line 334) and naturally within the limitations mentioned in the discussion.
The integration of the discussion with relevant literature concerning Cryo-EM and trypanosomatid cytoskeletal studies feels somewhat limited; for example, no mention or comparison with the formative Höög et al. 2012 study for Cryo-EM within trypanosome flagella and basal bodies.
Reviewer #3 (Public review):
Summary:
The authors use the advanced technique of cryo-electron tomography to understand the connections between the basal body and the mitochondrion in trypanosomes. They show that this connection is mediated by flexible and extendable filaments.
Strengths:
The authors use an array of advanced cryo-electron microscopy techniques that provide solid evidence that p197 forms the filaments between the basal body and the mitochondrion and demonstrate that these filaments are flexible and able to span a wide range of distances. This provides an important insight into how the trypanosome basal body, which has a dynamic localisation through the cell cycle, can manage its critical connection to the mitochondrion. The filaments provide the necessary flexibility to maintain the basal body - mitochondrion connections as the basal body is repositioned.
Weaknesses:
The authors were unable to directly connect the filaments to the structural models for their protein component, p197 and how the filament flexibility relates to the changes in the repeat region of the protein. In part, this is due to the flexibility of these filaments and demonstrates the limitation of cryo-EM to resolve such structures, but this in itself is an important outcome, as it will inform the community's use of this technology.
eLife Assessment
This study reveals valuable links between cell-penetrating peptide uptake and the cell cycle, membrane voltage and an endogenous potassium channel. The evidence is solid, and further experiments would strengthen the proposed mechanism and the degree of the mechanism's generalization. The findings will interest researchers working on delivery of therapeutic molecules into cells.
Reviewer #1 (Public review):
Summary:
Ribeiro et al. investigate the molecular mechanisms affecting the efficiency of synthetic cell-penetrating peptide (CPP) uptake by cultured mammalian cells. CPPs are frequently coupled to cargoes that can change cell function and are highly useful to investigators studying cell biology. Moreover, CPPs are potentially amenable to delivery of cargoes that can affect cells in clinically useful ways, with several under investigation as potential anti-cancer treatments. A challenge to the therapeutic use of CPPs is their inefficient delivery to their intended molecular targets, with poor cellular uptake being a major barrier.
The present study presents new insights into the cellular and plasma membrane properties that can affect CPP uptake. In the studied cell line, the authors find that cell cycle-dependent changes in the plasma membrane potential correlate with uptake of a CPP conjugated to a fluorescent dye. These changes correlate with mRNA abundance of a voltage-gated potassium channel, KCNH1. RNA silencing of KCNH1 expression or pharmacological blockade of KCNH1 currents impairs CPP uptake, supporting the proposed model.
Strengths:
(1) Identified KCNH1 expression and function as a specific modulator of CPP uptake in U2OS cells and LAN1 cells.
(2) Data show a clear correlation between the plasma membrane potential, cell cycle stage, and efficiency of intracellular CPP delivery in these cells.
Weaknesses:
(1) Some of the claims are more generalized than warranted by the presented evidence. Is KCNH1-dependent modulation of cell cycle-dependent changes in the plasma membrane potential and ensuing effect on CPP uptake a niche property of this channel in some cell lines? Or a more broadly generalized function? I think this can be addressed merely through more careful or caveated language in the manuscript. However, an additional experiment that would be potentially informative and complementary to the chronic high-potassium depolarization experiments would be to examine the impact of heterologous expression of a hyperpolarizing potassium channel (for example, Kir2.1).
(2) Please provide more details on how endosomes were identified and measured.
Reviewer #2 (Public review):
Summary:
The article by Ribeiro et al. provides further mechanistic understanding of the abilities of cell-penetrating peptides (CPPs) to cross directly through the membrane. The authors performed a transcriptomic evaluation to conclude that CPPs enter via direct translocation through cells undergoing cell division, particularly in G2/M phase cells. They reasoned that this is due to the expression of KCNH1, a potassium channel, which is hypothesized to fluctuate throughout the cell cycle and peak at G2/M phase. This expression, in turn, hyperpolarizes the cells, enabling the direct translocation of CPPs.
The uptake mechanism of CPPs has long been debated. The authors and other groups have previously mentioned the importance of cell membrane potential and the involvement of specific ion channels in its regulation. The current work further adds a layer of understanding by connecting the occurrence of direct translocation to a specific cell cycle phase, which may be of great relevance when considering the use of CPPs for cells that are constantly dividing, such as cancer cells.
Strengths:
The work took advantage of transcriptomics and the unique use of the PIP-FUCCI system to determine the different uptake of CPPs in different cell cycle stages. All the data are in agreement with each other, especially when observing the membrane potential change through the cell cycle stages. The experiments are well executed and are all complementary to the authors' hypothesis.
Weaknesses:
The article could be further strengthened through:
(1) Implementation of additional confirmatory experiments
(2) Positioning against other recently reported channel proteins
(3) Confirming whether CPPs with different physicochemical properties are also affected by the fluctuation of KCNH1 expression
(4) Determining whether the concept of cell cycle-dependent membrane potential change is applicable not just in model cells, but also in cells that undergo rapid cell cycles, and
(5) Relating CPP direct translocation to known biological phenomena that are physiologically similar.
eLife Assessment
The study reports on the relative lipid permeabilities of different ribonucleotide isomers. The findings are valuable for prebiotic chemistry and protocell research because they offer an additional possible explanation for how the canonical ribonucleotide isomers were selected. The strength of the evidence is solid, being supported by experimental and computational approaches, although the step from differential permeability to selection is less firmly supported.
Reviewer #1 (Public review):
Summary:
The authors have carried out a study of lipid membrane permeabilities on different nucleoside/tide isomers. The study primarily relied on PAMPA as the experimental technique for assessing permeabilities, although interestingly, the authors found that there was a correlation between permeability and retention time on a C18 column. The authors specifically looked at the permeabilities of furano and pyrano alpha/beta nucleosides/tides, since a major question is why biochemistry uses only the beta furano isomers, when three other isomers based on anomeric configuration and ring size are also possible. The authors found that for all A, U, G or C nucleoside isomers, the beta-furanoside is consistently the most lipid-permeable. These findings were also supported by computational (DFT and MD) studies. The authors conclude that such a property could plausibly have been a chemical selection mechanism that favoured these isomers when they first started interacting with protocell membranes.
Strengths:
There are two major strengths to this study: (1) it offers an alternative/additional explanation backed by experimental and computational evidence as to how the canonical ribonucleotides (namely, the beta-furanoside isomers) were selected by interactions with protocell membranes; (2) it makes the important observation that retention time on a C18 column is correlated with permeabilities determined by PAMPA, the "industry standard", suggesting that LC could also be a convenient way to assess relative permeabilities of analytes of interest. With this report, there are many promising follow-up studies that can be carried out in the context of synthetic lipid protocells in order to further support the hypothesis that selective permeability played a major role in early chemical/Darwinian evolution.
Weaknesses:
The biggest weakness is that there needs to be more of a critical discussion of chemical vs Darwinian selection. There are two basic hypotheses for how the canonical ribonucleotides came to be: they were the products of prebiotic synthesis and chemical selection pressures, or they emerged as a product of natural selection after Darwinian evolution already became established. While the experimental evidence in terms of ribonucleotide prebiotic synthesis is discussed well, the work of Eschenmoser in particular, which supports the alternative hypothesis that RNA is the product of Darwinian evolution, should be discussed in more detail on page 4.
The authors should also acknowledge somewhere in the discussion/conclusion that there is a certain amount of bias that the field has towards extant biology, and this bias could be influencing how certain results are interpreted as beneficial for ribonucleotide selection, when the same result might be used to argue against it. For example, the fact that the 2',3'-cyclic phosphates are the most permeable among the phosphate regioisomers might be used to argue these were the preferred substrates for oligomerization in a protocell, given it had a catalyst which could do it, like a short peptide. While the differences in permeabilities are quite solid, what it means in terms of selection could be applied to multiple, specific scenarios, each with its own interpretations and implications.
Reviewer #2 (Public review):
Summary:
The authors ask whether membrane transport could have contributed to the emergence of the molecular configuration characteristic of extant RNA. They examine the permeability of α- and β-furanosides and of different nucleotide phosphate isomers using fatty-acid-based PAMPA systems, supported by partitioning experiments and molecular simulations.<br /> The central finding is that closely related molecular species show different transport behaviour: β-furanosides are preferentially transported, while 5′-phosphorylated nucleotides are less permeable than 2′- and 3′-phosphates. These observations are potentially interesting because membrane transport could influence the composition of an evolving compartment.
Strengths:
(1) The study addresses an interesting and relatively underexplored role for membranes in prebiotic chemical evolution.
(2) The comparison of closely related molecular species provides a potentially useful demonstration that membrane transport can discriminate between different nucleotide configurations. The experimental and computational approaches complement each other.
(3) The analysis of 5′-, 2′-, 3′- and cyclic phosphate species is particularly interesting and could provide a useful framework for thinking about how membrane transport interacts with nucleotide chemistry.
Weaknesses:
(1) The principal concern is that the manuscript moves from differential permeability to evolutionary selection rather too quickly. The experiments demonstrate differences in transport, but do not establish that these differences would produce sufficient intracellular enrichment, persistence or functional consequences to drive selection of the canonical RNA configuration.
(2) The relevance of the PAMPA system to primitive vesicles is also not fully established. Membrane permeability depends strongly on membrane composition and environmental conditions, and the relationship between the membrane compositions used in the principal experiments and plausible primitive vesicles needs to be clearer.
(3) Similarly, the pH-gradient experiments show an effect of pH on transport, but a gradient between two macroscopic PAMPA reservoirs does not demonstrate that primitive fatty-acid vesicles could maintain a comparable gradient.
(4) Finally, some of the key results are presented as changes in relative composition rather than direct measurements of molecular flux. The distinction between compositional enrichment, partitioning and permeability should therefore be clearer. The molecular simulations provide useful support but, given the small calculated free-energy differences and acknowledged limitations in the nucleotide simulations, do not independently establish the mechanism.
Overall, the work contains potentially valuable observations, but I think the conclusions should be framed more cautiously. The most compelling interpretation is that membrane transport could have been one component of a multifactorial chemical-selection process, alongside prebiotic synthesis, stability and nucleotide chemistry, rather than that membrane permeability itself explains the emergence of canonical RNA.
eLife Assessment
This valuable study compares hippocampal-cortical functional connectivity to various other brain measures and examines their development across youth. It uses sophisticated analyses replicated in multiple datasets, providing solid, albeit cross-sectional, evidence to support the primary claim that hippocampal-cortical connectivity relates to cognitive maturation. The manuscript would benefit from a more nuanced consideration of the biological basis of some of the derived imaging measures and the limitations of the cross-sectional design. This work will be of interest to neuroimaging specialists and cognitive neuroscientists.
Reviewer #1 (Public review):
Summary:
The authors studied the development of hippocampal connectivity gradients based on open datasets and performed correlation analyses with other MRI features as well as gene expression information from other datasets.
Strengths:
The hippocampus is a key region in understanding large scale brain organization and cognition, and the authors applied advanced and suitable analytics to study its development. The paper is overall well organized and well written, and the findings are relevant for studying large-scale brain development.
Weaknesses:
While sophisticated, several of the analyses appear mainly correlational, cross-sectional, and rely on cross-dataset contextualization, which should also be stated as a limitation of the current work.
Comments on revised version.
I thank the authors for their revisions which have improved the paper.
Reviewer #2 (Public review):
Summary:
In this manuscript, the authors aim to assess how the functional organisation of the hippocampus is related to the geometry and neurobiological differences of the hippocampus. In particular, the authors focus on the first three eigenvectors of hippocampal-cortical functional connectivity, based on non-linear dimensionality reduction on resting-state functional MRI data. Furthermore, the work aims to describe changes in these functional axes, and their relation to other factors, throughout youth and evaluate whether they are predictive of individual variations in cognition.
Strengths:
A major strength of this study is the attempt to replicate key findings across multiple developmental cohorts.
Weaknesses:
T1w/T2w signal intensity is incorrectly used as synonymous with myelin.
Appraisal of Aims and Conclusions:
The authors partially achieve their aims by illustrating certain age-related changes in hippocampal function; however, the correlative study design is not equipped to examine how these changes are "shaped" by geometry, myelination or gene expression (especially the latter two). Furthermore, conclusions were often overstated based on small effect sizes.
Context and Field Impact:
This work adds to a growing body of literature focused on gradient-based representations of hippocampal topology. By applying these methods across a wide developmental age bracket, it provides a useful reference point for how the hippocampus and wider cortex interact during maturation. To improve utility to the neuroimaging and cognitive neuroscience communities, the nesting of subfields within the eigenvector topology should be addressed too.
Comments on revised version.
I appreciate the authors' revisions and their clear efforts to improve technical transparency throughout the manuscript. I believe the manuscript could have been more strongly improved were the authors more constructive than defensive in their responses, but the manuscript is nevertheless more clear.
Author response:
The following is the authors’ response to the original reviews.
eLife Assessment
This valuable study compares hippocampal-cortical functional connectivity to various other brain measures and examines their development across youth. It uses sophisticated analyses replicated in multiple datasets, but provides incomplete evidence to support the primary claim that hippocampal-cortical connectivity relates to cognitive maturation. The manuscript would benefit from a more nuanced consideration of the biological basis of some of the derived imaging measures and the limitations of the cross-sectional design. This work will be of interest to neuroimaging specialists and cognitive neuroscientists.
We thank the Editors for credits on the value of the multimodal, multi-dataset design and the study’s relevance to cognitive neuroscience. We have revised the manuscript in three corresponding ways:
(1) Connectivity–cognition interpretation. We now state explicitly that the hippocampal-cortical connectivity–cognition findings are age-related associations with cognitive measures during youth. They do not demonstrate that hippocampal-cortical connectivity causes, mediates, or fully explains cognitive maturation.
(2) Biological interpretation of imaging measures. We refined the interpretation of derived imaging measures throughout the manuscript. T1w/T2w is described as an indirect, myelin-sensitive but non-specific MRI contrast that may reflect multiple tissue and acquisition factors, rather than direct myelin content. Similarly, transcriptomic findings are described as cross-dataset spatial associations rather than evidence of developmental molecular mechanisms. Consistent with recent guidance on neuroimaging data contextualization, the interpretation of these associations is constrained by the characteristics of the reference atlas and by cross-dataset spatial alignment[1].
(3) Limits of cross-sectional evidence. We added an explicit limitation that the cross-sectional design does not measure within-person developmental change and cannot establish temporal ordering or causal relationships.
Together, these revisions address the evidential limitations identified in the Assessment while preserving the supported contribution of the replicated multimodal associations.
Reviewer #1 (Public Review):
Summary:
The authors studied the development of hippocampal connectivity gradients based on open datasets and performed correlation analyses with other MRI features as well as gene expression information from other datasets. Although the main findings are correlational and cross-sectional, the analyses are overall sophisticated and replicated in several datasets.
Strengths:
The hippocampus is a key region in understanding large-scale brain organization and cognition, and the authors applied advanced and suitable analytics to study its development. The paper is overall well-organized and well-written, and the findings are relevant for studying large-scale brain development.
Weaknesses:
While sophisticated, several of the analyses appear mainly correlational, cross-sectional, and rely on cross-dataset contextualization, which should also be stated as a limitation of the current work.
We are grateful to Reviewer #1 for recognizing the importance of the hippocampus, the suitability of the analytical framework, and the value of the replicated developmental design. We agree that the public-review concern about correlational, cross-sectional, and cross-dataset evidence is central to how the findings should be interpreted. In particular, consistent with recent recommendations for neuroimaging data contextualization, we now explicitly identify the transcriptomic component as group-level spatial contextualization and state that its interpretation is limited by the properties of the adult post-mortem reference atlas and by cross-dataset spatial alignment[1]. It therefore cannot provide individual-level developmental inference or establish causal molecular mechanisms. We have addressed these boundaries in detail in our response to Reviewer #1, Recommendation 1, and in the revised Limitations section, so that the same qualification is applied consistently throughout the manuscript.
Revised manuscript text (Page 29-30): “Although key findings were replicated across two independent longitudinal datasets, the primary developmental analyses were cross-sectional and therefore did not measure within-person change. The absence of pubertal hormone measures in the HCP-D dataset also limited assessment of hormonal contributions to the observed age-related associations. The observed patterns should be interpreted as age-related associations rather than individual developmental trajectories or causal developmental mechanisms. T1w/T2w intensity was treated as an indirect, myelin-sensitive MRI proxy influenced by multiple tissue and acquisition factors, not as direct histological myelin content. Transcriptomic results were based on post-mortem samples from a separate dataset and were interpreted as cross-dataset spatial associations rather than developmental mechanisms. More generally, the interpretation of such cross-dataset map correspondence is constrained by the properties and representativeness of the source atlas and by cross-dataset spatial alignment[1]. These limitations motivate future longitudinal, multimodal, and biologically validated studies.”
Reviewer #1 (Recommendations for the Authors):
(1) Some language suggests causality, but the associated findings were mainly derived from correlational analysis. I would suggest verifying the wording and toning down causal terms ('shape', 'impact', 'linked' ...).
We thank the reviewer for identifying this important mismatch between the strength of the wording and the correlational nature of the evidence. We have completed a manuscript-wide terminology review and replaced causal or mechanistic verbs with relational language throughout the Abstract, Introduction, Results, Discussion, and figure captions. We have used terms such as associated with, corresponds to, covaries with, and is consistent with when describing relationships among gradients, geometry, T1w/T2w intensity, transcriptomic measures, and cognition. Stronger terms have been retained only when they describe a mathematical operation or the construction of an analysis, and the revised Discussion now explicitly states that cross-sectional associations do not establish directionality or causation.
(2) Please also provide the following citations when referring to hippunold: https://elifesciences.org/articles/88404 (which describes additional advanced registrations). If the authors have the ability to quickly rerun some processing, then I would recommend a sensitivity analysis based on hippunfold v2, which also provides more homogenous vertex spacing, fewer distortions during unfolding, and some additional improvements.
We thank the reviewer for pointing out the missing registration reference and for recommending a HippUnfold v2 sensitivity analysis. We have added the requested eLife study on surface-based hippocampal registration and have included the relevant surface-generation and registration procedures for our study.
We carefully considered reprocessing the data with HippUnfold v2. The official release notes indicate that v2 includes important changes to surface precision, surface re-tessellation, Laplace coordinate generation, and robustness to topological breaks. These changes could affect surface coordinates and the precise spatial sampling of imaging data. A meaningful v2 sensitivity analysis would therefore require reprocessing the original MRI data and repeating the downstream connectivity, gradient, developmental, cognitive, and cross-dataset analyses. We did not perform this complete reprocessing in the present revision because it would represent a new version-wide analysis rather than a limited sensitivity check.
We did not consider a partial rerun of selected downstream analyses to provide a sufficiently informative or internally consistent test of version sensitivity. Rather than present an incomplete comparison, we retained the original HippUnfold v1.2.1 pipeline, explicitly identified version dependence as a limitation, and removed wording that could imply that the findings are fully insensitive to preprocessing choices.
We appreciate that a preregistered comparison across HippUnfold versions would be valuable in future work, particularly for analyses in which fine-scale surface correspondence or registration precision is the primary scientific focus.
Revised manuscript text (Page 33): “We employed HippUnfold (v1.2.1)[2] (https://hippunfold.readthedocs.io; advanced surface-registration procedures are described by DeKraker et al.[3]), which integrated a deep learning method with a topological constraint strategy.”
(3) Are there changes in motion parameters across developmental time windows/age? Are findings consistent when fMRI motion parameters are regressed out in the statistical models, and do the 3 derived gradients correlate with head motion?
We thank the reviewer for raising this important potential confound. We noted that the original developmental GAMs already included mean framewise displacement (mFD). First, we tested whether head motion changed with age. In the 601 included HCP-D participants, we fitted mFD as the outcome using mFD ~ s(Age, k = 4) + Sex + site + s(hemi,bs = “re”) with REML estimation. mFD showed a robust nonlinear decrease with age (edf = 2.66, F = 125.13, smooth-term p < 2.2 × 10<sup>-16</sup>; likelihood-ratio p < 2.97 × 10<sup>-78</sup>; Supplementary Fig. 3a). Importantly, the original developmental models for hippocampal gradient statistics already included mFD as a covariate. We therefore directly examined the mFD coefficients from those same 18 models, which used Feature ~ s(Age, k = 4) + Sex + site + s(hemi,bs = “re”) + <mFD.
After Benjamini–Hochberg correction across the 18 mFD coefficients, mFD was associated with explained variance for lLG (β = −0.0412, SE = 0.0147, FDR q = 0.047) and qLG (β = 0.0280, SE = 0.00818, FDR q = 0.011), but with nocIG feature (all FDR q ≥ 0.101; Supplementary Fig. 3b). Thus, head motion is strongly age-related and is associated with a limited subset of the analyzed gradient-distribution features, supporting its inclusion as a covariate. The developmental effects reported in Figure 2e–f are already estimated after adjustment for mFD; we do not claim that every gradient feature is motion-independent.
Revised manuscript text
(Page 9): “We fit the model with a smooth term for age as a fixed effect; sex, scanning site, and in-scanner head motion (mean framewise displacement, mFD) as linear covariates”
(Page 10): “The age-related mFD trajectory and direct mFD coefficients from the original gradient feature models are shown in Supplementary Fig. 3.”
(4) Did the authors also observe morphological changes in the hippocampus? Are there changes in hippocampal cortical thickness that could explain some of the changes in functional coupling as well?
We thank the reviewer for this important suggestion. We performed additional HippUnfold surface-morphology analyses and a thickness-adjusted reanalysis of the same gradient features examined in Figure 2 and Extended Data Fig. 4. Morphological measures were calculated separately for each hemisphere after pooling the hippocampal-proper (label-hipp) and dentate-gyrus (label-dentate) surfaces. Thickness, curvature, and gyrification were native-surface-area-weighted, surface area was summed across both labels, and left-hemisphere curvature was sign-oriented to match the right-hemisphere convention.
For each morphology measure, we fitted Metric ~ s(Age, k = 4) + Sex + site + s(hemi,bs = “re”) in 601 included participants, using the same GAM framework as the primary gradient-feature analyses but without mFD. Thickness, oriented curvature, gyrification, and surface area all showed age-related change after FDR correction (all q ≤ 0.000304). Hippocampal thickness decreased from childhood through adolescence and then plateaued, whereas surface area and gyrification increased before reaching late-adolescent plateaus.
To test whether morphology accounted for the functional findings, we refitted each of the 18 original gradient feature models with hemisphere-matched thickness added: Feature ~ s(Age, k = 4) + Sex + site + s(hemi,bs = “re”) + <mFD + thickness. No individual thickness coefficient survived FDR correction across the 18 features (minimum q = 0.061). Thus, developmental changes in hippocampal and dentate morphology are present, but surface thickness does not account for the principal age-related functional gradient effects.
Revised manuscript text (Page 10):
“Given the age-related variation in hippocampal functional-gradient features, we next assessed whether concurrent changes in global hippocampal morphology could account for these patterns. Surface-area-weighted thickness, curvature, gyrification, and surface area of the pooled hippocampal proper and dentate gyrus showed significant age-related associations (Supplementary Fig. 5). However, after adding hemisphere-matched thickness to the corresponding hippocampal gradient feature models, no thickness association survived FDR correction (minimum q = 0.061). Thus, although global hippocampal morphology changes across youth, these results do not support global thickness as an explanation for the primary functional-gradient patterns.”
(5) The authors corrected for sex in several of their models. It would be interesting to show, in a supplemental form, sex-disaggregated analyses that outline the consistency and/or potential shifts in developmental patterns in males and females.
We thank the reviewer for this valuable suggestion. We first performed formal sex-by-age interaction tests in the three existing main-text age-GAM families, rather than introducing new outcomes. For each outcome, we compared the original additive model, Measure ~ s(Age, k = 4) + Sex + site + s(hemi,bs = “re”) + <mFD, with a sex-specific age-trajectory model, Measure ~ s(Age, k = 4) + Sex + site + s(hemi,bs = “re”) + <mFD. Nested-model interaction tests were conducted using ML fits, whereas the displayed trajectories were estimated with REML, consistent with the primary analyses.
After BH correction within each original analysis family, sex-by-age interactions were detected only for three Figure2 gradient-distribution features:lLG explained variance (q = 0.00272), cIG range (q = 0.0449), and cIG standard deviation (q = 0.0449). No interaction survived correction among the six Figure 3 hippocampus–cortical-hierarchy measures or the three Figure 4 geometry–function coupling measures.
In the sex-specific trajectories, the clearest difference concerned lLG explained variance: females showed a lower childhood level followed by a steeper increase and higher adolescent values, whereas males exhibited a more gradual increase. For cIG range and standard deviation, both sexes showed broadly similar inverted-U-shaped trajectories; the male trajectory was modestly higher through mid-to-late adolescence, with substantial overlap of the confidence intervals. The original combined-sex trajectory lay between, or closely followed, the two sex-specific curves. Thus, the principal developmental patterns are broadly consistent across sexes, while sex-related shifts are limited to a small subset of gradient-distribution features.
Revised manuscript text (Page 10):
“We additionally tested sex-by-age interactions across the previously reported developmental models. After FDR correction, significant interactions were identified for lLG explained variance and the range and standard deviation of cIG, whereas no sex-by-age interaction survived correction for the remaining primary developmental outcomes. Sex-disaggregated trajectories for these significant interactions are shown in Supplementary Fig. 4.”
(6) There have been reports of some distortions in inferior temporal regions after non-linear alignment to standard space in the HCP processing, which may affect hippocampal data sampling. Did the authors verify registration accuracy and absence of such distortions in their samples?
We thank the reviewer for raising this important concern regarding potential registration-related distortion. We clarified the registration and sampling chain and now explicitly distinguish native-space signal sampling from the subsequent use of a common coordinate system for group-level correspondence.
Crucially, the hippocampal fMRI time series were sampled in each participant’s native space: BOLD data were mapped back to the native anatomical space and sampled within the participant-specific hippocampal grey-matter ribbon onto the native HippUnfold mid-thickness surface. Likewise, hippocampal T1w/T2w measurements were sampled from the participant-specific native hippocampal surfaces. Thus, neither the fMRI nor the T1w/T2w hippocampal measurements was obtained by sampling voxels after nonlinear transformation into a standard medial-temporal-lobe space. This design avoids standard-space deformation influencing the extraction of these hippocampal signals and intensity measures. The subsequent common surface coordinate is used to establish vertex correspondence across participants for group-level analyses, rather than to define the native-space signal samples themselves.
We also performed visual quality control of the HippUnfold registration and surface reconstruction, inspecting the alignment of each participant’s hippocampal segmentation and reconstructed surfaces with the native T1-weighted anatomy. The documented mask-overlap quality-control procedure further showed Dice coefficients above 0.7. The paired inner and outer surfaces define the mid-thickness sheet used for ribbon-constrained sampling, which is particularly suitable for the variable folding geometry of the hippocampus.
(7) Can the authors clarify how gene expression was analyzed for the hippocampus? Were HippUnfold models run across all donors separately, and probes were then sampled and co-registered? How did the authors handle data sparsity?
We thank the reviewer for this request for clarification. The full hippocampal transcriptomic workflow is described in the Methods subsection “Transcriptomic association analysis of the hippocampal gradients”. We clarify the procedure here to make the correspondence between the Allen Human Brain Atlas (AHBA) samples and the hippocampal gradients explicit.
We used AHBA microarray data from six donors. Donor-specific age- and sex-related effects were regressed from each probe before the normalized residual expression values were combined. We selected samples annotated as CA1, CA2, CA3, CA4, subiculum, or dentate gyrus from both hemispheres, yielding 188 hippocampal samples. The participant-specific native HippUnfold mid-thickness surfaces were transformed to MNI space and averaged to generate the group template mid-thickness surfaces. Each AHBA hippocampal sample was then matched to its nearest vertex on this template. Fifteen samples located more than 6 mm from the template surface were excluded, leaving 173 samples from six donors. The group-averaged hippocampal gradient value at the matched template vertex was assigned to each retained tissue sample.
We then used LASSO-PCR to model each gradient from the 173 × 58,692 normalized expression matrix. Principal-component analysis was first used to represent the correlated probe-expression structure, and LASSO regression was fitted on the retained component scores. The penalty parameter was selected by nested five-fold cross-validation, and generalizability was evaluated using ten repeated ten-fold cross-validation. Probe weights were subsequently reconstructed from the fitted component weights for the feature-deconstruction and enrichment analyses.
Thus, the transcriptomic analyses are group-template spatial associations between adult post-mortem AHBA samples and the group-averaged hippocampal gradients. They are not donor-specific developmental measurements, nor do they establish a causal molecular mechanism. We have retained the complete methodological description in the Methods subsection cited above and clarify the workflow here for ease of reference.
Reviewer #2 (Public Review):
Summary:
In this manuscript, the authors aim to assess how the functional organisation of the hippocampus is related to the geometry and neurobiological differences of the hippocampus. In particular, the authors focus on the first three eigenvectors of hippocampal-cortical functional connectivity, based on non-linear dimensionality reduction on resting-state functional MRI data. Furthermore, the work aims to describe changes in these functional axes and their relation to other factors throughout youth and evaluate whether they are predictive of individual variations in cognition.
Strengths:
A major strength of this study is the attempt to replicate key findings across multiple developmental cohorts.
Weaknesses:
The major weaknesses of the manuscript center on gaps in technical transparency and several conceptual inaccuracies. The machine learning methodology used for cognitive prediction is scarce, leaving little means to evaluate whether the behavioral results suffer from data leakage or overfitting. The introduction sets up an oversimplified historical premise regarding the field's understanding and appreciation of hippocampal connectivity, and contains several incorrect references that throw doubt on the argumentation. Additionally, T1w/T2w signal intensity is incorrectly used as synonymous with myelin, despite gold-standard histological validation showing a non-significant correlation between T1w/T2w and myelin staining (Sandrone et al., 2013).
Appraisal of Aims and Conclusions:
The authors partially achieve their aims by illustrating certain age-related changes in hippocampal function; however, the correlative study design is not equipped to examine how these changes are "shaped" by geometry, myelination, or gene expression (especially the latter two). Furthermore, conclusions were often overstated based on small effect sizes.
Context and Field Impact:
This work adds to a growing body of literature focused on gradient-based representations of hippocampal topology. By applying these methods across a wide developmental age bracket, it provides a useful reference point for how the hippocampus and wider cortex interact during maturation. To improve utility to the neuroimaging and cognitive neuroscience communities, the nesting of subfields within the eigenvector topology should be addressed, too.
We thank Reviewer #2 for the careful assessment and for recognizing the value of replicating key findings across multiple developmental cohorts. We have addressed the specific concerns regarding cognitive prediction, historical framing, biological interpretation, effect-size interpretation, citation placement, and T1w/T2w signal in the detailed responses below. These revisions expand the prediction methods, clarify the interpretation of hippocampal gradients and subfields, recalibrate the developmental claims, and improve the precision of the supporting citations.
We also thank the reviewer for raising the need for a more precise interpretation of T1w/T2w signal. We agree that T1w/T2w intensity should not be used synonymously with myelin, and we have revised the manuscript throughout to refer to it as a myelin-sensitive but non-specific MRI contrast that may reflect multiple tissue properties. We respectfully clarify the interpretation of the cited Sandrone et al. study. We believe that the intended reference is Sandrone et al. (2023)[4], rather than 2013. This study tested the correspondence between T1w/T2w and Luxol Fast Blue staining specifically in corpus-callosum white matter; it did not include cortical gray matter or hippocampal tissue. Its callosal analysis did not show the expected positive correspondence between T1w/T2w and histological myelin staining (indeed, it reported a significant inverse correlation), but this result cannot establish the validity or invalidity of T1w/T2w as an index of myelin-sensitive microstructural variation in hippocampal gray matter. The authors also distinguished this white-matter result from prior evidence of cortical co-localization between high T1w/T2w signal and myelin-rich regions.
Accordingly, we do not interpret our hippocampal T1w/T2w findings as direct measurements of myelin content or cellular myelination. Rather, they indicate associations with an indirect MRI contrast that is sensitive to myelin-related, as well as other, tissue properties. We have revised the relevant Introduction, Results, Discussion, figure text, and Methods language to reflect this distinction.
We thank the reviewer for highlighting the importance of considering how hippocampal subfields are nested within the connectivity-eigenvector topology. This question has been addressed in our related subfield-specific study using the same HCP-D cohort and HippUnfold surface framework. We explicitly examined the distribution of the two long-axis gradients across the cytoarchitectonic subfields (Sub, CA1, CA2/3, and CA4–DG), their distinct cortical-connectome organization, and their associations with episodic memory (Zeng et al., 2025[5]). We have added a citation to this work to provide the relevant subfield-level context. The present study focuses on the continuous whole-surface organization and developmental trajectories of the three hippocampal functional gradients.
Revised manuscript text (Page 29):
“Continuous hippocampal gradients and cytoarchitectonic subfields
The continuous whole-surface topology examined here should not be viewed as separate from hippocampal cytoarchitectonic subfield architecture. Rather, it characterizes hippocampal-cortical connectivity organization across the unfolded hippocampal surface, including variation that extends across subfield boundaries. The proximal-distal association of cIG provides one link between this continuous topology and the Sub-CA-DG axis. Complementing the present whole-surface analysis, our recent subfield-specific study of the same HCP-D cohort examined how the two long-axis gradients are distributed across Sub, CA1, CA2/3, and CA4-DG. It further showed that their subfield-specific cortical-connectome organization was associated with episodic memory[5]. Taken together, these findings indicate that hippocampal functional gradients provide a unifying framework that relates continuous topology and cytoarchitectonic architecture to hippocampal-cortical network organization, developmental reconfiguration, and cognition.”
Reviewer #2 (Recommendations for the Authors):
(Methodological and Statistical Revisions)
(1) Prediction of Cognitive Performance:
Provide a comprehensive, step-by-step description of the machine learning pipeline used for cognitive prediction. Explicitly specify the features extracted, the exact estimator employed, the cross-validation strategy, and the precise measures taken to prevent data leakage.
We thank the reviewer for requesting clarification of the cognitive prediction analysis. We have added a dedicated Methods subsection describing the prediction framework and its safeguards against data leakage. We used hippocampal–cortical projection features to estimate episodic memory, cognitive flexibility, working memory, and inhibitory control, while including age, sex, scanning site, and mean framewise displacement as linear covariates.
To address the high dimensionality and correlation structure of the projection features, we used a PCA–LASSO framework. PCA generated component scores from the hippocampal–cortical projection features, and LASSO regression was fitted to these scores. The LASSO regularization parameter was selected using nested five-fold cross-validation.
Predictive performance was evaluated with age-balanced 10-fold cross-validation, using rounded age solely to stratify the outer folds. To prevent information leakage, all data-adaptive steps were completed using only the outer training data. These steps included feature transformations, estimation of linear covariate effects, PCA, and LASSO penalty selection. The fitted processing steps and model were then applied to the held-out fold. Performance was quantified by the squared Pearson correlation between cross-validated predicted and observed cognitive scores. We applied FDR correction across the 12 resulting gradient–cognition tests.
Revised manuscript text (Page 37-38):
“Cognitive prediction analyses
We used hippocampal–cortical projection features to predict episodic memory, cognitive flexibility, working memory, and inhibitory control. Age, sex, scanning site, and mean framewise displacement were included as linear covariates.
The model used a PCA–LASSO framework to address the high dimensionality and correlation structure of the projection-feature matrix. Principal-component analysis generated component scores from the hippocampal–cortical projection features. LASSO regression was then fitted to these scores.
Predictive performance was evaluated using age-balanced 10-fold cross-validation. Rounded age was used solely as the stratification label for the outer folds. Within each outer training fold, feature transformations, principal-component estimation, linear covariate effects, and LASSO penalty selection were fitted using training data only. These fitted transformations and model parameters were then applied to the held-out fold. Thus, held-out participants did not contribute to feature preprocessing, covariate estimation, component construction, or model tuning.
The LASSO regularization parameter was selected using nested five-fold cross-validation within each outer training set. Predictive performance was quantified as the squared Pearson correlation between cross-validated predicted and observed cognitive scores. We tested 12 gradient–cognition pairs from three hippocampal gradients and four cognitive measures. False-discovery-rate correction was applied across all resulting p-values.”
(2) Integration of Core Statistics:
Move the relevant test statistics, degrees of freedom, and FDR-corrected p-values for primary developmental claims from the Extended Data figures into the main text body to ensure transparency.
We thank the reviewer for emphasizing the need to make the statistical evidence underlying the extended developmental displays directly traceable. We have now added quantitative reporting for the three gradient-distribution features displayed in Extended Data Fig. 4 but not in Figure 2e, f. Specifically, the revised Results text reports the FDR-corrected age-related changes in range, skewness, and kurtosis.
We also created Supplementary Data 1 for the parcel-wise developmental trajectories shown in Extended Data Fig. 5f. This table provides the GAM statistics for all 360 cortical parcels for each hippocampal gradient, including the age-smooth F statistic, Δ adjusted R<sup>2</sup>, raw and FDR-adjusted p values, and Yeo seven-network assignment. We added an explicit reference to this table immediately after the Extended Data Fig. 5f citation in the Results. These changes enable readers to inspect the numerical evidence underlying the Extended Data displays without inferring statistics from the figures alone.
Revised manuscript text
(Page 10):
“Beyond these features shown above, the ranges of lLG and qLG changed with age (lLG: ΔAdj. R<sup>2</sup> = 0.0098, p = 0.0017; qLG: ΔAdj. R<sup>2</sup> = 0.0044, q = 0.0148). lLG skewness and lLG / qLG kurtosis also changed with age, whereas qLG skewness and all three cIG measures did not survive FDR correction across the six distributional features of each gradient (Extended Data Fig. 4).”
(Page 15):
“Complete parcel-wise GAM statistics are provided in Supplementary Data 1.”
(3) Signal-to-Noise Ratio (SNR) Justification:
Provide an analytical justification demonstrating why vertex-wise probing at a 483-vertex density within the hippocampus does not suffer from too severe spatial oversampling or low SNR problems.
We thank the reviewer for raising the important concern that a dense hippocampal surface representation could create spatial oversampling or unstable vertex-level measurements. We have now added a quantitative assessment of the native HippUnfold mesh spacing. The 483-vertex representation is the den-2mm HippUnfold template, comprising 419 hippocampal and 64 dentate vertices per hemisphere. We measured distances between adjacent vertices on all 2,608 un-smoothed native T1w mid-thickness surfaces from the 652 HCP-D participants. Across 1,765,616 unique mesh edges, the pooled median distance was 2.01 mm (IQR, 1.43–2.57 mm), and 78.3% of edges were between 1 and 3 mm. Thus, the native mesh is centered on the 2-mm spatial scale of the BOLD acquisition rather than constituting a submillimeter sampling grid. The remaining local variability reflects individual anatomy and the triangular surface geometry. These results are shown in Supplementary Fig. 7.
We noted that BOLD data were sampled in each participant’s native anatomical space using partial-volume-weighted ribbon-constrained mapping between the paired inner and outer hippocampal surfaces. Each vertex value is therefore a weighted estimate from voxels within the hippocampal grey-matter ribbon, rather than a nearest-voxel sample. Importantly, the 483 vertices were not treated as independent fMRI observations. Spatial dependence among neighboring vertices is expected, whereas the developmental inference was performed on participant-level gradient-derived features and cortical-projection measures, with sex, site, hemisphere, and mean framewise displacement included as covariates. Replication of the principal organizational patterns in the independent NKI-RS and CBD samples provides additional evidence that the reported gradient findings are unlikely to be materially driven by low-SNR or spatial-oversampling artefacts.
Revised manuscript text (Page 35): “The vertex spacing was set at approximately 2 mm, close to the 2.0 mm isotropic fMRI voxel size. The distribution of distances between adjacent vertices across all native-space mid-thickness surfaces was centered near this target spacing (median, 2.01 mm; interquartile range, 1.43–2.57 mm; 78.27% of edges, 1–3 mm; Supplementary Fig. 7).”
(4) Justification for Eigenvector Exclusion:
Strengthen and clarify the rationale for excluding eigenvectors 4 and 5 from subsequent analyses, particularly since they appear above the "knee" of the variance explained plot.
We thank the reviewer for pointing out that the variance-explained profile and the decision about biological interpretability are distinct considerations. We have clarified that G4 and G5 were excluded from the primary interpretation because their biological meaning and spatial interpretation were not sufficiently clear, not because their variance contribution was negligible. We have added this distinction to the Results.
Revised manuscript text (Page 9): “We did not include G4 and G5 in the current interpretation because their biological significance was not sufficiently clear; this decision reflected interpretability rather than negligible explained variance.”
(Interpretation)
(1) Eigenvector Interpretation and Topography:
Delineating how continuous gradients relate to or interact with discrete histological subfields would provide a more biologically grounded interpretation than relying strictly on geometry. Naming the eigenvectors based on weak geometric associations (especially in the case of G3) risks obfuscating the complex variation within the data. For full transparency, all linear and quadratic associations with all anatomical axes should be systematically reported, and a more descriptive or neutral nomenclature should be considered.
We thank the reviewer for this important comment. We agree that the biological interpretation of continuous hippocampal gradients should not be framed as geometry in isolation. We have therefore clarified that the posterior–anterior and proximal–distal coordinates are complementary anatomical descriptions of hippocampal biological organization, rather than biologically empty geometric axes.
In particular, the posterior–anterior axis captures the hippocampal long-axis organization, whereas the proximal–distal axis follows the iso-to-allocortical topology from subiculum through the Cornu Ammonis fields to dentate gyrus. This latter direction is topologically aligned with the discrete hippocampal subfields, which are themselves defined by differences in cytoarchitecture. Thus, the geometric coordinate system and the cytoarchitectonic organization provide two complementary descriptions of the same underlying hippocampal topology, rather than mutually exclusive interpretations. We revised the text to make this relationship explicit and to avoid implying that the observed gradients provide a direct histological measurement.
We also clarify that the reported gradient–axis relationships were not selected post hoc from a single geometric comparison. We systematically evaluated linear and quadratic associations between each of the three functional gradients and both intrinsic anatomical axes, and additionally assessed higher-order relationships where warranted. Figure 2b–d shows the strongest statistically supported relationship for each gradient: a linear posterior–anterior association for G1, a quadratic posterior–anterior association for G2, and a cubic proximal–distal association for G3. The remaining axis associations are reported in Extended Data Fig. 2. Accordingly, the descriptive names are based on comparative assessment of the full set of candidate geometric models, while G1–G3 remain the primary neutral identifiers throughout the manuscript.
Revised manuscript text (Page 9): “This gradient follows the medial–lateral, iso-allocortical organization of the hippocampus and is topologically aligned with cytoarchitectonically differentiated subiculum, CA fields and dentate gyrus.”
(2) Balanced Nomenclature:
Elevating the cortical component to a higher-order conceptual term ("hierarchy") while leaving the hippocampal counterpart strictly geometric ("gradients"), given that both metrics are derived using equivalent non-linear dimensionality reduction.
We thank the reviewer for highlighting this asymmetry in terminology. We agree that applying the term “hierarchy” to the cortical component without defining its evidential basis could be confusing, particularly because both cortical and hippocampal maps are obtained using nonlinear dimensionality-reduction methods.
The term cortical functional hierarchy does not arise simply from estimating a cortical gradient. It originates from longstanding models of cortical organization in which sensory systems occupy relatively lower-order positions and progressively integrated association systems occupy higher-order positions. This framework has subsequently been formalized at the macroscale as the reproducible unimodal-to-transmodal axis of functional connectivity and has been supported by convergent evidence from cortical microstructure, structure-function coupling, connectivity architecture, and transcriptomic specialization[6-9]. We have clarified this historical and evidential basis in the revised manuscript.
In the present study, the cortical functional-hierarchy map is therefore used as an independently established reference map of continuous cortical organization, rather than as a new hierarchy inferred from our hippocampal data. We use this term operationally and non-causally: it indexes each cortical region’s position along the established unimodal-to-transmodal functional-connectivity axis.
Crucially, this terminology does not assign greater biological importance to the cortical measure. Hippocampal functional gradients are the primary object of the present study, and determining their structural and functional significance is a central aim. We therefore examine their associations with hippocampal anatomical coordinates and geometric modes, myelin-sensitive MRI, transcriptomic variation, cortical functional organization, and cognition. We retain the neutral labels lLG, qLG, cIG and “hippocampal functional gradients” because they describe empirically derived axes without presupposing a biological interpretation. As convergent anatomical, microstructural, functional, and comparative evidence accumulates, a more specific organizational terminology for particular hippocampal gradients may become justified. Thus, the present distinction between “cortical functional hierarchy” and “hippocampal functional gradients” reflects the evidential role of each measure in this study, rather than a difference in scientific importance or in the legitimacy of their structure-function interpretation.
Revised manuscript text (Page 14): “Here, FH denotes the established unimodal-to-transmodal ordering of macroscale cortical functional organization, rooted in longstanding models of hierarchical cortical processing and subsequently formalized using functional-connectivity gradients.”
(3) Refinement of Textual Tone and Claims:
- Remove causal verbs (e.g., "sculpts", "constrains") when describing correlations between functional gradients, structural metrics, and cognitive scores, replacing them with appropriate relational or correlational language.
We thank the Reviewer for reinforcing the need for this terminology correction. We agree that the request to remove causal verbs describes the same manuscript-wide claim-calibration issue addressed in our response to Reviewer #1, Recommendation 1. We have therefore applied the consolidated terminology revision across the overlapping sections, including the descriptions of functional gradients, structural measures, MRI intensity, and cognitive scores.
- Temper the interpretation of the ~5% variance explained in the maturation analyses, ensuring that this modest effect size is not framed as a massive macroscale reorganization or "decoupling."
We thank the reviewer for noting that the age-related variance explained by some global gradient features is modest. We agree that the reported ∆Adjusted R<sup>2</sup> values quantify age-related variation in a specific descriptive summary—for example, the explained variance or standard deviation of a gradient, or a summary measure of geometry–function coupling—after accounting for the covariates in the model. They should therefore not be interpreted as the proportion of the entire hippocampal functional-connectivity organization that changes with age, nor as a standalone measure of the magnitude of hippocampal reorganization.
Our interpretation of developmental reorganization does not rely on these global summary statistics alone. The spatially resolved hippocampal–cortical connectivity organization shows complex variation across the hippocampal surface (Extended Data Fig. 3). In addition, the age-related analyses of hippocampal gradient projections identify distributed and system-specific developmental changes across the cortex (Figure3e–g; Extended Data Fig. 5f). Together, these complementary analyses support the interpretation that hippocampal–cortical functional organization undergoes a distributed developmental reorganization. We thus retained the description of a distributed developmental reorganization of hippocampal–cortical functional organization, while clarifying that this conclusion is supported by the spatially resolved analyses rather than by the magnitude of any individual global-summary effect size.
Revised manuscript text (Page 15): “This distributed developmental reorganization of hippocampal–cortical functional organization, which correlates with increasingly complex hierarchical cortical processing, may link to the maturation of episodic memory and executive function (EF)—a set of top-down mental processes potentially associated with cortical hierarchical organization.”
Correction of Inaccuracies
(1) Historical Context:
Revise the introduction to accurately reflect that distributed network connectivity has long been recognized as central to hippocampal function and memory, rather than framing it as a highly novel realization.
We appreciate the opportunity to clarify the historical framing. The manuscript did not intend to present distributed hippocampal-cortical connectivity as a newly discovered concept; its intended contribution is a developmental, topographic analysis of hippocampal functional gradients. We revised the surrounding wording to make this distinction explicit while retaining the established literature on distributed hippocampal connectivity.
Revised manuscript text (Page 4): “Distributed hippocampal-cortical connectivity has long been central to models of memory and cognition[10-15].”
(2) Citation Accuracy:
Address several inaccuracies where cited literature does not support the accompanying claims. I would like to highlight that this is one of the most concerning aspects of the paper, as it suggests that (at best) the primary authors made several mistakes in their citations that went unnoticed by co-authors, or (worse) it could appear as if the authors have not read the papers they cited, despite them being key to the field.
We thank the reviewer for emphasizing the importance of citation accuracy and take this concern seriously. We re-checked the source papers and the placement of the citations underlying the three statements raised below. We identified one genuine citation-placement error introduced during iterative revision: Shohamy and Turk-Browne (2013) was inadvertently retained with a developmental statement after being moved from the preceding statement on the hippocampus’s broad cognitive involvement. We have corrected this placement.
The other two cases were not instances of uncited or unread literature, but the cited papers did not support the precise wording used in the manuscript. Strange et al. was relevant to longitudinal-axis functional organization but not to the proximal–distal cytoarchitectonic context, and Vos de Wael et al. identified medial–lateral variation within separately analyzed subfields but did not establish a whole-hippocampus myelin gradient. We have therefore removed or replaced these citations, revised the associated claims, and now use Genon et al. (2021)[16] for the appropriate anatomical and microstructural-organizational context. These revisions distinguish direct evidence, anatomical interpretation, and indirect MRI-based microstructural measures more clearly.
a) "It was traditionally assumed that the hippocampus undergoes its most significant development in infancy and early childhood and achieves full maturity by the end of this period13" - Shohamy et al. do not discuss hippocampal development.
We thank the reviewer for identifying this citation mismatch. We agree that Shohamy and Turk-Browne (2013) does not support a claim about hippocampal developmental timing. During iterative manuscript revision, this reference was inadvertently retained with the developmental statement after being moved from the preceding sentence concerning the hippocampus’s broad involvement in cognition. We have restored it to that preceding sentence.
The revised historical statement is now supported by and Ghetti and Bunge (2012)[17] and Bauer (2007)[18], which describe earlier accounts emphasizing hippocampal development in infancy and early childhood. We further clarify immediately thereafter that more recent evidence supports continuing, regionally differentiated hippocampal development beyond early childhood.
Revised manuscript text (insert on Page 4): “Earlier accounts proposed that hippocampal development was most consequential in infancy and early childhood[17-18].”
b) "This gradient exhibits similarities to the previously established cytoarchitectonic variations within the hippocampus4,27. We hence termed it the cubic iso-allocortical gradient (𝑐𝐼𝐺)." - Strange et al., does not mention cytoarchitecture, and Genon et al., is only a secondary source on the matter.
We thank the reviewer for this important citation correction. We agree that Strange et al. primarily concerns functional specialization along the hippocampal longitudinal axis and does not provide appropriate support for a claim regarding the proximal–distal or medial–lateral cytoarchitectonic organization. We have therefore removed this citation from the relevant sentence.
We retained Genon et al.[16] because this review systematically synthesizes the major dimensions of human hippocampal organization and identifies the medial–lateral dimension as a key organizational axis associated with cytoarchitectonic differentiation. In our data, cIG showed a significant cubic association with the proximal–distal coordinate (Figure 2b). This coordinate is topologically aligned with the medial–lateral, iso-allocortical organization spanning subiculum, CA fields and dentate gyrus. We have clarified that Genon et al. is cited to provide this anatomical-organizational context, rather than as direct histological validation that the fMRI-derived cIG measures cytoarchitecture.
Revised manuscript text (insert on Page 9): “This gradient follows the medial–lateral, iso-allocortical organization of the hippocampus and is topologically aligned with cytoarchitectonically differentiated subiculum, CA fields and dentate gyrus[16].”
c) "Moving to the mesoscopic scale, myelin content has been found to exhibit a significant medial-lateral gradient along the hippocampus26." - this is not evidenced by the referenced paper (Vos de Wael et al.), and it is not true (see Abrahim et al., 2010).
We thank the reviewer for raising this important distinction. We agree that Vos de Wael et al. does not support a claim of a continuous medial–lateral gradient in hippocampal myelin content. Although that study identified medial–lateral variation in functional-connectivity organization when hippocampal subfields were examined separately, it was not designed to establish a whole-hippocampus myelin gradient. We have therefore removed Vos de Wael et al. as support for this statement and revised the text to remove the unsupported myelin interpretation.
We now cite Genon et al. (2021), which synthesizes evidence that local hippocampal microstructural variation, including T1w/T2w-intensities-based estimates that are sensitive to myelin, is organized predominantly along the medial–lateral and ventrodorsal axes. The review further notes partial spatial overlap between medial–lateral functional gradients, myelin-sensitive MRI estimates, and subfield differentiation.
We respectfully note that Ábrahám et al. (2010) does not contradict this spatial interpretation. Its histological study examined the developmental timing and layer-specific distribution of myelination across ages; it did not test a surface-wide medial–lateral spatial pattern across the hippocampal formation. It is therefore not evidence for a continuous medial–lateral myelin gradient, but neither does it provide evidence against such an organizational pattern.
Revised manuscript text (insert on Page 5): “local microstructural variation in the hippocampal formation, including myelin-sensitive T1w/T2w intensities, is organized predominantly along the medial–lateral and ventrodorsal axes along the hippocampus and partially overlaps with subfield differentiation[16].”
(3) Other:
a) Clearly define or rephrase ambiguous phrasing, such as "Myelination during nurturing process."
We thank the reviewer for identifying this ambiguous wording. We agree that “nurturing process” was imprecise in this general background statement. We have therefore revised the Introduction to state: “Myelination during brain development has been identified as a key factor in neural plasticity and circuit refinement, affecting neural activity synchronization[19-23].”
b) Re-evaluate the evolutionary claim in the discussion. It's unclear why the authors assert that the longitudinal axis is uniquely more conserved. The iso-to-allocortical axis is highly conserved, observed even in reptiles.
We thank the reviewer for this important clarification. We agree that the original Discussion conflated the evolutionary conservation of hippocampal anatomical axes with the interpretation of the functional gradients. We have therefore removed the statements that the linear long-axis gradient is more evolutionarily conserved and may consequently depend more on shared gene-expression patterns, as well as the statement that cIG reflects the portion of the hippocampal–cortical connectome influenced by cytoarchitectonic differentiation.
We retained the broader comparative interpretation, but revised it to make its evidential status explicit. G2 showed the strongest association with cortical functional hierarchy. Given the greater prominence of hierarchical cortical processing in humans and macaques than in marmosets or rodents, this association raises the possibility that G2 is less evolutionarily conserved than G1 and G3. However, this is a hypothesis motivated by the observed cortical association, rather than a conclusion directly tested by the present human developmental data. We now state this limitation explicitly and identify direct cross-species comparisons as necessary to evaluate it.
Revised manuscript text (insert on Page 25): “Given the greater dominance of hierarchical processing in humans[24] and macaques[25] in contrast to other species such as marmosets[26] or rodents[27], the strong association of quadratic long-axis gradient with cortical functional hierarchy raises the possibility that it is less evolutionarily conserved than linear long-axis gradient and cubic iso-allocortical gradient. Future cross-species comparisons could test this hypothesis.”
c) Frame the T1w/T2w intensity variations cautiously as an indirect proxy reflecting multiple tissue properties, incorporating recent literature (e.g., Sandrone et al., 2023) to explicitly acknowledge the confounding effects (e.g., transmit field bias).
We thank the reviewer for emphasizing the need for a cautious interpretation of T1w/T2w intensity. We agree that it should not be treated as synonymous with myelin or as a direct measure of myelin content. Following its established use as a myelin-sensitive cortical MRI contrast (Glasser and Van Essen, 2011), we now consistently describe T1w/T2w as myelin-sensitive measures. Its variation may reflect myelin-related tissue properties, as well as other tissue and acquisition factors, including transmit-field bias.
We have incorporated Sandrone et al. (2023)[4] as an important qualification. We respectfully note that this study evaluated corpus-callosum white matter and did not examine cortical gray matter or hippocampal tissue. It therefore supports caution in interpreting T1w/T2w, but cannot itself establish or invalidate the hippocampal gray-matter association reported here. Accordingly, we interpret our result only as an association between hippocampal functional gradients and an indirect MRI contrast, not as evidence of myelin, cellular myelination, or a myelin-mediated developmental mechanism. We revised the relevant Introduction, Results, Figure 5 legend, Discussion, and Limitations accordingly.
Revised manuscript text (insert on Page 30): “T1w/T2w intensity was treated as an indirect, myelin-sensitive MRI proxy influenced by multiple tissue and acquisition factors, not as direct histological myelin content.”
Reference
(1) Royer, J. et al. Opportunities and pitfalls of data contextualization in neuroimaging. Nat. Rev. Neurosci. 27, 416-434, doi:10.1038/s41583-026-01038-0 (2026).
(2) DeKraker, J. et al. Automated hippocampal unfolding for morphometry and subfield segmentation with HippUnfold. eLife 11, e77945, doi:10.7554/eLife.77945 (2022).
(3) Evaluation of surface-based hippocampal registration using ground-truth subfield definitions. doi:10.7554/eLife.88404 (2023).
(4) Sandrone, S. et al. Mapping myelin in white matter with T1-weighted/T2-weighted maps: discrepancy with histology and other myelin MRI measures. Brain Struct. Funct. 228, 525-535, doi:10.1007/s00429-022-02600-z (2023).
(5) Zeng, D., Li, D., Li, Q., He, Y. & Li, S. Distinct cortical connectome organization of hippocampal subfields is associated with episodic memory. Sci. Bull. 70, 33-37, doi:https://doi.org/10.1016/j.scib.2024.05.014 (2025).
(6) Margulies, D. et al. Situating the default-mode network along a principal gradient of macroscale cortical organization. Proceedings of the National Academy of Sciences 113, 12574-12579, doi:10.1073/pnas.1608282113 (2016).
(7) Huntenburg, J. M., Bazin, P.-L. & Margulies, D. S. Large-Scale Gradients in Human Cortical Organization. Trends Cognit. Sci. 22, 21-31, doi:10.1016/j.tics.2017.11.002 (2018).
(8) Burt, J. B. et al. Hierarchy of transcriptomic specialization across human cortex captured by structural neuroimaging topography. Nat. Neurosci. 21, 1251-1259, doi:10.1038/s41593-018-0195-0 (2018).
(9) Wang, X.-J. Macroscopic gradients of synaptic excitation and inhibition in the neocortex. Nature reviews. Neuroscience 21, 169-178, doi:10.1038/s41583-020-0262-x (2020).
(10) Genon, S., Reid, A., Langner, R., Amunts, K. & Eickhoff, S. B. How to Characterize the Function of a Brain Region. Trends in Cognitive Sciences 22, 350-364, doi:10.1016/j.tics.2018.01.010 (2018).
(11) Shohamy, D. & Turk-Browne, N. Mechanisms for widespread hippocampal involvement in cognition. Journal of Experimental Psychology: General 142, 1159-1170, doi:10.1037/a0034461 (2013).
(12) Moscovitch, M., Cabeza, R., Winocur, G. & Nadel, L. Episodic memory and beyond: the hippocampus and neocortex in transformation. Annual review of psychology 67, 105-134 (2016).
(13) Ranganath, C. & Ritchey, M. Two cortical systems for memory-guided behaviour. Nature reviews neuroscience 13, 713-726 (2012).
(14) Barnett, A. J. et al. Hippocampal-cortical interactions during event boundaries support retention of complex narrative events. Neuron 112, 319-330.e317, doi:10.1016/j.neuron.2023.10.010 (2024).
(15) Eichenbaum, H. A cortical–hippocampal system for declarative memory. Nature Reviews Neuroscience 1, 41-50, doi:10.1038/35036213 (2000).
(16) Genon, S., Bernhardt, B. C., La Joie, R., Amunts, K. & Eickhoff, S. B. The many dimensions of human hippocampal organization and (dys)function. Trends in Neurosciences 44, 977-989, doi:https://doi.org/10.1016/j.tins.2021.10.003 (2021).
(17) Ghetti, S. & Bunge, S. A. Neural changes underlying the development of episodic memory during middle childhood. Developmental Cognitive Neuroscience 2, 381-395, doi:10.1016/j.dcn.2012.05.002 (2012).
(18) Bauer, P. J. Recall in infancy: A neurodevelopmental account. Current Directions in Psychological Science 16, 142-146, doi:10.1111/j.1467-8721.2007.00492.x (2007).
(19) Bonetto, G., Belin, D. & Káradóttir, R. T. Myelin: A gatekeeper of activity-dependent circuit plasticity? Science 374, eaba6905, doi:doi:10.1126/science.aba6905 (2021).
(20) Pease-Raissi, S. E. & Chan, J. R. Building a (w)rapport between neurons and oligodendroglia: Reciprocal interactions underlying adaptive myelination. Neuron 109, 1258-1273, doi:10.1016/j.neuron.2021.02.003 (2021).
(21) McGee, A. W., Yang, Y., Fischer, Q. S., Daw, N. W. & Strittmatter, S. M. Experience-Driven Plasticity of Visual Cortex Limited by Myelin and Nogo Receptor. Science 309, 2222-2226, doi:doi:10.1126/science.1114362 (2005).
(22) de Faria, O. et al. Periods of synchronized myelin changes shape brain function and plasticity. Nature Neuroscience 24, 1508-1521, doi:10.1038/s41593-021-00917-2 (2021).
(23) Fields, R. D. A new mechanism of nervous system plasticity: activity-dependent myelination. Nature Reviews Neuroscience 16, 756-767, doi:10.1038/nrn4023 (2015).
(24) Dong, H.-M., Margulies, D. S., Zuo, X.-N. & Holmes, A. J. Shifting gradients of macroscale cortical organization mark the transition from childhood to adolescence. Proceedings of the National Academy of Sciences 118, e2024448118, doi:doi:10.1073/pnas.2024448118 (2021).
(25) Xu, T. et al. Cross-species functional alignment reveals evolutionary hierarchy within the connectome. NeuroImage 223, 117346, doi:https://doi.org/10.1016/j.neuroimage.2020.117346 (2020).
(26) Tong, C. et al. Multimodal analysis demonstrating the shaping of functional gradients in the marmoset brain. Nature Communications 13, 6584, doi:10.1038/s41467-022-34371-w (2022).
(27) Huntenburg, J. M., Yeow, L. Y., Mandino, F. & Grandjean, J. Gradients of functional connectivity in the mouse cortex reflect neocortical evolution. NeuroImage 225, 117528, doi:https://doi.org/10.1016/j.neuroimage.2020.117528 (2021).
eLife Assessment
This study provides a valuable resource describing the size, density, and synaptic proximity of mitochondria across thousands of neurons in the Drosophila brain, drawn from a well-annotated electron microscopy dataset. The evidence supporting the main descriptive findings is mostly solid, though some measured parameters contain internal inconsistencies and the proposed links between mitochondrial properties and neural activity or energy use remain speculative. The work will interest neuroscientists studying mitochondrial biology, energetics, and connectomics, and lays a foundation for future studies connecting mitochondrial organization to neural circuit function.
Reviewer #1 (Public review):
Summary:
In this manuscript, the authors build a connectome-scale resource for studying mitochondria in the Drosophila Hemibrain. The authors segment and classify millions of mitochondria, quantify their size, abundance, appearance, and spatial relationship to synapses across thousands of neuron types, and extend neuPrint so that these organelles can be queried together with neuronal connectivity. The analyses identify substantial mitochondrial variation across cell types, systematic differences between pre- and postsynaptic mitochondrial organization, associations with neurotransmitter class, and changes in mitochondrial properties with synaptic fan-out.
Strengths:
The scale and accessibility of the resource are the major strengths of this work. The mitochondrial annotations are integrated with an established connectomic dataset and can be queried through neuPrint, which should make the study useful to researchers beyond the immediate biological questions addressed here. The manuscript also describes the underlying data model and how mitochondrial information is linked to neurons and synapses.
The manuscript compares mitochondrial organization across multiple levels, including individual cells, cell types, brain regions, synaptic polarity, neurotransmitter class, and synaptic fan-out. Several descriptive results are particularly clear, including the tendency for mitochondria to be closer and larger at presynaptic than postsynaptic sites and the identification of cell-type-specific patterns.
The biological interpretations are generally placed in the context of prior experimental literature on synaptic energy demand, calcium handling, and mitochondrial positioning. In many places, the authors appropriately use cautious language such as "may," "possible," and "consistent with," which helps distinguish mechanistic hypotheses from direct measurements. The cited literature therefore provides a reasonable biological framework for interpreting the structural observations, even where the corresponding physiological variables are not measured in this specimen.
The authors also acknowledge important limitations of the Hemibrain sample, including incomplete neurons, regional differences in reconstruction quality, and the fact that the analysis represents a single physiological snapshot. The public code and data infrastructure further increase the utility and reproducibility of the study.
Weaknesses:
The dataset has extraordinary cellular coverage but derives from one Hemibrain specimen. Claims about "natural ranges," generality across Drosophila neurons, or organizational principles extending beyond this specimen should therefore be phrased with some caution. This is particularly relevant for cell types represented by only one or a few instances. For example, MBON13 is represented by a single cell in this dataset, so its unusually high mitochondrial investment is well supported for that instance, but variability of this property within the cell type cannot be assessed here. This is consistent with the authors' own acknowledgement in the Discussion that detailed conclusions for individual cell types may require larger numbers of cells.
The mitochondrial appearance classes support several biological conclusions, particularly the differences associated with brain region and neurotransmitter class. The manuscript describes how dark, medium, and light training labels were generated and reports strong validation of mitochondrial detection and nearby-mitochondrion recall, but I could not find a corresponding independent validation of the three appearance labels themselves. Some assessment of classification reproducibility, for example using a held-out manually assessed set, a confusion matrix, or inter-annotator agreement, would strengthen confidence in conclusions that depend on these categories.
Reviewer #2 (Public review):
Summary:
Electron microscopy (EM) offers resolution of neuronal ultrastructure not available through any other technique, but it has a limited "field of view". Also, an EM study's value is limited by the ability to identify neuron types within that field of view. This data set largely overcomes both limitations, first, through the analysis of contiguous blocks of SEM-FIB data spanning almost half a millimeter in extent, serially milled at an exceptionally fine 8 nm to create isotropic 8 nm voxels, and, secondly, through efforts that identified most neurons within the sectioned volume. In overcoming the usual limitations, the authors can offer an overview of the capacity of different classes of neurons to control the size and distribution of their mitochondria - a non-trivial outcome considering the essential role of mitochondria in generating cellular ATP and neurotransmitter precursors. The science community is doubly served by the authors' website "neuPrint" which is accessible to the public, where the data can be visualized and interrogated according to various hypotheses. As it stands, this study fails to reach its otherwise exceptional potential as analysis errors throw a shadow over the findings.
Strengths:
(1) This is the first comprehensive analysis of mitochondrial morphometrics and ultrastructural positioning across a substantial portion of an insect brain. The adult fruit fly brain was previously milled at 8 nm intervals and imaged at 4 nm/pixel (Scheffer et al., 2020), offering an unprecedented data set for mining mitochondrial parameters relevant to neuronal function (this manuscript).
(2) The value of the data set has been amplified through systematic identification of neuronal identities and likely neurotransmitter types. The numbers of mitochondria (>6,000,000), neurons (>20,000), and neuron types (5,555) included in the study are similarly unprecedented.
(3) The language of the manuscript is straightforward and accessible.
Weaknesses:
(1) The data describing mean median (mean of the medians derived from many neuron types) mitochondrial volume (Fig.2B) and mean median length (Fig.2C) are not internally consistent, i.e. an ellipsoid with volume 0.15 cubic microns requires a longest axis of at least ~0.66 microns to remain a valid prolate shape (i.e., for the long axis to exceed the two minor axes); the reported value of 0.55 microns falls well short of this, indicating the two panels are not describing a geometrically coherent object. This inconsistency can be reconciled only by looking at the supplementary data, where the data are reported for each cell type and the relevant column is titled "Median miitochondrion long axis (r0)" (sic). Here it is evident that the parameter being reported on each plot in Figure 2C is in fact the radius of the long axis and not mitochondrion "length" as currently stated.
(2) However, even with the correction referred to above, the mean median mitochondrial diameter, i.e., minor axes, would be 0.5 microns, a robust mitochondrion for otherwise fine central processes, and double the diameters reported from C.elegans motor neurons (Riboul et al., 2024) and interneurons within the optic lobe of the miniature parasitic wasp (Fischer et al., 2017). While it is a dubious exercise to reverse engineer possible ellipsoid body dimensions from the mean of medians of both longest axis and volume, the inconsistency remains when referring to the mean of medians of smaller groups of neurons in the supplementary data.
(3) The estimates of mitochondrial volume density are uncommonly high, i.e., 4%-24%, with a mean median of 10%. These values are 2-3 fold greater than estimates from comparable EM studies, e.g. 4-6% for C.elegans motor neurons (Riboul et al., 2024); 4-6% for Drosophila larval motor neuron terminals (Justs et al., 2022); and 4-8% for lamina-medulla and medial-medulla interneurons in the miniature parasitic wasp (Fischer et al., 2017). Estimates from rat brain are similarly small, not exceeding 8% (e.g., Santuy et al., 2018), leading to concerns that, just like the two-fold error in reporting mitochondrial lengths, a two-fold error has propagated in mitochondrial volume density.
(4) The internal inconsistency in reported mitochondrial axis length can be identified from the manuscript's own numbers alone, using simple arithmetic. By contrast, the plausibility of the resulting mitochondrial minor-axis diameter (~0.5 microns), and of the reported mitochondrial volume density (4-24%, median 10%), can only be assessed by comparison with the wider literature - where comparable EM-based estimates across several species consistently fall in the 0.18-0.24 micron (diameter) and 4-8% (density) range, respectively. Both this literature comparison and the simple arithmetic used to establish the axis-length inconsistency were available to a reviewer. The other parameters reported throughout the manuscript are inscrutable, i.e., not readily checked by arithmetic or by comparison with existing literature. Nevertheless, the errors and implausibilities identified above throw a shadow over every other piece of data.
(5) Putting errors to one side, if the authors wish to report values in an unprecedented range, they need to do so with due diligence and defend the novelty of their data in the context of precedence in the field.
Assessment of whether aims were met:
The authors did not achieve their aims, but those aims remain well within their reach, as processing data sets is generally trivial relative to the physical act of repeating experiments and recapturing ideal experimental conditions.
Discussion of the likely impact:
This manuscript has great bones, and should it receive the attention/rigor it deserves, it is likely to become the go-to reference for mitochondrial morphometrics and positioning across whole central neurons in Drosophila, specifically, and in insects in general.
Several of the findings in this study, should they persist through reanalysis, are likely to be impactful. A particularly interesting finding is that the size of the closest mitochondrion is larger at pre- rather than post-synaptic compartments (line 309; Figure 7). The dogma in neuroscience, mostly stemming from studies on vertebrate neurons and partly supported by the misguided use of fluorescence microscopy to assess mitochondrial size and distribution, is that mitochondria are larger in dendrites than in axons. The ability to assess a compartment as either pre- or post-synaptic in the present study gets to the heart of the matter about pre- vs post-synaptic dependence on mitochondrial function without getting drawn into arguments about more macroscopic cell structures that may or may not be preserved between insects and vertebrates. This finding is worth discussing in the context of the mouse data from Sager et al., 2026, Science, but it is a discussion that requires consideration of the relative volume of the pre- vs post-synaptic compartments, and whether mitochondria have the same opportunity to closely attend postsynaptic compartments.
A second finding, and one that might have received more attention in the discussion, was the observation: "KC presynapses with smaller median distances to mitochondria have larger nearest mitochondria, ... . For KCs and MBONs, the mitochondrial volume is inversely correlated with the distance". There are several avenues here that might have been explored. Firstly, in the context of data processing, and secondly, in the context of bioenergetics, the implication of greater intermittence of input at any particular postsynapse, and less intermittence of output at any particular presynapse. Lastly, implications with regard to the relative reliance on mitochondria for neurotransmitter synthesis between pre- vs post-synaptic compartments, an implication that might be addressed through reference to the particular neurotransmitters of each neuron class and whether that neurotransmitter is directly dependent on neighboring mitochondria for precursors.
Reviewer #3 (Public review):
Summary:
This manuscript presents a resource describing the morphology and distribution of mitochondria across a large part of the central brain of Drosophila. In addition to providing a valuable resource for the community, the authors analyze their dataset in great detail. These analyses reveal interesting differences in mitochondria across cell types and brain regions with regard to morphological parameters, their spatial relationship to pre- and post-synapses, differences in mitochondrial characteristics between neurons that use different neurotransmitters, and between synapses with multiple postsynaptic partners.
Strengths:
The main strength of the paper is the large size of the dataset comprising ca. 6 million segmented mitochondria distributed across more than 20000 neurons. This dataset provides enough statistical power to detect differences in mitochondrial characteristics between brain regions and cell types. An additional strength is the easy accessibility of the data through the neuPrint web interface, enabling the community to explore the dataset and perform further analyses.
Weaknesses:
The study is primarily descriptive. While this is not necessarily a weakness, the manuscript could more clearly discuss potential applications of the dataset beyond pure characterization. Potential improvements include neurotransmitter predictions, cell type classification, or sites of gap junction coupling.
Conclusion:
Taken together, I consider this manuscript to provide a highly valuable resource and analysis that will be useful to researchers interested in mitochondria and neuronal cell biology in Drosophila. Furthermore, the dataset provides potential for researchers to link mitochondrial characteristics to neuronal physiology.
Reviewer #4 (Public review):
Summary:
Longden et al. present a comprehensive analysis of all the mitochondria in the hemibrain dataset of the Drosophila brain. They quantify variations in mitochondrion size, spatial distribution, and proximity to pre- and post-synapses that reveal stark differences between cell types and brain regions. Similar to findings from other model systems, their results indicate that mitochondria are larger and closer to presynapses than to postsynapses. They also show that inhibitory neurons have more mitochondria by volume than excitatory neurons. Furthermore, they have made all this data available through neuPrint, making it a valuable resource for the broader research community. The evidence for their conclusions is compelling based on the unprecedented breadth of analysis across cell types and brain regions.
Strengths:
The comprehensive nature of their analyses across all of the mitochondria in the dataset enables Longden et al. to describe each observation in the context of individual neurons, cell types, and brain regions. The authors use this analysis structure to highlight cell types that possess unique mitochondrial metrics, such as the Kenyon cells
The reconstructions and results from this paper have been incorporated into neuPrint, making these data accessible to a wider set of researchers and scientific investigations.
Weaknesses:
The motivation for this study could be enhanced with a more general overview in the introduction. Answering some of the following questions would provide more context as to the interpretation of their results and the value of extending neuPrint to include mitochondria. How does the distribution of mitochondria in a cell enable function? What is the biological implication of having mitochondria near a synapse? What do we know about mitochondria from other modalities, and how does the information from the connectome extend or corroborate what we already know?
Similarly, this paper would benefit from more contextualization of their results within the broader study of mitochondria, both in the field of fly neuroscience and connectomics across species. A few of these studies are rightfully referenced; however, a more direct discussion of how the results from Longden et al. sit within the current landscape would highlight the value of this effort and broaden the readership of the manuscript.
Author response:
We thank the Editors and reviewers for their close reading of our work and valuable comments. We appreciate their recognition of the foundational nature of our study. The reviews suggest multiple improvements that we plan to implement in a revised version, including the following: validation of appearance labels; short self-contained example code for analysing the dataset; greater comparison with prior work in the discussion; more details in the methods. The reviews also make many suggestions for improving the presentation, wording, and clarity of the work. In particular, we acknowledge that we used incorrect terminology in referring to the axes rather than the semi-axes of fitted ellipsoids. The revision will correct this error and provide the full parameters of the ellipsoid fits (r0, r1, r2) in the supplementary data, allowing the consistency of these fits with mitochondrial volume to be appreciated. We did identify that the values of one of the mitochondrial variables not used for the analysis in the paper (r2) had been corrupted in the uploading of the publicly available hemibrain:v1.2.1 dataset to neuprint. Accordingly, we have generated hemibrain:v1.2.3 which restores the original, correct and uncorrupted values of this variable.
eLife Assessment
This is a valuable study that uses several operant paradigms to test Drosophila for the appetitive or aversive behavioral effects of self-activation of specific dopaminergic neuron (DAN) populations. The work provides solid data, although the range of paradigms and genetic driver lines will require more specific future evidence in case-by-case assays. The results suggest that a more nuanced view of the role of these neurons in ongoing behaviour and learning is necessary. The results will be of broad interest to the neuroscience community.
Reviewer #1 (Public review):
This study addresses an important question: whether dopaminergic activity in Drosophila functions as a general value signal across behavioral contexts. Its main strength is the breadth of the experimental program, combining multiple DAN driver lines, several operant paradigms, rescreening, and more targeted follow-up experiments. The cross-paradigm design is well suited to test whether self-stimulation effects generalize.
The main weakness is that several conclusions extend beyond what the experiments directly establish. The genetic tools define populations rather than the same identified neurons at single-cell resolution. In the key time-course experiments, the data more clearly support a reduction of initial avoidance toward neutrality than a consistent aversive-to-appetitive sign reversal. Similarly, the non-contingent comparison uses continuous pre-exposure rather than a stimulation-history-matched open-loop condition; it therefore shows that simple non-contingent exposure is insufficient, but does not isolate agency itself as the causal variable.
The general claims are also constrained by incomplete overlap among the initial screens and by design changes between the original and follow-up joystick experiments. Overall, the study provides useful evidence that the behavioral effects of DAN activation are weak, context-dependent, and time-dependent, while the claims about neuronal identity, positive valence, agency, and a general dopamine "common currency" require more cautious framing.
Reviewer #2 (Public review):
Summary:
Broadly, the authors use a wide range of lines targeting dopaminergic neurons in different behavioral paradigms to determine their role in behavior.
Strengths:
There is a comprehensive use of lines and assays by a single group, such that internally consistent comparisons can be made.
Weaknesses:
It is not clear to what extent these broad assays will provide concrete answers that enable a "unified hypothesis" of how dopamine functions in behavior.
The manuscript has a feel of being a mix of an original paper, commentary, hypothesis-generating outlook, a focused summary, and a broad review of the field. As someone outside this field, I find that it has its advantages and disadvantages. For example, the final sentence of the Abstract, despite multiple readings, is not obvious to someone outside the field. Plus, due to the large number of references for any one view, it is not clear which work is saying exactly what. For example, which of the fly papers on dopamine clearly say they favor the "common currency" model? Do these references also use this term? Even for a reference that occurs multiple times (Jovanoski and Waddell, 2025), it's not clear what this paper actually proposes.
Also, what does "a reporter for a specific goal" mean (Line 16 of Abstract, and lines 1986-1987 of Discussion)? They provide a nice example of a tethered fly and what happens when it is suddenly released, vs a free-flying fly that has entered a maze--what is the difference here in terms of dopamine function and the fly's "goal" and how does the fly know the difference? When a fly is free flying in a maze, how does it know that it has just been released from a tether, or whether it has entered a maze as it was freely flying outside it?
And for their "unified hypothesis", how does a fly know that it is in "controls its environment" versus "lack of control"?
In sum, I found this paper to be valuable in bringing up the different views of dopamine signaling. However, it was a bit unclear at the end as to what dopamine is actually doing in the fly.
Reviewer #3 (Public review):
Summary:
Dopaminergic neurons (DANs) are known to play a role in conveying rewarding and punishing signals that influence future behaviour. This study uses optogenetic manipulation of a range of subpopulations of DANs to assess their influence in several operant learning paradigms, such that the animal's own behaviour determines the DAN activation. The results show surprisingly inconsistent effects and do not support the widespread assumption that specific DANs consistently convey either positive or negative valence. In one case, the apparent valence of the DAN activation switches over a time frame of minutes from aversive to attractive, and the relevant circuits are further dissected to show this effect is dopamine-dependent, for a specific DAN subset, and contingent on operant control of behaviour.
Strengths:
This study has two major strengths. The first is to address the relatively neglected field of operant learning in Drosophila. The detailed unpicking of learning circuits in the Drosophila brain that has been made possible by neurogenetics and other advances has almost exclusively focussed on classical conditioning paradigms, but these are often not representative of the real behavioural situations that animals face. Cross-comparison of the behavioural results of identified neuron activation in a wider range of paradigms is very welcome. The second is the use of multiple paradigms and a wide-ranging screen to survey the effects across many lines (targeting different DAN populations), including many that have been previously tested in classical conditioning. A further strength is the useful review (in the introduction and discussion) of existing research that already raises problems with the view that a single dimensional positive and negative "value" (as conceived of in standard reinforcement learning theory) can be directly mapped to the activity of two broad clusters of DANs in the brain.
Weaknesses:
The main weakness is that it is difficult from the manuscript to understand what the authors believe is the underlying mechanism by which DAN activation should affect behaviour in operant paradigms, and how this should relate to the better-known role of DAN activity in classical conditioning. For example, they refer to Heisenberg and Wolf's 1991 distinction between operant activity and operant learning. That paper proposed a mechanism involving a goal, some degree of random variation in action, comparison of sensory input to a goal, correlation of the efference copy of the action to the output of this comparison, with the strength of this correlation subsequently shaping the action probabilities. It would help to be explicit about what DAN activity is assumed to represent in this scheme - the outcome of comparison to a goal? The actual assumptions behind the work seem more simplistic: if different groups of DANs 'signal' reward or punishment, animals should persist in actions that activate the reward DANs and desist in actions that activate the punishment DANs. This might be justified by the standard way that innately rewarding/punishing stimuli are operationally defined in learning theory. But what brain circuitry is assumed to underlie the variation and choice of actions, and how is this influenced by DANs?
Perhaps what is most clearly shown is that avoidance behaviour to activation of innately punishing sensory neurons is not directly a consequence of downstream activation of specific DANs, such that DAN activation can be substituted to produce the same behavioural phenomena (at least for the DANs tested here). Indeed, this is how the core results are summarised at the start of the discussion. But should this have been expected? There are a lot of assumptions needed to connect the expected effect of a reinforcement signal in classical conditioning to its effect on operant behaviour that are not made sufficiently explicit.
Appraisal:
The authors met their aim to carry out a (fairly) comprehensive test of the effects of optogenetic activation of a variety of DAN subpopulations under several operant paradigms that differ in relevant respects (group vs. individual testing, freely moving vs. tethered). The lack of consistent results across these paradigms, and generally weak effects (with limited statistical analysis) make it difficult to interpret whether the outcome here is mostly 'noise', but perhaps this is an example where the failure to find a clear effect (as predicted) is itself informative. The paper appropriately describes the limitations (such as the mixed range and specificity of DAN subsets manipulated across the paradigms) which were probably unavoidable given the tools available.
Likely impact on the field:
This work should have an impact in countering assertions that dopamine neurons in the fly brain can be understood as (collectively) conveying a single dimension of valence, from positive to negative, across all behavioural and learning situations. It also provides strong motivation to more clearly understand the nature of operant learning and how/where it might be implemented in the insect brain.
Author response:
We thank the reviewers and the Reviewing Editor for their careful and constructive assessments. We find the reviews fair and helpful, and we plan to revise the manuscript along the lines they suggest. In brief:
Scope of the conclusions (Reviewer #1):
We can see how a reader may get the impression that some of our conclusions go beyond what the experiments directly establish. In the revision we will:
- Describe the genetic tools as targeting neuronal populations (DAN clusters) rather than identified single neurons, and adjust statements about neuronal identity accordingly.
- Rephrase the description of the time-course results, to emphasize that the shift from aversive to appetitive only appears across experiments and that some single experiments may only show a shift to neutrality. This should help moderate claims about positive valence.
- State explicitly that our non-contingent comparison used continuous pre-exposure rather than a yoked control. It therefore shows that simple non-contingent exposure is insufficient but does not isolate agency as the causal variable.
- Better motivate the design changes between the original screen and follow-up joystick experiments.
- Frame the “common currency” hypothesis more cautiously throughout.
Clarity and accessibility (Reviewer #2):
We appreciate that the manuscript is hard to follow for readers outside the field. We will revise the final sentence of the Abstract and clarify what we mean by a dopaminergic “reporter for a specific goal”. We will also spell out how the contrast between a tethered fly and a free-flying fly in a maze bears on this, and what we mean by “control” versus “lack of control” of the environment. In the Introduction and Discussion we will make clearer which cited studies propose or favor the “common currency” view, including Jovanoski and Waddell (2025), and which ones merely bear on it.
Conceptual framework (Reviewer #3):
We will make our assumptions linking the role of DANs in classical conditioning to their expected effects in operant paradigms more explicit. The reference to the Heisenberg and Wolf (1991) scheme as a suitable level of explanation was particularly helpful and we will thrive to develop our hypotheses at the same level. At least for the designers of the original screen (BB and CR), the lack of consistent effects was clearly unexpected, or we would have designed the experiments differently. We will justify these expectations and from this why we consider this negative result to be so informative.
Code availability (Reviewing Editor): We will deposit the code in a repository with a permanent DOI, such as Zenodo, in addition to GitHub.
eLife Assessment
This important study provides a detailed analysis of the transcriptional landscape of the mouse hippocampus in the context of various physiological states. The main conclusions are convincingly supported: that most transcriptional targets are generally stable, with notable exceptions in the dentate gyrus and with regard to circadian changes. The authors comprehensively addressed the concerns raised in the first round of review.
Reviewer #1 (Public review):
Olmstead et al. present a single-cell nuclear sequencing dataset that interrogates how hippocampal gene expression changes in response to distinct physiological stimuli and across circadian time. The authors perform single nucleus RNA sequencing on mouse hippocampal tissue after (1) kainic acid-induced seizure, (2) exposure to an enriched environment, and (3) at multiple circadian phases.
The dataset is rigorously collected, and a major strength is the use of the previously established ABC taxonomy from Yao et al. (2023) to define cell types. The authors further show that this taxonomy is largely independent of activity-driven transcriptional programs. Using these annotations, they examine activity-regulated gene expression across neuronal and glial subclasses. They identify ZT12, corresponding to the transition from the light to the dark period, as transcriptionally distinct from other circadian time points, and show that this pattern is conserved across many cell types. Finally, they test how circadian phase influences activity-dependent gene expression by exposing mice to an enriched environment at different times of day, and report no significant interaction between circadian phase and enriched environment exposure.
Comments on revised version.
The authors have addressed the comments from the first round of review.
Reviewer #2 (Public review):
Olmstead et al. present ACT-DEPP, a single-nucleus RNA-sequencing atlas of the mouse hippocampus. The atlas profiles transcription after environmental enrichment, after kainate-induced seizure, and at four circadian time points, and all nuclei are registered to the Allen Brain Cell taxonomy. The authors report three main findings. First, seizure and novel experience induce transcriptional programs that differ greatly in magnitude. Second, ZT12 is a transcriptional inflection point shared across cell types. Third, induction of classical immediate early genes (IEGs) is broadly maintained across circadian phases, with only a few genes showing circadian-activity interactions.
The study is ambitious in scope, and the dataset is rigorously collected. Registering the data to the ABC taxonomy makes them easy to reuse and to integrate with other brain-wide resources. In the revision, the authors substantially strengthened the analyses and refined their interpretations. Overall, the descriptive conclusions are well supported.
Locomotor activity was not recorded, so some ZT-dependent effects may partly reflect behavioral state as well as intrinsic clock regulation. The authors clearly acknowledge this in the Limitations section. With relatively few replicates per ZT, small circadian-activity interactions may go undetected. The pathways highlighted in Fig. 6I-J are best viewed as hypothesis-generating. These points are appropriately framed and do not detract from the value of the work. This study will be a useful reference for investigators studying IEG biology, experience-dependent plasticity, and circadian regulation in the brain.
Author response:
The following is the authors’ response to the original reviews.
Public Reviews:
Reviewer #1 (Public review):
Olmstead et al. present a single-cell nuclear sequencing dataset that interrogates how hippocampal gene expression changes in response to distinct physiological stimuli and across circadian time. The authors perform single-nucleus RNA sequencing on mouse hippocampal tissue after (1) kainic acid-induced seizure, (2) exposure to an enriched environment, and (3) at multiple circadian phases.
The dataset is rigorously collected, and a major strength is the use of the previously established ABC taxonomy from Yao et al. (2023) to define cell types. The authors further show that this taxonomy is largely independent of activity-driven transcriptional programs. Using these annotations, they examine activity-regulated gene expression across neuronal and glial subclasses. They identify ZT12, corresponding to the transition from the light to the dark period, as transcriptionally distinct from other circadian time points, and show that this pattern is conserved across many cell types. Finally, they test how circadian phase influences activity-dependent gene expression by exposing mice to an enriched environment at different times of day, and report no significant interaction between circadian phase and enriched environment exposure.
A crucial consideration for users of this dataset is the potential confounding effect between circadian phase and locomotor activity. This is particularly relevant because dentate gyrus activity is strongly modulated by locomotion. The authors acknowledge this issue in the Discussion and provide useful guidance for how to interpret their findings, considering this confound.
We thank the reviewer for their accurate and thoughtful summary of the ACT-DEPP dataset and the key experimental design and analytical choices that form the basis of the study. We agree that circadian phase and locomotor activity are major confounders that are irreconcilable in the current study. We have modified our Discussion of this point to consider future studies which would be able to delineate the two (lines 403-406)
Taken together, this dataset represents a useful resource for the neuroscience community, particularly for investigators interested in how novel experience and circadian phase shape activity-related and immediate early gene expression in the hippocampus
Reviewer #2 (Public review):
This manuscript presents the ACT-DEPP dataset, a comprehensive single-nucleus RNA-sequencing atlas of the mouse hippocampus that examines how activity-dependent and circadian transcriptional programs intersect. The dataset spans multiple experimental conditions and circadian time points, clarifying how cell-type identity relates to transcriptional state. In particular, the authors compare stimulus-evoked activity programs (environmental enrichment and kainite-induced seizures) with circadian phase-dependent transcriptional oscillations. They also identify a transcriptional inflection point near ZT12 and argue that immediate early gene (IEG) induction is broadly maintained across circadian phases, with minimal ZT-dependent modulation.
Strengths:
The study is ambitious in scope and data volume, and outlines the data-processing and atlasregistration workflows. The side-by-side treatment of stimulus paradigms and ZT sampling provides a coherent framework for parsing state (activity) from phase (circadian) across diverse neuronal and non-neuronal classes. Several findings - especially the ZT12 "inflection" and the differential sensitivity of pathways across subclasses - are intriguing.
Weaknesses:
(1) The authors acknowledge, but do not adequately address, the fundamental confounding factor between circadian phase and spontaneous locomotor activity. The assertion that these represent "orthogonal regulatory axes," based on largely non-overlapping DEGs, may be overstated. The absence of behavioral monitoring during baseline is a major limitation.
We agree that circadian phase and locomotor activity are irreconcilable confounders in the current study. We have removed the phrase “orthogonal regulatory logic.” This point is disclaimed as the leading point in our Limitations section. We expand on this section to consider future studies which would be able to delineate the two (lines 403-406).
(2) The statement "Thus, novel experiences and seizures trigger categorically distinct transcriptional responses-with respect to both magnitude and specific genes-in these hippocampal sub regions" is overstated, given the data presented. Figure 2A-B shows that approximately one-third of EE-induced DEGs at 30 minutes overlap with KA DEGs, and this overlap increases substantially at 6 hours in CA1 (where EE and KA responses become "fully shared"). This suggests the responses are quantitatively different rather than "categorically distinct."
We have modified the sentence (lines 127-129) to state: “Thus, novel experience and seizure induce transcriptional programs that differ greatly in magnitude and, apart from CA1 after 6 hours, contain stimulus-specific genes.” We have additionally modified the section title to de-emphasize stimulus-specificity.
(3) In Figure 4B, "active cells" are defined as those with >=3 of 15 IEGs above the 90th percentile, with thresholds apparently calibrated in CA1. Because baseline expression distributions differ across subclasses, this rule can bias activation rates across cell types.
The baseline expression thresholds were computed for each CA1 Supertype, e.g. CA1 69-1, 702, etc. (Figure 4D).
(4) Few genes show significant ZT × stimulus (EE or seizure) interactions, concentrated in neuronal populations. Given unequal nucleus counts and biological replicates across subclasses, small effects may be underpowered.
We agree with the observation that our measurements may be underpowered to observe small effects. The nuances of our DGE methodology are discussed below under Reviewer 2’s Recommendation #3.
(5) In Figure 6 I, J, the relationship between the highlighted pathways/functions and circadian phase is not yet explicit.
We agree that there is no obvious relationship for ZT-based changes in expression among genes in shared sets highlighted in Figures 6I-J. However, their enrichment among the small set of 13 significantly interacting genes genome-wide merits remark as we have done so here. What functional implications these observations have is the subject of future experiments.
(6) Line 276-280: The enrichment of lncRNAs at ZT12 in CA1 is intriguing but underdeveloped. What are these lncRNAs, and what might they regulate?
Gene symbols were ascertained by mapping reads to the GRCm39-113 mouse genome in Parse Bioscience’s Trailmaker pipeline (Methods). GRCm39 returns gene biotype annotations, which is how we determined these genes were lncRNAs. A literature search for these gene symbols found that all 12 are predicted or uncharacterized loci without functional experimental data.
Overall, most descriptive conclusions are supported (e.g., broad phase-robustness of classical IEGs; an inflection near ZT12). Claims about the separability/orthogonality of activity vs circadian programs, and about categorical distinctions between EE and KA responses, would benefit from more conservative wording or additional analyses to rule out behavioral and power-related alternatives.
Recommendations for the authors:
Reviewer #1 (Recommendations for the authors):
I have a few suggestions for additional analyses and clarifications that could further strengthen the study:
(1) KA responsive populations and spatial context
In Figure 2, the authors show that KA-responsive cells cluster within two supertypes, which may 99 represent seizure-specific populations. Could the authors leverage the spatial transcriptomic data 100 from Yao et al. to ask whether there is a specific cell type or subregion within the dentate gyrus 101 that is predisposed to express the seizure-associated gene programs identified in Figure 2?
We thank the reviewers for this interesting analysis suggestion. We have added Figure 2—figure supplement 6 as an exploration of the spatial distribution of DG Supertype 138-3. We find that these cells are predominantly located in dorsal hippocampus, although are sparsely distributed throughout all of hippocampus.
(2) Baseline circadian modulation of canonical IEGs
In the Discussion, the authors note that canonical immediate early genes do not overlap with circadian differentially expressed genes, citing results from Figure 5. However, the current 109 analyses do not directly show how canonical IEG expression varies across the circadian cycle at 110 baseline. It would be helpful to include a focused visualization, for example, a heatmap or line plot 111 of IEG expression across circadian time points, to more directly support this conclusion.
We have added Figure 6—figure supplements 1-2. These plots demonstrate low variation of IEG expression within the standard environment across ZTs. EE-induction is likewise similar across ZTs in CA1. DG demonstrates less consistent induction relationships, likely due to dramatically low abundances leading to noisy expression estimates.
(3) Neurogenesis-related signatures
Given that dentate gyrus cells are included in the dataset, it may be informative to comment on neurogenesis-related transcriptional programs. For example, can the authors report whether markers of adult-born granule cells or progenitor populations are detected, and whether genes associated with neurogenesis are modulated by experience (seizure or enriched environment) or circadian phase? Even a brief analysis or summary would likely be of interest to the community.
We do capture a small population of immature neuronal cells (IMNs) in our dataset that are likely the product of dentate neurogenesis (Figure 1B). Our experiment was not adequately designed to detect the origin of these cells, nor determine whether EE induces changes in rate of neurogenesis or cell fates. However, we can measure the effects of acute EE on the preexisting IMN population present at the time of experimentation. We have added Figure 2—figure supplement 7 to examine this topic. We observe that hours of EE exposure instead of minutes is necessary to induce transcriptional changes. Interestingly, many of the DEGs observed encode neurotransmitter receptor subunits or intercellular adhesion proteins.
Reviewer #2 (Recommendations for the authors):
Behavioral confounding factor: Please provide locomotor activity measures aligned to ZT for the relevant cohorts (e.g., wheel counts or video tracking), or at minimum, discuss how ZT-dependent changes in arousal/locomotion could contribute to the observed transcriptional differences and how your analyses mitigate this confounding factor. Including locomotion as a covariate in pseudobulk models or presenting matched restraint controls would be helpful.
We did not collect locomotion data on the mice and so are unable to include it as a covariate in any analyses. We place the Limitations subsection at the beginning of our Discussion and disclose the confounding effect of locomotion/arousal as a main limitation of this study.
It is well established that mice increase locomotor activity in the hour preceding the dark phase (i.e. during ZT 11-12) [1]. Similarly, in vivo neurophysiological studies of hippocampus during locomotion consistently demonstrate that neuronal cells in hippocampus exhibit spatially tuned activity that increases as an animal navigates an environment compared to when it is stationary [2]. If locomotion were to exhibit major influence on gene expression in our experiment, it is reasonable to assume that we would observe such effects in IEGs. Nevertheless, we do not observe changes in baseline IEG expression across ZTs (Figure 6—figure supplement 1-2). Based on this albeit circumstantial evidence, we conclude that circadian influences on gene expression are likely to predominate over locomotor ones in the absence of other manipulations.
What may explain this lack of IEG induction across dramatically different locomotor phases? It is possible that locomotion-dependent effects on neuronal activity are insufficient to drive activity-dependent changes without instructive learning signals within the hippocampal circuit. One speculative mechanism for this could be that, in the standard environment (i.e. the animal’s home cage), the lack of strong plasticity events such as dendritic plateau potentials [3] fails to overcome the signaling threshold(s) necessary to induce activity-dependent transcriptional programs. It also remains possible that locomotion-induced transcriptional changes involving hundreds of genes such as that observed in Figure 5 are coordinated through processes independent of canonical IEG induction.
(2) "Active cell" definition: Please provide (i) a sensitivity analysis varying the percentile cutoff (e.g., 80/85/95%) and the number of IEGs required (2-4), and (ii) a complementary analysis using continuous activity scores (e.g., AUCell or ssGSEA) with concordance to the binary call reported (ROC/PR or agreement metrics). Subclass-specific thresholding or normalization would further mitigate bias; reporting the fraction of "active" cells per subclass under each parameter set would be helpful.
We have added Figure 4— figures supplements 1-2 to incorporate all these requested analyses. Rather than rely on an external library to construct an IEG module score, we employ a simple z-scoring index of the 15 IEGs used in Figure 4B, D, E to overcome the technical challenges unique to this set of low-abundance genes.
From these requested analyses, we find that ROC curves were extremely accurate at classifying active and inactive cells based on a continuous activity IEG score (AUC > 0.985, Figure 4 — figure supplement 2). Precision-recall curves were also constructed, which demonstrated decreased—although still robust—classification of cell activity state in SE. This is consistent with the performance of precision/recall tradeoff between conditions when there is an imbalance of active/inactive cells in the underlying populations (see Figure 4B). When investigating the decreased PR-accuracy in SE, we observed that discordant classifications were likely to arise when a single IEG was elevated enough to significantly increase the composite score, leading to a spurious classification as “active” with the continuous scoring method when it would not meet the 3-gene threshold in the other method.
Finally, we note that we already employed supertype-specific thresholding to set baseline thresholds for each CA1 sub-population.
(3) Interaction effects and power: Please (a) report per-subclass effective sample sizes and replicate numbers, (b) re-analyze with a complementary framework (limma-voom or edgeR) and, where possible, (c) include sample-level mixed-effects models (animal as a random effect; pseudobulk counts or cell-fraction outcomes). Presenting effect sizes with uncertainty (e.g., forest plots) would improve interpretability beyond binary FDR thresholds.
We have added Figure 2—figure supplements 1-5; and Figure 3—figure supplement 1 to address these requests and included a new paragraph in the Results section (lines 130-146) describing how these different methods perform comparatively on our dataset. We observe that these methods perform similarly under most conditions, but that DESeq2 shrinkage—the method used for all DGE analysis in the main figures—aggressively forces estimated fold-change toward zero in certain conditions with high variance. We maintain our original conclusion that Type I error rate is adequately minimized. However, Type II error rate in our DGE analyses may be higher with DESeq2 under certain conditions compared to other frameworks.
We modified our Discussion to expand on these findings (lines 415-424) and disclose to the reader that “Our findings should be interpreted as a conservative observation of activity- and circadian-dependent changes in gene expression.” We thank the reviewer for these suggestions which have significantly improved the analytic rigor of the project and revealed its potential technical shortcomings.
(4) Figure 6 I,J : Please make clear whether these pathways are simply enriched among DE genes or whether they exhibit bona fide circadian modulation.
The 13 significant genes ascertained through the interaction analysis presented in Figure 6I-J were manually annotated through literature review, given the feasibility from such a small number of DEGs. The pathways described—endocannabinoid synthesis, signaling and excitability, and chromatin remodeling/gene expression—were all cellular processes implicated by literature review of the gene list. From this, we conclude that these processes are influenced by an interaction between circadian- and activity-dependent induction of gene expression after controlling for their individual effects.
References
(1) Eckel-Mahan, K. L. et al. Circadian oscillation of hippocampal MAPK activity and cAMP: implications for memory persistence. Nat Neurosci 11, 1074–1082 (2008).
(2) Buzsáki, G. Theta Oscillations in the Hippocampus. Neuron 33, 325–340 (2002).
(3) Bittner, K. C., Milstein, A. D., Grienberger, C., Romani, S. & Magee, J. C. Behavioral time scale synaptic plasticity underlies CA1 place fields. Science 357, 1033–1036 (2017).
eLife Assessment
This important study combines cortex-wide activity imaging and optogenetics to characterize large-scale changes in functional connectivity following administration of the psychedelic DOI, as well as the circuit mechanisms underlying them. It provides solid evidence that DOI reduces the modularity of cortical networks and has concurrent effects on both cortico-cortical and cortico-thalamic functional connectivity. The study will be of broad interest to basic scientists working on the circuit mechanisms of psychedelics, as well as to researchers studying humans who may wish to reverse-translate their findings into mice.
Reviewer #1 (Public review):
Summary:
Doronin et al. performed wide-field calcium imaging to investigate the cortex-wide activity dynamics in mice after the administration of the psychedelic DOI.
Strengths:
The lab has extensive experience with this approach. The analyses are not trivial, but here the authors progressively step through the rich data to identify salient effects. The observed effects at the level of cortical networks will be of great interest to the field, both for basic scientists interested in circuit mechanisms of psychedelics and for human researchers who may want to reverse translate from humans to mice.
Weaknesses:
The analyses were extensive, but the choices made for the analyses are quite specific, and there are places where other analyses may be more appropriate and logical. For example, instead of comparing activity with serotonergic axonal density, compare with Htr2a transcript because the receptor is the drug target. Moreover, data were combined for day 1 and day 5 for an unclear reason. Furthermore, a few statements in the text did not seem fully supported by the results and may be over-interpreted.
Reviewer #2 (Public review):
Summary:
In this paper, the authors report a quantitative change in the functional connectivity between cortical domains in mice, due to the influence of the psychedelic, DOI. Additionally, they report that specific cell types (and circuits) within the brain may be involved in the changes in the observed functional connectivity. This will have important implications - both for basic neuroscience and with potential translational implications.
Strengths:
This is a fascinating result - especially since the work on psychedelics has predominantly been focused on humans - and therefore a deeper understanding of cell-type-specific responses due to psychedelics has remained poorly studied. Given that psychedelics result in a variety of distinct changes within the circuit - both affecting plasticity and the steady-state dynamics - it is critical to characterize these changes within a system where we can measure changes in individual cell types.
Weaknesses:
The authors attempt to model the types of analyses that have been performed in humans within mice. Perhaps because of their attempt to translate from one system to another, the quantitative justification for each of the specific analyses that have been performed appears unclear within the mouse system. This makes some of their analyses, as currently structured, appear incomplete, and we have detailed the resulting questions in the following comments:
(1) The concept of a "cortical domain" is used in the abstract and is a fundamental component of the paper. We were not familiar with the concept prior to its use in this paper and, given the importance of the concept, it would be helpful to clearly delineate what the authors define as a domain at the outset, so that readers can build their expectations. Further, both AL and PM have been used as acronyms for secondary visual areas (which the authors are very familiar with). This leads to further confusion for the reader. As such, it would be helpful to clearly delineate that these are distinct acronyms for a distinct class of spatial feature within the brain - even within the abstract.
(2) The authors define the AL and PM cortical domains through the use of a threshold on the seed correlation maps. They chose a threshold of 35%, but it's not clear why that threshold was chosen - and what would occur if the authors used a different threshold instead. Would the boundaries change significantly, and would the rest of their results then be affected by this change?
(3) The authors report that the behavioral events that show a maximally distinguishable response in the PM and AL domains are behavioral fidgets. However, the authors do not define either (a) the set of all possible behavioral events which they compared against (and how those behavioral events were assayed), and (b) the quantitative definition of fidgets that was used. This is particularly critical since the authors use the fidget response as a metric of responsiveness in the rest of the paper.
(4) The authors report that the AL and PM domains correlate with the projection patterns of specific neuromodulatory systems. Given the (apparently) arbitrary threshold used in the definition of the AL and PM domains, and the fact that the authors mention that they examined "some of the major neuromodulatory systems", it is unclear how to interpret this result.
(5) The authors claim that the DOI-induced increase in the correlation between the cortical domains is 5-HT2A dependent, and their evidence for this depends on a block of the 5-HT2A receptor via ketanserin, in the presence of DOI. However, the result that they report from ketanserin is a decrease in the correlations below baseline, and not simply a removal of the effect observed in DOI. This suggests that the effect of DOI on the network cannot be simplistically explained through 5-HT2A dependence, and requires some further characterization and / or explanation.
(6) The authors introduce the concept of "amplitude of activity variation". However, they do not define the specific quantification that they are using in the text. Additionally, their justification for the use of the concept is also unclear. Why should the activity of input variation change "on first approximation"? If this is a quantifiable prediction, what is the approximation? Increasing the precision of their introduction of the concept would greatly aid the reader in interpreting the results in Figure 4.
(7) The authors report that EEG responses do not directly reflect the opposing responses observed via widefield calcium imaging. They conclude that "the opposing effects we observe in calcium activity are difficult to relate to psychedelic-induced EEG effects in humans". However, their apparent effects are also not observed in mouse EEG recordings. As such, their claim is unclear - are they reporting that calcium recordings are not reflected in EEG responses, or is there a specific effect present in mice that is absent in humans?
(8) The authors compare the widefield response resulting from stimulation of specific populations of cortical and subcortical neurons prior to and under the influence of DOI. They claim that the resulting changes provide "candidate circuit mechanisms" for the effects observed during DOI. However, the observed changes due to stimulation do not match the specific changes observed during DOI, making it hard to justify their claim. Indeed, one could hypothesize that the observed changes are only due to downstream effects in a secondary population, only indirectly stimulated through the authors' paradigm. As such, clarification of their results and increased precision of their claims would be helpful. As one example, the authors could provide a model for how the specific changes that they observe during stimulation may directly result in the domain-specific changes observed due to DOI.
(9) Following up on question (8), it would also be helpful for the authors to characterize the space of stimulation paradigms within the selected neuronal populations. How specific are the changes that they observe to the specific stimulation paradigm used? Are these stimulation paradigms biologically plausible? And, if not, what precisely do we learn through this stimulation?
eLife Assessment
This study provides valuable insights into how the timing of dopamine signals can influence associative learning and memory in fruit fly larvae at the single-cell level. Combining optogenetic manipulation of individual dopaminergic neurons and computational modeling of circuit computations provides some convincing evidence for the underlying circuit-level mechanisms. However, there are also aspects of incomplete evidence that require key control and validation experiments. The work will be of interest to researchers studying the neural mechanisms of learning and memory.'
Reviewer #1 (Public review):
Summary:
This study by Weber et al. investigates how artificial activation or inhibition of dopaminergic neurons innervating the Drosophila larval mushroom bodies in the context of learning paradigms influences individual properties of acquired memories. Similar approaches have been undertaken in the past (e.g., in the adult); however, the reduced complexity of the larval system makes this study valuable, especially as it eases the possibility for informed modelling approaches, as done here. A special aspect shown here is that artificial training overrides the need for context typically observed for recall of larval aversive memories.
Strengths:
This study adds insights into how the larval mushroom bodies store memories. The behavioral data using genetically precise manipulation of identified neurons is combined with computational modelling. Information on the physiological basis of the underlying mechanism is not provided. In light of the claim of ‚temporal precise optogenetic manipulation' and the authors ‚manipulating the intrinsic physiology' of DANs, this is important and should be stated clearly. While this reviewer agrees that the ‚temporal precision' is provided for memory phases, this might not be true for physiological parameters, especially when using tools like ChR2-XXL.
Thus, this study offers interesting behavioral and cellular insights; however, this reviewer also recommends specifying claims on temporal and physiological precision more carefully.
Weaknesses:
(1) The authors should be more precise about temporal and physiological windows/properties. It is recommended to discuss the published properties of the optogenetic tools used and how this might influence the obtained results, especially when comparing the results to 'naturally' reached learning after pairing stimuli in the absence of optogenetic intervention.
(2) The authors should comment more directly as to why pairing odors with activation or inhibition of DANs does not lead to measurable learning (Figure 1).
(3) A key finding is that DL DAN activation leads to context-independent memories (as also somewhat indicated in the title). To make this claim, the 'natural' training experiments should be conducted in parallel, or these potential limitations should be mentioned. However, if the authors state that the results using regular conditioning approaches are so clear that this is not necessary, the should state that.
(4) Potential caveats of activating chloride channels over prolonged time periods should be mentioned.
(5) The appropriate literature showing odor-evoked synaptic depression to MBONs (e.g., cited in lines 431/432) in the context of learning should be cited throughout.
Reviewer #2 (Public review):
Summary:
This study investigates how the timing of dopaminergic signals influences associative learning and memory in Drosophila larvae. Using optogenetic activation and inhibition of the dopaminergic neurons DAN-f1 and DAN-g1, the authors examine their roles during different phases of aversive olfactory learning and memory testing. The study reports that manipulating these neurons can alter behavioral performance not only when the manipulation is paired with punishment, but also during non-punished periods and during memory testing. The finding that behavior induced by artificial activation of these neurons can be expressed across different test conditions is also interesting. Overall, the study provides valuable observations concerning the timing-dependent roles of dopaminergic signals in associative learning.
Strengths:
A major strength of the study is the systematic manipulation of the same dopaminergic neurons at different stages of the behavioral paradigm. This provides a useful framework for examining whether dopaminergic signals have functions beyond their conventional role as reinforcement signals during training. The study also combines optogenetic activation and inhibition with behavioral assays and measurements of neuronal responses. Several of the observations are potentially informative for understanding how dopaminergic signals contribute to memory formation and expression in a context- and trimming-dependent manner.
Weaknesses:
The main weakness is that several conclusions are stronger than warranted by the current evidence. The optogenetic control experiments do not fully distinguish the effects of light, genotype, and opsin expression, and some experimental and control genotypes already differ in naïve behavior.
In addition, some conclusions rely on whether individual groups differ from zero rather than on significant differences between experimental groups. This affects key claims concerning loss of memory formation or expression.
The comparison between optogenetic reinforcement and natural punishment is also confounded by a fixed training order, making it difficult to separate reinforcement strength from recency or interference effects.
Finally, the proposed role of baseline dopaminergic activity is not directly supported by physiological measurements in the relevant neurons under the experimental conditions.
Overall, these issues make the evidence for several of the main mechanistic conclusions incomplete.
Reviewer #3 (Public review):
Summary:
Weber et al. focused on a punishment learning paradigm in Drosophila larvae using primarily high salt as an aversive stimulus. They examined the DL1-cluster of dopaminergic neurons (DANs) consisting of four neurons, mostly important for encoding such punishment. Using temporally precise optogenetic activation and inhibition during both training and testing, the authors show that particularly DAN-f1/g1 activity can bidirectionally alter memory strength and valence of high salt learning. Depending on the timing of the manipulation, artificial DAN activation could also strengthen, weaken, or abolish expression of learned behavior without altering innate preferences. By combining behavioral experiments with computational modeling, the authors developed a connectome-based circuit model in which activity of the DL1 compartment causes output neurons (MBONs) to recruit reward-like DANs via an intrinsic feedback loop. In contrast, models based on independent compartmental plasticity for reward and punishment failed to reproduce the memory valence reversals observed in behavioral experiments. Thereby, the study identified an important feature of the larval mushroom body circuit relying on intercompartmental feedback signals to process learned signals and provide appropriate behavioral outputs.
Strengths:
(1) The study benefits from temporal precision across both memory acquisition and retrieval phases using an optogenetic approach. DAN-f1/g1 function is addressed thoroughly and manipulated across several paradigms, convincingly validating that optogenetic manipulations specifically alter learning but not innate preferences.
(2) The study benefits from treating the DL1 cluster as functionally heterogeneous. By examining individual neurons during retrieval, the authors solidify a prominent role for DAN-c1 in memory retrieval and DAN-f1/g1 in aversive reinforcement.
(3) Computational modeling nicely complements the behavioral results, showing that simpler circuit architectures fail to reproduce the valence reversals observed during behavioral experiments. The success of the cross-compartmental feedback model therefore narrows the range of plausible circuit explanations and supports the authors' circuit-level rebalancing theory.
Weaknesses:
(1) The PI is calculated from the number of larvae occupying each side of the arena after a 5-minute period of free exploration. This is a reasonable summary and standard method to measure the behavioral results, but it provides limited information about how preferences and memory retrieval dynamics may develop over time. While standard in the field, trajectory-based parameters as odor approach behavior could differ between the experiments, so not all information can be retrieved.
(2) Their model argues for cross-compartment interaction in the mushroom body circuit, which is, however, not corroborated experimentally. Silencing individual DL1 neurons during the testing phase had little behavioral effect aside from DAN-c1, leaving the specific retrieval roles of DAN-f1 and DAN-g1 somewhat unresolved, particularly given that their simultaneous inhibition was shown to enhance memory recall.
(3) Another limitation concerns how neuronal dynamics are represented in computational models: whereas the four DL1-DANs are represented individually, the four-neuron pPAM cluster is modelled as a single unit. Together with the limited temporal resolution of the calcium signals, as well as the unknown output dynamics in MBONs, the current model simplifies the cellular organization and physiological dynamics underlying the proposed circuit mechanism. The response parameters of the four DL1-DANs and the pPAM cluster were fitted to previously measured calcium responses evoked by brief high-salt or fructose stimulation. Although these fits reproduce the distinct temporal profiles and stimulus sensitivities observed experimentally, it remains uncertain how faithfully such stimulus-evoked responses capture DAN dynamics.
(4) Blue light activation used in the optogenetic paradigm adds an additional (aversive) stimulus that might interfere with the behavioral outcome. While the authors included all reasonable controls and point out the benefits of being consistent with their previous work, it does add a potential confounder.
eLife Assessment
This paper provides fundamental data on the ultrastructural signatures of neuromodulatory axons in mouse cortex obtained from exceptionally high-quality correlated light- and volume electron microscopy. They report unique vesicle and synapse configurations of cholinergic and serotonergic axons, providing compelling structural support for volume transmission.
Reviewer #1 (Public review):
Summary:
This study investigates the structural basis of neuromodulatory transmission in the cerebral cortex. Using genetic labelling, correlative light and electron microscopy (CLEM), and focused ion beam scanning electron microscopy (FIBSEM), the authors reconstruct cholinergic axons originating from the basal forebrain and serotonergic axons from the raphe nuclei in the mouse somatosensory cortex, and compare their ultrastructural organisation with that of glutamatergic and GABAergic axons.
The study provides detailed quantitative evidence that cholinergic and serotonergic axons have an axonal architecture fundamentally different from that of classical fast-transmitting glutamatergic and GABAergic axons. In particular, vesicles in cholinergic and serotonergic axons are distributed more diffusely along the axon and are only weakly associated with morphologically identifiable synaptic specialisations. These findings provide strong structural support for the idea that volume transmission represents a major mode of signalling by these neuromodulatory systems in the cortex.
Although the paucity of conventional synaptic contacts made by cortical neuromodulatory axons has been reported previously, including in classical work by Descarries and colleagues, the present study provides substantially more detailed and quantitative three-dimensional evidence. The study also reveals characteristic differences among neurotransmitter systems in vesicle density, vesicle distribution relative to synaptic membranes, and vesicle morphology.
Strengths:
A major strength of this study is the combination of genetic labelling, CLEM, and high-resolution FIBSEM, which allows molecularly identified cholinergic and serotonergic axons to be reconstructed without the electron-dense reaction products used in conventional immunoelectron microscopy. This enables detailed analysis of vesicles and membrane specialisations along extended axonal segments.
Another important strength is that the analysis is not restricted to individual boutons or varicosities. Instead, the authors analyse vesicle distributions continuously along reconstructed axons and compare them quantitatively with glutamatergic and GABAergic axons. This provides a more comprehensive assessment of the relationship between vesicle distribution and morphologically identifiable synaptic sites.
Overall, the data are convincing and support the authors' main conclusions. In particular, the finding that most vesicles in cholinergic and serotonergic axons are spatially dissociated from identifiable synaptic specialisations provides strong structural evidence for predominantly non-synaptic or volume transmission. The work should therefore be of broad interest to researchers studying neuromodulation, cortical circuitry, and connectomics.
Weaknesses:
I do not identify any major methodological weakness. The manuscript is well prepared, and the experimental results generally support the authors' conclusions. I have only a few points that could further improve the interpretation and discussion.
First, in the Introduction, ionotropic receptors are emphasised in the context of rapid actions of acetylcholine and serotonin. However, these neuromodulators can also produce relatively rapid effects on cortical neurons through G-protein-coupled receptors. This could be acknowledged, for example, with reference to Gulledge et al. (2009) and Tanaka & North (1993).
Second, the implications of the findings for volume transmission could be discussed somewhat more explicitly. If acetylcholine and serotonin are released extrasynaptically and are not immediately degraded or removed, they may diffuse through the extracellular space and activate receptors located at some distance from the release site. Thus, their actions need not be restricted to cellular elements immediately adjacent to vesicle-rich regions of the axon. A brief discussion of the sequence from action-potential-dependent Ca²⁺ entry and transmitter release to extracellular diffusion and activation of spatially distributed receptors would strengthen the interpretation.
Reviewer #2 (Public review):
Summary:
This is an important, timely and extremely well-conducted study, providing 3D EM data of the highest quality for identification of modulatory axons in the neocortex. The field of connectomics, in which ever larger 3D EM volumes are being acquired, has so far focused on glutamatergic and GABAergic synaptic circuits in the cerebral cortex. These obviously constitute the major fraction of synapses and are known to form high-specificity networks. However, it has been clear that modulatory systems originating in subcortical structures are an important additional defining property of neocortical circuits. Without a clear identification of these specific axon types, their identification in 3D EM data has so far not been possible at scale.
Strength:
The present study addresses this issue and resolves the problem of identifying modulatory axons in 3D EM data of the cerebral cortex. Given the fast pace of development in the field of connectomics, such high-quality information is extremely relevant and of substantial impact for the field. The study is conducted beautifully, and the figures are very clear.
Slight weakness:
For the quantitative analyses, it would be highly appreciated to plot the raw data distributions as scatter plots. This is particularly important because, for many readers, these data will serve as a benchmark for classifying and identifying such synapses. Thus, any parameter distributions that are (close to) disjunct will be extremely informative. Along those lines, even at the level of the already extracted parameters, it would be very interesting to see whether the authors can already train a simple classifier that would allow the identification of cholinergic and serotonergic synapses or axons in the low-dimensional parameter space that has already been sampled.
Reviewer #3 (Public review):
Summary:
This paper uses correlative FIB-SEM to reconstruct cholinergic and serotonergic axons in mouse somatosensory cortex. The authors wanted to examine how these neuromodulatory axons are organized at the ultrastructural level, specifically whether their vesicles are concentrated at synaptic structures like fast transmitters (glutamate and GABA) or distributed differently. They compare the two types against glutamate and GABA axons in the same datasets, and found that the synaptic vesicles are distributed more homogeneously in the neuromodulatory axons, supporting volume transmission.
Strengths:
The main strength is the dataset. FIB-SEM gives isotropic resolution, so vesicles can be counted and positioned accurately in 3D. The correlative light-EM approach lets them identify genetically labeled axons without using electron-dense stains that would obscure internal structures. This is a clear improvement over previous EM studies that relied on immunolabeling and only looked at varicosities.
The analysis is quantitative and systematic. They don't rely on identifying boutons or varicosities as the unit of analysis. Instead, they sample along the axon centerline at regular intervals, which avoids the bias of only looking at swellings. This is a cleaner way to ask where vesicles actually sit relative to synapses and other structures.
The comparison with glutamate and GABA axons is important. It shows that the differences are not just about neuromodulatory axons being "less synaptic" in a graded way, but qualitatively different in how vesicles are organized. Glutamate and GABA axons concentrate vesicles at synapses; ACh and 5-HT axons don't.
Weaknesses:
An intrinsic weakness of the EM approach is the missing of molecular information. The paper finds vesicles to be distributed homogeneously regardless of potential "post-synaptic" contact elements. However, protein machinaries for vesicle release might still statistically bind to certain postsynaptic elements.
eLife Assessment
This important study examines the relationship between climatic niche limits and extinction risk across more than 23,000 terrestrial vertebrate species. The large comparative dataset and phylogenetic analyses indicate solid evidence that species with lower upper thermal niche limits are more likely to be classified as threatened, whereas associations with aridity limits are less consistent. However, the treatment of data-deficient species and unclear definitions of niche metrics warrant further scrutiny, and the analyses do not establish that climate or land-use change drives the observed patterns. The findings will interest researchers in macroecology, global change biology, and conservation.
Reviewer #1 (Public review):
Summary and strengths:
This manuscript examined whether realized thermal and aridity niche limits were associated with extinction risk across more than 23,000 terrestrial vertebrate species. The authors found that climatic niche limits were phylogenetically conserved and that species with lower upper thermal limits were more likely to be threatened, while relationships with aridity niche limits were weaker and less consistent. They then used simulations based on current extinction-risk patterns and found taxon-specific changes in the distribution of climatic niche limits and loss of phylogenetic diversity under simulated future extinction scenarios.
The study addresses an important question at an impressive taxonomic scale, and the underlying dataset and analyses have the potential to provide valuable insights into how climatic niche limits relate to patterns of extinction risks.
Weaknesses:
(1) One main concern is a disconnect between the central framing of the study and what the analyses can establish. The manuscript proposes that climatic niche limits increase species' vulnerability to climate and land-use change, yet extinction risk is measured using overall IUCN threat status, which integrates pressures from many different causes, including exploitation (e.g. hunting; pet trade), disease, and invasive species. For several of these threats, there is no clear mechanistic basis for why climatic niche limits should influence extinction risk. Consequently, the observed association does not establish that climate or land-use change is the underlying driver of the relationship between climatic niche limits and extinction risk. Analysing whether climatic niche limits are associated with extinction risk from climate- and habitat-related threats would provide a more direct test of the proposed mechanism.
(2) The treatment of Data Deficient species is another important concern. The authors classified Data Deficient species as Not Threatened in their analysis. However, previous studies have shown that many may in fact be threatened (Borgelt et al., 2022, Communications Biology; Jetz and Freckleton, 2015, Philosophical Transactions of the Royal Society B). This issue is particularly important here given the large number of Data Deficient species, especially among reptiles and amphibians, for which extinction risk may itself be underestimated (Caetano et al., 2022, PLoS Biology; González-del-Pliego et al., 2019, Current Biology). Treating these species as Not Threatened could therefore influence the reported relationships and should be evaluated through additional analyses.
(3) The role of the phylogenetic diversity analysis in addressing the central question is also unclear. The primary question concerns the relationship between climatic niche limits and extinction risk, whereas the subsequent analysis examines phylogenetic diversity loss following simulated extinctions. Disproportionate losses of phylogenetic diversity following non-random extinction are already well studied (e.g. Hughes et al., 2022, Current Biology; Oliveira et al., 2020, Global Ecology and Biogeography; Davis et al., 2018, PNAS). The authors should therefore clarify what this analysis adds beyond existing work and how it provides a meaningful extension of the climatic niche-extinction framework.
(4) The interpretation of the future extinction scenarios is also stronger than warranted by the simulations. The authors appropriately acknowledge that these scenarios do not explicitly incorporate climate or land-use change, but subsequently interpret the results as suggesting that climatic niche constraints shape future extinction pathways and that climatically vulnerable species will be disproportionately affected. Identifying species with lower climatic niche limits as climatically vulnerable is reasonable within the framework of the study. However, because the simulations are based on present-day extinction-risk transition probabilities rather than future climate- or land-use-driven processes, they do not establish that future climate or land-use change will preferentially drive the extinction of these species.
Overall, the manuscript addresses an important question and provides potentially valuable data and analyses. However, several of the central conclusions currently extend beyond what the analyses can directly support. Addressing the issues above would substantially strengthen the manuscript and clarify the scope of its conclusions.
Reviewer #2 (Public review):
Summary:
The manuscript presents a comprehensive and useful evaluation of the association between extinction risk and niche limits. If these results were clearly articulated, then the manuscript would make a valuable contribution to the literature. However, I struggled to understand the manuscript, as the terms and methodology are not clearly defined. I realize that methods do not belong in the main text, but as written, it is often very difficult to understand what was calculated.
To give an example. In the abstract, the main result is: "Species with colder upper thermal (niche) limits are more likely to be classified as threatened across taxa." In the 1st paragraph of the discussion, we read: "We show that... species with narrower and colder thermal niches face greater threats to extinction...". These are three different variables: colder upper thermal limits, narrower thermal niches and colder thermal niches. I really did not see where these were all defined in the manuscript, why so many different variables are considered and why the results are not consistent. I think this is likely a case of insufficient care in term definition and usage, but I found it rather difficult to follow the manuscript as a result. Clearer short explanations of terms (including a sense of how they were calculated/what they actually mean) and consistent use of terms would make the manuscript much more understandable.
Next, there are extra bits of information - for example, much of a paragraph in the introduction devoted to dispersal, many references to land-use (though I did not see that land-use was considered in this manuscript), and overly long discussion paragraphs that are highly speculative (or at least lack citations).
Finally, while I am a big proponent of identifying macro-ecological patterns, I would appreciate a critical discussion of what assumptions are made (i.e., everything about a microclimate; no consideration of extremes, etc.) and what the approach may be missing. This is not to invalidate the work, but to give the reader greater confidence that such broad-scale analyses are informative and are not glossing over too much species-specific ecology or finer spatial or temporal scale heterogeneity.
In summary, I think the paper has great potential but would benefit from a very conscious and careful revision.
eLife Assessment
This work provides an important set of findings using sensory preconditioning procedures that advance our understanding of the circumstances that influence the integration of discrete learning experiences. The evidence supporting the conclusions is solid, as the study used multiple different conditions to differentiate between different theories of memory integration. Moreover, the findings should be of broader interest to the learning and memory field for considering the role of the medial temporal lobe in how separate experiences are integrated.
Reviewer #1 (Public review):
Summary:
Sensory preconditioning protocols have been used to understand how the brain controls aversive or appetitive behavioral responses towards stimuli that have never been directly reinforced. In this study, the authors combine pharmacological NMDA blockade in a specific brain region previously involved in sensory preconditioning (i.e., perirhinal cortex) and different versions of a light-tone sensory preconditioning task in rats (i.e., different number of associations during preconditioning, distinct temporal separation between stage 1 and stage 2, stimuli order) to better understand how the integration between separated stimuli occurs. Overall, the main conclusion of the study is that the number, order, and time determine how distinct experiences are encoded and stored in the brain, which clearly advances the understanding of how these complex cognitive processes are regulated by the brain.
Strengths:
(1)The manuscript is very well written, and the methodology is well detailed, fully aligned with previous work by the research group.
(2) Results from each pharmacological manipulation and/or protocol configuration are clear, demonstrating the robustness of their findings.
(3) The overall conclusions are very interesting and clearly advance our understanding of higher-order conditioning processes.
Weaknesses:
(1) The Discussion is overly redundant with the Results and should be rewritten to place the findings more clearly within the existing literature and add specific important topics for the field (e.g., type of stimulus presentation, other brain circuits involved, etc).
(2) Before-after and sex-disaggregated data should be presented to better show statistical differences between periods with or without cue during the test and between sexes.
(3) The authors should broaden their coverage of recent literature on light-tone higher-order conditioning beyond their own previous work.
Reviewer #2 (Public review):
Summary:
In a series of experiments, the authors addressed the effect of the number of CS1-CS2 pairings, their order, and the time between their pairings and the second phase (CS2-US) in a sensory-preconditioning procedure. They hypothesized that:
(1) A higher number of CS1-CS2 pairings would favor the integration of CS1-CS2 and CS2-US associations via chaining at Test of CS1, as the CS2 develops a stronger prediction of the absence of CS1 in Phase 1, thus preventing the formation of a CS1-US association in Phase 2 and a chaining mechanism at Test of CS1 (CS1-->CS2-->US). In contrast, when the order is reversed (CS2-->CS1), a higher number of pairings should not change the mechanism of integration, as CS2 becomes a predictor of the presence of CS1. This would preserve the retrieval of CS1 and the formation of a CS1-US association in Phase 2 (mediated learning), while weakening the CS1-->CS2 link required for chaining at Test.
(2) A longer interval between the two phases of sensory-preconditioning would disrupt the retrieval of the order of the CS1-->CS2 association. Consequently, CS2 is no longer a good predictor of the absence of CS1, and the latter gets associated with the US in Phase 2, thus favoring a chaining mechanism at Test of CS1.
(3) A reminder of the CS1-CS2 order after a long interval between the two phases of sensory-preconditioning should restore the retrieval of the order of the CS1-->CS2 association, thus reversing the effect of time, and favoring a chaining strategy at Test of CS1.
To distinguish the two mechanisms, the authors blocked NMDA receptors in the PRh with DAPV before Stage 2. If integration occurs through a mediated CS1-US association formed in Stage 2, DAPV should reduce freezing to CS1 at test; if it occurs through chaining at test, DAPV should have no effect.
Strengths:
The strategies developed by the authors to address this issue are elegant and well controlled. While the sensory-preconditioning procedure has been extensively investigated, the mechanisms underlying the integration of separately acquired associations remain unclear. This study determines - in a convincing manner - the conditions under which a conditioned response is expressed via a chaining or mediated learning mechanism.
Weaknesses:
The role of NMDA antagonism is somewhat ambiguous throughout the manuscript. Does it prevent retrieval or acquisition or both?
(The Introduction (lines 98 to 101) refers to retrieval, while the Results (lines 128 to 132) and Discussion (lines 329 to 331) refer to acquisition.
eLife Assessment
This study makes an important contribution by broadening the range of eukaryotic model systems and establishing Blastocystis, the most prevalent microeukaryote in the human gut, tractable for reverse-genetics investigations. The presented imaging data are convincing and informative. The work should interest readers studying host-microbe interactions in the human gut, as well as those developing new systems for eukaryotic research.